# *IN VIVO* CELL BIOLOGY OF CEREBRAL CORTICAL DEVELOPMENT AND ITS RELATED NEUROLOGICAL DISORDERS

EDITED BY: Takeshi Kawauchi, Margareta Nikolic´ and Yoko Arai PUBLISHED IN: Frontiers in Cellular Neuroscience

#### *Frontiers Copyright Statement*

*© Copyright 2007-2016 Frontiers Media SA. All rights reserved. All content included on this site, such as text, graphics, logos, button icons, images, video/audio clips, downloads, data compilations and software, is the property of or is licensed to Frontiers Media SA ("Frontiers") or its licensees and/or subcontractors. The copyright in the text of individual articles is the property of their respective authors, subject to a license granted to Frontiers.*

*The compilation of articles constituting this e-book, wherever published, as well as the compilation of all other content on this site, is the exclusive property of Frontiers. For the conditions for downloading and copying of e-books from Frontiers' website, please see the Terms for Website Use. If purchasing Frontiers e-books from other websites or sources, the conditions of the website concerned apply.*

*Images and graphics not forming part of user-contributed materials may not be downloaded or copied without permission.*

*Individual articles may be downloaded and reproduced in accordance with the principles of the CC-BY licence subject to any copyright or other notices. They may not be re-sold as an e-book.*

*As author or other contributor you grant a CC-BY licence to others to reproduce your articles, including any graphics and third-party materials supplied by you, in accordance with the Conditions for Website Use and subject to any copyright notices which you include in connection with your articles and materials.*

> *All copyright, and all rights therein, are protected by national and international copyright laws.*

*The above represents a summary only. For the full conditions see the Conditions for Authors and the Conditions for Website Use.*

ISSN 1664-8714 ISBN 978-2-88919-962-4 DOI 10.3389/978-2-88919-962-4

# About Frontiers

Frontiers is more than just an open-access publisher of scholarly articles: it is a pioneering approach to the world of academia, radically improving the way scholarly research is managed. The grand vision of Frontiers is a world where all people have an equal opportunity to seek, share and generate knowledge. Frontiers provides immediate and permanent online open access to all its publications, but this alone is not enough to realize our grand goals.

# Frontiers Journal Series

The Frontiers Journal Series is a multi-tier and interdisciplinary set of open-access, online journals, promising a paradigm shift from the current review, selection and dissemination processes in academic publishing. All Frontiers journals are driven by researchers for researchers; therefore, they constitute a service to the scholarly community. At the same time, the Frontiers Journal Series operates on a revolutionary invention, the tiered publishing system, initially addressing specific communities of scholars, and gradually climbing up to broader public understanding, thus serving the interests of the lay society, too.

## Dedication to Quality

Each Frontiers article is a landmark of the highest quality, thanks to genuinely collaborative interactions between authors and review editors, who include some of the world's best academicians. Research must be certified by peers before entering a stream of knowledge that may eventually reach the public - and shape society; therefore, Frontiers only applies the most rigorous and unbiased reviews.

Frontiers revolutionizes research publishing by freely delivering the most outstanding research, evaluated with no bias from both the academic and social point of view. By applying the most advanced information technologies, Frontiers is catapulting scholarly publishing into a new generation.

# What are Frontiers Research Topics?

Frontiers Research Topics are very popular trademarks of the Frontiers Journals Series: they are collections of at least ten articles, all centered on a particular subject. With their unique mix of varied contributions from Original Research to Review Articles, Frontiers Research Topics unify the most influential researchers, the latest key findings and historical advances in a hot research area! Find out more on how to host your own Frontiers Research Topic or contribute to one as an author by contacting the Frontiers Editorial Office: researchtopics@frontiersin.org

# *IN VIVO* **CELL BIOLOGY OF CEREBRAL CORTICAL DEVELOPMENT AND ITS RELATED NEUROLOGICAL DISORDERS**

#### Topic Editors:

**Takeshi Kawauchi,** Institute of Biomedical Research and Innovation & Keio University School of Medicine & Japan Science and Technology Agency, Japan **Margareta Nikolic´,** University of Hertfordshire, UK **Yoko Arai,** INSERM & Institut Jacques-Monod, Université Paris Diderot, France

Sequential images of cerebral cortical development. The images show mouse cerebral cortices at embryonic days 15 (E15) (left panel), E17 (middle panel) and postnatal day 0 (P0) (right panel), which were electroporated with EGFP-expressing vectors at E14. EGFP and DAPI signals are shown in green and blue, respectively. In the left panel, most of EGFP-positive cells are neural progenitors. Ki67, a proliferative cell marker, is shown in red. In the middle panel, EGFP-positive cells are postmitotic neurons migrating along radial fibers. Nestin, a marker of radial fibers, is shown in red. In the right panel, EGFP-positive cells are post-migratory neurons or those undergoing terminal translocation. MAP2, a dendritic marker, is shown in red.

Image by Takeshi Kawauchi (Institute of Biomedical Research and Innovation & Keio University & JST)

The brain consists of a complex but precisely organized neural network, which provides the structural basis of higher order functions. Such a complex structure originates from a simple pseudostratified neuroepithelium. During the developing mammalian cerebral cortex, a cohort of neural progenitors, located near the ventricle, differentiates into neurons and exhibits multi-step modes of migration toward the pial surface. Tight regulation of neurogenesis and neuronal migration is essential for the determination of the neuron number in adult brains and the proper positioning of excitatory and inhibitory neurons in a specific layer, respectively. In addition, defects in neurogenesis and neuronal migration can cause several neurological disorders, such as microcephaly, periventricular heterotopia and lissencephaly. Recent advances in genetic approaches to study the developing cerebral cortex, as well as the use of a number of novel techniques, particularly *in vivo* electroporation and time-lapse analyses using explant slice cultures, have significantly increased our understanding of cortical development. These novel techniques have allowed for cell biological analyses of cerebral cortical development *in vivo* or *ex vivo*, showing that many cellular events, including endocytosis, cell adhesion, microtubule and actin cytoskeletal regulation, neurotransmitter release, stress response, the consequence of cellular crowding (physical force), dynamics of transcription factors, midbody release and polarity transition are required for neurogenesis and/or neuronal migration.

The aim of this research topic is to highlight molecular and cellular mechanisms underlying cerebral cortical development and its related neurological disorders from the cell biological point of views, such as cell division, cell-cycle regulation, cytoskeletal organization, cell adhesion and membrane trafficking. The topic has been organized into three chapters: 1) neurogenesis and cell fate determination, 2) neuronal migration and 3) cortical development-related neurological disorders. We hope that the results and discussions contributed by all authors in this research topic will be broadly useful for further advances in basic research, as well as improvements in the etiology and care of patients suffering from neurological and psychiatric disorders.

**Citation:** Kawauchi, T., Nikolic´, M., Arai, Y., eds. (2016). *In vivo* Cell Biology of Cerebral Cortical Development and Its Related Neurological Disorders. Lausanne: Frontiers Media. doi: 10.3389/978-2-88919-962-4

# Table of Contents


Chan Choo Yap and Bettina Winckler


# **Chapter 3. Cellular insights into cortical development-related neurological disorders**

*196 Morphological and functional aspects of progenitors perturbed in cortical malformations*

Sara Bizzotto and Fiona Francis


Aslam Abbasi Akhtar and Joshua J. Breunig

# Editorial: In vivo Cell Biology of Cerebral Cortical Development and Its Related Neurological Disorders

Takeshi Kawauchi 1, 2, 3 \*, Margareta Nikolic´ <sup>4</sup> and Yoko Arai 5, 6

<sup>1</sup> Laboratory of Molecular Life Science, Institute of Biomedical Research and Innovation, Foundation for Biomedical Research and Innovation, Kobe, Japan, <sup>2</sup> Department of Physiology, Keio University School of Medicine, Tokyo, Japan, <sup>3</sup> Precursory Research for Embryonic Science and Technology, Japan Science and Technology Agency, Saitama, Japan, <sup>4</sup> Department of Biological and Environmental Sciences, School of Life and Medical Sciences, University of Hertfordshire, Hatfield, UK, <sup>5</sup> PROTECT, INSERM, Unversité Paris Diderot, Sorbonne Paris Cité, Paris, France, <sup>6</sup> Institut Jacques-Monod, Centre National de la Recherche Scientifique UMR 7529, Université Paris Diderot, Paris, France

Keywords: neurogenesis, neuronal migration, cell division, cytoskeleton, endocytosis, microcephaly, periventricular heterotopia, lissencephaly

#### **The Editorial on the Research Topic**

#### **In vivo Cell Biology of Cerebral Cortical Development and Its Related Neurological Disorders**

The brain consists of complex but precisely organized neural networks, which determine the structural basis of higher order functions. Remarkably, this complex structure originates from a simple pseudostratified neuroepithelium. How it is formed is best seen in the elegant example of the cerebral cortex. In the developing mammalian cerebral cortex, polarized neural progenitors are arranged in a pseudostratified structure that forms the mitotically active ventricular zone. At the onset of neurogenesis, a cohort of neural progenitors differentiates into neurons and through multistep modes of migration generates a six-layered structured cerebral cortex. Defects in neurogenesis and neuronal migration can cause several neurological disorders, including microcephaly and lissencephaly. Importantly, recent advances in not only human and mouse genetic approaches but also the use of a number of novel techniques, particularly in vivo electroporation and timelapse analyses of explant slice culture, have significantly increased our understanding of cortical development. In addition, these novel techniques have allowed us to open a new avenue for cell biological analyses of cortical development in vivo or ex vivo.

The aim of this research topic is to highlight important mechanisms underlying cerebral cortical development and associated neurological disorders, with a specific focus on cell biology, including cell division, cell cycle regulation, cytoskeletal organization, cell adhesion, endocytosis, and membrane trafficking. The topic has been organized into three sections: (1) neurogenesis and cell fate determination, (2) neuronal migration, and (3) cortical development-related neurological disorders.

The first section highlights cellular insights into neurogenesis and cell fate determination. In the developing cerebral cortex, apical neural progenitors (radial glial progenitors) exhibit cell cycledependent nuclear movement, termed interkinetic nuclear migration (INM). Rho family small GTPases, including Rac1 and Rnd3, are known to control the proliferation and INM of apical progenitors (Azzarelli et al). While a physiological significance of INM remains unclear, it is thought that INM is associated with the nuclear traffic control in the ventricular zone (Miyata et al).

At M phase, apical progenitors undergo proliferative symmetric or neurogenic asymmetric division along the ventricle. The precise control of the balance between symmetric and asymmetric

Edited and reviewed by: Christian Hansel, University of Chicago, USA

\*Correspondence: Takeshi Kawauchi takeshi-kawauchi@umin.ac.jp

Received: 30 May 2016 Accepted: 02 June 2016 Published: 21 June 2016

#### Citation:

Kawauchi T, Nikolic M and Arai Y ´ (2016) Editorial: In vivo Cell Biology of Cerebral Cortical Development and Its Related Neurological Disorders. Front. Cell. Neurosci. 10:162. doi: 10.3389/fncel.2016.00162 cell division is required for the proper production of neurons during cortical development. Recently, spindle size asymmetry has been suggested as a core component to regulate the asymmetric cell division (Delaunay et al). Interestingly, spindle size asymmetry in neural progenitors is observed both in mouse and macaque cerebral cortices, suggesting a conserved, and thus important mechanism during mammalian cortical evolution. Another feature of asymmetric cell division is a release of the midbody, a cytoplasmic bridge between two daughter cells at the end of mitosis. While midbody retention is for the maintenance of proliferative capacity of the cells, the midbody release is associated with the progression of differentiation/neurogenesis. Proliferative cancer cells retain the midbody carrying a phospholipid, phosphatidylserine (PS), geometrically inside of the plasma membrane, however, differentiating cells and apical progenitors release the midbody carrying PS outside of the plasma membrane (Arai et al). This PS asymmetry in midbody allowed cells to take midbody up by themselves or loose them to regulate their differentiation process.

Several transcription factors are involved in controlling the balance between proliferation and differentiation of neural progenitors. Among them, Pax6 is highly expressed in apical progenitors in the ventricular zone of the developing cerebral cortex and regulates their proliferation and cell-cycle exit to generate cortical excitatory projection neurons (Manuel et al.). In the ganglionic eminence and preoptic area (POA), Nkx2.1 and Nkx5.1 transcription factors play pivotal roles in the production of cortical inhibitory interneurons (Peyre et al.). Recent reports show that a temporal oscillatory nature of basic helix-loop-helix (bHLH) transcription factors, such as Ascl1/Mash1, Hes1, Neurogenin2, and Olig2, play important roles in fate determination. Apical progenitors co-express several bHLH transcription factors in an oscillatory expression pattern. Upon the determination of cell fate, one of bHLH transcription factors dominates and is continuously expressed (Imayoshi et al.).

Apical progenitors give rise to either neuronally-committed intermediate progenitor cells or immature cortical excitatory neurons by losing their cell polarity. An Axin-GSK3β complex controls the generation and proliferation of intermediate progenitor cells (Ye et al). Interestingly, the delamination from the pseudostratified neuroepithelium or ventricular zone resembles an epithelial–mesenchymal transition (EMT) and requires the down-regulation of several cell adhesion molecules, such as cadherins, nectins, and junctional adhesion molecules (JAMs) (Singh and Solecki).

The second section highlights cellular insights into the migration of excitatory neurons and inhibitory interneurons from the dorsal ventricular zone or ventral ganglionic eminence, respectively. Newly generated excitatory neurons from the dorsal ventricular zone first display a multipolar morphology, which requires Cx43, a gap junction protein, and p27 that acts as a cytoskeletal regulator rather than a cell cycle inhibitor (Cooper). Subsequently, neurons transform into a bipolar form by extending an axon and a pia-directed leading process and retracting all other processes. Many molecules regulating the axon formation during the multipolar-to-bipolar transition have been identified (Cooper). LKB1 and its associated molecules, Stk25 and STRADα, are reported to regulate axon outgrowth. Interestingly, under the control of an atypical cyclin-dependent kinase (Cdk5), Axin, and GSK3β also play a role in axon formation (Ye et al.).

The bipolar neurons, called locomoting neurons, migrate along apical progenitor-derived long radial fibers with unique morphological changes (Kawauchi). The attachment of neurons to the radial fibers and the neuron-specific migration mode require N-cadherin-mediated adhesion and the Cdk5–Dcx/p27 pathway, respectively. At the final phase of neuronal migration, neurons change their migration mode into a radial fiberindependent terminal translocation mode, which is controlled by a secreted molecule, Reelin, and its downstream cytoplasmic adaptor, Dab1 (Yap and Winckler).

In contrast to the excitatory projection neurons, immature inhibitory interneurons, born at the ganglionic eminence or POA, migrate tangentially to the cerebral cortex (Luhmann et al.; Peyre et al.). Immature cortical inhibitory and excitatory neurons partly share migration mechanisms at least on a molecular level. For example, Cdk5, p27, Dcx, N-cadherin, and Rho family small GTPases are known to regulate both types of neuronal migration (Azzarelli et al.; Cooper; Kawauchi; Luccardini et al.; Peyre et al.; Ye et al.). However, inhibitory interneurons display branched leading processes and their tangential migration is not dependent on radial fibers, suggesting that specific mechanisms are also required for their movement. Interestingly, neurotransmitters, glutamate, and GABA, and their receptors are known to control neuronal migration (Luhmann et al.), and glycine α2 receptor is required for the migration of cortical inhibitory interneurons via the fine-tuning of acto-myosin contraction during nucleokinesis (Peyre et al.).

Neuronal migration depends on dynamic regulation of a huge number of intracellular and membrane proteins. Microtubule and actin cytoskeletal organization are essential for the morphological changes of migrating neurons and dysregulation of the cytoskeletons can result in several neurological disorders (Lian and Sheen, Peyre et al.). For example, axophilic migration of GnRH neurons requires cooperation of cortical actin flow and microtubule organization, defect in which causes Kallmann syndrome, a neuroendocrine disorder (Hutchins and Wray). In addition, recent reports have indicated that endocytosis and membrane trafficking pathways regulate several steps of neuronal migration (Kawauchi, Yap and Winckler). Endocytic pathwaymediated regulation of N-cadherin plays an essential role in the radial fiber-dependent migration of cortical excitatory neurons (Kawauchi). The tangential migration of cortical inhibitory interneurons, which is independent of radial fibers, also requires N-cadherin (Luccardini et al.). Thus, the multi-step neuronal migration and morphological changes rely on coordinated regulation of various cellular events.

The third section highlights cellular insights into cortical development-related neurological disorders. Disruption of the proper balance between proliferation and differentiation results in cortical malformations, such as microcephaly or megalencephaly (Bizzotto and Francis). Increased cell death of neural progenitors can also lead to microcephaly. At least 12 causative genes have been linked to autosomal recessive primary microcephaly. The loci are numbered by MCPH1–MCPH12. MCPH1 controls the centrosome cycle through the Chk1–Cdc25 pathway and DNA damage repair, whose defects may lead to the development of a small brain (Pulvers et al.). Mutations of ASPM (MCPH5) are the most common cause of autosomal recessive primary microcephaly in humans, and ASPM protein shows important roles in maintaining the spindle positioning during the mitosis of neural progenitors (Bizzotto and Francis, Pulvers et al.).

Abnormalities in apical progenitors can also result in cobblestone (type II) lissencephaly or periventricular heterotopia, caused by defects in the attachment of basal processes (radial fibers) to the pial surface or from disruptions in the apical (ventricular) surface, respectively. Globular heterotopia, which was observed in Eml1 homozygous mutant or Ncadherin Emx1Cre conditional knockout mice, occurs when apical progenitors detach from the ventricular surface (Bizzotto and Francis).

Filamin A and ArfGEF2, whose gene products regulate actin cytoskeleton and membrane trafficking, respectively, are reported as causative genes for periventricular heterotopia (Lian and Sheen). Mutations in genes encoding microtubuleregulatory proteins, such as Lis1 and Dcx, result in type I lissencephaly (Kawauchi). In addition to these cell intrinsic factors, prenatal environmental stresses, such as alcohol, hypoxia, and exposure to heavy metals, can induce cortical malformation and the impairment of cognitive and memory functions (Ishii and Hashimoto-Torii). Interestingly, these types of environmental stress activate intracellular stress response signaling, including the heat shock protein (HSP)-mediated pathway. Thus, dysregulation of various cellular events is closely associated with neurological disorders.

One of the strategies for overcoming these neurogenesisor neuronal migration-related neurological disorders is to reactivate neurogenesis in the postnatal cerebral cortex. Despite the obvious challenges, reprogramming of astrocytes (or neurons) into specific neuronal subtypes may be one important approach for brain repair (Akhtar and Breunig).

This research topic aims to provide multidisciplinary approaches, encompassing developmental neuroscience and cell biology to understand mechanisms of cortical development as well as an etiology of neurological and psychiatric disorders. We hope that results and knowledge provided by all authors in this research topic will be useful for patients' care, as well as future advances in basic research.

# AUTHOR CONTRIBUTIONS

TK, MN, and YA wrote the manuscript.

# ACKNOWLEDGMENTS

We are grateful to all authors and reviewers for their outstanding contributions to this research topic.

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2016 Kawauchi, Nikoli´c and Arai. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Mitotic spindle asymmetry in rodents and primates: 2D vs. 3D measurement methodologies

# *Delphine Delaunay1,2\*, Marc C. Robini <sup>3</sup> and Colette Dehay1,2\**

*<sup>1</sup> Stem Cell and Brain Research Institute, Institut National de la Santé et de la Recherche Médicale, U846, Bron, France*

*<sup>2</sup> Université de Lyon I, Lyon, France*

*<sup>3</sup> CREATIS (CNRS Research Unit UMR5220 and INSERM Research Unit U1044), INSA-Lyon, Villeurbanne, France*

#### *Edited by:*

*Yoko Arai, Université Paris Diderot, France*

#### *Reviewed by:*

*Fumio Matsuzaki, RIKEN Center for Developmental Biology, Japan Nicolas Minc, Centre National de la Recherche Scientifique, France*

#### *\*Correspondence:*

*Delphine Delaunay and Colette Dehay, 2- Université de Lyon, Université Lyon I, 69100 Villeurbanne, France e-mail: delphine.delaunay@ inserm.fr; colette.dehay@inserm.fr*

Recent data have uncovered that spindle size asymmetry (SSA) is a key component of asymmetric cell division (ACD) in the mouse cerebral cortex (Delaunay et al., 2014). In the present study we show that SSA is independent of spindle orientation and also occurs during cortical progenitor divisions in the ventricular zone (VZ) of the macaque cerebral cortex, pointing to a conserved mechanism in the mammalian lineage. Because SSA magnitude is smaller in cortical precursors than in invertebrate neuroblasts, the unambiguous demonstration of volume differences between the two half spindles is considered to require 3D reconstruction of the mitotic spindle (Delaunay et al., 2014). Although straightforward, the 3D analysis of SSA is time consuming, which is likely to hinder SSA identification and prevent further explorations of SSA related mechanisms in generating ACD. We therefore set out to develop an alternative method for accurately measuring spindle asymmetry. Based on the mathematically demonstrated linear relationship between 2D and 3D analysis, we show that 2D assessment of spindle size in metaphase cells is as accurate and reliable as 3D reconstruction provided a specific procedure is applied. We have examined the experimental accuracy of the two methods by applying them to different sets of *in vivo* and *in vitro* biological data, including mouse and primate cortical precursors. Linear regression analysis demonstrates that the results from 2D and 3D reconstructions are equally powerful. We therefore provide a reliable and efficient technique to measure SSA in mammalian cells.

**Keywords: asymmetric cell division, cerebral cortex, mouse, primate, corticogenesis**

## **INTRODUCTION**

Asymmetric cell division (ACD)—unequal division producing two daughter cells with distinct fates—generates cell diversity in prokaryotes and eukaryotes. Significant progress in elucidating the key mechanisms underlying ACD has revealed a high degree of conservation between invertebrates and vertebrates (Knoblich, 2010; Li, 2013).

The conserved mechanisms include sibling cell size asymmetry, which refers to physical asymmetry and has been shown to occur in various cell types and species (including *Saccharomyces cerevisiae* cells, *Drosophila* and *C. elegans* neuroblasts and sensory organ precursor cells). The cellular and molecular machinery responsible for sibling cell size asymmetry is complex and not fully understood (reviewed in Roubinet and Cabernard, 2014). One major player in physical ACD in invertebrates is the asymmetry in spindle poles geometry (Kaltschmidt et al., 2000; Betschinger and Knoblich, 2004; Knoblich, 2010). Recently, we have shown that spindle shape asymmetry (SSA) is a highly conserved mechanism that also operates in the mouse developing mammalian cerebral cortex (Delaunay et al., 2014), where it plays a major role in the tight spatiotemporal control of selfrenewal and differentiation during corticogenesis. In the present study, we extend these findings to primates by showing that SSA occurs during the division of macaque monkey cortical precursors. We also demonstrate that SSA magnitude is not biased by the orientation of the spindle with respect to its substrate.

ACD in cortical development occurs in the germinal zones including the apical progenitors of the ventricular zone (VZ) and serves to generate differentiating neurons while amplifying the progenitor pool through self-renewal (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004; Kriegstein et al., 2006). SSA in apical cortical progenitors is characterized by the unequal organization of the two spindle poles which appear asymmetric in size during metaphase and throughout division, leading to the generation of two daughter cells of distinct size and fate (Delaunay et al., 2014).

Although SSA is easily delineated in invertebrates, its amplitude is smaller in cerebral cortex, making it harder to quantify. Here we present two simple methods based on regular confocal stack acquisitions, which allow accurate SSA measurements using 3D volume estimation and 2D surface area calculation. We describe the procedures for both methods and demonstrate theoretically and empirically that they give similar results. These findings allow us to conclude that, compared to the 3D method, 2D measurement is an efficient and preferred methodology for SSA assessment.

# **METHODS**

#### **CELL CULTURE**

Surgical procedures and animal experimentation were in accordance with European requirements 2010/63/UE. The protocol C2EA42-12-11-0402-003 has been approved by the Animal Care and Use Committee CELYNE (C2EA #42). E13.5 to E14.5 of one mouse brains were electroporated *ex-vivo* (3x 50–70 V pulses of 100 ms duration and 100 ms interval) with 0.1.8 to 2.5μg/μl DNA. Cortex were dissected in HBSS, cell dissociated with trypsin 1X (Invitrogen) and plated at 4.5. 104 cells per 12 mm diameter poly-D-Lysine (Sigma, 40μg/ml) coated glass cover slips. Cells were maintained in culture for 1–1.5DIV in DMEM/F12 supplemented with B27 (1:50*<sup>e</sup>* , Invitrogen) and N2 (1:100*<sup>e</sup>* , Invitrogen) and fixed with 37◦C 2%PFA for 2–5 min.

#### **MONKEY TISSUE PREPARATION**

Fetuses from timed-pregnant cynomolgus monkeys (Macaca fascicularis, gestation period 165 days) were delivered by caesarian section as described elsewhere (Lukaszewicz et al., 2005). All experiments were in compliance with French national and European laws as well as with institutional guidelines concerning animal experimentation. Surgical procedures were in accordance with European requirements 2010/63/UE. The protocol C2EA42- 12-11-0402-003 has been reviewed and approved by the Animal Care and Use Committee CELYNE (C2EA #42). Lethally anesthetized primate fetuses (E63–E80) via intraperitoneal injection of Sodium Pentobarbital 60 mg/kg were perfused through the heart with buffered 4% paraformaldehyde (PFA) for 30 min. After cryoprotection in PBS/Sucrose (10 then 20%), brains were embedded in Tissue-Tek. 20μm-thick parasagittal cryosections were cut, mounted on superfrost glass slides and immunostained.

#### **IMMUNOCYTOCHEMISTRY AND IMMUNOHISTOCHEMISTRY**

Cultured cover slips were saturated for 1 h in PBS1X/10% goat serum and incubated with the primary antibody overnight or up to 30 h at 4◦C: mouse anti-α-tubulin (sigma, 1:500), rabbit anti-pericentrin (Covance, 1:1000). Sections were then washed in PBS, followed by incubation with appropriate goat fluorescence-conjugated secondary antibodies at room temperature for 2 h (Alexa 488 goat anti Mouse (Invitrogen, 1:1000), Alexa Fluor 555 goat anti-rabbit IgG (Invitrogen, 1/1000). Nuclei were stained with DAPI (0.5μg/ml). Monkey cryosections were air-dried (30 min) and hydrated in Tris-buffered saline (TBS) for 15 min. Heat-mediated antigen retrieval was performed at 95◦C for 15 min. Slices were permeabilized with TBS + TritonX (0.5%) and saturated by incubation in TBS + BSA 1% (= TBSb) + normal donkey serum (10%) + 0.5% triton in TBS for 30 min. Primary antibodies were co-incubated 1 day and 2 nights in TBSb + 0.5% triton at 4◦C as follows: mouse anti-alphatubulin (sigma, 1:200), rabbit anti-pericentrin (Covance, 1:2000), sheep anti-EOMES (R&D 1:800). Secondary antibodies were coincubated in Dako Diluent (Dako) 1 h at RT, at the following concentrations: Alexa 488 donkey anti mouse (Invitrogen, 1:200), Alexa 555 donkey anti rabbit (Invitrogen, 1:200), Alexa 647 donkey anti sheep (Invitrogen, 1:200). Nuclear staining was performed using Dapi (Invitrogen, D1306, 2μg/mL in TBS) 10 min at RT. Sections were mounted in Fluoromount G. Specimens were analyzed with a confocal microscope (Leica DM 6000 CS) allowing acquisition of axial image sequences ("z-stacks") for 3D quantification. Z-stack images were acquired using a Plan-Apochromat 63 × 1.40 NA oil objective. Excitation wavelength was 488 nm and emission was detected using a long pass filter from 505 nm. Image pixel size was 0.045μm (x and y) and bit depth 0.998, z-step size 0.3–0.5μm, and pinhole diameter from 100μm.

# **3D VOLUME QUANTIFICATION**

The 3D volume calculation was based on the original serial confocal acquisitions. The "VolumeJ" program was designed by Denis Ressnikoff (SFR Lyon-Est, CNRS UMS3453– INSERM US7, Centre Commun de Quantimétrie) based on the 3D object counter plugin [Fabrice Cordelieres (fabrice.Cordelieres@curie.u-psud.fr); Jonathan Jackson (j.jackson@ion.ucl.ac.uk)] and the 3D volume viewer [Benjamin Schmid (Bene.Schmid@gmail.com)]. The program is available at https://www.labex-cortex.com/en/users/delphine-delaunay and as .txt in "Supplementary Material."

#### **2D AREA QUANTIFICATION**

Serial sections of metaphase cells from dissociated mouse cortical precursors and from *in situ* monkey VZ precursors, were acquired from 0.2 to 0.6μm intervals from the top to the bottom of the cells (back to back) in order to measure the entire spindle apparatus. Only metaphase cells presenting equal sized centrosomes were taken into consideration. The area of each spindle pole was measured using Image J on maximal stack projections based on the alpha-tubulin staining. The area of each spindle pole was reported as the percentage difference between the two spindle areas. The folded normal distribution and Permutation test "utilFuncs.R" script as well as data test and instructions are available at https://www.labex-cortex.com/en/ users/delphine-delaunay and as .txt in "Supplementary Material."

#### **3D vs. 2D COMPARATIVE ANALYSIS**

Two independent sets of data were used to compare 2D and 3D measurements. Eighty one mouse dividing apical precursors in dissociated culture were analyzed. Ninety three monkey apical precursors were analyzed *in situ* on parasagittal sections. The regression analyses were implemented using the MATLAB software.

#### **RESULTS**

#### **3D SSA QUANTIFICATION**

In dividing apical cortical progenitors, spindle size is correlated with daughter cell identity. The daughter cell inheriting the large spindle gives rise to a neuron, and the daughter cell inheriting the small spindle a self-renewing apical progenitor (Delaunay et al., 2014; **Figure 1**). 3D reconstruction of the two spindle poles, which allows calculation of the volume of each pole, appears as the method of choice for the accurate determination of SSA (Delaunay et al., 2014). Dissociated mouse apical precursors were fixed and stained for α-tubulin, pericentrin and DAPI to reveal the condensed nuclei. For each metaphase cell, optical sections were acquired every 0.2 to 0.5μm from top to bottom, using confocal microscopy. To avoid problems with tilted or incomplete spindles, we only considered the metaphases displaying centrosomes of

**FIGURE 1 | SSA in the developing cortex. (A–D)** Schematic summarizing the link between spindle shape asymmetry and asymmetric cell division in the cortical ventricular zone. **(A)** Dividing apical progenitor (AP) presenting asymmetric spindle in metaphase. The bigger spindle is highlighted in red and the smaller in blue. The dashed white line indicates the cell shape. **(B)** The cell divides asymmetrically and gives rise to two distinct daughter cells: a neuron (red) and a new dividing AP (blue). The newly born neuron arises from the cell that inherits the bigger spindle. **(C)** Example of symmetric dividing AP displaying spindle of equal sizes at metaphase (dark and light blue). This cell will give rise to daughter cells of equal fate **(D)**. Scale bars: 10μm.

equal size (based on the pericentrin staining). We measured each spindle pole volume and named the larger of the two spindles "*Left spindle*" (green, **Figure 2A**) and the smaller "*Right spindle"* (yellow, **Figure 2A**). The difference between the left and right spindle poles, called the "3D spindle pole difference," denoted by -<sup>V</sup> and expressed as a percentage, revealed the SSA magnitude:

$$
\Delta\_{\rm V} = \left(\frac{V\_{\rm L} - V\_{\rm R}}{V\_{\rm L} + V\_{\rm R}}\right) \times 100,
$$

where VL and VR denote the volumes of the left and right spindle poles, respectively.

The volume was calculated using a hand designed ImageJ program (VolumeJ, **Figure 2**; see Methods). The sequential steps of the program are detailed in **Table 1**. Briefly, the spindle apparatus is extracted from the optical stack (**Figures 2B,C**) and the signal transformed in pixels after appropriate thresholding (**Figures 2D,E**). For each optical section, one side of the spindle pole is selected using the wand tool (**Figure 2F**). The program then considers the non-selected pixels as belonging to the same structure and will create the second spindle pole. A mask appears, displaying one spindle pole in green and the other in yellow. This allows comparison between the mask and the original picture in order to avoid any mistakes. In a final step, the program calculates the spindle volume for each pole based on the extracted voxels (**Figure 2H**). In the cortex, SSA is consistently maintained in anaphase and throughout mitosis. An example of SSA is shown in **Figure 2**, where the 3D spindle pole difference, -Vis 20.7, typical of an asymmetric spindle (see Delaunay et al., 2014; **Figure 1** -<sup>V</sup> ≤ 10%: symmetric spindle; -<sup>V</sup> > 10%: SSA). Hence, 3D SSA quantification is easy to apply, although its implementation is time consuming. We therefore searched for an alternative, equally reliable method that will allow high scale quantification of SSA and explored the capacity of 2D SSA determination to recapitulate 3D SSA measurements.

#### **2D SSA QUANTIFICATION**

To design a reliable 2D SSA quantification method, we analyzed the same data set as for the 3D analysis, that is, dissociated primary cortical precursors from E10.5 to E16.5 for a mean period of 1 day *in vitro* (DIV) (**Figure 3**). As for the 3D analysis, cells were fixed and stained for α-tubulin, pericentrin and DAPI to reveal the condensed nuclei. For each metaphase cell, optical sections were acquired from top to bottom with 0.2–0.5μm intervals using a Leica DM6000 confocal microscope. The same criteria as for the 3D quantification were applied (**Figures 3A–C**). To quantify the SSA, we measured each spindle pole area and named the larger of the two "*Left spindle*" (green, **Figures 3D,H**) and the smaller "*Right spindle"* (yellow, **Figures 3D,H**). The difference between the left and the right spindle poles (called the "2D *spindle pole difference"*), denoted by and expressed as a percentage, reveals the SSA magnitude. Using the ImageJ software, for each cell, optical sections were transformed into maximal intensity stack projections (**Figures 3E,I**). The resulting left and right spindle pole domains were then manually drawn as ROI (**Figures 3F, J**) and their respective surface area estimated. Let *A*<sup>L</sup> and *A*R, respectively denote the left and right surface areas, the 2D spindle-pole difference is defined by

$$
\Delta = \left(\frac{A\_{\rm L} - A\_{\rm R}}{A\_{\rm L} + A\_{\rm R}}\right) \times 100 \,\text{J}
$$

Examples of symmetric vs. asymmetric spindle are illustrated **Figures 3G,K**. In particular, a spindle-pole difference ≥ 20 was often measured in highly asymmetric cells (**Figure 3K**).

To analyze SSA evolution during cortical development, the spindle-pole difference was evaluated at five distinct developmental stages between E10 and E18. 322 metaphases cells were analyzed and the SSA variations reported using a *folded normal distribution* (see Methods, "2D area quantification"). The *folded normal distribution* represents the distribution of the absolute value of a given variable [the probability measure of the normal

both sides is represented by the red dashed line and both spindle poles rotate along the same revolution axis (gray). **(B–H)** Spindle volume determination using the VolumeJ program. **(B,C)** Stack projection of an *in vitro* metaphase

displaying the selected Left and right spindle pole. **(H)** 3D volume rendering and calculation. The volume is independently calculated for each spindle pole. Here, the difference in volume (v) is 20.7, typical of an asymmetric cell.

distribution on (−∞,0) is folded over to (0,∞)]. The probability density function is reported for each developmental stage and reflects SSA magnitude. This task was performed using an R script specifically designed in our laboratory and freely available (see Methods for details). From E10.5 to E16.5, spindle-pole difference was found to follow the neurogenesis kinetics, with the mean and standard deviation increasing up to E14.5 (neurogenesis) and then decreasing (as illustrated in **Figure 3L**). The significance values were confirmed using a permutation test. Under the permutation hypothesis, it is assumed that is distributed evenly across ages, so that randomly permuting the labels of the ages across the data set should not change observed differences. We randomly permuted 10,000 times the group membership labels between the control stage (E10.5) and E14.5 (**Figure 3M**). This analysis reveals a significant shift in SSA magnitude at E14.5 compared to earlier stages. These results are in accordance with previously published data using the 3D quantification method (Delaunay et al., 2014).

To ensure that the spindle orientation had no effect on SSA measurement, we plotted the spindle angle deviation against the SSA magnitude. The spindle angle measurement was carried out on dissociated precursors (used in Delaunay et al., 2014) as described in Toyoshima and Nishida (2007). The angle between the axis of the metaphase spindle and the substrate surface was measured using the linear distance (X) and the vertical distance (Y) between the two poles of the metaphase spindles revealed by the pericentrin staining. The calculated angle was denominated α and expressed in degrees. α was measured on a representative sample of metaphase cells at two extreme developmental stages: E10.5+1 DIV (when SSA magnitude is the lowest) and E14.5+1.5 DIV (when the SSA values peak). For each metaphase, the value of α was compared to the value of the 2D SSA. As expected, the spindle deviation was very low and mostly distributed between 0 and 5 degrees at both E10.5+1DIV and E14.5+1.5 DIV. No correlation was observed between α and SSA magnitude at both stages. At E14.5, a stage characterized by high SSA values, the spindle orientations stay close to 5 degrees for most of the population but the number of individuals with greater α increased. However, these individuals did not display a higher SSA. Inversely, at both ages, metaphase displaying a greater angle

#### **Table 1 | ImageJ program for 3D spindle reconstruction.**

#### **The program is downloaded as a stand-alone program and run in ImageJ version 1.47T**

Optical Stacks are taken with a X63 objective, with a minimal pixel resolution of 90 × 90 nm (format 512 × 512, bidirectionnal) and with *z*-value of 0.5μm.

• Under ImageJ software, open the α-tubulin channel and rename it with a simple name.

• Start the "volume quanti 1\_0.ijm" program.

• To avoid background, make a ROI selection close to the spindle, and select "Ok."

• The macro create two sets of picture: one named ".tif\_ROI,"

(visualization picture) and the other named ".tif\_mask" (3d skeleton).

• On ".tif\_mask," check the correlation between the spindle pixelation and the observed α-tubulin channel (on ".tif\_ROI").

• If it match, click apply to the "threshold" windows.

• Click "ok" on the macro windows to observe simplification of the pixelated shapes.

#### **Clean non-desired structures recognized as signal (If necessary)**

• Using ".tif\_ROI" as models, define the most accurate pixel shape as possible.

• Select non-significant areas on the ".tif\_ROI" window, report this selection on ".tif\_mask" window and delete it.

• Perform this for all the non-significant areas on each picture of the stack until each one shows only the precise shape of the spindle, as it is observed on the ".tif\_ROI" pictures.

• With the "wand" tool, select one side of the spindle and record the selection using "T" button (ROI manager opens automatically). Small fragments can be added to the selection, maintaining "maj" button. Do this for each frame.

• When it's done, select all the ROI at the same time and click "ok" on the macro windows.

• The macro will calculate the volume of selected structures and deduced the volume of the other spindle pole from the non-selected pixels.

• The macro creates two excels files, named.tif\_Mask\_1

and.tif\_Mask\_2, giving the volume of each part of the spindle apparatus, plus a colored 3d reconstruction of the spindle (".tif\_color\_mask").

of deviation exhibits a low level SSA level. For both stages— E10.5 and E14.5—the standard deviations of the spindle angle were respectively 8.14 and 8.5. Altogether, these results formally demonstrate the independence between the spindle deviation and the asymmetry of the mitotic spindle. A deviation from the substrate surface at metaphase will not result in an increase in the measured SSA.

To conclude, the 2D surface area measurements reliably capture SSA distributions as well as efficiently quantifying changes in magnitude during corticogenesis.

#### **THEORETICAL RELATIONSHIP BETWEEN 2D AND 3D SPINDLE POLE DIFFERENCES: SHAPE INDEPENDENCE**

To explore the correlation between 2D and 3D quantification, we investigated the theoretical relationship between the 2D and 3D spindle-pole differences and -V. To do so, we started by modeling the left and right spindle poles as half-spheroids and as right-circular cones having the same axis of revolution. For these two simple models, we found that -<sup>V</sup> is nearly linear in with a slope bounded by 1 and by the ratio of the left-to-right spindlepole diameter. We then showed that this result holds in the general case where the spindle poles are scaled versions of an arbitrary solid of revolution.

#### *Spheroidal and conical models*

**Figure 4A** shows a cross-section of these two simple models in a plane containing the axis of revolution. Assume the spindle poles are half-spheroids, and let *P* be a pole with equatorial diameter *<sup>d</sup>* and polar radius *<sup>w</sup>*. The volume of *<sup>P</sup>* is *<sup>V</sup>* <sup>=</sup> <sup>2</sup>π*d*2*w*/3, and the cross-sectional area of *P* (that is, the area of a cross-section of *P* in a plane containing the axis of revolution) is *A* = π*dw*/2. So *V* is proportional to *Ad* (we have *V* = 4*Ad*/3), and hence it follows from the definition of -<sup>V</sup> that

$$
\Delta\_{\rm V} = \frac{\delta A\_{\rm L} - A\_{\rm R}}{\delta A\_{\rm L} + A\_{\rm R}} \times 100 \qquad \text{with} \qquad \delta = \frac{d\_{\rm L}}{d\_{\rm R}}, \tag{1}
$$

where *d*<sup>L</sup> and *d*<sup>R</sup> are the diameters of the left and right poles, respectively. The conical model leads to the same result; indeed, a right-circular cone with diameter *d* and height *w* has volume *<sup>V</sup>* <sup>=</sup> <sup>π</sup>*d*2*w*/12 and cross-sectional area *<sup>A</sup>* <sup>=</sup> *dw*/2, and so *<sup>V</sup>* is proportional to *Ad*, as in the spheroidal model.

Equation (1) can be rewritten as

$$
\Delta\_{\rm V} = \left( \delta \frac{A\_{\rm L} - A\_{\rm R}}{\delta A\_{\rm L} + A\_{\rm R}} + (\delta - 1) \frac{A\_{\rm R}}{\delta A\_{\rm L} + A\_{\rm R}} \right) \times 100 \quad (2)
$$

$$\text{for} \quad \Delta\_{\text{V}} = \left( \frac{A\_{\text{L}} - A\_{\text{R}}}{\delta A\_{\text{L}} + A\_{\text{R}}} + (\delta - 1) \frac{A\_{\text{L}}}{\delta A\_{\text{L}} + A\_{\text{R}}} \right) \times 100. \tag{3}$$

Since *A*<sup>L</sup> ≥ *A*R, we have

$$\frac{A\_{\rm L} - A\_{\rm R}}{\delta A\_{\rm L} + A\_{\rm R}} \in \begin{cases} \left[ \triangle / \delta, \,\bigtriangleup \right] \text{ if } \,\delta > 1\\ \left[ \triangle, \,\bigtriangleup \right] \text{ if } \,\delta < 1 \end{cases} \tag{4}$$

$$\text{and} \quad \frac{A\_{\text{R}}}{\delta A\_{\text{L}} + A\_{\text{R}}} \le \frac{1}{\delta + 1} \le \frac{A\_{\text{L}}}{\delta A\_{\text{L}} + A\_{\text{R}}} \tag{5}$$

We deduce from Equations (2)–(5) that

$$
\min\left(\delta, 1\right)\,\Delta + \varepsilon\left(\delta\right) \le \Delta\_V \le \max\left(\delta, 1\right)\,\Delta + \varepsilon\left(\delta\right)
$$

$$
\text{with}
\quad \varepsilon\left(\delta\right) = \frac{\delta - 1}{\delta + 1} \times 100.\tag{6}
$$

In practice, the left-to-right diameter ratio δ is close to one (for example, in our data, the sample mean and standard deviation of δ are 1.09 and 0.12, respectively). Therefore, it follows from Equation (6) that -<sup>V</sup> is a nearly linear function of for both the spheroidal and conical models.

#### *General case*

The equivalence between the spheroidal and conical models in terms of the relationship between and -<sup>V</sup> motivates a generalization: we now assume that the left and right spindle poles are solids of revolution with generating curves obtained by scaling the value and the argument of an arbitrary function *<sup>f</sup>* : [0, 1] <sup>→</sup> <sup>R</sup>+.

pericentrin staining). **(B)** Maximum intensity stack projection showing that the entirety of the spindle apparatus is taken into consideration thanks to the equal sized centrosomes (pericentrin, red). **(C)** Maximum intensity stack *(Continued)*

#### **FIGURE 3 | Continued**

projection of the same cell revealing the tubules only. **(D–K)** Detailed methods for 2D area determination. **(D,H)** Schematic representation of a symmetric **(D)** and an asymmetric metaphase cells **(H)**. When the spindles are symmetric, each part are of equal sizes (Green = Left spindle, Yellow = Right spindle, arbitrarily consider), conversely, when the spindles are asymmetric, the left spindle area is significantly larger than the right one. **(E,I)** For each cells, a primary reconstruction is made to verify that the centrosome are of equal sizes (**E,I**, yellow dots). **(F,J)** The SSA intensity is determined on maximal intensity stack projection reconstructed under the ImageJ software. Each spindle pole is manually drawn and the corresponding area (1 and 2)

This general model is schematized in **Figure 4B**. The left and right spindle poles have the same axis of revolution and their shapes differ only in the values of the diameter *d* and the width *w* (for example, we obtain the spheroidal and conical models by respectively setting *<sup>f</sup>*(*x*) <sup>=</sup> <sup>√</sup> 1 − *x*<sup>2</sup> and *f*(*x*) = 1 − *x*). Our only assumptions on the shape function *f* are that it is continuous and such that *f* (0) = 1 and *f* (1) = 0.

Let us temporarily drop the subscripts "L" and "R" for simplicity. Formally, a spindle pole *P* is obtained by rotating the region

$$D = \left\{ (\mathbf{x}, r) \in \mathbb{R}\_+ \times \mathbb{R}\_+ \mid \mathbf{x} \le \mathbf{w} \text{ and } r \le (d/2)f(\mathbf{x}/\mathbf{w}) \right\} \tag{7}$$

about the *x* axis. According to Pappus' centroid theorem, the volume of *P* is

$$V = 2\pi A(D)\overline{r} \tag{8}$$

where *A*(*D*) denotes the area of *D* and *r* is the distance of the centroid of *D* to the axis of revolution. By definition,

$$A(D) = \frac{d}{2} \int\_0^\infty f(\mathbf{x}/\mathbf{w})d\mathbf{x} \tag{9}$$

$$\mathbf{i} \text{ and } \quad \overline{r} = \frac{1}{A(D)} \int\_0^\mathbf{w} \left( \int\_0^{(d/2)f(\mathbf{x}/\mathbf{w})} r d\mathbf{r} \right) d\mathbf{x}, \tag{10}$$

or equivalently,

$$A(D) = \frac{d\nu}{2} \int\_0^1 f\left(u\right) \,\mathrm{d}u\tag{11}$$

$$\text{and}\quad \overline{r} = \frac{d^2 w}{8A \left(D\right)} \int\_0^1 f^2 \left(\mu\right) \,\text{d}u. \tag{12}$$

Let us now reintroduce the subscripts "L" and "R" to distinguish the left and right poles. From Equation (8), we have *V*<sup>L</sup> = π*A*L*r*<sup>L</sup> and *VR* = π*A*R*r*R. Substituting these two expressions into the definition of -<sup>V</sup> gives

$$
\Delta\_{\rm V} = \left(\frac{\rho A\_{\rm L} - A\_{\rm R}}{\rho A\_{\rm L} + A\_{\rm R}}\right) \times 100 \quad \text{with } \rho = \frac{\overline{r}\_{\rm L}}{\overline{r}\_{\rm R}}.\tag{13}
$$

calculated. Arbitrarily, the bigger area will be defined as the Left spindle and the smallest as the Right spindle. **(G,K)** The difference between the Left (green) and the Right area (yellow) expressed in percentage will be the unit of measurement, delta (-). (**L**) SSA distribution at different time points during *in vitro* cortical cells development. Cells were respectively taken at E10, E11.5, E13 and cultured for 1 day to 1.5 day *in vitro* (DIV). Consistent with previous report (Delaunay et al., 2014), the SSA variation follows a folded normal distribution and parallels the asymmetric cell division kinetics: first an increase with a peak at E14.5 followed by a decrease at E16.5. **(M)** The change in SSA at E14.5 appears highly significant, as demonstrated by the permutation test (*<sup>p</sup>* <sup>=</sup> <sup>5</sup>.10−4). Scale bars: **(A)** <sup>5</sup>μm E, I 10μm.

Furthermore, by Equation (11), *A*R/*A*<sup>L</sup> = *d*R*w*R/(*d*L*w*L), and so it follows from Equation (12) that

$$
\rho = \left(\frac{d\mathbf{l}}{d\mathbf{k}}\right)^2 \frac{\mathbf{w} \mathbf{L} A\_\mathbf{R}}{\mathbf{w} \mathbf{R} A\_\mathbf{L}} = \frac{d\mathbf{l}}{d\mathbf{k}} = \delta.\tag{14}
$$

In other words, Equation (1) holds when the spindle poles are solids of revolution defined by an arbitrary shape function. Consequently, the bounds given in Equation (6) remain valid in the general case, and so we conclude that 2D and 3D measurements have the same discriminative ability for SSA assessment.

#### **EXPERIMENTAL VALIDATION OF THE QUASI-LINEAR RELATIONSHIP BETWEEN 3D AND 2D SSA**

We performed linear regression analyses to validate the quasilinear relationship (Equation 6) between 3D and 2D spindle-pole differences. Two separate samples were analyzed: (i) dissociated cortical progenitors (from E10 to E16, **Figure 3**) and (ii) monkey VZ precursors *in situ* (from E63 to E80, **Figures 5A–F**). The results are summarized in **Figure 5**. The green and magenta curves respectively delimit the 95% simultaneous and pointwise confidence bands; that is, the true regression lines lie between the green curves with a probability of 95%, and given a 2D measurement -∗, there is a 95% probability that the corresponding 3D measurement is bounded by the magenta curves at - = -∗. In accordance with the bounds given in Equation (6), the slopes of the regression lines are close to one: the regression line L1 of the *in vitro* mouse data has a slope of 1.009 with a standard deviation of 0.095, and the regression line L2 of the *in vivo* monkey data has a slope of 0.831 with a standard deviation of 0.084 (the intercepts of L1 and L2 are smaller than 4%). The 95% confidence intervals for the true slopes of L1 and L2 are (0.82, 1.20) and (0.66, 1.00), respectively. That is, we estimate with 95% confidence that if the 2D spindle-pole difference increases by 10%, then the mean 3D spindle-pole difference increases by somewhere between 8.2 and 12% in the case of the *in vitro* mouse data, and between 6.6 and 10% in the case of the *in situ* monkey data—this further confirms the proportionality between 2D and 3D measurements, and hence their equivalent discriminative power.

#### **DISCUSSION**

In the present study, we provide two distinct methods to achieve accurate SSA measurements—3D volume determination and 2D area measurement. Theoretical and empirical comparisons of the two methods show a nearly linear relationship. Using

Pappus'centroid theorem, we demonstrated that this relationship is independent of the spindle shape. This structural property rules out any potential bias of spindle deformation on SSA determination, thereby further supporting the validity of SSA 2D measurement. Finally, we confirmed our theoretical findings by performing linear regression analyses on *in vitro* (mouse) and *in situ* (monkey) metaphase cell samples.

#### **METHODOLOGICAL CONSIDERATIONS**

3D volume measurement requires the analysis of spindle contours on approximately ten individual optical sections. Therefore, minute errors in delineating the pixel contours (2D) are amplified when summing the results. By contrast, errors in spindle contour delineation will have a smaller impact on 2D surface measurement, where the whole spindle apparatus is reduced to a single plane. These technical discrepancies could explain the minute variations in linearity observed experimentally when comparing 2D and 3D SSA measurements (**Figure 5**).

For both methods, we selectively sampled the metaphase cell populations. Only cells displaying equal sized centrosomes were taken into consideration, a configuration that favors cells harboring a spindle aligned parallel to the acquisition plane. *In vitro*, the spindle apparatus is easily accessible and cells mostly divide parallel to the coated surface—our observations show that 86% (E10) and 82% (E14) of precursors exhibit a spindle angle deviation which ranges between 0 and 5 degrees. When we compared the spindle angle deviation with the SSA magnitude, we found no correlation between those two parameters, demonstrating the independence between the spindle angle deviation and the spindle size asymmetry (SSA). *In situ* however, cells could potentially divide along all axis, causing a bias in the representation of rostrocaudally dividing cells. In the neuroepithelium, apical progenitor have been described as aligning along the planar axis before rotating along the rostro-caudal axis during metaphase (Peyre et al., 2011). Such a rotation pattern has also been observed under live imaging in mouse embryonic cortex, in dividing apical progenitors expressing alpha-tubulinEGFP at metaphase (Delaunay et al., 2014). The spindles were aligning along the planar axis, docking, rotating around the caudo-rostral axis, coming back to their original planar position, moving around the planar axis or staying

at the same place, rotating again along the rostro-caudal axis and so on until the beginning of anaphase. We quantified the 2D and 3D spindle pole differences for each sequential moment of planary aligned spindles–between the rounds of rostro-caudal rotations and observed that the spindle pole size difference was stable (Delaunay et al., 2014; **Figure 1**). In all cases, asymmetric spindle are observed to remain asymmetric and a symmetric spindle remains symmetric, independently of the rostro-caudal rotations. Therefore, population sampling is unlikely to affect SSA measurements. This is further supported by the fact that SSA can be observed at similar frequencies on coronal (mouse) (Delaunay et al., 2014) and parasagittal (monkey) sections. Thus, SSA can be unambiguously determined in 2D. Taken together, our results provide a reliable method for SSA quantification in cortical apical progenitors, a method that can be extended to other cell types.

#### **2D vs. 3D SSA QUANTIFICATION**

Alongside SSA, changes in plane of division orientation have always been considered as major determinants of ACD in the cerebral cortex. Spindle orientation—although controversial could regulate the fate of cortical progenitors by controlling the balance between proliferation and differentiation (Chenn and McConnell, 1995; Yingling et al., 2008; Godin et al., 2010). Generally quantified in 2D, spindle orientation varies between two extremes: horizontal divisions (0–15◦ angle, relative to the referential axis) or vertical divisions (75–90◦). Horizontal divisions are associated with symmetric divisions and vertical divisions with asymmetric divisions. A recent work from the Knoblich group reports a new method for 3D analysis of the mitotic plane orientation (Juschke et al., 2014). The authors approximate the mitotic cell by a sphere and mathematically define the spindle orientation by elongating the spindle axis so that it interacts with the surface of the sphere. Under these conditions, randomness results in a predominance of horizontally oriented spindles (close to 50%), a result which could be explained by true stochasticity. To refine their 3D analysis, the authors have introduced two novel parameters: λh and λv, respectively the horizontal and vertical enrichment. This method excludes the effects of planar cell polarity (important in numerous epithelia) and assumes symmetry around the z axis. Juschke et al used this method to assess the role of two proteins: PP4C and mInsc on the plane of division orientation. Interestingly, they report an equivalence between 2D and 3D results for the PP4C-KO with no distinction between randomization or horizontal vs. vertical enrichment. The same analysis performed with the mInsc protein reveals a horizontal enrichment in the KO and a vertical enrichment in mInsc overexpressing cells. Of note, the instructive effect of mInsc overexpression on vertical divisions has been reported by another group, using 2D analysis (Konno et al., 2008; **Figure 3**). The congruence between 2D and 3D spindle orientation analysis argues in favor of the robustness of the 2D SSA assessment methodology.

SSA has recently been documented in the developing mouse cortex, highlighting its importance in Vertebrates and Invertebrates ACD regulation (Delaunay et al., 2014). Previously our evidence for a role of SSA in ACD was in rodents. The present data provide the first evidence of SSA in primate apical progenitors. This indicates that despite the difference in the basic cellular (Betizeau et al., 2013) and molecular regulation (Arcila et al., 2014) between rodent and primate corticogenesis, SSA operates in similar fashion in both orders. Likely, SSA will play a crucial role in controlling ACD in the VZ but also in the OSVZ in the primate lineage. This maintenance of SSA in asymmetrically dividing progenitors argues in favor of its crucial importance during cortical development, thus, highlighting the need for an accurate yet methodically simple quantification process such as the one proposed here.

# **AUTHOR CONTRIBUTIONS**

Delphine Delaunay and Colette Dehay conceived and designed experiments. Delphine Delaunay performed experiments and analyzed the data. Marc C. Robini performed the mathematical and statistical analysis. All authors wrote the paper.

#### **ACKNOWLEDGMENTS**

We thank Cedric Manesse for technical help with the monkey data set acquisition, Ken Knoblauch for computation and statistical expertise as well as for the R manuscript and Henry Kennedy for critical reading of the manuscript. This work received support from LABEX CORTEX (ANR-11-LABX-0042), LABEX DEVWECAN (ANR-10-LABX-0061) of Université de Lyon (ANR-11-IDEX-0007), ANR PRIMACOR.

#### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at: http://www.frontiersin.org/journal/10.3389/fncel.2015. 00033/abstract

The VolumeJ program,the "utilFuncs.R" script and the related instructions are provided as text format.

## **REFERENCES**


Yingling, J., Youn, Y. H., Darling, D., Toyo-Oka, K., Pramparo, T., Hirotsune, S., et al. (2008). Neuroepithelial stem cell proliferation requires LIS1 for precise spindle orientation and symmetric division. *Cell* 132, 474–486. doi: 10.1016/j.cell.2008.01.026

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 28 October 2014; accepted: 19 January 2015; published online: 09 February 2015.*

*Citation: Delaunay D, Robini MC and Dehay C (2015) Mitotic spindle asymmetry in rodents and primates: 2D vs. 3D measurement methodologies. Front. Cell. Neurosci. 9:33. doi: 10.3389/fncel.2015.00033*

*This article was submitted to the journal Frontiers in Cellular Neuroscience.*

*Copyright © 2015 Delaunay, Robini and Dehay. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Interkinetic nuclear migration generates and opposes ventricular-zone crowding: insight into tissue mechanics

#### **Takaki Miyata \*, Mayumi Okamoto, Tomoyasu Shinoda and Ayano Kawaguchi**

Anatomy and Cell Biology, Nagoya University Graduate School of Medicine, Nagoya, Aichi, Japan

#### **Edited by:**

Takeshi Kawauchi, Keio University School of Medicine, Japan

#### **Reviewed by:**

Federico Calegari, Center for Regenerative Therapies Dresden, Germany Caren Norden, The Max Planck Institute of Molecular Cell Biology and Genetics, Germany

#### **\*Correspondence:**

Takaki Miyata, Anatomy and Cell Biology, Nagoya Graduate School of Medicine, 65 Tsurumai, Showa, Nagoya, Aichi 466-8550, Japan e-mail: tmiyata@med.nagoya-u.ac.jp The neuroepithelium (NE) or ventricular zone (VZ), from which multiple types of brain cells arise, is pseudostratified. In the NE/VZ, neural progenitor cells are elongated along the apicobasal axis, and their nuclei assume different apicobasal positions. These nuclei move in a cell cycle–dependent manner, i.e., apicalward during G2 phase and basalward during G1 phase, a process called interkinetic nuclear migration (INM). This review will summarize and discuss several topics: the nature of the INM exhibited by neural progenitor cells, the mechanical difficulties associated with INM in the developing cerebral cortex, the community-level mechanisms underlying collective and efficient INM, the impact on overall brain formation when NE/VZ is overcrowded due to loss of INM, and whether and how neural progenitor INM varies among mammalian species. These discussions will be based on recent findings obtained in live, threedimensional specimens using quantitative and mechanical approaches. Experiments in which overcrowding was induced in mouse neocortical NE/VZ, as well as comparisons of neocortical INM between mice and ferrets, have revealed that the behavior of NE/VZ cells can be affected by cellular densification. A consideration of the physical aspects in the NE/VZ and the mechanical difficulties associated with high-degree pseudostratification (PS) is important for achieving a better understanding of neocortical development and evolution.

**Keywords: interkinetic nuclear migration, cortical development, time-lapse imaging, neural progenitor cells, cell division, slice culture, crowding, mechanical processes**

## **UNDIFFERENTIATED NEURAL PROGENITOR CELLS EXHIBIT INTERKINETIC NUCLEAR MIGRATION**

The neural tube and walls of the early embryonic brain vesicles are composed entirely of undifferentiated progenitor cells (or "matrix cells" (Fujita, 1963)) and are referred to collectively as the neuroepithelium (NE) (reviewed in Götz and Huttner, 2005; Miyata, 2008; Taverna et al., 2014). Structurally, the NE is pseudostratified; that is, although there are several layers of nuclei in the NE wall (**Figure 1**, left), one layer of cells can host multiple layers of nuclei (**Figure 1**, right). Each cell extends to contact both the apical and basal surfaces of the wall, resulting in a bipolar cellular morphology with apical and basal processes. Nuclei of progenitor cells born at the apical surface of the NE move toward the basal side of the NE during G1 phase of the cell cycle. After completing S-phase in the basal portion of the NE, the nuclei return to the apical surface, where they undergo division as their parent cells did. Collectively, these processes are referred to as interkinetic nuclear migration, INM (or IKNM), (Schaper, 1897; Sauer, 1935; Sauer and Walker, 1959; Sidman et al., 1959; Fujita, 1962, reviewed in Taverna and Huttner, 2010; Kosodo, 2012; Reiner et al., 2012; Spear and Erickson, 2012; Lee and Norden, 2013).

As development proceeds, brain walls thicken and the bipolarshaped progenitor cells grow in length (matrix cells (Fujita, 1963) or radial glial cells (reviewed in Rakic, 2003; Götz and Huttner, 2005; Lui et al., 2011; Taverna et al., 2014)). In these thickening brain walls, a large neuronal zone emerges along the outer pial surface, whereas the inner ventricular zone (VZ) consists mainly of progenitor cell somata. Together, NE cells and undifferentiated VZ cells are referred to as apical progenitors (based on their division at the apical surface, Götz and Huttner, 2005; Lui et al., 2011; Shitamukai and Matsuzaki, 2012; Taverna et al., 2014). Similar to the early NE, the VZ is pseudostratified due to the INM behaviors of progenitor cells. The thickness of the VZ is defined as the range of INM, leaving the outer neuronal territory free of progenitors' nuclei. In the neocortical VZ, the basal region is dominated by nuclei of cells in S-phase and late G1-phase, whereas more apical parts are filled with nuclei of early G1-phase and G2-phase cells (Takahashi et al., 1994; Hayes and Nowakowski, 2000). To preserve normal tissue integrity, cells in the NE/VZ need to maintain an apically attached morphology with apical localization of the centrosome. INM is affected, due to abnormal delamination, when molecules involved in this process do not function, e.g., in the absence of Cdc42 (Cappello et al., 2006) or aPKCλ (Imai et al., 2006; Baye and Link, 2007), which control the adherens junction, or SAS-4 (Insolera et al., 2014), which regulates centriole biogenesis.

# **INM-MEDIATED PSEUDOSTRATIFICATION IN THE DEVELOPING NEOCORTEX IS EXTENSIVE AND INCREASES CELL PRODUCTION PER UNIT APICAL AREA**

INM is observed in a wide variety of epithelia, including those of non-nervous system tissues or non-vertebrate animals (Sauer, 1936; Fujita, 1960; Grosse et al., 2011; Meyer et al., 2011; Rujano et al., 2013; Yamada et al., 2013). NE of the zebrafish retina and hindbrain (Leung et al., 2011; Lee and Norden, 2013) and the mouse retina (Baye and Link, 2007) do not exhibit clear intra-NE segregation of nuclei of cells at different cell-cycle phases (as seen in the mouse neocortex, with S-phase nuclei localized basally and non–S-phase nuclei localized apically). Instead, nuclei of all cell-cycle phases (except M-phase) intermingle and are seen all along the apicobasal axis of the epithelium. This "intermingling" (non-segregation) pattern also arises in the Drosophila wing disk (Meyer et al., 2011) and the mouse embryonic ureteric tube (Yamada et al., 2013). The difference between the neocortical NE/VZ and non-neocortical NE or non-NE pseudostratified epithelia could be explained by that the trajectory of INM (i.e., the extent of basalward nucleokinesis) differs depending on the subtype of progenitor cells. In the retina, Baye and Link (2007) found that the more basal the nucleus moves, the more likely it becomes that the next division will lead to production of neurons. Another intriguing possibility is that the collective INM pattern in the neocortical NE/VZ reflects physical conditions, such as tissue volume, cell number, and cellular traffic/flow in a given space. Histological comparisons in embryonic mice have revealed that the NE/VZ is thicker and more persistently maintained in the neocortex than in the brain stem (Miyata, 2007). From an evolutionary standpoint, it is noteworthy that the neocortical VZ is much thicker in human than in mouse (Zecevi´c, 1993; Bayer and Altman, 2006). These size-related observations suggest that pseudostratification (PS) in the neocortical primordium is the most extensive (i.e., in the apicobasal range of INM), and that neocortical NE/VZ would therefore be a good model to study how physical or mechanical issues or parameters at the level of tissue or cell communities may affect neural progenitor behaviors.

What is the biological significance of high-degree INMmediated PS of the type observed in the neocortical NE/VZ?

In **Figure 2**, a simple cuboidal epithelium and two differently pseudostratified columnar (two-nuclei–and four-nuclei–deep) epithelia are compared. In the simple cuboidal epithelium, the length of each side of a cell is *a*. In the pseudostratified epithelia, the longer side of each apical endfoot remains *a*, whereas the other side of the apex shortens and the apicobasal length of each cell increases. The comparison reveals that increasing the degree of PS along the apicobasal axis may horizontally densify neural progenitors (i.e., increase the number of progenitors per unit of subapical volume and increase the number of mitoses per unit [5*a* 2 ] of apical surface area). Therefore, highdegree PS allows an epithelial system to increase its productivity at the apical surface (Smart, 1972; Fish et al., 2008; Miyata, 2008).

This discussion should be coupled with consideration of why NE/VZ progenitor cells divide at the apical surface. The centrosome is located in the apical endfoot due to the presence of a primary cilium (Paridaen et al., 2013; Insolera et al., 2014). Primary cilia are implicated in Wnt and Shh signaling as well as

cell-cycle regulation (reviewed in Bisgrove and Yost, 2006; Fuccillo et al., 2006; Marshall and Nonaka, 2006), suggesting that progenitor cells need to possess an apical endfoot with a primary cilium in order to maintain their developmental potential and stem cell– like proliferation (Götz and Huttner, 2005; Cappello et al., 2006). Furthermore, the Delta–Notch interaction, which is important for the maintenance of stem-like cells, occurs at the apical surface (adherens junction) (Ohata et al., 2011; Hatakeyama et al., 2014). For undifferentiated NE/VZ cells connected to the apical surface, it seems beneficial to send the nucleus/soma to the apical endfoot in order to make the centrosome available for mitosis. Also, integration of newly generated daughter cells into the apical surface is easily achieved through apical mitoses. In the developing mouse neocortex, most apical mitoses occur with a cleavage furrow perpendicular to the apical surface, dividing each apical endfoot (Smart, 1973; Konno et al., 2008; Shitamukai and Matsuzaki, 2012; **Figure 3**). Consequently, daughter cells can easily and immediately join the apical meshwork. Thus, localizing mitoses to the apical surface is a favorable cytogenetic strategy for efficient expansion of undifferentiated NE/VZ cells in brain primordia. INM-mediated PS facilitates apical divisions, thereby supporting the maintenance/expansion of undifferentiated stem-like cells.

# **WHAT DIFFICULTIES CONFRONT HIGH-DEGREE PS AND COLLECTIVE INM?**

Once we understand the aforementioned benefits of PS, we can then consider how it is efficiently achieved. In other words, we need to understand how INMs of NE/VZ cells are coordinated and assembled in an orderly manner. As an approach to addressing this question, it is useful to discuss several potential difficulties that NE/VZ cells need to overcome. **Figure 3** illustrates, schematically but as faithfully as possible based on microscopic observations, cells in the VZ of the midembryonic mouse neocortex. All cells have an apical process that, together with the neighbor cells' apices, constitutes the apical junction meshwork. An M-phase cell (green colored) undergoes cytokinesis horizontally (in an orientation parallel to the apical surface) to give rise to two daughter cells. The right panel highlights the relationship between initial volume of the pair-generated daughter cells and the space that is allowed as a route for the daughter cells' nuclear migration. The outflow tract (a canal for basalward nucleokinesis by daughter cells) seems to be less than one-cell diameter due to the existence of other cells surrounding it at high density, while there are two nuclei (somata) to flow out. How can such a potential bottleneck problem be solved? Our time-lapse observations on slices prepared from H2B-mCherry transgenic mice (in which all nuclei are visualized) yielded a clear answer: pair-generated daughter cells usually move their nuclei basalward in a sequential, rather than simultaneous, manner (Okamoto et al., 2013). However, this observation raises a further question: how can daughter cells initiate basalward nucleokinesis sequentially?

# **BASAL PROCESS: A MOTHER'S KIND GIFT HELPS DAUGHTERS' TRAFFIC AND BRAIN FORMATION**

Time-lapse monitoring of daughter cells generated at the apical surface of VZ in slice culture was followed by quantitative analysis of nuclear movements. Measurement of mean-squared displacement (MSD) was used to determine the relationship between the morphology of daughter cells and the directionality of their initial nuclear movement. If the MSD for a tracked nucleus has a linear relationship with elapsed time (i.e., the MSD graph exhibits a linear pattern), the movement of the tested nucleus is considered to have random tendencies (i.e., non-directional and fluctuating motion). If the MSD graph instead exhibits a positive curvature, the movement is considered to be directional or persistent (Norden et al., 2009; Leung et al., 2011). The basal process of each apically dividing M-phase progenitor in the NE/VZ with a certain minimum thickness (>50 µm) is inherited by one of its daughters (Miyata et al., 2001; Noctor et al., 2001). In nuclear MSD profiles of daughter cells generated from a single progenitor, a more directional pattern was observed in the processinheriting daughter cell than in its sister cell (Okamoto et al., 2013). This mechanism, in which the process-inheriting daughter cell moves its nucleus more quickly than its sister cell, is analogs to "priority boarding" in air travel or "staggered commuting" in metropolitan railway systems, and may contribute to the normally prompt stratification of nuclei/somata in each outflow tract (**Figure 4**).

If inherited basal processes are strictly analogs to priority boarding passes that streamline the flow of passengers under crowded conditions, the loss of these processes should lead to traffic problems such as severe overcrowding and congestion. Acute knockdown (KD) of the cell-surface molecule TAG-1 was used to determine the functional importance of the basal process and address the role of normal INM in overall brain formation (Okamoto et al., 2013). Upon loss of TAG-1, which is normally expressed in the basal part of mouse neocortical walls at embryonic day 10 (E10)–E11, VZ cells lost their basal processes by E12. Consequently, their basalward nucleokinesis

was severely limited, causing their somata to accumulate near the apical surface (**Figure 5**, left part). By E13, these abnormally shortened VZ cells delaminated from the apical surface and invaded the basal area that should normally be occupied by neurons, thereby disrupting segregation of progenitor cells and neurons (**Figure 5**, center). The delaminated progenitors remained proliferative/undifferentiated (Pax6+) at heterotopic (far basal) positions until late in embryonic development (∼E17). Nevertheless, distribution of neurons that were generated sequentially from these malpositioned progenitors was quite abnormal. Instead of forming layers, ectopically generated neurons were scattered almost throughout the wall in a randomized pattern (**Figure 5**, right part). This dysplasia therefore indicates that appropriate control of nucleokinesis within the early NE/VZ is important for preventing intermingling of progenitors and neurons, and thereby contributes to normal brain formation. However, why do the acutely shortened VZ cells in TAG-1–KD cerebral walls observed at E12 detach from the apical surface by E13?

# **EXPERIMENTALLY INDUCED ACUTE OVERCROWDING INCREASES MECHANICAL STRESS IN VZ AND INDUCES ABNORMAL DELAMINATION**

Monitoring at E12 revealed that the shortened TAG-1–KD VZ cells were overcrowded (subapically about 20% denser than in the normal VZ) (**Figure 6**, left top corner). Prompted by the hypothesis that VZ cells leave the apical surface when mechanical

factors related to cell density increase to an intolerable level, reflecting high-degree proliferation (Smart, 1965, 1972), a series of experiments analyzed the physical condition of the overcrowded TAG-1–KD VZ. Microsurgical techniques such as laser ablation or making slices from hemispheric walls can be used to observe the mechanical conditions of cells or tissues of interest. If a certain portion is under tension or compression *in vivo*, the incision edges or freed tissue portions will then move according to the original mechanical conditions; these processes can be observed by microscopic monitoring. For example, laser ablation on the apical surface (as in *test 1*, **Figure 6**) results in centrifugal movement of the released vertices from the ablation point, revealing that the apical surface is contractile (as a result of the action of actomyosin-dependent mechanisms) and must therefore be under tension. Also, slicing cerebral hemispheric walls allows them to apically bend or curl (*test 2*, **Figure 6**). These techniques (destressing or stress-release tests) revealed that the subapical zone of the overcrowded TAG-1–KD VZ was indeed under excessive compression (as revealed in persistent separation of the tracked vertices in *test 1* and poorer bending/curling in *test 2*); this observation was further supported by *in silico* mechanical simulations (Okamoto et al., 2013). Thus, an overcrowdinginduced delamination mechanism, such as the one recently reported in the *Drosophila* epithelium (Mariani et al., 2012), may also function in the developing mammalian neocortex. Progenitors evacuate (or are forced to exit) from the VZ in response to excessive acute mechanical stress.

**FIGURE 6 | Mechanical tests used for comparing normal and TAG-1–KD cerebral walls (Okamoto et al., 2013)**. In test 1, a pulse of UV laser was applied on the midpoint of a boundary line formed by two polygonal apices of VZ cells. Vertices at both ends of the laser-targeted side were tracked, and their separation was quantitated. The TAG-1–KD group exhibited greater and more persistent separations. In test 2, bending and curling of slices freshly prepared from normal or TAG-1–KD hemispheres were monitored under a phase-contrast microscope. TAG-1–KD slices that subapically contained many overcrowded (shortened) VZ cells were stiffer, exhibiting no or poorer bending/curling.

Similar delamination of undifferentiated progenitors also occurred when another acute physical load on VZ cells was imposed by over-proliferation, induced by artificial expression of Wnt3a (Okamoto et al., 2013; **Figure 5**, left bottom corner). In this experiment, the nuclear horizontal packing density in the VZ increased (by about 6%), concomitant with a thickening of the VZ (by about 20%) and a nuclear densification along the apicobasal axis (by about 11%), and these VZ-densified Wnt3aoverexpressed cerebral walls bent or curled poorly. The molecular mechanisms by which VZ cells sense and respond to these acute mechanical loads should be investigated using mechanobiological approaches (Mammoto et al., 2012; Heisenberg and Bellaïche, 2013; Iskratsch et al., 2014). Wnt3a increases self-maintaining (non–neuron-producing) divisions during early cortical development through activation of β-catenin (Munji et al., 2011). Notably, telencephalic walls of transgenic mice expressing constitutively active β-catenin produce basal heterotopia of undifferentiated progenitors (Chenn and Walsh, 2002). Excessive FGF signaling also increased excessive basal mitosis (Inglis-Broadgate et al., 2005). Overproliferation induced in the VZ via artificial shortening of G1 phase of the cell cycle resulted in the expansion of non-VZ progenitors (Lange et al., 2009; Nonaka-Kinoshita et al., 2013). These previously reported heterotopic mitoses in rodent models may be better understood in light of progenitors' responsiveness to mechanical stress in NE/VZ. Abnormal expansion of VZ has also been reported in mice lacking Apaf1 (Cecconi et al., 1998; Yoshida et al., 1998), Caspase 3 (Kuida et al., 1996), or Caspase 9 (Kuida et al., 1998) (reviewed in Kuan et al., 2000). A more recent study, however, reported that both Apaf1-deficient and Caspase 9-deficient mice did not show NE/VZ overgrowth (Nonomura et al., 2013). If the expansion and abnormal fragmentation of the apoptosis-inhibited neocortical VZ is reproducible as reported in the initial studies (Kuan et al., 2000), it might be another useful material by which we could ask how delamination is induced by VZ densification.

# **PHYSIOLOGICAL THICKENING AND DENSIFICATION OF VZ DURING DEVELOPMENT**

The observation of Wnt3a-induced thickening and horizontal cellular densification of the VZ provides a good opportunity for further discussion of whether (and, if so, how) the VZ thickening/densification that occurs physiologically during development and evolution might affect VZ cell behaviors. The thickness of the VZ is defined by the extent of PS along the apicobasal axis, i.e., by how many nuclei exhibiting INM are staggered from the apical surface toward the basal side (Sauer, 1935; Smart, 1972). It is likely that as more nuclei are stratified within the VZ, net apicobasal nuclear movements per unit of apical surface area tend to increase. In other words, as the VZ thickens, apicobasal nuclear traffic per unit volume of VZ becomes heavier. **Figure 7** compares cell morphology between a VZ with 7-nucleus-deep PS and another with 12-nucleus-deep PS. The comparison is made within a cylinder-like hypothetical column, because timelapse monitoring of H2B-mCherry-labeled nuclei showed that all nuclei move almost purely apicobasally, rather than horizontally (Okamoto et al., 2013). Probable differences between the two VZ columns with different degrees of PS include reduction in the short diameter of the nucleus/soma (due to the existence of other cells' processes) and densification of the apical endfeet in the 12 nucleus-deep VZ.

During normal mouse embryonic development, the thickness of the neocortical VZ increases from E10 to E12 (Smart, 1973). Consistent with this, the density of apical endfeet also increases

from E10 to E12 (Nishizawa et al., 2007). Notably, basal processes of apically dividing M-phase progenitors in a thin NE/VZ (e.g., E10 mouse cerebral walls or 24-hr zebrafish neural tubes) are often split or bifurcated (Kosodo et al., 2008), which does not occur in mouse cerebral walls at E12 and later. In early zebrafish neural tubes, split basal processes can be inherited symmetrically by two daughter cells (4/9 cases in Kosodo et al., 2008). These results raise the possibility that certain stage-dependent mechanisms that determine whether or not the basal processes are split and/or inherited symmetrically might involve mechanical stimuli, which are presumably weaker at E10 than at E12 when the VZ may be denser. However, asymmetric inheritance of basal processes, as observed in mouse cerebral walls at E12 and later, occurs widely in many different epithelia exhibiting PS: zebrafish neural tube and retina (Das et al., 2003; Kosodo et al., 2008; Alexandre et al., 2010), mouse and rat retina (Cayouette and Raff, 2003; Saito et al., 2003), mouse intestine (Grosse et al., 2011), and mouse ureteric bud (Packard et al., 2013). It is possible that asymmetric inheritance of the basal process contributes almost ubiquitously to management (spatiotemporal dispersion) of tissue stress generated through nuclear currents, and such management could be modified depending on the mechanical situation, which changes as development proceeds in a tissuespecific manner.

# **FERRET–MOUSE DIFFERENCES IN PHYSIOLOGICAL VZ CROWDING AND INM**

The neocortical VZ is much thicker in human than in mouse (Zecevi´c, 1993; Bayer and Altman, 2006). To investigate whether different mammalian species have evolved different strategies for cellular management of VZ nuclear traffic, a recent study compared mouse and ferret VZ at equivalent neocortical developmental stages (E13.5 in mice and E29–30 in ferrets), and found that intra-VZ cellular dynamics differ concomitantly with the thickening and densification of the VZ (Okamoto et al., 2014). Apicobasally, ferret VZ is thicker and slightly denser than mouse VZ: 16 stacks of nuclei in about 120-µm–thick ferret VZ (13.5 nuclei per 100 µm) vs. 12 stacks of nuclei in about 100-µm–thick mouse VZ (12.6 nuclei per 100 µm). Horizontally, the density of apical endfeet is greater (144%) (i.e., each apex is smaller) in ferrets than in mice. Also, horizontal nuclear density in the basal part of VZ is significantly higher in ferret (28 nuclei per 1000 µm<sup>2</sup> ) than in mouse (22–24 nuclei per 1000 µm<sup>2</sup> ). Nuclei are significantly more slender in ferret: the major axis (16.2 µm) is longer than in mouse (11.7 µm), and the minor axis is shorter (5.7 µm vs. 6.0 µm). These differences, obtained by imagingbased quantitation, are in line with expectations schematically shown in **Figure 7**.

In the mouse neocortical VZ, apicalward nuclear movements exhibited by G2-phase progenitors are highly directional (i.e., quick and persistent until they reach the apical surface), with non-linear MSD profiles (Okamoto et al., 2013), very similar to the directional apicalward nuckeokinesis observed in zebrafish retina and brain stem (Norden et al., 2009; Leung et al., 2011). By contrast, the basalward nucleokinesis exhibited by G1-phase mouse VZ cells is less directional (i.e., nuclei exhibited non-linear

MSD profiles along the apicobasal axis), as is also the case for zebrafish cells (Norden et al., 2009; Leung et al., 2011). As mentioned earlier, initial basalward nucleokinesis is more directional in daughter cells that inherit the basal process ("BP") than in daughter cells that do not ("nonBP") (Okamoto et al., 2013; **Figure 4**). Accordingly, the directionality of nucleokinesis in the mouse neocortical VZ is ranked in the following order: apicalward > BP-basalward > nonBP-basalward (**Figure 8**, upper panel). Surprisingly, MSD analysis of ferrets revealed that directionality in the mid-embryonic ferret neocortical VZ is ranked in a different order: BP-basalward > nonBP-basalward > apicalward (Okamoto et al., 2014; **Figure 8**, lower panel). This finding suggests that although the basal process–mediated mechanism for differential initiation of nucleokinesis (Okamoto et al., 2013) is conserved between mice and ferrets, strategies for balancing flows to and from the apical surface differ between these species. Ferret– mouse comparisons at each phase of nucleokinesis suggested that the basalward phase is relatively accelerated, whereas the apicalward phase is decelerated, in ferrets. Future studies should investigate the molecular mechanisms underlying these differential nucleokinesis patterns between mice and ferrets. Whether physical conditions (such as elasticity or stiffness) vary between VZs of different nuclear density could be assessed quantitatively using atomic force microscopic (AFM) techniques (Iwashita et al., 2014).

#### **CONCLUSIONS AND PERSPECTIVES**

As discussed in the first part of this review, PS is an important means by which an epithelial system can increase its productivity at the apical surface (**Figures 2**, **9A**). The second part of this review discussed the difficulties of high-degree PS from the viewpoint of nuclear traffic (**Figures 5**, **6**). The apical surface is

**(B), and a possible revolutionary change in the strategy for cell production from "PS-based apical" under physical/traffic limitations to "non-PS-mediated basal", which is free from subapical traffic difficulties (C)**.

contractile and thus always spontaneously narrowing, although it receives M-phase somata that are expanding and voluminous (**Figure 3**). The co-occurrence of these two mechanically opposing phenomena is supported by the quick disappearance of newly generated G1-phase daughter cells' nuclei from the apical surface. The exclusive use of the mother (M-phase) cell's basal process by only one of its daughter cells facilitates the initial sequential (and thus non-competitive) nucleokinesis of the pair-generated sister cells' nuclei away from the subapical space (the third part of this review, **Figures 4**, **8**). This mechanism may collaborate with other mechanisms reported for basalward nucleokinesis during G1-phase: actomyosin-dependent (Schenk et al., 2009) and microtubule-dependent (Tsai et al., 2010) intracellular regulation, as well as passive basalward nuclear movements dependent on the apicalward nucleokinesis of other cells (Sauer, 1935; Norden et al., 2009; Kosodo et al., 2011; Leung et al., 2011).

The subapical space is physically limited, such that an acute 20% increase in the number of somata can result in abnormal delamination of undifferentiated cells (Okamoto et al., 2013; **Figure 9B**, left part). A similar increase in the load of nuclear traffic seems to be tolerable, if it occurs gradually during evolution. In the ferret VZ, which exhibits a higher level of PS than in the mouse VZ, the INM patterns (in both apicalward and basalward phases) are different from those in mice (Okamoto et al., 2014; **Figure 8**). This ferret–mouse difference suggests that modulation of INM may have allowed VZ cells to achieve high-degree PS, thereby increasing total cell production from the apical surface (**Figure 9B**, right part). However, such a thickened VZ would also encounter mechanical difficulties in the context of acute traffic problems, as shown experimentally in mice (Okamoto et al., 2013), probably limiting PS-based (apical) productivity. This discussion of traffic/mechanical difficulties in the PS system, based on recent live observations and experiments, prompted us to hypothesize that the mechanical conflicts may have caused the generation of new germinal layers during evolution (**Figure 9C**).

The idea that spatial limitations in the VZ may underlie the expansion of proliferative cells to the basal direction has previously been proposed based on histological observations using fixed specimens (Smart, 1965, 1972; Charvet and Striedter, 2011). Overproliferation induced in the VZ via artificial acceleration of the cell cycle resulted in the expansion of non-VZ progenitors (Lange et al., 2009; Nonaka-Kinoshita et al., 2013). Current research techniques allow us to quantitatively capture dynamic behaviors of cells, and to perform experimental manipulations that can change cells' mechanical condition either indirectly or directly (reviewed in Mammoto et al., 2012; Heisenberg and Bellaïche, 2013). Therefore, future studies using mechanobiological approaches should be able to elucidate how a non-PS (non-VZ) proliferative zone for stem-like cells have arisen during neocortical evolution (**Figure 9C**).

In the research field of urban engineering, mass transportation is studied and developed in order to achieve better (i.e., safer, more economical, and more sustainable) quality of life and greater productivity. The human neocortical VZ is much thicker than the mouse neocortical VZ (Zecevi´c, 1993; Bayer and Altman, 2006), giving us an impression that the former is more "urbanized." We speculate that in such an extremely "urbanized" VZ, efforts to become as productive as possible at the apical surface would inevitably face increasing mechanical difficulties in subapical INM traffic. A shift from relying only on the PS to elaborating a new, non-PS cytogenetic method seems to present a mechanically reasonable solution to this challenge (**Figure 9C**). Recent studies have demonstrated that the outer subventricular zone (OSVZ), which contains undifferentiated progenitor cells (OSVZ [or basal] radial glia-like cells, oRG [bRG] cells), is a characteristic developmental feature of the human neocortex (Zecevic et al., 2005; Fietz et al., 2010; Hansen et al., 2010; Reillo et al., 2011; LaMonica et al., 2012; Lewitus et al., 2013). The OSVZ is also evident in non-human primates (Smart et al., 2002; Kelava et al., 2012; Betizeau et al., 2013) and ferrets (Fietz et al., 2010; Reillo et al., 2011; Martínez-Cerdeño et al., 2012; Reillo and Borrell, 2012; Poluch and Juliano, 2013). Although rodent neocortical primordia do not have cytoarchitechtonically distinct OSVZ-like structures, they have oRG-like progenitors (although they are much less abundant than in primates and ferrets) in regions basal to the VZ (Shitamukai et al., 2011; Wang et al., 2011; Martínez-Cerdeño et al., 2012; Tabata et al., 2012).

Vertical mitotic spindle orientation (leading to cytokinesis perpendicular to the apical surface) can contribute to the supply of non–apically connected VZ cells that move basally and eventually adopt an oRG-like morphology (Konno et al., 2008; Postiglione et al., 2011; Shitamukai et al., 2011; LaMonica et al., 2013); therefore, regulation of the cleavage orientation of stem-like cells at the apical surface may underlie the evolutionary changes that have generated the OSVZ. Recent studies on the mechanism regulating cleavage orientation have shown that the intracellular molecular machinery can be influenced by extrinsic factors such as diffusible or extracellular matrix proteins (reviewed in Théry and Bornens, 2006; Lancaster and Knoblich, 2012; Peyre and Morin, 2012; Shitamukai and Matsuzaki, 2012; Williams and Fuchs, 2013). In light of these results, future studies should attempt to determine whether cleavage orientation is regulated by tissue-level mechanical factors or through VZ densification.

Physiological delamination is exhibited by neocortical VZ cells that have acquired non–stem-like (differentiation) properties (Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). This process can now be partly explained by a molecular mechanism similar to one that occurs during the epithelialto-mesenchymal transition: downregulation of E-Cadherin by the Scratch transcription factors, which belong to the Snail superfamily (Itoh et al., 2013). Whether this process is also mechanically regulated, as speculated by Smart (1973), is another question that should be addressed experimentally. It is possible that spatial segregation of different classes of progenitors, which are also seen in non-neocortical NE/VZ tissues such as the developing retina (Weber et al., 2014), occurs under mechanical regulation. Direct mechanical manipulations have been useful for dissecting the molecular mechanisms underlying delamination in several model systems. For example, *Drosophila* and zebrafish embryos undergoing gastrulation have been manipulated using magnetic force (Brunet et al., 2013). In addition, the involvement of uterus-mediated external force in the specification of visceral endoderm cells in early mouse embryos was assessed by a culture system in which embryos were placed in chambers made with gels of different stiffness and by compressing embryos with an AFM cantilever (Hiramatsu et al., 2013). Application of such experimental methods, coupled with quantitative measurement of mechanical forces (as exemplified in this review, **Figure 6**), will deepen our understanding of both physiological (developmental and evolutionary) and pathological delamination (i.e., withdrawal from PS-based apical cytogenesis).

Finally, we are still far from understanding how INM behaviors of all VZ cells are coordinated such that they are not abnormally synchronized, in terms of both cell-cycle progression and nucleokinesis. One possibility worth investigating is that progression of the cell cycle is fine-tuned by cellular sensing of mechanical factors in the environment, and that such mechanosensation-based cellcycle regulation might in turn regulate collective nucleokinesis. A combination of cell-biological experiments and *in silico* simulations should help to address this community-level question *in vivo*.

#### **ACKNOWLEDGMENTS**

This work was supported by a Grant-in-Aid for Scientific Research on Innovative Areas "Cross-talk between moving cells and microenvironment as a basis of emerging order" from the Ministry of Education, Culture, Sports, Science and Technology of Japan.

#### **REFERENCES**


controlling the spatial distribution of myosin II. *Nat. Cell Biol.* 15, 1294–1306. doi: 10.1038/ncb2858


zebrafish retina adapt to architectural changes upon tissue maturation. *Cell Rep.* 7, 386–397. doi: 10.1016/j.celrep.2014.03.014


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 17 October 2014; accepted: 31 December 2014; published online: 28 January 2015*.

*Citation: Miyata T, Okamoto M, Shinoda T and Kawaguchi A (2015) Interkinetic nuclear migration generates and opposes ventricular-zone crowding: insight into tissue mechanics. Front. Cell. Neurosci. 8:473. doi: 10.3389/fncel.2014.00473*

*This article was submitted to the journal Frontiers in Cellular Neuroscience*.

*Copyright © 2015 Miyata, Okamoto, Shinoda and Kawaguchi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms*.

# Regulation of cerebral cortical neurogenesis by the Pax6 transcription factor

*Martine N. Manuel †, Da Mi †, John O. Mason and David J. Price\**

*Centre for Integrative Physiology, The University of Edinburgh, Edinburgh, UK*

Understanding brain development remains a major challenge at the heart of understanding what makes us human. The neocortex, in evolutionary terms the newest part of the cerebral cortex, is the seat of higher cognitive functions. Its normal development requires the production, positioning, and appropriate interconnection of very large numbers of both excitatory and inhibitory neurons. Pax6 is one of a relatively small group of transcription factors that exert high-level control of cortical development, and whose mutation or deletion from developing embryos causes major brain defects and a wide range of neurodevelopmental disorders. Pax6 is very highly conserved between primate and non-primate species, is expressed in a gradient throughout the developing cortex and is essential for normal corticogenesis. Our understanding of Pax6's functions and the cellular processes that it regulates during mammalian cortical development has significantly advanced in the last decade, owing to the combined application of genetic and biochemical analyses. Here, we review the functional importance of Pax6 in regulating cortical progenitor proliferation, neurogenesis, and formation of cortical layers and highlight important differences between rodents and primates. We also review the pathological effects of *PAX6* mutations in human neurodevelopmental disorders. We discuss some aspects of Pax6's molecular actions including its own complex transcriptional regulation, the distinct molecular functions of its splice variants and some of Pax6's known direct targets which mediate its actions during cortical development.

Keywords: proliferation, cell cycle, differentiation, neuronal fate, neurotransmitter fate, cortical lamination, BAF complex, Meis2

# Introduction

The expansion of the cerebral cortex is a major hallmark of mammalian evolution, particularly in the primate lineages where it achieves its greatest complexity in humans (Martin, 1990). Despite great variation in size, there are many similarities in the structure and function of the cerebral cortex across all mammalian species. These similarities have encouraged neuroscientists to use relatively simple cortices, such as those of rodents, as models in which to investigate biological processes and mechanisms with likely relevance to humans. The strength of rodent models, in particular mice, derives in large part from the genetic and transgenic approaches that can be used to study molecular mechanisms. Much of the research described in this review comes from studies of mice and the view of cortical development and Pax6's role in that process presented here is strongly

#### *Edited by:*

*Margareta Nikolic, University of Hertfordshire, UK*

#### *Reviewed by:*

*Akiya Watakabe, National Institute for Basic Biology, Japan Marcos R. Costa, Federal University of Rio Grande do Norte, Brazil*

#### *\*Correspondence:*

*David J. Price, Centre for Integrative Physiology, The University of Edinburgh, Hugh Robson Building, George Square, Edinburgh EH8 9XD, UK David.Price@ed.ac.uk*

> *†These authors have contributed equally to this work.*

*Received: 21 November 2014 Paper pending published: 08 January 2015 Accepted: 18 February 2015 Published: 10 March 2015*

#### *Citation:*

*Manuel MN, Mi D, Mason JO and Price DJ (2015) Regulation of cerebral cortical neurogenesis by the Pax6 transcription factor. Front. Cell. Neurosci. 9:70. doi: 10.3389/fncel.2015.00070* biased toward mouse corticogenesis. It is important to recognize, however, that significant differences exist between the developing and mature cortices of primate and non-primate species that are more than just differences of scale, and we shall highlight some of these. It is remarkable that many high-level regulatory genes such as *Pax6* are themselves extremely highly conserved between primates and non-primates although they influence and control aspects of cortical development that differ: for example, the amino acid sequences of human PAX6 and mouse Pax6 are identical (Ton et al., 1992) although Pax6's expression in primate embryos extends to cortical structures that do not exist in mice. This implies that evolutionary changes in the functions of these regulators have been achieved by changes in the mechanisms regulating their expression and in the ways in which the downstream molecular networks respond to them, but we know almost nothing about these evolutionary changes at a molecular level.

The cerebral cortex is derived from the dorsal component (or pallium) of the embryonic telencephalon, which is itself the anterior-most subdivision of the forebrain (Kiecker and Lumsden, 2005). The cerebral cortex can be further sub-divided into distinct regions, including the neocortex, a novel acquisition of mammals that has evolved between the phylogenetically older archicortex (comprising entorhinal cortex, retrosplenial cortex, subiculum, and hippocampus) and paleocortex (olfactory piriform cortex). The evolutionary expansion of the neocortex is thought to account for much of the increase in overall brain size and complexity in more advanced species (Krubitzer and Kaas, 2005; O'Leary et al., 2007). The neocortex (hereafter referred to simply as cortex) contains an extraordinarily large number of neurons arrayed in a six-layered sheet, with neurons in each layer organized into a complex network of local circuits and subcortical connections. In primates, some of these layers are subdivided in some cortical areas: for example, in primary visual cortex, layer 4 is subdivided into layers 4a, b, and c. The primate cortex is also characterized by an expansion of the superficial layers of the cortex, layers 2 and 3 (also known as the supragranular layers), which have an important function in the transfer of information between cortical areas. Increased intracortical information processing is likely to have contributed to heightened cognitive ability. In all mammals, the cortex comprises two major groups of neurons: the majority are excitatory glutamatergic projection neurons (70–80%), which exhibit a characteristic pyramidal morphology and extend axons to distant intracortical, subcortical, and subcerebral targets; a minority are inhibitory GABAergic non-pyramidal interneurons (25–30% in primates, 15–20% in rodents), which have short axons and project locally (Hendry et al., 1987; Beaulieu, 1993). An appropriate balance between the excitatory and inhibitory circuitry of the cortex is critical for its normal function.

*Pax6* is a pivotal gene in CNS development. It is expressed when the major components of the developing CNS first emerge after neural tube closure and its expression patterns change considerably as major structures such as the cerebral cortex specialize and expand. Its expression plays a major role in the subsequent development of the regions that continue to express it. In this review we shall discuss how Pax6 plays critical roles in aspects of corticogenesis that include the early patterning of telencephalic subdivisions, control of cortical cell number, normal cortical layer formation and the development of the correct balance of excitatory and inhibitory neurons. We shall review what is known about the upstream control of *Pax6*'s transcription, the molecular basis of its functions and its actions on genes downstream of it and the cellular processes they regulate.

# Corticogenesis in Rodents and Primates and the Cortical Expression of Pax6

The closure of the neural tube is accompanied by its disproportionate expansion anteriorly to generate the early forebrain. The earliest cortical progenitors undergo symmetric proliferative divisions at the ventricular surface to amplify a pool of progenitors, an increasing proportion of which then divide asymmetrically to regenerate progenitors and to produce other cell types including neurons.

In mouse, the generation of excitatory cortical neurons occurs between embryonic day 11 (E11) and E18 (Gillies and Price, 1993; Price et al., 1997; Levers et al., 2001). Their progenitors are located in one of two pallial germinal epithelia, namely the ventricular zone (VZ), which lies adjacent to the ventricles, and the subventricular zone (SVZ), which lies just above the VZ (**Figure 1**). The early symmetrically dividing cortical neuroepithelial cells (NECs) that each produce two daughter NECs per division and whose population rapidly expands laterally and radially, transform into another type of progenitor called, for historical reasons, radial glial cells (RGCs; **Figure 1**). RGCs, whose long processes span the neuroepithelium, have been known for a long time to provide guidance for migrating neurons (Levitt and Rakic, 1980; Rakic, 1988; Hatten, 2002; Tan and Shi, 2013). Despite having morphological and molecular features associated with glial cells, they are progenitors capable of generating other types of progenitor, neurons and glial cells (Malatesta et al., 2000, 2003; Miyata et al., 2001; Noctor et al., 2001; Skogh et al., 2001; Heins et al., 2002; Tan and Shi, 2013). RGCs constitute the majority of the VZ progenitor population. They are often referred to as apical progenitors (APs) due to the location of their cell body close to the ventricular or apical surface of the VZ, to which they extend a short process while sending a longer basal process radially to the pial surface of the cortex (Cameron and Rakic, 1991; Bentivoglio and Mazzarello, 1999; Tan and Shi, 2013). As they progress through the cell cycle, RGCs undergo interkinetic nuclear migration. Their nucleus migrates radially through the cytoplasm such that mitosis occurs at the apical ventricular surface and S-phase at a basal position in the VZ. While at early stages of corticogenesis RGCs predominantly self-renew via symmetric divisions, they progressively switch to asymmetric divisions, which produce one daughter RGC and one daughter cell with a heightened level of commitment (Noctor et al., 2001, 2004; Haydar et al., 2003; Huttner and Kosodo, 2005; Tan and Shi, 2013). The differentiating daughter cell either migrates radially to the pial surface and differentiates into a neuron or migrates to the

SVZ to become an intermediate progenitor cell (IPC), also called a basal progenitor (BP) or basal IPC (bIPC; **Figure 1**; Haubensak et al., 2004; Miyata et al., 2004; Noctor et al., 2004). BPs accumulate in the SVZ and divide mainly symmetrically to generate two neurons. It is thought that most cortical projection neurons are generated by BPs (Farkas and Huttner, 2008). In addition to RGCs, the VZ contains another smaller subpopulation of APs called short neural precursors (SNPs), also known as apical IPCs (aIPC; **Figure 1**). Unlike RGCs, aIPCs do not self-renew and only generate pairs of neurons via symmetric divisions (Gal et al., 2006; Tan and Shi, 2013).

Significant progress in understanding the mechanisms underlying corticogenesis has been made through analysis of gene expression in both progenitor cells and post-mitotic neurons. Such studies have revealed a crucial role for transcription factors (TFs) as molecular markers of distinct progenitor and neuron types and as key regulators of progenitor cell proliferation and cell fate decisions. At the onset of neurogenesis, TFs including Pax6, Emx2, and Tlx expressed in the cortical neuroepithelium function mainly to influence areal patterning and regulate progenitor cell proliferation (Bishop et al., 2000, 2002; Muzio et al., 2002; Muzio and Mallamaci, 2003; Stenman et al., 2003a,b; Hevner, 2006). As neurogenesis proceeds, a large number of TFs are involved in regulating the balance between progenitor cell proliferation and neuronal differentiation. Regulating this balance is essential for the generation of the correct proportions of different classes of neurons and subsequent circuit assembly.

During the neurogenic period in the mouse cortex, Pax6 is expressed by VZ APs in a high rostro-lateral to low caudo-medial gradient (**Figures 2** and **3**). As well as being graded spatially, the expression of Pax6 is also graded temporally, with highest levels present at the onset of corticogenesis. APs that give rise to BPs transiently express the proneural TF Neurogenin 2 (Ngn2; Britz et al., 2006). Pax6 is not expressed in BPs which are characterized by their expression of another TF, Tbr2, or in postmitotic neurons which express Tbr1 (**Figure 1**; Englund et al., 2005). Thus a sequential Pax6 → Ngn2 → Tbr2 → Tbr1 expression correlates with the transition of APs to BPs to post-mitotic neurons.

In the macaque cortex, Rakic (1974) showed that cortical neurogenesis occurs between E45 and E100. Gestation in this species lasts 165 days and so cortical neurons are generated relatively earlier in gestation in primates than in rodents. The primate SVZ forms earlier and shows a much greater expansion than in rodents, becoming the predominant progenitor zone by midcorticogenesis (Smart et al., 2002). It is split into the inner and outer SVZs separated by an inner fiber layer (Smart et al., 2002; Lukaszewicz et al., 2005; Zecevic et al., 2005; Fietz et al., 2010; Hansen et al., 2010; Florio and Huttner, 2014). The inner subventricular zone (ISVZ) contains mainly IPCs similar to those in the rodent SVZ whereas the outer subventricular zone (OSVZ) contains mainly basal RGCs (bRGCs) similar to the APs present in the rodent VZ (**Figure 1**). bRGCs undergo proliferative divisions and self renewing asymmetric divisions to generate one bRGC daughter cell and one IPC that can proliferate further (Fietz et al., 2010; Hansen et al., 2010; Florio and Huttner, 2014). bRGCs have

also been observed in the rodent SVZ but, while they constitute about half of all progenitors present in the primate SVZ, they account for only a minute fraction of the SVZ progenitors in rodents (**Figure 1**). The OSVZ is the major source of supragranular layer neurons (Letinic et al., 2002; Lukaszewicz et al., 2005).

The sequential Pax6 → Ngn2 → Tbr2 → Tbr1 expression that correlates with the AP → BP → post-mitotic neuron transition in mice is not found during primate corticogenesis. In contrast to rodents, Pax6 is expressed by progenitors in the VZ, ISVZ, and OSVZ in primates (**Figure 3**; Fietz et al., 2010; Betizeau et al., 2013; Florio and Huttner, 2014) and many progenitors co-express Pax6 and Tbr2 (**Figure 1**). In the macaque, the majority of VZ progenitors (60–80%) express only Pax6 during early and mid-stages of corticogenesis (up to E79), but at later stages (after E79) 40% of them co-express Tbr2 (Betizeau et al., 2013; Florio and Huttner, 2014; **Figure 1**). In the ISVZ, 60–80% of progenitors co-express Pax6 and Tbr2, 5–30% express only Tbr2 and less than 15% express only Pax6. In the OSVZ, Pax6, and Tbr2 are co-expressed by 25–50% of progenitors while 20–35% express only Pax6 and 10–20% only Tbr2 (**Figure 1**; Betizeau et al., 2013; Florio and Huttner, 2014).

In mice, cortical GABAergic interneurons originate from distant germinal domains in the ganglionic eminences and follow tangential migratory routes to reach the developing cortex (Gelman and Marin, 2010). In primates, the origin of cortical interneurons is controversial. It has been proposed that while many interneurons have a ventral telencephalic origin, a significant fraction are produced in the progenitor layers of the cortex itself during the second half of corticogenesis (Letinic et al., 2002; Zecevic et al., 2005; Petanjek et al., 2008; Jakovcevski et al., 2011). However, a recent study by Hansen et al. (2013) found no evidence of interneuron production in the cortical wall. Instead, their analysis suggests that, as in rodents, the vast majority of human cortical interneurons are produced in the ganglionic eminences (Hansen et al., 2013).

In all mammalian species, the positions adopted by neurons migrating from the cortical progenitor zones to the overlying developing cortical plate (CP) are related to their birthdate. Each successive generation of newly born projection neurons bypasses earlier-born neurons and settles close to the pial edge of the CP. Thus, cortical layers (with the exception of layer 1) are formed in a deep-first superficial-last sequence (Angevine and Sidman, 1961; Berry and Rogers, 1965; Rakic, 1974; McConnell, 1995; Tan and Shi, 2013). When projection neurons arrive in their final laminar positions, they undergo terminal differentiation including elaboration of their dendrites and axons to establish connections and eventually form the cortical circuitry. Projection neurons in each layer tend to exhibit similar gene expression patterns, morphologies and organization of afferent and efferent connections (Stiles and Jernigan, 2010).

# Human Brain Disorders Associated with *PAX6* Mutations

In humans, heterozygous loss-of-function mutations of *PAX6* cause sight-threatening congenital eye defects, typically including severe hypoplasia of the iris (aniridia) and retina. These mutations are also associated with a range of neurological and psychiatric conditions including nystagmus, impaired auditory processing and verbal working memory, autism, and mental retardation (Malandrini et al., 2001; Bamiou et al., 2004, 2007a,b; Davis et al., 2008; Hingorani et al., 2009; Maekawa et al.,

2009). These conditions are linked to structural brain defects including reduced size of the corpus callosum and anterior commissure, abnormalities of the cerebral cortex and cerebellum and absence of the pineal gland (Sisodiya et al., 2001; Free et al., 2003; Mitchell et al., 2003; Bamiou et al., 2004, 2007b; Ellison-Wright et al., 2004).

Only four cases of children with mutations in both *PAX6* alleles (compound heterozygotes) have been reported (Glaser et al., 1994; Schmidt-Sidor et al., 2009; Solomon et al., 2009). Two of them survived past birth, one only for about 1 week (Glaser et al., 1994), the other until at least 4 years (Solomon et al., 2009). The former had anophthalmia, the latter microphthalmia, and both had numerous defects in the CNS including agenesis of the corpus callosum and microcephaly. Glaser et al. (1994) described a disturbed stratification of the cerebral cortex with heterotopic islands of germinal and ependymal cells, as well as focal polymicrogyria of the cerebral cortex. Two other cases were sibling fetuses with the pregnancies terminated at 21 and 23 weeks (Schmidt-Sidor et al., 2009). In both cases the brain was very small, filling only 1/3 of the cranial cavity, and displayed completely disorganized structures of the brain hemispheres

and cerebellum. Microscopic analysis showed that the cerebral hemispheres contained an enormous amount of germinal matrix both in the inner parts and at the surface of the hemispheres. The cortex was very narrow with a paucity of cells with an irregular distribution of large neurons. The structure of the cortex was disturbed with a thick layer of germinal cells on its surface, a poorly defined marginal zone and no normal stratification. In the entire cortex the cells were mainly in clusters. The development of the white matter was also severely disturbed. The intermediate zone was absent. In several places of the cortex and also between clusters of neuroblasts inside the brain hemispheres there were fascicles of axons which did not form normal tracts. The heterotopia of germinal cells observed in the human compound heterozygotes (**Figure 4A**) are reminiscent of the clusters of germinal cells found in the intermediate zone of the cortex of *Pax6*−*/*<sup>−</sup> mouse embryos (**Figure 4B**; Caric et al., 1997). In these Pax6 mutant embryos it is thought that the clusters form as a consequence of a cell non-autonomous migration defect (Caric et al., 1997).

# Structure and Transcriptional Regulation of the *Pax6* Locus

In both human and mouse, the *Pax6* gene has 16 exons distributed over a 30 kb genomic region including alternatively spliced exons *alpha* and 5a (Glaser et al., 1992; Williams et al., 1998; Kammandel et al., 1999; Plaza et al., 1999a; Xu et al., 1999b). Four transcription start sites have been identified in mouse *Pax6*, associated with the P0, P1, P*alpha*, and P4 promoters respectively (**Figure 5**; Kammandel et al., 1999; Xu et al., 1999b; Kleinjan et al., 2004; Morgan, 2004). Transcriptional regulation of the *Pax6* locus is particularly complex. A number of shortrange regulatory elements have been identified in the vicinity of the *Pax6* coding region which control tissue-specific *Pax6* expression (**Figure 5**; Kammandel et al., 1999; Plaza et al., 1999b;

FIGURE 4 | Histology of *Pax6***−***/***<sup>−</sup>** developing cortex in human and mouse. (A) Coronal section through the cortex of a fetus with a compound heterozygosity for two *PAX6* mutations showing a layer of germinal cells on the surface of the cortex (arrow) and heterotopia of germinal cells within the cortex (arrowhead). Photograph taken from Schmidt-Sidor et al. (2009). (B) Coronal section through the cortex of a *Pax6*−*/*<sup>−</sup> mutant mouse embryo showing clusters of germinal cells in the intermediate zone (arrowhead). Photograph from Caric et al. (1997).

Kleinjan et al., 2001, 2004; Griffin et al., 2002). Some of these elements exhibit overlapping tissue specificity, particularly in the eye, telencephalon and diencephalon, suggesting that they exert functions through cooperative interactions.

Although these short-range regulatory elements account for much of *Pax6*'s normal expression pattern in mice, genetic analyses in both humans and mice revealed that they are insufficient to drive the full pattern of *Pax6* expression, particularly in the eye and forebrain (Kim and Lauderdale, 2006; Kleinjan et al., 2006; McBride et al., 2011). The first evidence that PAX6 expression is influenced by distant regulatory elements located far downstream of *PAX6*'s coding exons came from studies of aniridia patients, a subset of which harbor chromosomal rearrangements whose breakpoints are located far downstream of the *PAX6* transcriptional unit (Fukushima et al., 1993; Fantes et al., 1995; Lauderdale et al., 2000; Kleinjan et al., 2001; Crolla and Van Heyningen, 2002). The most distant of these breakpoints, designated "SIMO," is located 124 kb downstream of the *PAX6* polyadenylation site (Fantes et al., 1995; Kleinjan et al., 2001; **Figure 5**). The functional importance of regulatory regions around distant breakpoints was confirmed by the finding that a yeast artificial chromosome (YAC) comprising 420 kb of the human *PAX6* coding sequence and flanking regions stretching beyond the SIMO breakpoint rescues the mouse *small eye* (*Sey*) phenotype, whereas a YAC transgene with 110 kb less DNA sequence at the 3 end fails to rescue (Kleinjan et al., 2001). Subsequent analyses revealed the presence of essential 3 distant regulatory elements within a 75 kb region termed the *PAX6* downstream regulatory region (DRR) located 3 of the SIMO breakpoint (**Figure 5**; Schedl et al., 1996; Kleinjan et al., 2001; McBride et al., 2011). A number of other conserved regulatory elements, including E60, E100, and RB (**Figure 5**), located either upstream or downstream of the DRR region are thought to be crucial for the induction of *PAX6* expression in the eye and forebrain (Griffin et al., 2002; Kleinjan et al., 2002, 2006; Kim and Lauderdale, 2006; McBride et al., 2011).

We recently reported a detailed examination of a series of novel transgenic reporter mice lines, carrying versions of either the full length 420 kb YAC described above or truncated versions that lacked putative regulatory elements. The YAC transgenes were modified by the introduction of a tau-GFP reporter cassette into the first coding exon of *PAX6*, such that GFP expression marks cells in which the transgenes are expressed. We found that regulatory elements that lie outside the 420 kb YAC transgene are required for correct *PAX6* expression in sub-regions of the telencephalon and diencephalon, indicating that recapitulation of the full *PAX6* expression pattern requires even more elements than had previously been thought (Mi et al., 2013b).

The *Pax6* locus is subject to autoregulation – *Pax6* has been shown to regulate its own expression. Putative Pax6 binding sites have been identified in *Pax6* regulatory elements in several species. For example, in *Drosophila*, two intronic enhancer elements containing putative Pax6 binding sites mediate autoactivation of *eyeless* (a *Drosophila Pax6* homolog) in the nervous system and eye (Hauck et al., 1999). These elements may serve the same function in vertebrates as they exhibit significant sequence identity between *Drosophila* and vertebrate species (Chow et al., 1999). In addition, Pax6 directly interacts with putative Pax6-responsive elements within the head surface ectoderm-specific enhancer of mouse *Pax6*, positively regulating its own transcription (Aota et al., 2003). Similarly, Pax6 autoregulation in the forebrain was revealed through the identification of a highly conserved regulatory element (CE2) in intron 7 of mouse *Pax6* (**Figure 5**). Overexpressing either Pax6 or its alternative spliced isoform Pax6(5a) positively autoregulated expression of the endogenous *Pax6* locus in Neuro2D and NIH3T3 cell lines (Pinson et al., 2006). The *Pax6* locus is also subject to negative autoregulation, whereby Pax6 represses transcription of the *Pax6* locus, as shown in the developing telencephalon (Manuel et al., 2007). In summary, these studies highlight the fact that appropriate levels of Pax6 expression are regulated by both positive and negative autoregulation during development.

# Pax6 Regulates Cell Cycle Length and Cell Cycle Exit

Generating the correct number of cortical neurons of the correct types involves changes in both the mode of division of cortical progenitors and length of their cell cycle. As discussed above, cortical progenitors switch progressively from proliferative symmetrical divisions to differentiative asymmetrical divisions during corticogenesis. At the same time, the length of the cell cycle in AP cells in mouse cortex increases from 8 h at the onset of neurogenesis (E10) to 18 h at the end (E18), largely due to lengthening of the G1 phase (Takahashi et al., 1993, 1995; Estivill-Torrus et al., 2002). Subsequent studies showed that G1 lengthening acts to promote the genesis of BP cells which have much longer cell cycle length (around 26 h in E14.5 mouse cortex) due to increase in the G1 length (Arai et al., 2011). In the macaque, AP cell cycle length is much longer than in rodents and it follows a different developmental course (Kornack and Rakic, 1998). It lengthens linearly from 23 h at E40 to 54 h at E60, then shortens linearly to 27 h by E80 (Kornack and Rakic, 1998). As the neurogenic period is 10 times longer than in rodents, macaque cortical progenitors

undergo many more rounds of division even though the duration of the cell cycle is longer (Kornack and Rakic, 1998), generating a greater number of cortical cells and a larger cortex.

There is mounting evidence that TFs exert region-specific control of cortical progenitor cell cycle progression. Pax6 is one of a number of TFs that are expressed in distinct gradients across cortical areas (Zaki et al., 2003; Sansom and Livesey, 2009; Salomoni and Calegari, 2010; Georgala et al., 2011a) and several studies have implicated Pax6 in the temporal and spatial control of cell cycle duration in cortical progenitors. Loss of Pax6 during early cortical development (E12.5) in mice *in vivo* led to a shortening of the cell cycle of progenitors coupled with higher proportions of asymmetrical divisions, resulting in a temporary increase in the production of post-mitotic neurons (Warren et al., 1999; Estivill-Torrus et al., 2002; Walcher et al., 2013). Experiments with cultured *Pax6*−*/*<sup>−</sup> mutant cortical cells showed that they too exhibit accelerated proliferation, indicating that Pax6's effects on the cell cycle are cell autonomous (Estivill-Torrus et al., 2002). In contrast, at the mid-corticogenesis stage (E15.5) *Pax6*−*/*<sup>−</sup> progenitor cells proliferated more slowly than wild type controls, showing that the effects of Pax6 on the cell cycle of cortical progenitors are context-dependent.

Gain of function studies also support the idea that Pax6 controls cell cycle progression. For example, forced expression of Pax6 impairs progenitor cell proliferation *in vitro* (Heins et al., 2002; Hack et al., 2004; Cartier et al., 2006). In *PAX77* mice, which express PAX6 protein in its normal pattern but at approximately double the wild type level, there is a reduction in the number of proliferating progenitors in the rostral and central cortex, the areas where levels of Pax6 expression are normally highest (Manuel et al., 2007). Further, APs located in rostral cortical areas proliferate more slowly in these mice (Georgala et al., 2011a,b). Thus, both gain and loss of Pax6 function studies have revealed powerful context-specific effects of Pax6 on the cell cycle of progenitor cells in a variety of cortical areas and stages.

Early in corticogenesis, expression levels of Pax6 vary between different cortical regions but its expression levels become increasingly uniform with embryonic age (Mansouri et al., 1994; Stoykova and Gruss, 1994; Manuel et al., 2007). Some of the evidence cited above indicates that the actions of Pax6 on the cell cycle of cortical progenitors are associated with its expression levels. This assumption is strongly supported by our recent study in which cell cycle parameters were systematically examined in different cortical regions at different developmental stages using mouse models with either constitutive or conditional loss of Pax6 function (Mi et al., 2013a). During early corticogenesis (E12.5), when the Pax6 gradient is steepest, areas of highest expression correlate with regions where the cell cycle duration is longest (**Figure 6**). Loss of Pax6 causes shortening of the cell cycle only in these areas (**Figure 6**). In normal embryos at older ages, the Pax6 gradient becomes progressively more uniform across the cortex, as does cell cycle length, and loss of Pax6 causes shortening of the cell cycle in all areas (**Figure 6**; Mi et al., 2013a). Taken together, these studies indicate that Pax6 primarily exerts a repressive action on the cell cycle progression of cortical progenitors, and distinct expression levels of Pax6 confer

embryos. At E12.5 Pax6 is normally expressed in a high rostral to low caudal

and caudally while the S-phase length remains unaffected.

its region and age -specific role in regulating progenitor cell proliferation.

As well as regulating the length of the cell cycle in cortical progenitors, Pax6 is involved in the control of cell cycle exit. In the *Pax6sey/sey* mutant cortex, the number of progenitor cells that exit the cell cycle during early corticogenesis (E12.5) is increased, leading to a smaller progenitor pool and an increased proportion of newborn neurons (Estivill-Torrus et al., 2002; Quinn et al., 2007). A similar effect has been seen in the eye and the spinal cord, where loss of Pax6 caused precocious neuronal differentiation due to premature cell cycle exit (Philips et al., 2005; Bel-Vialar et al., 2007). Interestingly, a study using *PAX77* mice revealed that increasing Pax6 levels also causes an increase in the proportion of progenitor cells exiting the cell cycle during late corticogenesis (E15.5; Georgala et al., 2011b), indicating that there is no simple dosage-dependent relationship between Pax6 level and progenitor cell cycle exit rate. Recently, we found that loss of Pax6 led to fewer progenitors exiting the cell cycle at the PSPB (pallial–subpallial boundary) at E12.5, the area of the telencephalon where Pax6 expression is highest, contrasting with the previous finding that *Pax6*−*/*<sup>−</sup> cortical progenitors show increased cell cycle exit at E12.5 (Estivill-Torrus et al., 2002). It appears that, although a general function of Pax6 is to influence cell cycle exit, the exact nature of its effect depends on the molecular and cellular context within which it acts. This context may change corresponding to diverse Pax6 levels in different regions at different developmental stages. In support of this assumption, a number of independent analyses demonstrated that Pax6 exerts different control functions on progenitor proliferation in different cellular contexts: Pax6 enhances proliferation in the neural retina of vertebrates (Marquardt et al., 2001) and *Drosophila* (Dominguez et al., 2004), whereas it inhibits proliferation of human glioblastoma cells (Zhou et al., 2005), cultured corneal epithelial cells (Ouyang et al., 2006) and cortical progenitors in primary cell cultures (Heins et al., 2002; Haubst et al., 2004) as well as *in vivo* (Estivill-Torrus et al., 2002; Berger et al., 2007). Therefore, it is difficult to conclude a simple universally consistent role of Pax6 in the regulation of cell cycle exit.

# Pax6 Direct Targets Regulating Progenitor Cell Proliferation

As described above, Pax6 is important for the regulation of cortical progenitor proliferation (Estivill-Torrus et al., 2002; Manuel et al., 2007; Quinn et al., 2007; Georgala et al., 2011b; Mi et al., 2013a,b). Cell cycle progression is driven by the concerted action of cyclin-dependent kinases (Cdks) and their activating partners, cyclins (Malumbres and Barbacid, 2005). Low levels of Cdk activity are sufficient for cells to transit from G1 into S phase, whereas high-levels of Cdk activity are required for the cell to undergo mitosis (Dehay and Kennedy, 2007; Kaldis and Richardson, 2012). During G1, cells integrate and respond to extracellular cues that either allow the cell to continue cycling or promote its withdrawal from the cell cycle and begin differentiation (Zetterberg et al., 1995). Progression through G1 is mainly driven by the activation of Cdk4/6 and their activating partners, D-type cyclins (Cyclin D1 and D2) as well as Cdk2 and its partner Cyclin E (Malumbres and Barbacid, 2001, 2005). Cdk activity is negatively regulated by members of the cyclin-dependent kinase inhibitors family (CKIs; Polyak et al., 1994; Sherr and Roberts, 1995; Vidal and Koff, 2000). The primary substrate of Cdks during G1 phase is the retinoblastoma protein (pRb). pRb acts as a repressor of E2F TFs through direct interaction. In quiescent cells, the activities of E2Fs are repressed through direct interaction with hypophosphorylated pRb. In cycling cells, mitogenic stimuli that elevate Cyclin levels promote pRb phosphorylation by Cyclin/Cdk complexes, which in turn prevents pRb from binding to E2Fs. Consequently, pRb/E2F complexes are dissociated and free E2Fs are now active and ready to drive the transcription of genes encoding proteins that are involved in G1 phase progression and DNA replication. Gene expression profiling studies showed that Pax6 is capable of regulating the expression of a number of cell cycle genes involved in G1 to S phase transition, such as *Cdk4, Cdk6, Ccnd1, Ccnd2, p27kip1, Cdc6, Mcm3, Mcm5, Mcm6, Cdca2* and *Cdca7*, with many of them being repressed by Pax6 (Sansom et al., 2009; Mi et al., 2013a,b). These findings are in line with the notion that Pax6 mainly acts as an important regulator of progenitor cell proliferation during early cortical development as reviewed above. Among the cell cycle genes whose expression is regulated by Pax6, Pax6 direct targets identified to date include *Cdk4*, *Cdk6*, *Mcm3*, *Cdca2*, *Cdca7,* and p27*kip1* (**Table 1**; Duparc et al., 2007; Sansom et al., 2009; Mi et al., 2013a,b).

It is interesting to note that in addition to its repressive effect, Pax6 can also activate expression of some cell cycle genes (Holm et al., 2007; Osumi et al., 2008; Sansom et al., 2009; Mi et al., 2013a,b; Xie et al., 2013). For example, Pax6 directly promotes the expression of *Cdk4* in neural stem cells in early cortical development (Sansom et al., 2009). This supports the concept that Pax6 has dual roles to both promote and inhibit cell proliferation in the developing cortex, possibly through the divergent functions of its DNA-binding subdomains (see below; Walcher et al., 2013). Using transcriptome analysis of the cortex in embryos with either gain and loss of Pax6 function, Sansom et al. (2009) showed that Pax6 regulates a core transcriptional network that controls cortical neural stem cell proliferation in a dosage-dependent manner. They found that under normal conditions *in vivo*, Pax6 has the potential to both promote and inhibit stem cell proliferation by exhibiting opposing effects on the expression of different sets of cell cycle genes, but when Pax6 levels are increased, as in transgenic cortex overexpressing Pax6, the neurogenic functions of Pax6 are dominant over its ability to promote proliferation. This study is in line with our finding that at the onset of corticogenesis, the repressive effect of Pax6 on progenitor cell proliferation and the Cdk/Cyclin mediated phosphorylation of pRb is localized to regions of cortex where Pax6 expression is normally highest, indicating a relationship between the level of Pax6 expression and its antiproliferative effect (Mi et al., 2013a). Given that Pax6 is expressed in a gradient during early cortical development, the dosagedependent effects of Pax6 on cell proliferation are important in the context of understanding how the cerebral cortex becomes divided into regions with specific cytoarchitectures and functions.

# Pax6 is an Intrinsic Determinant of Neuronal Fate in the Developing Cortex

The observations that Pax6 is specifically expressed by radial glial progenitor cells in the developing cortex and is required for them to acquire their normal morphology, gene expression pattern and cell cycle characteristics (Gotz et al., 1998), suggested that Pax6 may control their ability to generate neurons. To address this question, Heins et al. (2002) used a combination of loss and gain of function approaches. They found that cultured RGCs isolated from the developing cortex of *Pax6*−*/*<sup>−</sup> mutant mouse embryos generated half as many pure neuronal clones as RGCs from control embryos. Using a tau-GFP transgene to label and quantify neurons by fluorescence activated cell sorting (FACS), they showed that the number of neurons in the cortex of *Pax6*−*/*<sup>−</sup> mutant mouse embryos was reduced by half at E14.5 and by a third at E16.5 compared to controls, consistent with the previously reported thinner CP in *Pax6*−*/*<sup>−</sup> cortex (Schmahl et al., 1993). Forced expression of Pax6 in cells dissociated from E14 cortex or in astrocytes from postnatal cortex, via transduction of a retroviral vector containing full length Pax6 cDNA, resulted in a significant increase in the number of pure neuronal clones generated, demonstrating that Pax6 is an intrinsic determinant of neuronal fate in murine cortical progenitors (Heins et al., 2002). Mo and Zecevic (2008) showed that siRNAmediated knockdown of Pax6 expression in cultured human fetal RGCs led to a significant decrease in the number of neurons produced, suggesting that the role of Pax6 in cortical neurogenesis is maintained from rodents to humans (Mo and Zecevic, 2008).

TABLE 1 | Genes that have been identified as direct Pax6 targets in the mouse developing cortex grouped according to the developmental process that they affect.


# Pax6 Confers both Regional and Neurotransmitter Fates

The mammalian CP comprises mainly glutamatergic neurons born in the cortical VZ and GABAergic interneurons (INs). In mouse, most INs are born in the medial ganglionic eminence (MGE; ∼70%) with others coming from the caudal ganglionic eminence (CGE) and areas including the preoptic area (POA; Anderson et al., 2001; Wonders and Anderson, 2006; Fogarty et al., 2007; Gelman et al., 2009; Miyoshi et al., 2010). INs born in the LGE populate the olfactory bulb (OB) and the ventral telencephalon. Most cortical glutamatergic neurons and GABAergic interneurons are generated between E12.5 and E16.5. By E18.5, GABAergic INs are distributed throughout the cortex of wild type embryos. In contrast, the cortex of *Pax6*−*/*<sup>−</sup> mutant embryos contains large subpial ectopias composed of GABAergic INs (Kroll and O'Leary, 2005). These ectopias develop from E15.5 onward and express markers of LGE-derived INs, including Sp8, rather than markers of MGE-born INs, such as Lhx6. Their development parallels a progressive ventralization of the dorsal telencephalic VZ in *Pax6*−*/*<sup>−</sup> mutant embryos. At the start of neurogenesis (E12.5–E13.5) the most ventrolateral part of the *Pax6*−*/*<sup>−</sup> mutant cortex ectopically expresses the ventral telencephalic markers Ascl1 and Gsh2. As neurogenesis progresses ventral telencephalic identities extend dorsally such that by E14.5 most of the dorsal telencephalic VZ exhibits ventral characteristics. This progressive ventralization is accompanied by a dorsal shift in expression of markers of GABAergic INs such as Dlx1/2 and GAD67. Importantly, the INs that form the ectopias are derived from Emx1-expressing cortical progenitors which normally generate glutamatergic neurons, demonstrating that in *Pax6*−*/*<sup>−</sup> mutant mice, cortical progenitors are re-specified to generate GABAergic INs instead of glutamatergic neurons (Kroll and O'Leary, 2005).

Dlx1/2, Ascl1, and Gsh2 have been shown to promote GABAergic fate, such as the expression of Gad67 (Long et al., 2009a,b; Wang et al., 2013). Thus it is possible that Pax6 normally represses GABAergic identities in the cortex by blocking the expression of these TFs.

In *Pax6*−*/*<sup>−</sup> embryos, early born CP neurons are correctly specified as dorsal telencephalic neurons, expressing dorsal markers Math2 and Tbr1 (Schuurmans et al., 2004). They fail, however, to adopt a correct glutamatergic phenotype. Interestingly, these early born neurons are not GABAergic either (Schuurmans et al., 2004). Thus, during early corticogenesis, Pax6 participates in the specification of a correct glutamatergic phenotype but is not required to suppress GABAergic identities. Schuurmans et al. (2004) analyzed double mutants for Ngn1 and Ngn2, and showed that, during early stages of corticogenesis (E13.5), neurogenins are required to specify a glutamatergic, cortical phenotype, and repress GABAergic identities. Conversely, cortical neurons born at later stages of corticogenesis (E15.5) acquire a normal cortical and glutamatergic phenotype in *Ngn1*−*/*−*; Ngn2*−*/*<sup>−</sup> double mutants whereas in *Pax6*−*/*<sup>−</sup> embryos and in mutants lacking the orphan nuclear receptor Tlx, they express reduced levels of cortical and glutamatergic markers, compared to controls, and instead acquire a subcortical, GABAergic phenotype (Schuurmans et al., 2004). This study shows that distinct genetic programs operate during early and late corticogenesis to specify cortical regional identity and neurotransmitter fate: neurogenins are required during early stages while Pax6 and Tlx are required during later stages of corticogenesis (Schuurmans et al., 2004).

# Pax6 Confers Laminar Fate

Throughout corticogenesis, newborn neurons migrate to the CP in an inside-out sequence such that deep layers are formed first and superficial layers form later. The final laminar position and subtype of cortical projection neurons is dependent on the time at which progenitors exit the cell cycle. In *Pax6*−*/*<sup>−</sup> mutant embryos, the SVZ and superficial cortical layers are defective, as shown by absence of expression of Svet1 (which marks the SVZ and superficial layers), while the VZ and deep CP layers are unaffected, as shown by the expression of Otx1 (Tarabykin et al., 2001).

Examination of cortical lamination in *Ngn1*−*/*−; *Ngn2*−*/*<sup>−</sup> mutants showed that the expression of markers of deep cortical layers is reduced (Tbr1, Er81) while markers of superficial layers are expressed normally, indicating that late-born cortical neurons are correctly specified (Schuurmans et al., 2004). In contrast, no defects are detected in the deeper cortical layers of *Pax6*−*/*−, *Tlx*−*/*−, or *Pax6*−*/*−; *Tlx*−*/*<sup>−</sup> double mutants. Markers of upper layers, however, are downregulated in Pax6 and Tlx single mutants and completely lost in *Pax6*−*/*−;*Tlx*−*/*<sup>−</sup> embryos suggesting that Pax6 and Tlx cooperate to specify the identity of late-born cortical neurons (Schuurmans et al., 2004).

Study of cortical lamination in *Pax6*−*/*<sup>−</sup> mutants is limited by the fact that they die perinatally while cortical neurons do not reach their final location until around postnatal day 8 (P8). Therefore, more recent studies on the role of Pax6 in cortical lamination have used conditional mutants in which *Pax6* is inactivated specifically in the cortex and at later stages of cortical development and which survive postnatally. When *Pax6* was deleted specifically in the cortex from the onset of corticogenesis, markers of superficial layers were either downregulated or lost completely (Tuoc et al., 2009). In contrast, when *Pax6* was deleted in RGCs after the generation of deep cortical layers (between E14.5 and E16.5) using an hGFAPcre transgenic line, the specification and number of late born neurons was unaffected (Tuoc et al., 2009), leading to the conclusion that Pax6 is not required for the specification of late born cortical neurons and that the defects of superficial layers observed in *Pax6*−*/*<sup>−</sup> and *Pax6*cKO mutants are a result of earlier defects in RGC proliferation. In a more recent study, Georgala et al. (2011b) used an *Emx1-creErT2* transgenic line to remove Pax6 specifically in the cortex between E12.5 and E13.5, before superficial layer neurons are born. The resulting cKO mutant brains showed a dramatic reduction in the thickness of the superficial cortical layers, compared to control embryos. Late-born neurons were birth-dated by injecting BrdU at E15.5 and stained for expression of Cux1, a marker of upper layer (II to IV) pyramidal neurons. There was a substantial reduction in the number of E15.5 born neurons in the superficial cortical layers of the *Pax6-cKO* mutants and many fewer late-born neurons expressed Cux1 (Georgala et al., 2011b), indicating that Pax6 plays a crucial role in generation and specification of late born cortical neurons. Note that Pax6 protein was lost slightly earlier in the conditional mutants generated by Georgala et al. (2011b) than in those used by Tuoc et al. (2009), potentially explaining the difference in outcomes between the two studies.

# Direct Transcriptional Targets of Pax6 Regulate Basal Progenitor Genesis, Neurogenesis, and Cell Fate Specification

Pax6 exerts many of its effects, at least in part, by directly regulating the expression of specific target genes (**Table 1**). As described above, work in the mouse has shown that Pax6 expressing APs that give rise to BPs transiently express the proneural TF Neurogenin 2 (Ngn2). BPs do not express Pax6 but do express the TF Tbr2. Loss of Pax6 function results in loss of Ngn2 expression in the cortical VZ and downregulation of Tbr2 in the SVZ (Stoykova et al., 2000; Toresson et al., 2000; Yun et al., 2001; Scardigli et al., 2003; Holm et al., 2007; Quinn et al., 2007; Sansom and Livesey, 2009), suggesting that the development of BPs is impaired. Conversely, Pax6 overexpression results in a marked increase in both Ngn2 and Tbr2 expression (Scardigli et al., 2003; Sansom and Livesey, 2009), each of which have been shown to be directly regulated by Pax6 (Scardigli et al., 2003; Sansom and Livesey, 2009; Tuoc et al., 2013a). Scardigli et al. (2003) showed that Pax6 directly binds and regulates the activity of the E1 enhancer of *Ngn2*, which regulates *Ngn2* expression in the cortex and the ventral spinal cord.

Another direct target that is positively regulated by Pax6 is the anti-neurogenic transcriptional co-repressor Tle1, a member of the Groucho/transducin-like enhancer of split (Gro/Tle) family (Sansom et al., 2009; Tuoc et al., 2013a). Gro/Tle proteins are expressed in proliferating neural progenitor cells where they promote maintenance of the undifferentiated state by inhibiting/delaying neuronal differentiation (Buscarlet and Stifani, 2007; Jennings and Ish-Horowicz, 2008).

Pax6 also directly inhibits the expression of Olig2 (Jang and Goldman, 2011), a TF critical for glial cell fate determination (Marshall et al., 2005). Olig2 is expressed in a subset of cells in the rat neonatal SVZ, which differentiate into glia. Forced expression of Pax6 in these Olig2+ cells, using a retrovirus, caused downregulation of Olig2 and a shift toward neuronal fate (Jang and Goldman, 2011). These authors showed that Pax6 can directly bind to Olig2's promoter and repress its activity.

Pax6 directly regulates markers of specific neuronal subtypes including Er81 and Cux1 (Tuoc and Stoykova, 2008a; Sansom et al., 2009; Tuoc et al., 2013a). Er81 is a TF expressed in cortical progenitors, in a rostrolateral-high to caudomedial-low gradient, and in a subset of pyramidal cells scattered through layer 5 (Yoneshima et al., 2006; Tuoc and Stoykova, 2008a). In the absence of Pax6, cortical progenitors do not express Er81 and the rostrolateral cortex lacks Er81 positive layer 5 neurons. Cux1 is a marker of upper layer (II to IV) pyramidal neurons and functions as a negative regulator of dendritic complexity for these neurons (Cubelos et al., 2010, 2014; Li et al., 2010). As mentioned above, loss of Pax6 results in a dramatic reduction of Cux1+ upper layer neurons, suggesting that the specification of late born neurons is impaired (Georgala et al., 2011b).

How does a single gene exert such a wide range of biological effects? Part of the answer to this question is suggested by the findings that the Pax6 protein contains two distinct DNAbinding domains, one of which is regulated by alternate splicing and that Pax6 is able to interact with a number of co-factors, which influence its activity. This issue is discussed in the next section.

# Complexity of Pax6 DNA-Binding Properties and its Distinct Functions

Pax6 binds to target DNA sequences through one or both of two DNA-binding domains, the paired domain (PD) and the homeodomain (HD; Bertuccioli et al., 1996; Jun and Desplan, 1996; Sheng et al., 1997; Singh et al., 2000). Transcriptional regulation of Pax6 target genes is mediated by a carboxy terminal proline/serine/threonine (PST) rich transactivation domain (Singh et al., 1998, 2001; Tang et al., 1998; **Figure 7**).

The PD is an evolutionarily conserved 128 amino acid DNA-binding domain shared by all members of the paired box (Pax) family of TFs (Pax1-Pax9; Chalepakis et al., 1992; Gruss and Walther, 1992). The Pax6 PD can be structurally and functionally separated into two independent DNA-binding globular helix-turn-helix subdomains, PAI and RED (Puschel et al., 1992; Czerny et al., 1993; Jun and Desplan, 1996; **Figure 7A**). The well-established DNA-binding motif for canonical Pax6 PD, known as P6CON, is a bipartite 20 base pair DNA sequence whose 5 -half is recognized by the PAI subdomain and 3 -half by the RED subdomain (Czerny et al., 1993; Epstein et al., 1994; **Figure 7A**). The crystal structure of the Pax6 PD – P6CON complex revealed direct interactions between both PAI and RED subdomains, the linker regions of PAI and RED subdomains and the N-terminal β-turn subdomain with individual bases and phosphate residues of cognate DNA (Czerny et al., 1993; Epstein et al., 1994; Xu et al., 1999a). Pax6 also contains an internal, paired-type HD (Czerny et al., 1999; Eberhard and Busslinger, 1999) which binds DNA in the form of a homodimer recognizing a symmetric binding site (Wilson et al., 1993; Eberhard and Busslinger, 1999; Mikkola et al., 2001; **Figure 7B**).

Although different Pax6 DNA-binding domains can independently bind to their "optimal" binding sites, using individual consensus binding motif to predict and confirm candidate Pax6 binding sites has proven to be very difficult (Shimoda et al., 2002; Visel et al., 2007; Wolf et al., 2009). Unlike sequence specific DNA-binding proteins which harbor only one DNAbinding domain, Pax6 exhibits increased DNA-binding complexity because of its multiple DNA-binding subdomains (PAI, RED, and HD) which can be used in varying combinations when binding to DNA (Cvekl et al., 2004; Grapp et al., 2009). A recent study further reinforced the complexity of Pax6 DNA-binding properties by systematically examining coordinated actions of different Pax6 binding subdomains and identified a number of novel variants of Pax6 DNA-binding motifs which can be bound by different combinations of Pax6 binding subdomains (Xie and Cvekl, 2011). As Pax6 can function either as a transcriptional activator or repressor (Wolf et al., 2009), it is possible that some of these binding motif variants may induce structural changes in Pax6 proteins that enable them to switch their function between transcriptional activation and repression.

The complexity of the DNA binding properties of different Pax6 subdomains implies that the subregions exert distinct functions during embryonic cortical development. Pax6 mutant mouse lines with disruption of either the PD or the HD have been used to investigate the impact of different Pax6 binding domains on telencephalic development. Haubst et al. (2004) showed that the PD alone is necessary and sufficient for the regulation of most aspects of telencephalic development including neurogenesis, progenitor cell proliferation, and patterning in the developing cortex. In contrast, HD-specific mutations affected only subtle aspects of telencephalic development, such as establishment of the PSPB, and had no obvious effect on neurogenesis or proliferation (Haubst et al., 2004). This suggests that the majority of Pax6's effects during telencephalic development are mediated by the PD with potentially different contributions by its subdomains. A recent study further dissected the roles of the different

subdomains of the Pax6 PD during telencephalic development by examining mice with point mutations in its individual subdomains PAI and RED (Walcher et al., 2013). This showed that both PAI and RED act largely in a redundant manner in patterning of the telencephalon, with only the PAI mutation resulting in the ectopic expression of subpallial genes such as *Gsx2* and *Olig2* into the pallium, suggesting that RED subdomain on its own is not sufficient to regulate telencephalic patterning. In addition, neurogenesis was affected only by the PAI subdomain mutation, phenocopying the neurogenic defects observed in full *Pax6*−*/*<sup>−</sup> mutants. Strikingly, this study also revealed that subdomains of Pax6 PD have distinct roles in regulating cortical progenitor cell proliferation, with mutations affecting the PAI and RED subdomains respectively reducing and increasing the number of mitoses.

# Alternatively Spliced Isoforms of Pax6 and their Functions

In addition to Pax6 and Pax6(5a) a number of other splice variants of Pax6 have been reported, named p43, p33 and p32, each of which is less abundant and mostly localized in the cytoplasm (Carriere et al., 1993; Jaworski et al., 1997; Pinson et al., 2006). The 14 amino acid insertion in Pax6(5a) destroys the DNA-binding capacity of the PAI subdomain, leaving the RED subdomain of the PD and the intact HD for DNA binding (Epstein et al., 1994; Kozmik et al., 1997). The PD in Pax6(5a) binds to a distinct DNA sequence from that bound by the canonical Pax6 PD, that has been named 5aCON (Epstein et al., 1994). Direct comparison of the DNA binding sites recognized by Pax6 PD and PD(5a) showed partial sequence homology between the 3' halves of P6CON and 5aCON (Epstein et al., 1994; Duncan et al., 1998, 2000). This homology allows Pax6 (which contains the canonical PD) to bind to both P6CON and 5aCON sites, whereas Pax6(5a) can only interact with the 5aCON site (Epstein et al., 1994).

The majority of published molecular studies have focused on Pax6, as it contains the canonical PD and appears to be more abundant than Pax6(5a) (Carriere et al., 1993; Richardson et al., 1995; Koroma et al., 1997). During early development, *Pax6* and *Pax6(5a)* transcripts are expressed in a ratio of 8:1 in the mouse forebrain, falling to 3:1 between E12.5 and E14.5 (Kozmik et al., 1997; Pinson et al., 2006). A reduced ratio between Pax6 and Pax6(5a) was also seen during chick retina development (Azuma et al., 2005). *PAX6* and *PAX6(5a)* transcripts are equally abundant in the human adult lens epithelium and cornea, as well as in monkey retina (Zhang et al., 2001). These findings raise the possibility that expression of the two main Pax6 isoforms are regulated in a tissue- and temporal-specific manner, thereby affecting which target genes they recognize.

Although Pax6 and Pax6(5a) are commonly co-expressed, there is evidence that they exert distinct functions during embryonic development. For example, overexpression of either Pax6 or Pax6(5a) in chick retina showed that both isoforms led to increased retinal progenitor cell proliferation, but Pax6(5a) induced ectopic differentiation of the retina to a stronger degree than Pax6 (Azuma et al., 2005). Similarly, Pax6(5a) strongly induces neural differentiation of murine embryonic stem cells *in vitro*, while Pax6 does not have such a strong effect (Shimizu et al., 2009). In the embryonic mouse forebrain, the different roles of the two Pax6 isoforms in regulation of cell proliferation and differentiation were also established from studies of Pax6 mutants. It was shown that mutations that abolish the binding property of the PD affect both proliferation and differentiation, while a PD(5a) mutation only affects proliferation and has no effect on cell fate/differentiation in mouse developing telencephalon (Haubst et al., 2004). In a more recent study, Pax6(5a) specific target genes were identified through microarray analyses of gene expression profiles in cell lines stably expressing either Pax6 or Pax6(5a) (Kiselev et al., 2012). This revealed that a number of genes involved in cell proliferation, differentiation, and migration are differentially regulated by Pax6 and Pax6(5a), providing a potential molecular basis by which Pax6 and Pax6(5a) could exert their distinct biological functions.

# Co-Factors Influence Pax6's Direct Regulation of Target Genes Involved in Embryonic and Adult Neurogenesis

It is becoming clear that transcriptional regulation is highly dependent on the molecular and cellular context, one reason for which is that TFs do not act on their own but are dependent on interactions with other co-factors to exert their functions (Oikawa and Yamada, 2003; Westerman et al., 2003). These interactions introduce more specificity into the regulatory function of a given TF in a particular cellular context. A number cofactors that can interact with Pax6 have been identified. For example, Pax6 cooperates with Sox2 to target the lens-specific enhancer of the δ*-crystallin* gene (Kamachi et al., 2001); with MDIA to influence Pax6's functions in cerebellar granule cells (Tominaga et al., 2002); with Trim11 to mediate Pax6 degradation through the ubiquitin proteasome system (UPS) and modulate Pax6 transcriptional activity (Cooper and Hanson, 2005; Tuoc and Stoykova, 2008b). More recently, Brg1/Brm associated factors complex (BAF) and Meis2 have been identified as two new co-factors that affect Pax6's direct regulation of target genes expression, in turn regulating its role in both embryonic and adult neurogenesis (Ninkovic et al., 2013; Tuoc et al., 2013a,b; Agoston et al., 2014).

# BAF Complex

A number of chromatin remodeling factors are known to play an important role during mammalian neural development, including the BAF complex (Ho et al., 2009a,b; Wu, 2012; Yip et al., 2012; Tuoc et al., 2013a,b). Interestingly, the function of the BAF complex in regulating neural development is dependent on its subunit composition. In mammals, BAF complexes contain two ATPase subunits Brg1 or Brm (Brahma), which are mutually exclusive and essential for remodeling activity, in combination with at least 15 different BAF (Brg1/Brm-associated factor) subunits (Ho et al., 2009a; Singhal et al., 2010). Certain BAF subunits have restricted expression patterns and thus could define tissueor cell-type-specific BAF complexes. For example, the composition of BAF complexes in ES cells (esBAF) is defined by the incorporation of Brg1 but not Brm, BAF155 but not BAF170, and BAF60A but not BAF60C (**Figure 8**; Ho et al., 2009a,b). However, during the transition from ES cells to neural progenitors, the composition of BAF complexes is correspondingly changed, defining the neural progenitor specific BAF complex (npBAF; Lessard et al., 2007; Yan et al., 2008; Ho et al., 2009a). This process is accompanied by induction of BAF170 expression, replacing one of the BAF155 subunits and the recruitment of Brm as the ATP-ase subunit in the npBAF complex.

In two recent studies, Tuoc et al. (2013a,b), described an elegant model in which time-specific binding of BAF subunits including BAF170, BAF155, and Brm-ATPase to Pax6 modulates the expression of Pax6 target genes (*Tbr2, Cux1,* and *Tle1*) involved in the specification and generation of BPs and upper layer neuronal fate in the developing cortex. Interestingly, this model involves competition between BAF170 and BAF155 subunits in the BAF complex, which modulates chromatin modifications and binding of the Pax6/REST co-repressor complex to Pax6 target genes, thereby affecting their expression. During early neurogenesis (E12.5 to E14.5), the presence of BAF170 in the Pax6-BAF complex prevents the euchromatin state (a lightly packed form of chromatin) of the Pax6 target genes *Tbr2, Cux1* and *Tle1*, limiting the ability of Pax6 to bind them, thereby inhibiting their expression and promoting direct neurogenesis (RGCs directly give rise to neurons) at this developmental stage. In contrast, after E14.5 the expression of BAF170 decreases, accompanied by enhanced expression of BAF155, leading to a relaxed chromatin state of the promoters of the Pax6 target genes mentioned above, thereby enhancing their expression and in turn promoting indirect neurogenesis (RGCs indirectly give rise to neurons through generating BPs). This study provides novel insight into the interaction between BAF complex and Pax6 in the generation of BPs and the mode of neurogenesis in the developing cortex. In another recent study, Ninkovic et al. (2013) showed that Pax6 directly interacts with Brg1-ATPase of the BAF complex to modulate expression of genes involved in neuronal fate specification in adult neural progenitors. This study indicated that direct interaction of Pax6 with the Brg1/BAF complex enhances the expression of Pax6 target neurogenic TFs including *Sox11, Nfib,* and *Pou3f4* which form a cross-regulatory network driving neurogenic fate during adult neurogenesis. This study improved our understanding of Pax6-mediated adult neurogenesis and shed light on molecular mechanisms underlying the function of Pax6 in cell fate specification and conversion.

# Meis2

Meis2 belongs to the TALE (three amino acid loop extension) family of atypical HD-containing TFs. It can form heteromeric complexes with other transcriptional regulators to exert its functions during a range of biological processes (Moskow et al., 1995; Chang et al., 1997; Knoepfler et al., 1999; Moens and Selleri, 2006). Meis proteins control cell cycle progression and cell fate specification of different stem and progenitor cell types in a variety of tissues including liver, heart, retina, and brain during development (Mercader et al., 1999; Hisa et al., 2004; Bessa et al., 2008; Heine et al., 2008; Agoston and Schulte, 2009; Choe et al., 2009; Agoston et al., 2012, 2014; Paige et al., 2012). Recently the functional importance of Meis2 in adult neurogenesis has been discovered. It is involved in coordinated actions with Pax6 (Agoston et al., 2014). During mouse adult neurogenesis, Meis2 is strongly expressed in the SVZ and rostral migratory stream (RMS) and in a subset of OB interneurons (Agoston et al., 2014). Agoston et al. (2014) found that Meis2 can physically interact with Pax6 in cells in the SVZ and OB. This study also showed that the pro-neurogenic function of Pax6 in the adult SVZ requires direct interaction with Meis2. Meis2 has transcriptional activator function and Agoston et al. (2014) further confirmed that it acts as a Pax6 co-factor, binding to regulatory elements of the *Dcx* gene to promote its expression during adult SVZ neurogenesis.

# Conclusion

The TF Pax6 is one of the most intensively studied high-level developmental regulators, playing crucial roles in brain development and implicated in disease. Numerous studies, including those from our group, have shown that in both humans and mice many abnormalities resulting from mutations in *PAX6/Pax6* are in the cerebral cortex, where they arise largely because Pax6 is required for spatio-temporal control of many biological processes including cortical stem and progenitor cell proliferation and differentiation, neurogenesis, cell and laminar fate specification, neuronal migration, axon guidance and cortical arealization during early corticogenesis. However, we are a long way from having a comprehensive knowledge of exactly how Pax6 exerts such complex actions. In particular, there are at least three major challenges toward improved understanding of Pax6 functions. First, how does such highly complex expression pattern of Pax6 come about? As described above, the transcriptional regulation of *Pax6* is particularly complex and involves both short-range and long-range control mechanisms. Future studies are required to gain better knowledge of tissue- and time-specific regulatory elements responsible for spatiotemporally and quantitatively correct expression of Pax6. Secondly, how does Pax6 expression level influence its functional outputs? Our studies along with others have shown that Pax6 expression levels are crucial for its functions such as regulating progenitor cell proliferation and differentiation, and that mutations affecting Pax6 protein levels cause neurodevelopmental disorders in both humans and mice. Now it is time to ask exactly how Pax6 levels influence its actions at the molecular level. It will be of great interest to examine how varying levels of Pax6 influence its molecular actions at different cell cycle phases, which will be critical to obtain a clearer understanding of how Pax6 functions during the dynamics of cell division. Finally and probably most importantly, we are still far away from a comprehensive understanding of the complex gene networks that Pax6 regulates and how these networks change spatially and temporally during the course of cortical development. This represents a major challenge since Pax6 likely exerts direct and indirect regulation of different sets of downstream genes simultaneously, with context-dependent variation in the strength of their positive or negative effects.

# Acknowledgments

Research in the authors' laboratory was funded by the MRC (MR/J003662/1). Human embryonic and fetal material was provided by the Joint MRC/Wellcome Trust (grant # 099175/Z/12/Z) Human Developmental Biology Resource (www.hdbr.org).

# References


in the outer subventricular zone of the primate. *Neuron* 80, 442–457. doi: 10.1016/j.neuron.2013.09.032 S0896-6273(13)00863-5


cell cycle and specifies cortical cell identity by a cell autonomous mechanism. *Dev. Biol.* 302, 50–65. doi: 10.1016/j.ydbio.2006.08.035


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Copyright © 2015 Manuel, Mi, Mason and Price. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Real-time imaging of bHLH transcription factors reveals their dynamic control in the multipotency and fate choice of neural stem cells

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, Japan

#### Reviewed by:

Noelia Urbán, The Francis Crick Institute - Mill Hill Laboratory, UK Stewart Anderson, Children's Hospital of Philadelphia, USA William A. Harris, University of Cambridge, UK

#### \*Correspondence:

Itaru Imayoshi, The Hakubi Center, Institute for Virus Research, Kyoto University, Shogoin-Kawahara 53, Sakyo-ku, Kyoto 606-8507, Japan iimayosh@virus.kyoto-u.ac.jp; Ryoichiro Kageyama, Institute for Virus Research, Kyoto University, Shogoin-Kawahara, Sakyo-ku, Kyoto 606-8507, Japan rkageyam@virus.kyoto-u.ac.jp

> Received: 28 January 2015 Accepted: 13 July 2015 Published: 04 August 2015

#### Citation:

Imayoshi I, Ishidate F and Kageyama R (2015) Real-time imaging of bHLH transcription factors reveals their dynamic control in the multipotency and fate choice of neural stem cells. Front. Cell. Neurosci. 9:288. doi: 10.3389/fncel.2015.00288

#### Itaru Imayoshi 1, 2, 3, 4 \*, Fumiyoshi Ishidate<sup>3</sup> and Ryoichiro Kageyama2, 3, 5 \*

<sup>1</sup> The Hakubi Center, Kyoto University, Kyoto, Japan, <sup>2</sup> Laboratory of Growth Regulation, Institute for Virus Research, Kyoto University, Kyoto, Japan, <sup>3</sup> World Premier International Research Initiative–Institute for Integrated Cell-Material Sciences, Kyoto University, Kyoto, Japan, <sup>4</sup> Precursory Research for Embryonic Science and Technology, Japan Science and Technology Agency, Saitama, Japan, <sup>5</sup> Core Research for Evolutional Science and Technology, Japan Science and Technology Agency, Saitama, Japan

The basic-helix-loop-helix (bHLH) transcription factors Ascl1/Mash1, Hes1, and Olig2 regulate the fate choice of neurons, astrocytes, and oligodendrocytes, respectively; however, these factors are coexpressed in self-renewing multipotent neural stem cells (NSCs) even before cell fate determination. This fact raises the possibility that these fate determination factors are differentially expressed between self-renewing and differentiating NSCs with unique expression dynamics. Real-time imaging analysis utilizing fluorescent proteins is a powerful strategy for monitoring expression dynamics. Fusion with fluorescent reporters makes it possible to analyze the dynamic behavior of specific proteins in living cells. However, it is technically challenging to conduct long-term imaging of proteins, particularly those with low expression levels, because a high-sensitivity and low-noise imaging system is required, and very often bleaching of fluorescent proteins and cell toxicity by prolonged laser exposure are problematic. Furthermore, to analyze the functional roles of the dynamic expression of cellular proteins, it is essential to image reporter fusion proteins that are expressed at comparable levels to their endogenous expression. In this review, we introduce our recent reports about the dynamic control of bHLH transcription factors in multipotency and fate choice of NSCs, focusing on real-time imaging of fluorescent reporters fused with bHLH transcription factors. Our imaging results indicate that bHLH transcription factors are expressed in an oscillatory manner by NSCs, and that one of them becomes dominant during fate choice. We propose that the multipotent state of NSCs correlates with the oscillatory expression of several bHLH transcription factors, whereas the differentiated state correlates with the sustained expression of a single bHLH transcription factor.

Keywords: bHLH, neural stem cells, oscillation, fluorescent protein, imaging

# Introduction

Neural stem cells (NSCs) are multipotent and self-renewable cells that can give rise to neurons, astrocytes, and oligodendrocytes (Fishell and Kriegstein, 2003; Götz and Huttner, 2005; Kriegstein and Alvarez-Buylla, 2009). During brain development, basic helix-loop-helix (bHLH) transcription factors play pivotal roles in the self-renewal of NSCs and fate determination of neurons, astrocytes, and oligodendrocytes (Bertrand et al., 2002; Ross et al., 2003; Meijer et al., 2012; Wilkinson et al., 2013; Imayoshi and Kageyama, 2014a). bHLH transcription factors, such as Hes1 and Hes5, regulate the self-renewal of NSCs as downstream effectors of Notch signaling, whereas proneural bHLH transcription factors, such as Ascl1 and Neurog1/2, promote neuronal differentiation. Other bHLH transcription factors, e.g., Olig1 and Olig2, regulate oligodendrocyte differentiation. However, in addition to Hes1/Hes5, some bHLH fate determination factors, such as Ascl1 and Olig2, are also known to have roles in NSC maintenance or proliferation. In addition, Hes1 induces astrocyte formation at later stages. Currently, it is not understood completely how these various and sometimes opposing functions of each bHLH transcription factor in the self-renewal and fate-choice events of NSCs are achieved (Vasconcelos and Castro, 2014; Imayoshi and Kageyama, 2014a,b). We previously found that transcription of some bHLH genes is differentially controlled in NSCs and differentiating cells (Shimojo et al., 2008). Therefore, we decided to analyze the expression dynamics of bHLH transcription factors in more detail.

# Heterogeneous Expression of bHLH Transcription Factors in Cultured NSCs

NSCs can be cultured in an undifferentiated state in vitro. For instance, in the presence of epidermal growth factor and basic fibroblast growth factor, NSCs can be expanded from enzymatically dissociated brain tissues in a monolayer condition (NS cell culture) (Conti et al., 2003). NSCs can reportedly be established from brains at various stages, from embryos to adults. Furthermore, the purity of these cultures is very high; more than 99% of cells express NSC marker proteins, such as Nestin. Cultured NSCs also express bHLH transcription factors, including Hes1, Ascl1, and Olig2 (Imayoshi et al., 2013a).

In actively-dividing and self-renewing NS cells, Hes1, Ascl1, and Olig2 bHLH transcription factors are expressed (**Figure 1A**). Compared with the homogeneous expression of Sox2, which is a member of the high-mobility-group transcription factor family, the expression of all three bHLH transcription factors is variable among cells, indicating their dynamic expression (**Figure 1A**) (Imayoshi et al., 2013b). In order to elucidate the heterogeneous expression of bHLH transcription factors, we adopted a real-time imaging strategy to monitor the expression dynamics of bHLH proteins in cultured NSCs.

FIGURE 1 | Expression of bHLH factors in NS cells and fluorescent reporter-bHLH fusion constructs. (A) Self-renewing NS cells were immunostained with anti-Sox2, anti-Hes1, anti-Ascl1, and anti-Olig2 antibodies. Hes1, Ascl1, and Olig2 expression levels were variable while another NSC-specific factor, Sox2, was expressed at a relatively constant level. (B) Venus-Hes1 fusion knock-in mouse strain. (C) Venus-Ascl1 fusion BAC Tg mouse strain. (D) mCherry-Olig2 fusion BAC Tg mouse strain. Coding regions of bHLH factors are shown in gray. The pictures and graphs of this figure are reprinted from Supplementary Figures S1C,E; S2A; S6A–D of Imayoshi et al. (2013b).

# Fluorescent Reporter-fusion Knock-in or Bacterial Artificial Chromosome (BAC) Transgenic (Tg) Strategy for Monitoring Protein Expression Dynamics of bHLH Transcription Factors

Fusion reporter constructs are used to monitor the expression dynamics of proteins of interest in living cells (**Figures 1B–D**) (Miyawaki, 2011; Imayoshi et al., 2013a). In addition to fluorescent reporter proteins, bioluminescent reporters, such as luciferase, are also effective. One advantage of the reporter-fusion strategy is that the half-life and cellular localization of proteins basically depend on the fused target protein, at least in the case of bHLH transcription factors. Most bHLH transcription factors have very short half-lives, such as ∼20 min, and therefore reporter-bHLH fusion proteins are also very unstable.

In order to analyze precisely the fine expression dynamics of bHLH transcription factors, it is necessary that reporterbHLH fusion proteins are expressed under the control of their original promoter/enhancer sequences. One of the reliable ways to mimic the original regulation of bHLH gene expression is the knock-in strategy (Imayoshi et al., 2013a,b). By homologous recombination in embryonic stem cells, a reporter coding sequence is inserted into the N− or C− terminus of a bHLH gene so that the reporter-bHLH fusion protein is expressed from the original bHLH gene locus. As far as we know, the reporter-bHLH fusion protein is functional and, in most cases, the allele carrying the knocked-in fusion reporter functions as normally as the wild-type allele. Another strategy to express reporter-bHLH fusion proteins optimally is the generation of Tg mice with a BAC. Using homologous recombination in Escherichia coli, we can edit and modify BACs (Lee et al., 2001). BACs can contain very long DNA sequences, such as ∼200-kb. Therefore, by utilizing BAC clones with the promoter/enhancer sequences of a bHLH gene, it is possible to make transgene constructs designed to express reporter-bHLH fusion proteins under its original regulatory elements. From the Tg mouse founder lines carrying BAC transgene constructs, we can choose ones with a transgene copy number of only one by Southern blotting or real-time PCR analysis. These one-copy BAC Tg mice work as reporter strains for real-time imaging of the expression dynamics of bHLH transcription factors (Imayoshi et al., 2013b). One concsern regarding the reporter-fusion BAC Tg mouse strategy is that the number of functional bHLH gene alleles is increased from two to three. Ideally, reporter expression should be analyzed in a heterozygous-mutant background of the target gene by crossing with a mouse strain harboring a null allele. However, at least in the case of bHLH transcription factors, such as Hes1, Ascl1, or Olig2, the addition of the reporter-fusion transgenes dose not apparently affect the normal development of the nervous system (Imayoshi et al., 2013b).

As far as we know, the N-terminal fusion of a reporter proteins in both knock-in and BAC Tg mice shows better results than for a C-terminal fusion. The fluorescence or bioluminescence of a reporter protein is brighter when it is fused to the N-terminus than to the C-terminus of bHLH transcription factors. This is probably because translation starts from the N-terminus, so a reporter protein located at the N-terminus starts to work earlier during its expression.

For quantitative live-imaging of fluorescent reporters, we established NS cell cultures from each knock-in or BAC Tg mouse strain (Imayoshi et al., 2013b). In fluorescent protein-bHLH fusion reporter mice, Venus-Hes1 fusion knock-in, Venus-Ascl1 fusion BAC Tg, and mCherry-Olig2 fusion BAC Tg mouse strains (**Figures 1B–D**), the expression levels of fusion proteins and endogenous bHLH proteins were highly correlated to each other in cultured NSCs (Imayoshi et al., 2013b). Therefore, we conducted real-time imaging of fluorescent protein-bHLH fusion reporters in cultured NSCs with confocal fluorescence microscopy.

# Real-time Imaging of Fluorescent Reporter-fusion bHLH Proteins with Confocal Fluorescent Microscopy

For real-time imaging of fluorescent proteins with low expression levels, it is important to prepare a highly sensitive and lownoise imaging system. As we mentioned earlier, the mRNA and protein of most bHLH transcription factors have very short half-lives (e.g., ∼20 min): therefore, the reporter-bHLH fusion proteins are also very unstable (Imayoshi et al., 2013a,b). We can measure the decay time of reporter-bHLH fusion products by pharmacologically blocking the synthesis of new cellular proteins with cycloheximide. Indeed, the reporter activity of bHLH-fusion products are decayed with the same time course as the original bHLH proteins (data not shown). Furthermore, in cultured NSCs derived from knock-in or BAC Tg mice, reporterbHLH fusion proteins are expressed from a limited number of gene alleles. Therefore, the fluorescent signals of reporter-bHLH fusion proteins in NSCs are very dim and sometimes invisible by simple observation through ocular lenses.

To acquire these very weak fluorescent signals, we utilized a confocal microscope equipped with a highly sensitive detector (Imayoshi et al., 2013b). In some cases, the fluorescent signal intensity of reporter-bHLH fusion proteins is much weaker than that of the autofluorescence of cultured NSCs. To overcome this problem, a spectral imaging technique is very effective. Spectral imaging and linear unmixing has become an important tool in confocal fluorescence microscopy to discriminate between fluorescent signals with overlapping spectral characteristics. For example, when cultured NSCs are illuminated with a 514-nm argon laser, the autofluorescent signal appears at approximately 560 nm. Real signals from Venus-Hes1 or Venus-Ascl1 fusion proteins are observed approximately 530 nm, and their intensity is weaker than the autofluorescence of NSCs. However, when the fluorescent signals are acquired with a spectral GaAsP array detector, specific Venus-bHLH signals can be separated from autofluorescence by applying linear unmixing algorithms.

In addition to the highly sensitive and spectral imaging techniques, imaging systems should be optimized for lownoise and reliable quantitative analysis. As the fluorescent signal intensity of reporter-bHLH fusion proteins is very weak, contamination with noise signals impairs the reliable quantitative analysis of the expression dynamics of bHLH transcription factors. To prevent bleaching the fluorescent reporters and avoid photodamage to NSCs by the excitation lasers, the sensitivity of the imaging system is again critically important. Bleaching of the reporters impairs the reliability of the imaging results. During long-term continuous imaging, laser power should be as small as possible, because NSCs are delicate and very sensitive to exogenous perturbations, including photodamage by exposure to a strong laser.

# Oscillatory Expression of bHLH Transcription Factors in Multipotent NSCs

NS cell cultures established from Venus-Hes1 fusion knockin, Venus-Ascl1 fusion BAC Tg, and mCherry-Olig2 fusion BAC Tg mouse strains were used for real-time imaging with confocal fluorescence microscopy. This analysis revealed that the fluorescent fusion reporter proteins with Hes1, Ascl1, or Olig2 exhibited dynamic expression changes in NS cells (**Supplementary Movies 1**–**3**), as observed in the bioluminescence imaging experiments with luciferase-bHLH fusion reporters (Imayoshi et al., 2013b). By quantitative analysis of single cells, the levels of Hes1 and Ascl1 proteins seem to oscillate with a 2–3-h period (**Figures 2A,B**), while Olig2 oscillates with a longer period, such as 5–8 h (**Figure 2C**) (Imayoshi et al., 2013b). Although fluorescent reporter-bHLH proteins showed dynamic expression changes, oscillatory expression was observed more clearly in the bioluminescence imaging experiments with luciferase-bHLH fusion reporters.

Especially, the peak amplitude during signal changes was larger in the case of luciferase-bHLH fusion reporters (Imayoshi et al., 2013b), which is partly because the protein maturation time of luciferase is more rapid than that of fluorescent proteins, such as Venus and mCherry. Our observations revealed that, in actively dividing and multipotent NSCs, bHLH transcription factors are expressed in an oscillatory manner.

The oscillatory expression of bHLH transcription factors raised the possibility that NSCs may change their fate preferences over time. Indeed, when cultured NSCs are sorted by their expression levels of each bHLH transcription factor, their differentiation preferences are correlated with bHLH expression levels (Imayoshi et al., 2013b). For instance, Hes1-high NSCs preferentially differentiate into astrocytes, whereas Ascl1-high and Olig2-high NSCs preferentially differentiate into neurons and oligodendrocytes, respectively. These results suggest that different expression levels of bHLH transcription factors bias the fate choices of NSCs. However, such transient high expression of bHLH transcription factors may not be sufficient for cell fate determination, because these NSCs are still multipotent. Thus, we analyzed how the expression of these bHLH transcription factors changes during cell fate choice.

# Expression Dynamics of bHLH Transcription Factors in Multipotency and Fate Choice

In contrast to oscillatory expression of multiple bHLH transcription factors in self-renewing NSCs, one of the bHLH transcription factors is expressed in a sustained manner during cell fate choice, while the others are repressed (**Figure 3**) (Imayoshi et al., 2013b). For instance, we found that the transient down-regulation of Hes1 expression and the concomitant up-regulation of Ascl1 before cell division bias NSCs toward a neuronal fate choice, and the sustained expression of Ascl1 after cell division irreversibly determines neuronal fate (**Figure 3**). During astrocyte and oligodendrocyte differentiation, the expression of Hes1 and Olig2 is upregulated, respectively, although they still oscillate. However, even during the trough phases, both Hes1 and Olig2 levels are

higher than they are in NSCs, indicating that Hes1 and Olig2 expression continues in a sustained manner during astrocyte and oligodendrocyte differentiation (**Figure 3**). When Hes1 or Olig2 becomes dominant, the expression of the other two factors is downregulated. These results indicate that Hes1, Ascl1, and Olig2 are expressed in an oscillatory manner in multipotent NSCs, and that one of them becomes dominant during cell fate choice (Imayoshi et al., 2013b).

From these analyses, we propose the following model: the oscillatory expression of multiple bHLH transcription factors is correlated with the multipotent and self-renewable state of NSCs, whereas the sustained expression of a selected bHLH transcription factor regulates fate determination (**Figure 3**) (Imayoshi et al., 2013b; Imayoshi and Kageyama, 2014a,b). To validate this model, we applied a new optogenetic method (photo-activatable Gal4/UAS system) to manipulate artificially the expression patterns of bHLH transcription factors using blue light illumination, showing that oscillatory expression activates the proliferation of NSCs, whereas sustained expression induces Imayoshi et al. Oscillatory control of bHLH factors

cell fate determination. Detailed results are reviewed elsewhere (Imayoshi et al., 2013b; Imayoshi and Kageyama, 2014a,b).

# Conclusion

Previous reports have shown some evidence that the bHLH transcription factors Hes1, Ascl1, and Olig2 have multiple functions, such as NSC proliferation, self-renewal, and differentiation. Of course, protein modifications and/or partner co-factor variation may be involved in these different activities of bHLH transcription factors in a complex manner, but our studies revealed that oscillatory versus sustained expression dynamics also contribute to these different functions. More generally, our studies indicated that expression patterns, rather than simply the expression levels, of various transcription factors regulate whether stem cells proliferate or differentiate. In addition to Hes1, Ascl1, and Olig2, other bHLH transcription factors, such as Hes5 and Neurog2, are expressed in an oscillatory manner in NSCs or neural precursor cells (Shimojo et al., 2008; Imayoshi et al., 2013b). Dynamic expression is not necessarily confined to bHLH-type transcription factors, and many kinds of transcription factors are known to be involved in NSC regulation: therefore, it is expected that future imaging and manipulation studies will unveil the unique expression dynamics and their significance of various transcription factors in NSCs.

bHLH transcription factors regulate other kinds of somatic stem cells and pluripotent stem cells. For example, Ascl1 is essential for the differentiation of neuroendocrine cells from epithelial progenitors of the developing glandular stomach (Kokubu et al., 2008), and cyclical Hes1 expression in embryonic stem cells is important for their diverse differentiation abilities (Kobayashi et al., 2009). Therefore, the dynamic control of bHLH transcription factors, including their oscillatory expression, may regulate many regulatory processes of other kinds of stem cells.

Our studies suggest the importance and future possibilities of imaging techniques for the analysis of the expression dynamics of endogenous proteins at the single-cell level by coupling with various cellular events, such as cell cycle progression and cell differentiation (Miyawaki, 2011; Isomura and Kageyama, 2014).

# References


Under the current situation in developmental biology in which many biological processes start to be understood from the viewpoint of systems biology or single-cell biology (Levine et al., 2013; Sanchez and Golding, 2013; Zenobi, 2013), it is expected to become even more important to image subcellular localization and expression dynamics over time at a high resolution while keeping the original expression levels of target proteins.

# Acknowledgments

This work was supported by research grants from the Ministry of Education, Culture, Sports, Science and Technology of Japan (MEXT) and Japan Science and Technology Agency (JST): Grant-in-Aid for Scientific Research on Young Scientists (A) (JSPS 24680035; 15H05570) (II), Challenging Exploratory Research (JSPS 26640011) (II), Innovative Areas (JSPS 26117511; 15H01489) (II), and the JST PRESTO program (II). We also thank the supports by the Human Frontier Science Program, the Nakajima Foundation, the Suzuken Memorial Foundation, the Mochida Memorial Foundation, the Sumitomo Foundation, the Senri Life Science Foundation, the Uehara Memorial Foundation, The Waksman Foundation of Japan INC, the Research Foundation for Pharmaceutical Sciences, the Cell Science Research Foundation, the Brain Science Research Foundation, and Platform for Dynamic Approaches to Living System from the MEXT.

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fncel. 2015.00288

Supplementary Movie 1 | Venus-Hes1 expression in self-renewing NSCs. Time in h and min is indicated. Related to Figure 2A.

Supplementary Movie 2 | Venus-Ascl1 expression in self-renewing NSCs. Time in hr min, and sec is indicated. Related to Figure 2B.

Supplementary Movie 3 | mCherry-Olig2 expression in self-renewing NSCs. Time in hr min, and sec is indicated. Related to Figure 2C.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Imayoshi, Ishidate and Kageyama. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Lipidome of midbody released from neural stem and progenitor cells during mammalian cortical neurogenesis

Yoko Arai † , Julio L. Sampaio, Michaela Wilsch-Bräuninger, Andreas W. Ettinger † , Christiane Haffner and Wieland B. Huttner\*

Max Planck Institute of Molecular Cell Biology and Genetics, Dresden, Germany

#### Edited by:

Chao Deng, University of Wollongong, Australia

#### Reviewed by:

Xiao-Feng Zhao, University of Michigan, USA Yechiel Elkabetz, Tel Aviv University, Israel

#### \*Correspondence:

Wieland B. Huttner, Max Planck Institute of Molecular Cell Biology and Genetics, Pfotenhauerstrasse 108, D-01307 Dresden, Germany huttner@mpi-cbg.de

#### †Present Address:

Yoko Arai, Institut Jacques Monod, Centre National de la Recherche Scientifique UMR 7592, Université Paris Diderot, Sorbonne Paris Cité, Paris, France; Andreas W. Ettinger, University of California, San Francisco, San Francisco, CA, USA

> Received: 15 June 2015 Accepted: 06 August 2015 Published: 28 August 2015

#### Citation:

Arai Y, Sampaio JL, Wilsch-Bräuninger M, Ettinger AW, Haffner C and Huttner WB (2015) Lipidome of midbody released from neural stem and progenitor cells during mammalian cortical neurogenesis. Front. Cell. Neurosci. 9:325. doi: 10.3389/fncel.2015.00325 Midbody release from proliferative neural progenitor cells is tightly associated with the neuronal commitment of neural progenitor cells during the progression of neurogenesis in the mammalian cerebral cortex. While the central portion of the midbody, a cytoplasmic bridge between nascent daughter cells, is engulfed by one of the daughter cell by most cells in vitro, it is shown to be released into the extracellular cerebrospinal fluid (CF) in vivo in mouse embryos. Several proteins have been involved in midbody release; however, few studies have addressed the participation of the plasma membrane's lipids in this process. Here, we show by Shotgun Lipidomic analysis that phosphatydylserine (PS), among other lipids, is enriched in the released midbodies compared to lipoparticles and cellular membranes, both collected from the CF of the developing mouse embryos. Moreover, the developing mouse embryo neural progenitor cells released two distinct types of midbodies carrying either internalized PS or externalized PS on their membrane. This strongly suggests that phagocytosis and an alternative fate of released midbodies exists. HeLa cells, which are known to mainly engulf the midbody show almost no PS exposure, if any, on the outer leaflet of the midbody membrane. These results point toward that PS exposure might be involved in the selection of recipients of released midbodies, either to be engulfed by daughter cells or phagocytosed by non-daughter cells or another cell type in the developing cerebral cortex.

Keywords: neural stem cells, midbody, shotgun lipidomic analysis, phosphatydylserine, mammalian cortical neurogenesis

# Introduction

At the end of mitosis, after the ingression of the cleavage furrow, a dividing cell is partitioned into two daughter cells that remain connected through a cytoplasmic bridge, the midbody, which is composed of the remnants of the central spindle including a set of microtubule interacting proteins (Glotzer, 2009; Chen et al., 2013). Physical separation of daughter cells is achieved by abscission, which completes cell division/cytokinesis by severing the midbody-bridge, bypassing the central portion of the midbody that contains a dense matrix of antiparallel microtubules and electrondense material (Buck and Tidsale, 1962). Accumulating in vitro evidence from induced pluripotent stem and cancer-derived immortalized cells indicates that the post-abscission midbody, with its characteristic matrix composition, is subsequently either retracted asymmetrically by a daughter cell (single abscission) (Kuo et al., 2011) or released into the extracellular space (double abscission). Alternatively, released midbodies could get attached to the surface of a daughter cell and consequently be engulfed (Ettinger et al., 2011; Crowell et al., 2014). In culture cells, neural stem cells, embryonic stem cells, or cancer-derived cells, which are still responsive to differentiation agents, showed midbody-release into the extracellular space (Ettinger et al., 2011). Potentially, these midbodies might be taken up by non-daughter cells, which could result in a long-range dispersion of midbodies. A similar phenomenon is observed in neuroepithelial cells during mouse cortical development (Marzesco et al., 2005; Dubreuil et al., 2007).

At the onset of neurogenesis, neuroepithelial cells divide asymmetrically and preferentially release their midbody (NE midbody in short) into the extracellular ventricular fluid (Marzesco et al., 2005). The NE midbody release showed a strong correlation with the increase of neurogenesis (Dubreuil et al., 2007). Therefore, an efficient release of midbodies from neuroepithelial cells was postulated as a mechanism to reduce proliferative capacity of neuroepithelial cells. Midbody release ultimately results in loss of both the cytoplasmic and membraneous components present in the midbody including lipids from neuroepithelial cells. While the mechanism of the midbody release from neuroepithelial cells had been addressed through protein functions (Dubreuil et al., 2007), it is still unclear what is the fate of the NE midbody during neurogenesis.

In this study, we characterized the NE midbody for the first time through lipidome analysis, where we observed enrichments of specific classes of phosphatidylserine (PS), phosphatidylethanolamine (PE), and specific species of ceramide (Cer), and triacylglycerols (TAGs). Among these, PS was strikingly enriched in NE midbody. It has been proposed that midbodies with PS, particularly carrying externalized PS on their membrane bilayer, are cleaned up by phagocytes in Caenorhabditis elegans (Chai et al., 2012). This observation suggests that the PS status on the membrane bilayer is associated with the fate of the NE midbody, which may allow deducing its fate. Our data regarding PS status suggests at least two different categories of recipients, either a daughter cell for an engulfment pathway or non-daughter cells for phagocytosis. The distinct subtype of NE midbodies released upon differentiation (Dubreuil et al., 2007; Ettinger et al., 2011) and carry a stem cell marker (Marzesco et al., 2005; Corbeil et al., 2010) may be a novel cue in and contribute to the complexity of the proliferative territories in the developing cerebral cortex.

# Results

#### Midbody Enrichment for Lipidomic Characterization

To enrich midbodies released from neural progenitor cells of the developing mouse embryo ("NE midbody" in short), we collected the ventricular fluid [embryonic cerebrospinal fluid (CF)] of mouse embryos at E11.5 (the number of midbodies is high at this stage) (Marzesco et al., 2005) (**Figure 1A**). In order to separate the midbody particles (∼500 nm) from the

biggest lipid "contaminant," the lipoprotein particles (∼50 nm), which are both present in the ventricular fluid (Marzesco et al., 2005), we subjected the sample to a velocity sucrose gradient. We analyzed all fractions from the gradient by immunoblotting for marker proteins prior lipid analysis in order to check for the quality of the separation (**Figure 1B**). We used the midbody markers: citron rho-interacting kinase (CRIK) (Madaule et al., 1998), prominin-1 and α-tubulin (Dubreuil et al., 2007) in order to identify the fraction enriched in midbody particles in the sucrose fractions. Lipoparticles enriched fractions were identified with the marker apolipoprotein A1 (Furbee et al., 2002) (**Figure 1B**).

Total lipid quantified from cerebrospinal fluid of E11.5 mouse embryos present in the individual fractions 1–9 after velocity sucrose density gradient.

Comparing all fractions in terms of abundance of all markers, we were able to fractionate midbodies (L6) from lipoprotein particles (L1) and cell membranes (L8) (**Figure 1B**). We identified fraction 6 (L6) as the midbody particleenriched fraction due to the relative enrichment of midbody markers CRIK, prominin-1 and α-tubulin and the absence of apolipoprotein A1. While fraction 7 (L7) was also enriched with midbody markers, we did not present this fraction as a midbody fraction because of the increased contamination by apolipoprotein A1 as compared to fraction L6, and considered it being a mixed population of L6 and 8. Fraction 1 (L1) was a relatively pure lipoprotein particle-containing fraction judging from the enrichment of α-tubulin and apolipoprotein A1 and depletion of CRIK and prominin-1. Fraction 8 (L8) was used as a reference of a mixed fraction that included low amounts of midbody and plasma membrane of neural progenitor cells due to the presence of all protein markers (**Figure 1B**). Moreover, we could observe that lipoproteins were by far the most predominant lipid-containing entities present in the ventricular fluid where its purified fraction (L1) outweighs the midbody fraction (L6) by more than seven-fold (**Figure 1C**).

# Midbodies Have a Distinct Lipid Class Composition from Other Cellular Membranes

The lipid analysis of the all fractions confirmed that L1 is indeed enriched in lipoprotein particles. Lipoproteins possess a hydrophobic core composed by neutral lipids such as triacylglycerols (TAG) and cholesterol ester (CE) (Willnow et al., 2007) and we observe that this fraction is the most enriched in these lipid classes (**Figure 2A**). We then see a continuous decrease in the total lipid amount in lower gradient fractions that is accompanied by changes in several lipid classes (**Figures 1C**, **2A**). As expected from fractions more depleted from lipoprotein particles, CEs and TAGs decrease in lower gradient fractions as well as sphingomyelin (SM) (**Figure 2A**). Most importantly, we observe a sharp increase in phosphatidylethanolamine (PE) and phosphatidylserine (PS) in lower gradient fractions that peak precisely at fraction L6 where we observe the purest midbody fraction (L6) followed by a sharp decrease in the following fractions (**Figure 2A**). It is important to note that in the fractions below L6, we also observe a dramatic decrease in TAGs and PE plasmalogens (PE O-) and a slight increase in PC (phosphatidylcholine) suggesting high specificity in terms of lipid class composition of the midbody containing fraction compared to the remaining cell membrane fragments. Importantly, the mean PS amount in L6 was 27-fold higher than in L1 (L6, 2.64 ± 1.26; L1, 0.098 ± 0.11, **Figure 2B**) and 2.4-fold when L6 was compared to the intermediate fraction L8 (L8, 1.11 ± 0.8, **Figure 2B**). Respective to the amount of PE, the mean PE amount in L6 was 9.3 and four times higher than in L1 and L8, respectively (L6, 2.05 ± 0.66; L1, 0.22 ± 0.13; L8, 0.51 ± 0.15, **Figure 2B**).

In summary, we observe high specificity of lipid class composition in the midbody-enriched fraction most notably, PS and PE are highly enriched in the midbody-containing fraction compared to that of lipoprotein and cell membrane mixed fractions.

FIGURE 2 | The Lipidome of NE midbodies. (A) Lipid class profile of all sucrose fraction. The content of individual lipid classes was determined by summing up absolute abundances of all identified species and is expressed as mol %. We consider 100% to be all the membrane lipids, meaning that CE and TAG are excluded from the normalization. CE and TAG amounts are also normalized to the total membrane lipids. P-values were estimated by One-Way ANOVA; \*\*P < 0.01, \*\*\*\*P < 0.0001. Error bars correspond to SD (n = 3). CE amounts are divided by 10 and PC O-, PE, PE O-, PS, and Cer are multiplied by 10 for visualization purposes. CE, cholesterol ester; TAG, triacylglycerol; PC, phosphatidylcholine; SM, sphingomyelin; PI, phosphatidylinositol; Chol, Cholesterol; PC O- and PE O-, plasmalogens; PE, phosphatidylethanolamine; PS, phosphatidylserine, Cer, ceramide. (B) Enrichment of lipid constituency in layer 6 (L6) compared to layer 1 (L1) (L6/L1) and layer 8 (L8) (L6/L8), respectively. The y-axis is on a log2 scale.

# Specific Lipid Species are Enriched in Midbodies Compared to Other Cellular Membranes

It has been previously reported that midbodies collected from HeLa cells ("HeLa midbody" in short) showed the specific accumulation of lipid species of C22–24 ceramides, TAG [16:1, 12:0, 18:1] and PS [18:0–20:4] when compared to cells in cytokinesis (Atilla-Gokcumen et al., 2014). In agreement with the result of the HeLa midbody lipidome, the NE midbody lipidome displays an enrichment of TAG [46:2] in the midbody fraction and a specific enrichment of PS [38:4] and ceramides but in our case, we observed instead a specific increase in C16– 18 ceramides, Cer [36:1:2] (**Figure 3A**). Importantly, we could observe many more species enriched in the midbody fraction. Not only we found TAG [46:2] enriched in the midbody fraction, but also many other TAGs with similar short chain fatty acid compositions (C12–18) in different putative combinations, TAG [46:0], TAG [46:1], TAG [46:2], and TAG [50:4] (**Figure 3A**). We could also observe that most lipid species belonging to PS and PE irrespective of the fatty acid saturation and chain length level display a strong enrichment in the midbody fraction (**Figures 3B,C**). Interestingly, specific PE O- species, namely with longer and more unsaturated fatty acids, PE O- [38:6] and PE O- [40:6] are enriched in the midbody fraction (**Figure 3C**). Although PC as a class is not changing significantly between the different fractions, we can observe some changes in lipid

species. Shorter and more saturated PC lipid species (C30–34) are enriched in midbodies, PC [30:0], PC [32:0], PC [32:1], and PC [34:1], while longer and more unsaturated lipid species are depleted in the midbody fraction, PC [36:2], PC [36:4], and PC [38:4] (**Figure 3D**).

In conclusion, we could reproduce the previous published results on lipids enriched in the midbody of Hela cells but identified new midbobody specific lipids, namely within Cer, PS and TAG lipid classes, PE as a lipid class and long and unsaturated PE O- and a remodeling of the PC lipid class to shorter and more saturated lipid species in the midbody.

# NE Midbodies Carry Internalized and Externalized PS

From the lipidome analysis, we observed that mainly PC is the most depleted membrane lipid classes (aprox. 8 mol %) while PS and PE are the most enriched lipid classes in NE midbody (almost 5 mol % in total) compared to cell membranes (L8) (**Figures 3B–D**). If we take into consideration the plasma membrane lipid asymmetry where PC is on the extracellular leaflet of the plasma membrane at the cell surface, and both PS and PE face the cytoplasmic leaflet (Verkleij et al., 1973), it raises the interesting possibility that PS and/or PE might be externalized to keep the membrane integrity. In C. elegans, externalization of PS on midbodies released from Q neuroblasts was shown to be an engulfment signal for the midbody via clearance by phagocytes (Chai et al., 2012). These lines of evidence suggest that the status of PS externalization on the midbody might be a mark to infer the fate and/or recipient cell of released midbodies.

Therefore, we assessed the leaflet localization of PS in midbody membranes, taking advantage of its specific interacting partner, Annexin V (Schutte et al., 1998). To this end, we injected Annexin V directly coupled with Cy5 (Annexin V-Cy5) into the cerebral fluid of mouse embryos at the level of the forming cortex (E11.5), allowing binding for 30 min, followed by fixation and analysis by immunofluorescence. At this early stage of neurogenesis, the fluid filling the ventricular space contains a high amount of released NE midbodies. In this experimental setup, binding of Annexin V-Cy5 to particles is expected only if PS was externalized on the outer leaflet of the membrane bilayer. We confirmed Annexin V-Cy5 positive particles in the ventricular fluid as NE midbodies by a counterstaining with prominin-1, a midbody marker (Marzesco et al., 2005) (**Figures 4A–D**). In addition to particles in the lumen, we also detected Annexin V-Cy5 staining at the apical plasma membrane of neuroepithelial cells, which is the site of cytokinesis midbody formation between nascent daughter cells (**Figure 4C**). The presence of Annexin V-Cy5 and prominin-1 double positive midbody structures on the apical surface was confirmed by immunogold labeling and electron microscopy (**Figures 4E–J**). We found comparably electron-dense, ∼500 nm–large structures labeled individually or together by antibodies against prominin-1 or Cy5 (**Figures 4E–J**). These results indicate that PS externalization could occur at the end of cytokinesis of neuronal progenitor cells.

NE midbodies show three distinct populations defined by their decoration with Annexin V-Cy5 and presence of prominin-1. At the apical surface of the ventricular zone, we observed prominin-1 positive and Annexin V-Cy5 negative particles as a major population (57%, **Supplementary Figure 1**) and prominin-1 and Annexin V-Cy5 double positive particles as a second population (41%, **Supplementary Figure 1**). A third minor population (2%) contained only Annexin V-Cy5. Interestingly, in the CF, prominin-1, and Annexin V-Cy5 double positive particles became the major population (55%, **Figure 4K**)

and the prominin-1 positive and Annexin V-Cy5 negative particles were the second population (40%, **Figure 4K**). The third minor population (5%) contained only Annexin V-Cy5 was still observed. Taken together, these results strongly suggest that a fraction of neural progenitor cell midbodies externalized PS prior to their release in the ventricular fluid and ultimately both PS-externalized and -internalized midbodies were found after cytokinesis.

# PS Flipping Occurs Upon the Detachment of the Midbody after Cytokinesis

In order to determine the dynamics of flipping of PS on the membrane, we analyzed dividing neuroblastoma cells (Neuro-2a), which have been shown to have the capacity of both releasing and engulfing midbodies (Ettinger et al., 2011). We performed live-cell time-lapse microscopy using a transgenic Neuro-2a cell line expressing GFP- tagged mitotic kinesin-like protein-1 (MKLP1) (Poser et al., 2008; Ettinger et al., 2011), a core midbody and central spindle protein (Hu et al., 2012). This cell line allows the visualization of the forming and released midbody (Neuro2a midbody, in short). To visualize externalized PS, we added Annexin V-Cy5 to the culture medium at the beginning of the movie. We assessed PS externalization by appearance of the Cy5 signal upon binding of Annexin V-Cy5. Midbodies were identified as bright spots in the GFP channel.

As reported previously (Ettinger et al., 2011), MKLP1- GFP localized to the central portion of the midbody in Neuro-2a cells. PS was not externalized during the telophase (**Figure 5A**, "T"). The midbody was subsequently detached from the central portion of the midbody bridge between daughter cells after either a first or second abscission event of the cytoplasmic membrane. This detachment was identified by the displacement of the midbody from the point of contact of the daughter cell to one of the two daughter cells and PS was externalized at this moment, as determined from appearance of a Cy5 spot (**Figure 5A**, "A"). Approximately 1 h after the detachment, the midbody was released from a daughter cell (**Figure 5A**, "R"). These results indicate that PS was flipped out and exposed to the extracellular space at the end of the cytokinesis before midbody release (see Supplementary Movie 1).

We quantified Annexin V and MKLP1-GFP double positive midbodies in Neuro-2a MKLP1-GFP cell cultures in order to understand how widespread this phenomenon is (**Figures 5B–E**). During late telophase, a midbody bridge is formed between the nascent daughter cells. MKLP1-GFP was recruited to the

as z-stacks every 0.38-µm and are maximum intensity projections. Scale bars, 10µm. (B) Examples of bilateral midbodies. Fluorescence

central portion of the midbody (**Figure 5B**) as confirmed by immunostaining for α-tubulin to highlight the central spindle microtubules and for Aurora B kinase which localizes to the midzone in anaphase and decorates the inner part of the central spindle in telophase (**Supplementary Figure 2A**) (Hu et al., 2012). We almost never detected Annexin V-Cy5 staining in these newly formed midbodies (about 3.5% of total bilateral midbodies, **Figures 5B,E**). Instead, Annexin V was present in midbodies either adhering to one of the daughter cells after a first abscission or on interphase cells (**Figures 5B,C**, white arrowheads and see **Supplementary Figure 2B**). The majority of such monolateral midbodies were Annexin V-Cy5 positive (≈66% of total monolateral midbodies, **Figure 5E**). Fully released Neuro2a midbodies contained Annexin V-Cy5 in the majority of cases (≈90% of total released GFP positive particles, **Figures 5D,E**). Taken together, PS exposure occurred before the complete release of the midbody and it likely occurred after the first abscission step.

#### PS Status on HeLa Midbodies

and 137 for released); error bars indicate SD.

A previous study indicated that half of the total Neuro-2a midbodies are retained on daughter cells and the rest are released into the extracellular space (Ettinger et al., 2011). Conversely, in HeLa cells, midbodies were described to be engulfed by a daughter cell (Crowell et al., 2014). Therefore, we asked whether HeLa midbodies carry externalized PS. Similarly to the analysis of Neuro2a cells, we used a transgenic HeLa cell line expressing MKLP1-GFP to follow the midbody and determined the presence of PS in the outer leaflet of the midbody membrane with Annexin V-Cy5 (**Figure 6**). Midbody bridges formed during telophase were identified by recruitment of MKLP1-GFP and confirmed by immunostaining for α-tubulin and Aurora B (**Supplementary Figure 3**). In both monolateral inherited midbodies (**Figure 6A**) and in newly formed/bilateral midbodies (**Figure 6B**), we hardly detected PS externalization. These results suggest that PS exposure to the extracellular environment is a characteristic feature of neural-related cell lines and neuronal progenitor cells.

(total number of counted particles, 243 for bilateral, 234 for monolateral

# Discussion

The present study is the first quantitative analysis of the lipid composition of released NE midbodies from mouse embryonic CF. A comparison of lipidome data from HeLa and NE midbodies showed a common enrichment of several lipid species (Atilla-Gokcumen et al., 2014). Among them, the enrichment in specific PS, TAG and Cer species was observed both in NE and HeLa midbodies compared to both lipoprotein particles and dividing cells. In comparison to midbodies from HeLa cells, NE midbodies displayed a wider extent of enrichment of other lipid classes and specific lipid species. This apparent divergence may reflect the functional complexity of the NE midbody in developing mouse embryos, in contrast to cancerderived, cultured cells like HeLa. It has been shown previously that cultured cells showed a re-arrangement of protein levels involved in metabolic pathways at the expense of other pathways as compared to primary cells, which seems to be required for the adaptation to the cultured conditions (Pan et al., 2009). Particularly relevant for this study is the down-regulation of proteins involved in fatty acid metabolism in cell lines (Pan et al., 2009), a fact that can explain why we identified more varying lipid species in NE midbody when compared to HeLa midbody.

What could be the relevance of the combined enrichment of all the reported lipids for the midbody structure and function? Interestingly, both Ceramide species (C16 and C24-ceramide) identified in our and the previous study (Atilla-Gokcumen et al., 2014), respectively have been shown to contribute for apoptosis (Seumois et al., 2007). Moreover, the enrichment and exposure of PS to the outer leaflet of plasma membrane, a feature observed in midbody membranes, is a common feature of many apoptotic cells (Fadok et al., 1998). These two observations put together suggest that released midbody membranes might contain a potent apoptotic signal and, upon its releasing from the neural stem and progenitor cells and its engulfment by neighboring cells, it might alter the cell fate of both. The strong enrichment in unsaturated PE species in the midbody, is in line with its recognized fusogenic properties and the finding that PE species are exposed on the cell surface specifically at the cleavage furrow during late telophase of cytokinesis (Emoto et al., 1996). PE, due to its unique geometry (inverted cone) which is enhanced by the increase of the degree of unsaturation, in that it forms a non-bilayer structure, is thought to promote rapid phospholipid trans-bilayer movement (Ellens et al., 1989) and stalk membrane formation (Markin et al., 1984) facilitating membrane fusion and fission processes in cellular membranes a process required for midbody release after cytokinesis.

Cells in cytokinesis have been shown to possess more resilient cell membranes requiring three-fold higher forces to provoke their disruption when compared to cells in interphase (Atilla-Gokcumen et al., 2014). This observation is in line with the increase in relative saturation of the most abundant phospholipid in mammalian cells, PC (**Figure 2D**). We observe enrichment in PC species containing 0 and 1 unsaturations and a decrease in PC species containing two or more unsaturations (**Figure 2D**). Unsaturation level of lipids is known to alter dramatically the physical properties of cell membranes by changing the membrane fluidity and order [Gennis RB, Biomembranes: Molecular structure and function (Springer Advanced Texts in Chemistry), 1989]. The specific enrichment in short chain TAGs is intriguing. TAGs in general are known to be metabolic energetic reservoirs in cells but the putative different functions of TAGs with specific fatty acid composition are still unknown.

Together, the lipid specificity observed in midbodies, seems to be intimately related to its structure and function. From it, we can conclude that midbodies possess relatively sturdy membranes due to the decrease of saturation of PC species, they are enriched in fusogenic lipids, unsaturated PEs, which facilitate fission and release of midbody, and are also enriched in potent signaling lipids, specific Ceramides and PS species that might influence neighboring cells if engulfed by them. All these observations put together highlight the importance of lipid compositional complexity, specificity and cellular localization in cell biology and how it can be directly associated with structural and functional aspects of cellular processes (Simons and Sampaio, 2011).

Although differences in lipid composition exist, common lipid species suggest shared characteristics between NE and HeLa midbodies. In HeLa cells, midbodies are known to be engulfed by a daughter cell through an actin-dependent phagocytosislike mechanism (Crowell et al., 2014) while the release of midbody from neural-derived cells (Ettinger et al., 2011) seems not compatible with the engulfment by one of the daughter cells. PS is a structural phospholipid, which exists in the cytoplasmic leaflet of the plasma membrane bilayer and whose externalization has been associated with phagocytosis during apoptotic body clearance (Fadok et al., 1998), retinal rod and cone photoreceptors clearance (Kevany and Palczewski, 2010) and also in midbody clearance in C. elegans (Chai et al., 2012). In above examples, the phagocytosis of PS-externalized midbodies was carried out by non-daughter cells. Hence, it is possible that PS-externalized NE midbodies may be engulfed by phagocytes in the developing mouse embryo. It has been proposed, however, that other cell types in the developing mouse central nervous system such as microglia, astrocytes and even neural progenitor cells or neurons have the potential to behave as phagocytes due to the expression of phagocytic receptors (Sokolowski and Mandell, 2011). One of the phagocytic receptors, Gas6, which binds to PS, is indeed expressed highly in proliferative neural progenitor cells compared to neurogenic progenitor cells (Arai and Huttner, unpublished data). These lines of evidence support the idea that PS-externalized NE midbodies could be phagocytosed by sister neural progenitor cells through a phagocytosis-related mechanism using Gasp6 signaling. We previously reported that midbodies carrying the stem cell marker prominin-1 (Marzesco et al., 2005) were released from proliferative neural progenitor cells, which may contribute to their loss of stemness (Dubreuil et al., 2007). One possibility is that PS-externalized midbody could be engulfed by proliferative neural progenitor cells to maintain a balance for the stem cell capacity or serving as a means of communicating between the proliferative territories of the developing mouse brain. In addition to the PS externalized NE midbody, we also observed PS internalized NE midbodies (**Figure 4K**), which may reflect the heterogeneity of neural progenitor cells providing these two different types of midbodies. With the progression of neurogenesis, proliferative neural progenitor cells give rise to neuronally committed neurogenic progenitor cells (Haubensak et al., 2004). It would be interesting to identify which type of neural progenitor cells preferentially release either PS-externalized or -internalyzed midbodies.

What could be the fate of the biological components of NE midbodies after cells take them up? The engulfed midbody should be digested in lysosomes into monomolecular species, which are recycled if necessary. Fatty acids from the PS species of the NE midbody could also be recycled, if any, it participates on any metabolic pathways. Amongst the PS species identified in our lipidomics, the presence of omega-3 and 6 fatty acids that cannot be produced de novo have been strongly suggested, for instance, (PS [38:4] = PS [18:0–20:4], arachidonic acid [20:4]) (Sampaio et al., 2011). Therefore, the recycling of midbodies could participate to the essential lipid metabolism during cortical development, participating to the maintenance of a lipid species pool for controlling proper neurogenesis.

# Methods

## Animals

All mouse embryos used were wildtype C57BL/six mice obtained by overnight mating of wildtype C57BL/six males and females. Noon of the day on which the vaginal plug was observed was defined as embryonic day (E) 0.5. All animal studies were conducted in accordance with German animal welfare legislation, and the necessary licenses obtained from the regional Ethical Commission for Animal Experimentation of Dresden, Germany.

# Enrichment of the Midbody Fraction Using Velocity Sucrose Density Gradient Centrifugation

E11.5 C57BL/6 mouse embryos were dissected in ice-cold 150 mM NH4Ac dissolved in lipid mass spectrometry-grade water (Merck; LiChrosolv grade). CF was collected as described (Marzesco et al., 2005). 1–2µl of CF per embryo were collected and pooled in a 1.5 ml Eppendorf tube (Eppendorf) on ice. Seventy-six micro liter of CF were mixed with 4µl of 100x proteinase inhibitor mix (Roche) and 320µl of 2.5 M sucrose in water to yield a total of 400µl of CF mixture at a final concentration of 2 M sucrose. The CF mix was transferred into the bottom of an ultraclean ultracentrifugation tube (Beckman coulter). The gradient was prepared by adding layers of less dense sucrose solutions (400µl of 1.6 M, 400µl of 1 M, 300µl of 0.3 M, and 300µl of 0.1 M) upon one another. The samples were centrifuged in a TLS55 swinging rotor (Beckman coulter) at 50,000 rpm for 18 h at 4◦C with a no deceleration profile. Immediately after the centrifugation, 180µl fractions were collected into 1.5 ml Eppendorf tubes on ice. Ten micro liter of each fraction were used to determine the sucrose concentration in the fraction, 85µl were used each for Western blotting and lipid mass spectrometry. In case of electron microscopy (EM) analysis, 10µl of each fraction were fixed overnight at 4◦C in 0.5% of EM grade glutaraldehyde/4% paraformaldehyde in 0.12 M sodium phosphate buffer, pH 7.2.

# Protein and Lipid Precipitations from Sucrose Samples

For immunoblotting, samples were subjected to methanolchloroform precipitation by mixing samples in 1:4:1:3 ratio (sample:methanol:chloroform:water). Protein was precipitated by gently vortexing the tube followed by centrifugation at 13,000 rpm for 5 min at room temperature (r.t.). The upper layer was removed without touching the interface layer. Twohundred and fifty-five liter of methanol were added (3 × volume of the original sample volume, 85µl). Proteins were pelleted by centrifugation at 13,000 rpm for 5 min at r.t. and resuspended in 2x concentrated SDS Laemmli sample buffer. Protein samples were denatured by boiling at 95◦C for 5 min and stored at −20◦C.

For lipid extraction, 85µl sample from each fraction were transferred into a 1.5 ml Eppendorf tube filled with 157 mM of ammonium bicarbonate (NH4HCO3) up to 1.5 ml. Particles were pelleted in the TLA55 fixed-angle rotor (Beckman coulter) at 30,000 rpm for 4 h at 4◦C. Removed NH4HCO<sup>3</sup> and pelleted lipid samples were frozen in liquid nitrogen and stored at −80◦C.

#### Lipid Extraction for Lipid Mass Spectrometry

Lipid samples were mixed with 200µl of 157 mM of NH4HCO3 and vortexed at 1400 rpm at 4◦C. Ten micro liter of an internal standard lipid mix (Avanti Polar lipids; PC, PE, PS, PG, PA, Cer, SM, GalCer, LacCer, Chol, and DAG) was added to each sample. To extract all lipids, 1 ml of a 2.75:1 mixture of chloroform/methanol was added to each tube and vortexed at 1400 rpm for 1 h at 4◦C. Fractions with high sucrose content (>1 M) were "pre-washed" with 200µl of MS grade water before the extraction to remove excess sucrose. After extraction, sample tubes were centrifuged at 1000 g for 2 min at 4◦C. Bottom phases were transferred into fresh 1.5 ml Eppendorf tubes (10:1 lipid fraction). The organic solvent was evaporated using a vacuum centrifuge. Dried samples were re-suspended with 100µl of a 1:2 mixture of chloroform/methanol and vortexed for 10 min at 4◦C. This re-suspension solution was mixed with MS compatible polarity specific buffers to optimize for ionization of specific lipids as described previously (Carvalho, 2012). The data was acquired on a LTQ-Orbitrap using high mass resolution R<sup>m</sup> <sup>−</sup> <sup>z</sup> <sup>=</sup> <sup>400</sup> = 100, 000 and the spectra was analyzed with LipidXplorer software (Herzog et al., 2011, 2012).

#### Immunofluorescence, Electron Microscopy, Annexin V-Cy5 Injection, and Image Acquisition

Living embryos at E11.5 were dissected out of the uterus with the placenta intact and transferred into 1x Tyrode solution (Sigma-Aldrich). AnnexinV-Cy5 solution (BioVision) was diluted to 1:300 in 1x binding buffer (10 mM Hepes-NaOH pH7.4 containing 140 mM NaCl and 2.5 mM CaCl2) and injected into telencephalic vesicles of living embryos (1x concentrated binding buffer was used as control). Injected embryos were incubated at r.t. for 30 min (room air and protected from light). Whole embryos were fixed for 20–25 h at 4◦C in 4% paraformaldehyde in 0.12 M sodium phosphate buffer, pH 7.2. For cryosectioning, embryos were equilibrated for 12–24 h at 4 ◦C in 30% sucrose in PBS, embedded in Tissue-Tek O.C.T compound (Sakura Finetek), and stored at −20◦C. Cryostat sections (10µm) were rehydrated with PBS, quenched with 50 mM NH4CL for 30 min, washed with 0.01% Digitonin (Sigma-Aldrich) in PBS for 30 min and blocked with 3% BSA in 0.01% Digitonin/PBS for 30 min at r.t. Sections were washed with 1% BSA in 0.01% Digitonin/PBS and incubated overnight at 4 ◦C with first primary antibody, followed by incubation for 1 h at r.t. with fluorescently labeled secondary antibodies and DAPI (4′ ,6-diamidino-2-phenylindole, Roche) in 1% BSA in 0.01% Digitonin/PBS. Sections were mounted in Mowiol. For electron microscopy, E11.5 dorsal telencephalon was analyzed by transmission EM, with pre- or post-embedding prominin-1 immunolabeling.

The following primary antibodies were used: rabbit anti-Prominin-1 [αD, MPI-CBG; 1:20,000 for immunoblotting (IB)]; goat antibody against apolipoprotein A1 (Biodesign; 1:1000); mouse monoclonal antibodies against CRIK [BD bioscience; 1:200 for immunofluorescence (IF) and 1:4000 for IB], α-tubulin (Sigma; 1:300 for IF and 1:60,000 for IB) and Aurora B (BD bioscience; 1:300) and rat anti-Prominin-1 (13A4, MPI-CBG; 1:300 for IF). Secondary antibodies were: horseradish peroxidasecoupled antibodies (Jackson laboratory), Alexa Fluor 488, 555, and 647 labeled IgG antibodies (Invitrogen).

Fluorescence images were acquired by confocal microscopy (Zeiss LSM 700) and images were acquired with a Zeiss Plan-Apochromat 20x, 0.8 NA and a Zeiss C-Apochromat 40x, 1.2 NA objectives.

#### Live-cell Time-lapse Microscopy

The protocol for this time-lapse microscopy has been described previously (Ettinger et al., 2011). Briefly, Neuro-2a MKLP1- GFP cells were grown on chambered coverslips (Lab-Tek, Nunc) in DMEM-GlutaMax medium supplemented with 10% FCS and incubated for 24 h at 37◦C with 5% CO2. The medium was replaced with Annexin V-Cy5 containing medium (1:5000) and cells were imaged with a Zeiss Axiovert 200 M inverted epifluorescence microscope equipped with a motorized stage, incubator, humidifier, and carbon dioxide supply (Visitron Systems, Germany). Bright-field and fluorescence images were acquired using a C-Apochromat 40x water immersion objective (NA 1.4) and recorded with a Roper Scientific Cool SNAP ES CCD camera using MetaMorph software. Multiple fields were acquired in parallel per experiment at 12 min intervals for up to 24 h.

#### Annexin V-Cy5 Incubation with the Neuro-2a-MKLP1 Transgenic Cell Line

Neuro-2a MKLP1-GFP cells were grown on poly-L-lysine coated coverslips in DMEM-GlutaMax medium supplemented with 10% FCS and incubated for 24 h at 37◦C with 5% CO2. The medium was replaced with Annexin V-Cy5 containing medium (1:2500), incubated for 10 min at r.t. protected from light, and fixed with 4% paraformaldehyde in 0.12 M sodium phosphate buffer, pH 7.2. Cells were further processed for immunofluorescence like sections of mouse embryos.

# Author Contributions

YA designed and performed the experiments, analyzed the data and co-wrote the manuscript, JS designed and performed lipid mass spectrometry, analyzed the data and co-wrote the manuscript, MB performed EM and co-wrote the manuscript, AE helped and performed time-lapse imaging, and co-wrote the manuscript, CH performed experiments and WH analyzed data, co-wrote the manuscript and supervised project. The authors declare that they have no competing financial interests.

# Acknowledgments

We thank the animal and light microscopy facilities of MPI-CBG for excellent support, Drs. Vitary Matiash, Dominik Schwudke, Andrej Shevchenko, and Kai Simons for helpful discussion at the beginning of the lipid mass spectrometry and project. YA was supported by ARC (Association pour la Recherche sur le Cancer), FRM (Fondation pour la Recherche Médicale), and JSPS (Japan Society for the Promotion of Science) and WH was supported by grants from the DFG (SFB 655, A2; TRR 83, Tp6) by the DFGfunded Center for Regenerative Therapies Dresden, the Fonds der Chemischen Industrie and ERC.

# Supplementary Material

The Supplementary Material for this article can be found online at: http://journal.frontiersin.org/article/10.3389/fncel. 2015.00325

# References


Supplementary Figure 1 | Phosphatidylserine is on the outer-leaflet of the apical plasma membrane of the neuronal progenitor cells. (A,B) Micrographs of mouse embryonic E11.5 coronal cryosections of dorsal telencephalon injected with Annexin V-Cy5 binding buffer. Sections were consecutively stained by immunofluorescence for prominin-1 (green) and counterstained with DAPI (white). Images are z-projections of 32 consecutive 0.38-µm optical sections. (B) Higher magnifications of the apical plasma membrane of the ventricular zone [boxed region indicated in (A)]. Red-white arrows, Annexin V-Cy5 single positive particles, green arrows, prominin-1 single positive midbody particles, yellow arrows, Annexin V-Cy5 and prominin-1 double positive midbody particles. Scale bars, 50µm (A) and 10 µm (B). (C) Percentage of three different types of particles on the apical plasma membrane of the ventricular zone, classified by decoration with prominin-1 positive and Annexin V-Cy5 positive signal. Data are the mean of six 213-µm wide fields from three different embryos and litters; total number of counted particles, 455 for prominin+ AnnexinV–, 354 for prominin+ AnnexinV+ and 16 for prominin– AnnexinV+; error bars indicate SD; <sup>∗</sup><sup>P</sup> <sup>&</sup>lt; <sup>0</sup>.05, ∗∗∗∗<sup>P</sup> <sup>&</sup>lt; <sup>0</sup>.0001; unpaired t-test.

Supplementary Figure 2 | Expression of midbody markes in MKLP1-GFP Neuro-2a cells. (A,B) Fluorescence for MKLP1-GFP (green),

immunofluorescence for α-tubulin and Aurora B kinase (red) in Neuro-2a MKLP1-GFP cells during telophase (A) and interphase (B) combined with DAPI staining (white). Images are maximum intensity projections of z-stacks of optical sections taken every 0.38µm. Insets, magnifications of bilateral (A) and monolateral (B) midbodies. Scale bars, 10 µm and 1 µm (inset).

Supplementary Figure 3 | Expression of midbody markers in MKLP1-GFP HeLa cells. (A) Fluorescence for MKLP1-GFP (green) and immunofluorescence for α-tubulin in telophase (2 upper panels) and interphase (lower panels) cells combined with DAPI staining (white). Scale bars, 10µm. (B) Fluorescence for MKLP1-GFP (green) and immunofluorescence for Aurora B kinase in telophase (upper panels) and interphase (lower panels) cells having midbody combined with DAPI staining (white). Scale bars, 10µm. Insets, magnifications of newly formed midbodies with telophase-bridge between the two daughter cells intact [upper panels in (A,B)] or inherited, monolateral midbodies with no apparent cytoplasmic bridge to the other daughter cell [lower panels in (A,B)]. Scale bars, 1 µm.


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Arai, Sampaio, Wilsch-Bräuninger, Ettinger, Haffner and Huttner. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Regulation of cerebral cortex development by Rho GTPases: insights from *in vivo* studies

# *Roberta Azzarelli 1, Thomas Kerloch2,3 and Emilie Pacary2,3\**

*<sup>1</sup> Department of Oncology, Hutchison/MRC Research Centre, Cambridge Biomedical Campus, University of Cambridge, Cambridge, UK*

*<sup>2</sup> Institut National de la Santé et de la Recherche Médicale U862, Neurocentre Magendie, Bordeaux, France*

*<sup>3</sup> Institut National de la Santé et de la Recherche Médicale, Physiopathologie de la Plasticité Neuronale, Université de Bordeaux, Bordeaux, France*

#### *Edited by:*

*Takeshi Kawauchi, Keio University School of Medicine / Precursory Research for Embryonic Science and Technology (PRESTO), Japan Science and Technology Agency (JST), Japan*

#### *Reviewed by:*

*Troy Ghashghaei, North Carolina State University, USA Angeliki Louvi, Yale, USA*

#### *\*Correspondence:*

*Emilie Pacary, Neurogenesis and Physiopathology Group, INSERM U862 - Neurocentre Magendie, 146 rue Léo Saignat, 33077 Bordeaux cedex, France e-mail: emilie.pacary@inserm.fr*

The cerebral cortex is the site of higher human cognitive and motor functions. Histologically, it is organized into six horizontal layers, each containing unique populations of molecularly and functionally distinct excitatory projection neurons and inhibitory interneurons. The stereotyped cellular distribution of cortical neurons is crucial for the formation of functional neural circuits and it is predominantly established during embryonic development. Cortical neuron development is a multiphasic process characterized by sequential steps of neural progenitor proliferation, cell cycle exit, neuroblast migration and neuronal differentiation. This series of events requires an extensive and dynamic remodeling of the cell cytoskeleton at each step of the process. As major regulators of the cytoskeleton, the family of small Rho GTPases has been shown to play essential functions in cerebral cortex development. Here we review *in vivo* findings that support the contribution of Rho GTPases to cortical projection neuron development and we address their involvement in the etiology of cerebral cortex malformations.

**Keywords: Rho GTPases, cerebral cortex, neuronal development, cortical malformations, GAP, GEF**

#### **INTRODUCTION**

In humans, the cerebral cortex is responsible for the processes of thought, perception and memory and serves as the seat of advanced motor functions, social abilities and language. These functions relies on the proper development of two main populations of neurons, projection or pyramidal neurons, which are glutamatergic and excitatory, and interneurons, which are GABAergic and inhibitory. Neurons in the cerebral cortex are arranged into six distinct layers different in terms of connectivity, gene expression profile and birthdate (Molyneaux et al., 2007). Projection neurons originate from progenitors located in the cortex whereas interneurons are born in the ventral domains of the telencephalon and then migrate tangentially to reach the cortex (Kriegstein and Noctor, 2004).

After closure of the neural tube, the epithelium lining the ventricles becomes a specialized neuroepithelium. It consists of a single sheet of progenitor cells, called neuroepithelial cells (NEs). At early developmental stages, around E10 in mouse, NEs along the dorsal surface of lateral ventricles self-renew to expand the progenitor pool and then convert into cells expressing glial markers such as the astrocyte-specific glutamate transporter (GLAST) and brain lipid-binding protein (BLBP), the radial glial cells (RGs). The asymmetric divisions of these RGs are responsible for producing cortical projection neurons either directly or indirectly through intermediate progenitor cells (IPs) or outer radial glial cells (oRGs). Newborn projections neurons then migrate radially in a step-wise fashion to their final destination using RG fibers as a scaffold and finally undergo terminal differentiation to transmit and receive information. As neurogenesis progresses, diverse subtypes of projection neurons are generated sequentially and their migration occurs in an inside-out manner: neurons generated first occupy the deepest layers of the future six layered neocortex whereas later born neurons by-pass earlier born neurons and settle in more superficial layers (Gupta et al., 2002).

Unlike the projection neurons, the inhibitory interneurons of the cerebral cortex are generated from distinct progenitors in the germinal zones of the ventral telencephalon, mainly within the medial and caudal ganglionic eminences. Interneurons then undertake tangential migration toward the cortex using different routes according to the time and place of birth. Upon arrival in the cortex, interneurons switch their mode of migration from tangential to radial and reach their destination layer based on their molecular subtype, origin and birthdate, on cortical cues and make local connections with pyramidal cells (Bartolini et al., 2013; Guo and Anton, 2014).

The development of each category of cortical neurons is thus a multistep process, which involves dramatic morphological changes at each step of the process. These changes are mediated by an extensive and dynamic remodeling of the cytoskeleton. Among the major regulators of cytoskeletal properties, the small signaling molecules of the Rho GTPase family play essential functions in cerebral cortex development. In this review, we will address the role of Rho GTPases in cortical projection neuron development, focusing on *in vivo* studies, and discuss how dysfunctional Rho GTPase signaling may contribute to cortical malformations.

## **THE Rho GTPase FAMILY**

The Rho family of GTPases represents a subgroup of the Ras superfamily of small GTP binding proteins (Heasman and Ridley, 2008). The most extensively studied members of the Rho family are RhoA (Ras homologous member A), Rac1 (ras related C3 botulinum toxin substrate 1) and Cdc42 (cell division cycle 42) but this family actually includes 20 members which are subdivided into 8 subgroups based on amino-acid sequence similarities (**Figure 1**).

Like other small GTP-binding proteins of the Ras superfamily, most Rho GTPases cycle between GTP (active) and GDP (inactive)—bound states. The GDP/GTP cycle is promoted by the activity of two classes of molecules, guanine nucleotide exchanging factors (GEFs) and GTPase activating proteins (GAPs). GEFs facilitate the exchange of GDP with GTP, resulting in protein activation. GAPs instead stimulate the intrinsic enzymatic activity of the GTPases, which promotes hydrolysis of GTP into GDP. GAP activity therefore ends the cycle and returns the GTPases in their inactive state (Bos et al., 2007) (**Figure 2**). Over 80 GEFs and more than 70 GAPs have been reported, suggesting that Rho GTPase regulation is complex. In addition, Rho GTPases can bind to proteins known as guanine-nucleotide dissociation inhibitors

**FIGURE 1 | Members of the Rho GTPase family.** The column identity indicates the percentage of amino-acid sequence identity of a specific Rho GTPase compared with the first member of the corresponding subfamily (Heasman and Ridley, 2008). Among the atypical members, Rnd1, Rnd2, Rnd3, RhoH, RhoBTB1, and RhoBTB2 lack amino acids in the GTPase domain that are critical for GTPase activity, they are thus constitutively bound to GTP and do not detectably hydrolyse GTP (deficient Rho GTPase domain in dark blue). RhoV and RhoU harbor GTPase activity but they are atypical in this family as they display a high intrinsic guanine nucleotide exchange activity and are predominantly in the GTP-loaded conformation. In the C-terminal domain, the Hyper Variable Region (HVR in red) differs not only between the Rho GTPase subclasses but also within the same subclass in terms of the presence of either a polybasic region or a palmitoylation site (Roberts et al., 2008). The polybasic region and palmitoylation site present in the HVR are involved in targeting the GTPases to plasma membrane or

endomembrane compartment. The C-terminal CAAX-box (C, cysteine; A, Aliphatic Amino acid; X, any amino acid; in orange) contains a cysteine residue, which is crucial for prenylation that adds a farnesyl or geranylgeranyl group, enhancing the interaction with membranes and very often defining the localization to specific membrane compartments. The CAAX with a ∗ indicates that RhoV does not seem to have a functional CAAX box and the CAAX motif of RhoU is apparently not undergoing prenylation (Aspenstrom et al., 2007). The Rif members have a N-terminal extension (in yellow) that is unique to this subgroup. RhoUV proteins display a proline rich motif (PRM, in green), which is also present in RhoBTB1 and RhoBTB2. RhoBTB proteins contain two Broad complex/Tramtrack/Bric-a-brac domains (BTB). RhoBTB family members harbor different domains involved in protein-protein interaction (in gray: Coiled Coil, CC, only present in RhoBTB1 and PEST domain only present in RhoBTB2) and they display a nuclear localization signal (NLS, in pink).

(GDIs). RhoGDIs sequester RhoGTPase in their inactive state and protect them from degradation (Boulter et al., 2010) (**Figure 2**). When bound to GTP, Rho GTPases exhibit the correct structural conformation to interact with effectors and initiate downstream signaling to regulate actin and microtubule components of the cytoskeleton (Jaffe and Hall, 2005) (**Figure 2**). However, some members of the family do not follow this classical scheme of activation and are described as atypical. These atypical Rho GTPases are predominantly GTP bound, owing either to aminoacid substitutions at residues that are crucial for GTPase activity (for example in Rnd proteins) or owing to increased nucleotide exchange (for example in RhoU). Therefore, their expression, localization, stability and phosphorylation control their activity rather than the GDP/GTP switch (Aspenstrom et al., 2007) (**Figure 1**).

Experimental data indicating the importance of the Rho family of small GTPases in cerebral cortex development have been accumulated over the past few years. Much of our understanding on their role comes from *in vitro* studies (Govek et al., 2005). Nevertheless, in the last years, the use of conditional mutant mice and the development of techniques such as *in* *utero* electroporation have allowed to highlight and clarify their functions *in vivo*.

#### **EXPRESSION OF Rho GTPases IN THE DEVELOPING CEREBRAL CORTEX**

Most of these *in vivo* studies have focused on RhoA, Rac1, and Cdc42, and more recently on Rnd2 and Rnd3. The functions of the other members of the Rho GTPase family in cortical development remain largely unknown.

In the Rho subgroup, *RhoA* and *RhoB* are highly expressed in the embryonic cerebral cortex but with distinct patterns (Olenik et al., 1999; Ge et al., 2006; Heng et al., 2008). *RhoA* mRNA is mainly expressed in domains of cellular proliferation whereas *RhoB* mRNA is absent in the proliferative zones but highly expressed in the cortical plate (CP) where neurons migrate or settle at the end of their migration (**Figure 3**). *RhoC* mRNA is detected in the nervous system but its distribution in the developing cerebral cortex has not been thoroughly examined (Erschbamer et al., 2005). Three of the 4 vertebrate Racrelated genes, namely *Rac1*, *Rac3*, and *RhoG*, are expressed in the nervous system (**Figure 3**) (de Curtis, 2008). Most studies

**cortex.** Schematic representation of cortical domains depicting the expression pattern of Rho GTPase genes within the murine cerebral cortex at different developmental stages. To simplify the representation, this figure does not include spatial differences, which should be however kept in mind since all cortical areas do not develop at the same rate and timing. References cited in the Section EXPRESSION OF Rho GTPases IN THE

expression patterns and changes in expression during development. Dark gray and light gray indicate higher and lower relative levels of expression, respectively. VZ: ventricular zone; SVZ: ventricular zone; IZ: intermediate zone; CP: cortical plate; WM: white matter; PP: preplate; n.e.: not expressed in the brain; n.d.: expression in the cerebral cortex not determined.

on cortical neuron development have focused on Rac1, whereas only a few studies exist on Rac3 and RhoG, despite their expression in the developing cerebral cortex. However, their temporal expression is different as illustrated by low levels of *Rac3* mRNA in the embryonic cortex where *Rac1* and *RhoG* mRNA are highly expressed (Ishikawa et al., 2002; Bolis et al., 2003; Corbetta et al., 2005; Fujimoto et al., 2009). Instead, *Rac3* is mainly expressed in the postnatal cortex (P7) in layer V and to a lesser extent in layers II-III (Corbetta et al., 2005). In the Cdc42 subfamily, Cdc42 is expressed throughout the developing cerebral wall (Olenik et al., 1999; Yokota et al., 2010). Cdc42 protein is particularly enriched at the apical/ventricular side of the neuroepithelium and is present in basally located post-mitotic neurons (Cappello et al., 2006). In contrast to Cdc42, *TC10* expression is very low in the embryonic cerebral cortex (**Figure 3**) (Tanabe et al., 2000). Its expression in the brain however increases with development (Tanabe et al., 2000; Abe et al., 2003). Similarly, very low levels of *TCL* mRNA are detected in the brain, at least in adult murine tissue (Vignal et al., 2000). Concerning the last subfamily of classical Rho GTPases, although *RhoF* and *RhoD* are expressed in the adult brain (Murphy et al., 1996; Ellis and Mellor, 2000) and have been shown to regulate neuronal development *in vitro* (Hotulainen et al., 2009; Gad and Aspenstrom, 2010), their expression pattern in the developing cerebral cortex has not been described.

Among the atypical Rho GTPases, the expression profile of Rnd members in the developing cerebral cortex has been the most well-characterized (**Figure 3**). At E14.5, *Rnd1* mRNA levels are very low in the cortex but they gradually increase to peak at postnatal stages (Ishikawa et al., 2003). Conversely, the expression of *Rnd2* and *Rnd3* is high in the embryonic cerebral cortex but show different distribution in the cortical domains (**Figure 3**) (Azzarelli et al., 2014). *RhoV* mRNA is expressed in the human fetal brain (Katoh, 2002) and *RhoU* mRNA is found in the adult human cerebral cortex (Kirikoshi and Katoh, 2002) but their expression in the developing cerebral cortex has not been explored. Like Rac2, the expression of RhoH is restricted to hematopoietic stem cells (Troeger and Williams, 2013). Finally, the three members of the RhoBTB subfamily are expressed in the brain with *RhoBTB3* showing the highest expression levels in the adult tissue (Ramos et al., 2002). However, RhoBTB3 is not included in the Rho GTPase family since it does not seem to have a GTP-binding domain, at least it does not contain a consensus GTP-binding motif (Aspenstrom et al., 2007).

#### **Rho GTPases AND REGULATION OF CORTICAL PROJECTION NEURON DEVELOPMENT**

#### **Rho GTPases AND REGULATION OF ADHERENS JUNCTION INTEGRITY**

Like NEs, RGs are highly polarized along their apico-basal axis. They are attached to the luminal surface of the ventricle on their apical side, where they form adherens junctions (AJs) with neighboring RGs (**Figure 4**, ❶), and to the basal lamina via integrins (Gotz and Huttner, 2005). Their cell bodies are retained within the ventricular zone (VZ), a defined region next to the ventricles. AJs between RGs maintain VZ integrity and cortical architecture as well as RG behavior by anchoring a variety of proteins (N-cadherin, β-catenin, αE-catenin) to the actin cytoskeleton (Gotz and Huttner, 2005).

RhoA plays a critical role in the maintenance of these adherens junctional complexes. Indeed, the deletion of *RhoA* by *FoxG1Cre* (Katayama et al., 2011) or *Emx1-Cre* mediated recombination (Cappello et al., 2012) leads to a disorganization of the VZ surface and to a loss of catenin expression at the apical surface around E14.5. In these mutants, rings of intense catenin expression are instead observed inside the brain mass (Katayama et al., 2011; Cappello et al., 2012). Similarly, perturbation of Rho by electroporation either with the RhoA/B/C inhibitor C3 transferase or with *RhoA*, *RhoB*, *RhoC* shRNAs impairs the apical actin filament belt and the apico-basal polarity of electroporated cells (Thumkeo et al., 2011). Expression of dominant-active Rho also affects the actin structure and the apical localization of Ncadherin, suggesting that balanced Rho activity is necessary for maintaining AJ integrity in RGs (Thumkeo et al., 2011) (**Table 1**). While Rho is essential for the maintenance, Cdc42 seems to be essential for the initial formation of apical AJs since the disappearance of apical proteins (PAR6, aPKC, E-cadherin, β-catenin, F-actin and Numb) as well as the apico-basal polarity occurs as early as E10.5 in *FoxG1Cre Cdc42* null embryos (Chen et al., 2006). A similar phenotype is found after deletion of *Cdc42* by *Emx1Cre* or *Nestin-Cre* mediated recombinations (Cappello et al., 2006; Garvalov et al., 2007; Peng et al., 2013) (**Table 1**). Although Rac1 and Cdc42 share many effectors, junction formation and cell polarity during cerebral cortex development specifically require Cdc42 but not Rac1. In Rac1 mutants, VZ progenitors are not tightly packed or radially oriented as in controls. However, this phenotype is not due to an alteration of VZ progenitor polarity since the loss of *Rac1* does not affect the expression pattern of β-catenin and cadherin (Leone et al., 2010). Accordingly, the expression of phosphorylated PAK (p21- activated kinase), which is a direct downstream effector of both Rac1 and Cdc42, at the apical surface of VZ progenitors is affected by the loss of *Cdc42* but not *Rac1* (Leone et al., 2010), further demonstrating that these two Rho GTPases perform non-overlapping and nonredundant functions in the VZ. More recently, the atypical Rho GTPase Rnd3 has also been shown to maintain the integrity of the junctions between RGs through regulation of RhoA and the actin cytoskeleton (Pacary et al., 2013) (**Table 1**).

#### **Rho GTPases AND REGULATION OF INTERKINETIC NUCLEAR MIGRATION**

In addition to apico-basal polarity, another hallmark of NEs retained by RGs is interkinetic nuclear migration (INM), a process whereby nuclei change position along the apico–basal axis during the course of the cell cycle. In NEs, this INM spans the entire apical–basal axis of the cell, with the nucleus migrating to the basal side during the G1 phase of the cell cycle, staying at the basal side during S phase, migrating back to the apical side during the G2 phase and undergoing mitosis at the ventricular surface. In RGs, the same mitotic behavior occurs, except that is confined to the portion of the cell in the VZ (**Figure 4**, ❷). As a consequence of this movement, the neuroepithelium and the VZ appear pseudo-stratified (Taverna and Huttner, 2010). The precise role of INM during cortical neurogenesis is still an unresolved question. INM might allow packing an increasing number of progenitor cells within a limited ventricular surface or it might regulate progenitor fate by influencing the exposure of progenitor nuclei to proliferative vs. neurogenic signals (Taverna and Huttner, 2010; Spear and Erickson, 2012).

Both microtubule-based motors and actomyosin seem to participate in either direction of INM, although to a different extent depending on the system (Taverna and Huttner, 2010; Lee and Norden, 2013). In the developing cerebral cortex, a few studies have implicated the Rho GTPases in the regulation of this process.

**FIGURE 4 | Development of projection neurons in the mouse cerebral cortex.** The neural stem/progenitor cells of the cerebral cortex or radial glial cells (RGs) are highly polarized cells that are attached to one another in the ventricular zone (VZ) by apically located adherens junctions (AJ) ❶. Their nuclei migrate during cell cycle progression from a basal position during S phase to an apical position during mitosis (M), and the nuclei of the daughter cells migrate back to enter S phase on the basal side of the VZ, in a process called interkinetic nuclear migration (INM) ❷. During the peak of neurogenesis, most radial glial cells divide asymmetrically with a vertical cleavage plane ❸. In these divisions, one daughter remains a RG and continues to divide at the ventricular surface, whereas the other detaches from the ventricular surface, move radially away to the subventricular zone (SVZ)/lower intermediate zone (IZ) and acquires a multipolar shape ❹. Then, nascent neurons become bipolar, extending a leading process toward the pial surface and a trailing process in the opposite direction ❺. Upon multi to

In particular Cdc42, Rac1, and Rnd3 have been shown to control the basal-to-apical movement. Indeed, this movement is delayed in *Cdc42* deficient mice (Cappello et al., 2006) or after electroporation of a dominant negative (DN) form of Rac1 (Minobe et al., 2009) (**Table 1**). However, it is not known whether microtubule or actomyosin networks mediate the effects of Rac1 and Cdc42. In contrast, the impairment of basal to apical INM after *Rnd3* silencing in the embryonic cerebral cortex is rescued by coexpression of a constitutively active form of cofilin (cofilinS3A), demonstrating that Rnd3-mediated disassembly of actin filaments coordinates the cellular behavior of RGs during INM at least during the apical nuclear movement (Pacary et al., 2013).

#### **Rho GTPases AND REGULATION OF PROGENITOR CELL DIVISION, PROLIFERATION AND CELL FATE**

At early stages of corticogenesis, NEs divide symmetrically to self-renew and expand their pool. Following the transition to the bipolar transition, newborn neurons establish contacts with RG fibers and subsequently use them as a scaffold to migrate to the upper part of the cortical plate (CP) using a mode of migration called locomotion ❻. During this phase the trailing process becomes the axon and extends to its final destination. Once cortical neurons reach the upper part of the CP and right after their leading process makes contact with the marginal zone (MZ), they detach from the RG fibers and execute a terminal somal translocation ❼. The leading process then gives rise to the apical dendrite, which initiates local branching in the MZ ❽. Basal dendrites subsequently appear as well as oblique side branches emerging from the apical shaft ❾. At this stage, the cell body of early-born neurons translocate ventrally as neurons born at later stages bypass their predecessors. The final step in cortical projection development is the apparition and maturation of spines. For example, in layer V pyramidal neurons, spines are morphologically mature at P21 on apical dendrites ❿.

RG fate, some progenitor cells begin to divide asymmetrically to generate neurons directly or indirectly through the production of IPs or oRGs (Laguesse et al., 2014).

After their generation by asymmetric division of RGs, IPs, also called basal progenitors, retract their apical and basal processes, exhibit a multipolar morphology and migrate basally (**Figure 4**, ❸ and ❹) before they undergo mitosis. This second pool of proliferative progenitors undergo one or more symmetric cell divisions (Noctor et al., 2004), which significantly increases the yield of cortical neurons derived from a single RG. The accumulation of these dividing progenitors in basal regions starts around E13 and it determines the formation of the subventricular zone (SVZ). Whereas NEs and RGs express identical markers like the transcription factor Pax6 (Gotz et al., 1998), IPs are identified by the absence of Pax6 and by the expression of the transcription factor Tbr2 (Englund et al., 2005).

#### **Table 1 | Regulation of cortical projection neuron development by Rho GTPases (***in vivo* **studies).**


#### **Table 1 | Continued**


*VZ: ventricular zone, SVZ: subventricular zone, IZ: intermediate zone, CP: cortical plate, INM: interkinetic nuclear migration, CA: constitutively active, DN: dominantnegative.*

oRGs, also known as basal radial glia cells, were first discovered in human and ferret brains (Fietz et al., 2010; Hansen et al., 2010), and were initially proposed to be a specific feature of gyrencephalic brains. However, they have also been identified in the rodent brain, where they account for less than 10% of total cortical progenitors vs. 40% in human (Hansen et al., 2010; Wang et al., 2011), and in lissencephalic primate brains (Garcia-Moreno et al., 2012; Kelava et al., 2012). oRGs arise from the division of RGs as they delaminate from the apical surface and translocate their nuclei in the outer portion of the SVZ, where they start dividing. To note, this translocation of the soma along the basal fiber toward the CP, a process termed mitotic somal translocation, requires activation of the Rho effector ROCK (Ostrem et al., 2014). In contrast to IPs, oRGs maintain molecular characteristics of RGs such as expression of Pax6 and can divide either symmetrically to expand their number or asymmetrically to self-renew and give birth to new neurons (Hansen et al., 2010; Reillo et al., 2011).

The balance between proliferation and differentiation of these different categories of progenitors is tightly regulated and is fundamental for the generation of appropriate number of cortical neurons. Among the Rho GTPases, RhoA, Rac1, RhoG, and Rnd3 have a crucial role in the control of this balance. Indeed, conditional deletion of *RhoA* in cortical progenitors using *FoxG1Cre* mice causes hyperproliferation, which results in the expansion of the progenitor pool and exencephaly-like protrusions (Katayama et al., 2011). Similarly, the loss of *RhoA* by *Emx1-Cre* mediated recombination increases proliferation in a region-specific manner within the cerebral cortex, starting at occipital regions at E14 and later at E16 in rostral parts, and this phenotype is associated with an aberrant location of Pax6+ and Tbr2+ progenitors (Cappello et al., 2012) (**Table 1**). In contrast to RhoA, the forebrain-specific loss of *Rac1* by *FoxG1Cre* leads to a SVZ-specific reduction in proliferation, a concomitant increase in cell cycle exit and premature differentiation (Chen et al., 2009; Leone et al., 2010) (**Table 1**). How RhoA and Rac1 differently affect proliferation of cortical progenitors is not known. However, studies in other cellular systems have shown that RhoA and Rac1 influence the levels of cyclins during G1 progression. Interestingly, Rac1, but not Rho, stimulate cyclinD1 transcription when ectopically expressed in cells (Jaffe and Hall, 2005). In addition, RhoA might regulate proliferation of cortical progenitors during cytokinesis through its action on F-actin and myosin II into the actomyosin contractile ring (Marzesco et al., 2009) or through its action on actomyosin filaments at the cell cortex which influence mitotic spindles and the plane of cell division (see next paragraph) (Jaffe and Hall, 2005). Rnd3 has also been shown to control specifically the proliferation of basal progenitors via cyclinD1 but in an opposite manner, i.e., *Rnd3* silencing increases SVZ proliferation (Pacary et al., 2013). Finally, RhoG, another Rac-related Rho GTPase expressed in the VZ/SVZ (**Figure 3**), also promotes neural progenitor cell proliferation in the mouse cerebral cortex (**Table 1**) through phosphatidylinositol 3-kinase (PI3K) signaling (Fujimoto et al., 2009).

In the developing cerebral cortex, cleavage plane orientation remains predominantly vertical (planar division) during the period of symmetrical division prior to neurogenesis, and throughout the period of asymmetrical division during neurogenesis (**Figure 4**, ❸) (Morin and Bellaiche, 2011). In these asymmetric divisions, one daughter remains a RG and continues to divide at the ventricular surface, whereas the other loses its apical attachment and becomes an IP. Fewer RGs undergo oblique or horizontal divisions, and these divisions have been proposed to generate oRGs (Morin and Bellaiche, 2011; Shitamukai and Matsuzaki, 2012).

In the mouse cerebral cortex, Rnd3 is required to maintain the vertical orientation of the cleavage plane during RG divisions. Indeed, when *Rnd3* is knockdown the fraction of RGs dividing with an oblique or horizontal cleavage plane is increased (Pacary et al., 2013). Interestingly *Rnd3*-silenced cells prematurely leave the VZ, enter the SVZ while transiently maintaining their radial glial molecular phenotype, thus showing similarities with oRGs (Pacary et al., 2013). In addition, co-electroporation of cofilinS3A restores vertical cleavage-plane orientation in *Rnd3* silenced progenitors, thus indicating that Rnd3 maintains the vertical orientation of apical divisions by remodeling the actin cytoskeleton (Pacary et al., 2013). RhoA might be also important to determine the orientation of cortical progenitor divisions as suggested by a study in the chick neuroepithelium, in which the expression of a DN form of RhoA results in random spindle orientation (Roszko et al., 2006). Inversely, deletion of *Cdc42* does not influence spindle orientation in the developing cerebral cortex (Cappello et al., 2006). However, as mentioned previously, the loss of *Cdc42* causes defects in apical process maintenance, thereby leading to an increased number of progenitors dividing at basal rather than apical positions. These progenitors convert to an SVZ fate as shown by the increase of Tbr2+ progenitors and a decrease of Pax6+ population in the mutants, which ultimately leads to a higher rate of neuron generation (Cappello et al., 2006).

#### **Rho GTPases AND REGULATION OF RADIAL MIGRATION**

After detachment from the ventricular surface in the VZ, nascent neurons move radially away to the SVZ/lower intermediate zone (IZ), where they acquire a multipolar shape (**Figure 4**, ❹). During this phase, multipolar neurons actively extend and retract dynamic processes and tend to migrate tangentially in an apparent random fashion (Noctor et al., 2004; Jossin and Cooper, 2011). Then, neurons become bipolar, extending a leading process toward the pial surface and a trailing process in the opposite direction (nascent axon) (**Figure 4**, ❺). Upon multi to bipolar transition, neurons establish dynamic contacts with RG fibers and subsequently use them as a scaffold to migrate to the upper part of the CP using a mode of migration called locomotion (**Figure 4**, ❻). This movement is characterized by repetitive cycles of synchronized steps. First, a cytoplasmic dilation forms in the proximal region of the leading process. Second, the centrosome moves toward the swelling and finally the nucleus translocates toward the centrosome, a process known as nucleokinesis. This migration cycle then starts again confering a saltatory advancement to the locomoting neurons. Finally, once cortical neurons have reached the uppermost area of the CP and right after their leading process makes contact with the MZ, they detach from the RG fibers and execute a terminal somal translocation to settle in their appropriate final position (**Figure 4**, ❼) (Nadarajah et al., 2001). Rac1, Rnd2, and Rnd3 are three Rho GTPases with specific functions in the control of the migratory process in the developing cerebral cortex: Rac1 signaling regulates leading process formation (Kawauchi et al., 2003; Konno et al., 2005), Rnd2 is critical for the multi to bipolar transition (Heng et al., 2008; Pacary et al., 2011) and Rnd3 is important for nuclear-centrosome coupling during locomotion (Pacary et al., 2011).

Conditional knockout of *Rac1* in the forebrain, *in utero* electroporation of DN or constitutively active (CA) forms of Rac1 as well as *Rac1* shRNA or wild-type Rac1 have demonstrated a requirement for this Rho GTPase in radial migration (Kawauchi et al., 2003; Konno et al., 2005; Chen et al., 2007; Kassai et al., 2008; Yang et al., 2012a). *Rac1* deletion using the *FoxG1Cre* (Chen et al., 2007) or *Emx1Cre* line (Kassai et al., 2008) disturbs radial migration, but the defects observed in these mice are less severe than those observed after electroporation of the DN form of Rac1 (N17-Rac1). While the migration of nascent neurons seems to be only delayed in the mutants, the inhibition of Rac1 activity with N17-Rac1 leads to an accumulation of electroporated cells in the IZ (Kawauchi et al., 2003; Konno et al., 2005; Yang et al., 2012a). In addition, N17-Rac1 expressing cells in this cortical domain fail to extend a leading process and are instead round, with short and irregular processes (Kawauchi et al., 2003). Interestingly, the electroporation of *CA-Rac1* (*V12Rac1*) induces a similar phenotype (Konno et al., 2005; Yang et al., 2012a) indicating that cycles of Rac1 activation and inactivation and thus a fine regulation of Rac1 activity is important for proper morphological polarization and migration. The difference obtained between conditional knockout mice and the electroporation of dominant mutant might reflect the ability of this mutant as well as of CA mutant to interfere with the activity of other Rho GTPases, possibly through the competitive binding with regulators like RhoGDIs (Boulter et al., 2010). Nevertheless, the requirement of Rac1 for proper radial migration has been further confirmed recently with *Rac1* shRNA (Yang et al., 2012a). Indeed, the silencing of *Rac1* blocks radial migration and disrupts the formation of the proximal cytoplasmic dilation in the leading process of migratory cortical neurons (Yang et al., 2012a). Interestingly, in this study, the authors also show that electroporation of wild-type *Rac1* promotes neuronal migration and that POSH, a Rac1-interacting scaffold protein, recruits activated Rac1 to the plasma membrane. At this site, activated Rac1 regulates actin remodeling and controls the dilation of the leading process, two key events that promote centrosomal movement and soma translocation (Yang et al., 2012a). To note, another report provides the evidence that migration defects caused by loss of *Rac1* in *Foxg1Cre* mice may be due, at least in part, to defects in radial glial organization (Leone et al., 2010).

*In vivo* loss of function studies on the atypical Rho GTPase Rnd3 revealed that its knockdown in migrating neurons results in enlarged leading processes with numerous branches and increases centrosome-nucleus distance in the CP, indicative of disrupted nuclear-centrosome coupling during locomotion (Pacary et al., 2011). In contrast, *Rnd2*-deficient neurons fail to leave the IZ and display long processes at the multipolar stage, suggesting that Rnd2 is critical for the multipolar to bipolar transition that occurs in the IZ (Heng et al., 2008; Pacary et al., 2011). In addition, the two proteins fail to compensate for each other during neuronal migration, further indicating that they play distinct roles in this process (Pacary et al., 2011). Despite these different functions, both Rnd2 and Rnd3 regulate neuronal migration by inhibiting RhoA. Indeed, FRET analysis *in vivo* showed that *Rnd2* or *Rnd3* silencing increases RhoA activity in cortical cells and *RhoA* knockdown rescues the migratory defects associated with *Rnd2* or *Rnd3* loss of function. The inhibitory effect of Rnd3 on RhoA activity depends on its interactions with p190RhoGAP, whereas Rnd2's RhoA inhibitory activity does not. Further, although both Rnd2 and Rnd3 regulate actin dynamics in migrating neurons, only Rnd3 promotes neuronal migration by inhibiting RhoAmediated actin polymerization and remodeling (Pacary et al., 2011). Interestingly, the distinct subcellular localization of Rnd2 and Rnd3 and the resultant modulation of RhoA activity in different cell compartments underlie the difference in their effects. Rnd3 owes its distinct role in neuronal migration to its localization and interaction with RhoA at the plasma membrane. Rnd2 is expressed in early endosomes and can replace Rnd3 in migrating neurons if it is targeted to the plasma membrane by replacement of its carboxyl-terminal domain with that of Rnd3 (Pacary et al., 2011). However, the mechanisms by which Rnd2 promotes neuronal migration and inhibits RhoA remains unknown. The finding that Rnd3 and Rnd2 control different phases of radial migration by inhibiting RhoA in different cell compartments suggests that in cortical neurons, RhoA acts dynamically in different cellular domains to control different aspects of the migratory process.

The functions of the other most studied members of the Rho GTPase family, RhoA and Cdc42, in the control of radial migration are less well-understood. When *DN-Cdc42* (*N17Cdc42*) or *CA-Cdc42* (*V12Cdc42*) are electroporated *in utero*, radial migration is inhibited but this effect is not as strong as that seen with *DN-Rac1* or *CA-Rac1* (Konno et al., 2005). In addition, the role played by Cdc42 in migrating neurons might be different from that of Rac1 since the former is mainly localized to the perinuclear region on the side of the leading process whereas the latter is expressed at the plasma membrane (Konno et al., 2005). Similarly, strict regulation of RhoA levels and activity appear to be required for radial migration *in vivo* (Nguyen et al., 2006; Pacary et al., 2011; Cappello et al., 2012; Azzarelli et al., 2014; Tang et al., 2014). The general view proposes that RhoA activity must be downregulated to promote radial migration of pyramidal neurons. However, the analysis of *RhoA* knockout using *Emx1-Cre* suggests that RhoA is dispensable for radial migration. In this mutant, the deletion of *RhoA* generates migration defects that are only secondary to radial glia scaffold disruption. Indeed, when *RhoA* knockout cells are transplanted in a wild type environment, they migrate normally, suggesting that there is no cell-autonomous requirement for RhoA activity during radial migration (Cappello et al., 2012). How to reconcile the data showing the requirement of RhoA for radial migration with the fact that *RhoA*-depleted neurons normally migrate in a wild type environment is still an open question. One possibility is that compensatory mechanisms might occur to replace RhoA function in *RhoA* conditional knockout cortices. Indeed, it has been shown that the related GTPase RhoB is strongly up-regulated in the absence of RhoA (Ho et al., 2008). RhoB and RhoC might substitute for RhoA activity during cortical neuron migration. These compensatory mechanisms may be in operation only when *RhoA* expression is completely abrogated by knockout deletion.

#### **Rho GTPases AND REGULATION OF NEURONAL POLARIZATION**

Cortical neurons exist in a number of different shapes and sizes, although a mature neuron typically has several dendrites that receive inputs from presynaptic neurons and one axon that relays information to post-synaptic neurons. The formation of axondendrite polarity is thus crucial for a neuron to establish the precise information flow within the brain. Axons and dendrites differ in morphology, function, and protein and organelle composition. Although the development of these processes has been studied extensively *in vitro* using cultured embryonic hippocampal neurons, the corresponding developmental processes *in vivo* are still unclear. During corticogenesis, early electron microscopic studies revealed that projection neurons initiate their axons during migration whereas significant dendrite growth occurs after the cells have reached their final position (Shoukimas and Hinds, 1978).

#### *Rho GTPases and regulation of axon formation, growth, guidance and branching*

Time-lapse imaging revealed that multipolar cells in the IZ, after extending and retracting their short processes for several hours, suddenly elongate a long process tangentially. These cells then transformed into a bipolar shape, extending a pia-directed leading process (future apical dendrite), and migrate radially leaving the tangential process behind, forming an "L-shaped" axon (**Figure 4**, ❻) (Hatanaka and Yamauchi, 2013). Thus, during migration, the trailing process becomes the axon and extends while being guided to its final destination. Interestingly, a recent study has shown that the interaction between multipolar cells and the preexisting axons of early-born neurons is critical for axon specification. Indeed, once one of the neurites of a multipolar cell contacts the pioneering axons from the early-born cortical neurons, this neurite is stabilized, becomes the axon and extends rapidly (Namba et al., 2014). The duration of axon elongation is however variable according to the targeted area, which is more or less distant according to the final layer position of the cortical neuron (Molyneaux et al., 2007). For example, axons of corticofugal neurons in layer V reach the spinal cord around postnatal day P7 in mouse. Finally, upon reaching its target area, extensive axonal branching occurs during the formation of presynaptic contacts with specific post-synaptic partners (during the second and third postnatal week in the mouse cortex) (Lewis et al., 2013).

The regulation of axon development by Rho GTPases has been mainly studied *in vitro* and reviewed elsewhere (Govek et al., 2005; Hall and Lalli, 2010). However, their specific roles *in vivo* are less well-understood. Cdc42 seems however to be clearly required during cerebral cortex development for the efficient establishment of axonal polarity and growth. Indeed, cortices of *Cdc42* conditional knockout mice crossed with *Nestin-Cre* mice exhibit a widespread reduction of axonal tracts (Garvalov et al., 2007). This phenotype is accompanied by a specific increase in the phosphorylation (inactivation) of the Cdc42 effector cofilin (Garvalov et al., 2007). The axonal defects in *Cdc42* knockouts might be due to the increased levels of inactive cofilin since the deletion of this actin depolymerizing protein results in polarity defects analogous to the ones seen after *Cdc42* ablation (Flynn et al., 2012). In agreement with these data, the growth and organization of callosal axonal fiber tracts are also disrupted in *Cdc42* deficient mice obtained after mating *Cdc42* floxed mice with *hGFAP-Cre* line (Yokota et al., 2010). In addition to Cdc42, a role for Rnd2 in cortical axon extension has been suggested in a study showing that COUP-TFI, a transcription factor crucial for corticogenesis and arealization, promotes callosal axon elongation by finely regulating *Rnd2* expression levels (Alfano et al., 2011).

In contrast to Cdc42, the loss of *Rac1*, using *Foxg1–Cre* mice, does not prevent axonal outgrowth in cortical neurons (Chen et al., 2007). However, in these mutants, the anterior commissure is absent, and the axons of the corpus callosum and the hippocampal commissure fail to cross the midline. A similar phenotype is observed in *Rac1/Emx1Cre* knockout mutants (Kassai et al., 2008), demonstrating that Rac1 controls axon guidance rather than neuritogenesis. In addition, the thalamocortical and corticothalamic axons show defasciculation or projection defects in *Rac1/FoxG1–Cre* mutants (Chen et al., 2007), whereas corticospinal and corticothalamic projections are not affected in *Rac1/Emx1Cre* mice (Kassai et al., 2008). This phenotypic discrepancy might be due to the different pattern of Cre expression in these two Rac1 mutants. Indeed, in *FoxG1–Cre* mice, the recombinase is expressed in other regions than the telencephalon such as the thalamus (Hebert and McConnell, 2000) which might affect the development of thalamocortical projections. To note, Rac3, the other member of the Rac subfamily expressed in the brain, does not seem to have redundant functions with Rac1 since *Rac3* knockouts do not show any obvious developmental defects in the cortex (Corbetta et al., 2005).

The *in vivo* role of RhoA in cortical axon development has not been thoroughly examined but the cortical axons in *RhoA/Emx1– Cre* knockout mutants show correct morphology and projections (Cappello et al., 2012). Nevertheless, RhoA might act as a mediator for activity-dependent branch formation as suggested by a study performed in cortical explants (Ohnami et al., 2008).

#### *RhoGTPases and dendrite/spine formation*

In contrast to the axon, dendrites in cortical neurons form after migration ends. These structures are highly branched and this feature gives them the appearance of a tree. Initially, all excitatory cortical neurons exhibit the common shape of "pyramid," which is characterized by a prominent apical dendrite. This first dendrite derives from the leading process and branches out in an apical tuft that terminates in layer I (**Figure 4**, ❽). With time, basal dendrites appear as well as oblique side branches emerging from the apical shaft (**Figure 4**, ❾) (Whitford et al., 2002). The dendritic trees and consequently the overall neuronal shapes vary greatly within the cortex, with neurons of layers II, III, V, and VI acquiring a pyramidal morphology, whereas those of layer IV predominantly having non-pyramidal morphologies. Therefore, even if the initial stages of dendrite formation are very similar, further maturation determines the final neuronal morphology. For example, spiny stellate neurons in layer IV start out with a pyramidal morphology, but then acquire a stellate morphology, by retracting their apical dendrite at an early postnatal age (Vercelli et al., 1992).

The final step in the acquisition of a mature dendritic morphology is the development of spines (**Figure 4**, ❿). These small dendritic protrusions, which contain receptors and other proteins necessary for synaptic transmission, begin to appear in the first postnatal week in mice, when the arborization of the apical and basal dendrites becomes more complex. When pyramidal neurons reach their mature morphology, they have a highly complex dendritic arbor and are covered with spines. The timing of spine development is variable among neurons that occupy different layers or cortical areas (Huttenlocher and Dabholkar, 1997; Whitford et al., 2002).

The role of Rho GTPases in dendrite and spine formation has been mainly addressed in culture. By manipulating Rho, Rac1, and Cdc42 activities in a number of experimental systems, it has become clear that each of these Rho GTPases plays a prominent role in the development of dendrite structure and that interplay between them determines the complexity of the dendritic tree (Newey et al., 2005). Studies are generally consistent with a key role for RhoA in controlling dendritic length and for Rac and Cdc42 in regulating dendrite branching and remodeling. More precisely, RhoA activation has a negative effect on dendritic arbor growth, whereas Rac1 and Cdc42 activation promote this process. Similarly, RhoA inhibits, whereas Rac1 and Cdc42 promote spine formation and maintenance (Newey et al., 2005). However, considering the large body of *in vitro* work, it is surprising how little is known about their roles *in vivo* in the context of cortical development. To our knowledge, only one study has addressed this role directly *in vivo*. Rosario and colleagues have shown that *in utero* electroporation of a CA form of Cdc42 under the control of the post-mitotic promoter NeuroD1 decreases dendrite branching and complexity in layer II/III pyramidal neurons at postnatal stages (Rosario et al., 2012). The other evidences are indirect and come from *in vivo* studies showing involvement of Rho GTPase regulators, namely GAPs and GEFs, in this process. These regulators include GAPs, such as NOMA-GAP, srGAP2 (also called FNBP2), Myo9b and RICS, as well as GEFs like kalirin (see Section Upstream Regulators of Rho GTPases and Cortical Projection Neuron Development and **Table 2**).

#### **Rho GTPases AND REGULATION OF CORTICAL NEURON DEATH/SURVIVAL**

Programmed neuronal cell death, or apoptosis, is essential for proper cerebral cortex development, resulting in the refinement of nascent neuronal innervation and network formation (Nikolic et al., 2013). In the mouse, neuronal apoptosis takes place in the first 30 postnatal days, with a peak at P5, mostly pronounced in cortical layers II–IV. This wave of apoptosis accounts for a loss of approximately 30% of neuronal content in the cerebral cortex from birth to adulthood (Heumann et al., 1978; Heumann and Leuba, 1983).

Among the Rho GTPases, RhoA is of particular interest with respect to regulation of postnatal apoptosis in the cerebral cortex. Indeed, by engineering a mouse line in which a dominantnegative RhoA mutant (N19–RhoA) is specifically expressed in neurons, Sanno and colleagues have demonstrated that the inhibition of RhoA activity reduces the amount of apoptosis occurring in the postnatal cortex and results in a concomitant increase in the density and absolute number of neurons in the adult cortex (Sanno et al., 2010). Interestingly, the change in neuronal density in the N19–RhoA cortex is attributable to an increase in the number of excitatory projection neurons and not in that of the interneuron population, which originates in the ventral telencephalon (Sanno et al., 2010).

Besides the well-established programmed cell death naturally occurring in the postnatal brain, more recent studies indicate the existence of an earlier wave of programmed cell death affecting neural progenitors and nascent neurons (Yeo and Gautier, 2004). This early wave of cell death appears to play an even more critical role in determining the final size of the brain (de la Rosa and de Pablo, 2000; Kuan et al., 2000). In mouse, this death occurs between E12 and E16 within the VZ and IZ of the cerebral cortex (Blaschke et al., 1996, 1998; Thomaidou et al., 1997). Interestingly, the forebrain specific deletion of *Rac1* mediated by *FoxG1Cre* enhances apoptosis in VZ and SVZ progenitors, mainly around E14.5 (Chen et al., 2009; Leone et al., 2010); this effect partially contributes to a decrease in neural progenitors observed during mid-to-late telencephalic development (Chen et al., 2009). While Rac1 is required for survival of both VZ and SVZ progenitors, Cdc42 plays a dispensable role in cell survival during corticogenesis, as indicated by a comparable number of apoptotic cells in the cortex of control mice and *Cdc42/Nestin–Cre* knockout mutants (Peng et al., 2013).

#### **UPSTREAM REGULATORS OF Rho GTPases AND CORTICAL PROJECTION NEURON DEVELOPMENT**

A number of *in vivo* studies showing involvement of GEFs and GAPs in cortical projection neuron development further support a critical role of the Rho GTPase family in this process and are summarized in **Table 2**. Studying the functions of GAPs and GEFs not only provides an indirect way to clarify the roles played by Rho GTPases in specific steps of corticogenesis, but it also contributes to the overall understanding of the entire pathways activated downstream of Rho proteins. For example, the loss of *NOMA-GAP*, a Cdc42-specific GAP, leads to an oversimplification of cortical dendritic arborization, as well as an hyperactivation of Cdc42. Remarkably, these dendritic defects can be partially restored by genetic reduction of post-mitotic *Cdc42* levels, demonstrating that the post-mitotic inhibition of Cdc42, mediated by NOMA-GAP, is a necessary requirement for dendritic branching during cortical development (Rosario et al., 2012). In this study, the authors further show that *in utero* expression of active cofilin is sufficient to restore postnatal dendritic complexity in *NOMA-GAP*-deficient animals. Therefore, these data support a model, whereby, during cortical dendritic development, cofilin activation is positively regulated by NOMA-GAP through the inhibition of Cdc42 (Rosario et al., 2012). Interestingly, only the dendritic complexity of layer II/III neurons is affected by the genetic ablation of *NOMA-GAP*, whereas layer V pyramidal neurons develop normal dendrites. How NOMA-GAP regulates dendrite development only in specific layers, although this GAP is expressed in all cortical layers (Rosario et al., 2012), is still an open question. It is possible that other Cdc42-GAPs, which

#### **Table 2 | Regulation of cortical projection neuron development by GAPs and GEFs (***in vivo* **studies).**


**Table 2 | Continued**


are selectively expressed in the unaffected layers, might substitute for the loss of *NOMA-GAP* function. Alternatively, some neurons might be more sensitive to NOMA-GAP activity, because they exhibit higher active Cdc42 levels, which could be the result of layer-specific environmental signals or differentially expressed Cdc42-GEFs.

One difficulty with these studies is to determine whether the phenotype observed is due to the inhibition or activation of a specific Rho GTPase or of several members, since these regulators usually show activity against multiple Rho GTPases. For instance, RICS has been classified in **Table 2** as a Rho GAP, because the defects on dendrites induced by *RICS* knockdown are rescued by an inhibition of RhoA signaling (Long et al., 2013) and might thus be due to a modulation of Rho activity. However, it has been shown that RICS prefers Cdc42 over Rac1 or RhoA as a substrate (Simo and Cooper, 2012). Also, it is important to remember that some of the phenotypes described in **Table 2** might not be related to the GAP or GEF activity itself, but other protein domains might be involved. For example, α2-chimaerin, a Rac-GAP, has been shown to regulate radial migration, not through its GAP activity, which is instead dispensable, but through the association with the microtubule-associated protein CRMP-2 (Ip et al., 2012). Similarly, DOCK7 is a Rac-GEF that controls INM independently of its GEF activity (Yang et al., 2012b). The functions of GAPs and GEFs that are independent from GTPase activation or GDP-GTP exchange, respectively, have not been included in **Table 2**.

#### **Rho GTPases AND CEREBRAL CORTEX MALFORMATIONS**

As described previously, the development of the cerebral cortex is remarkably complex and tightly organized. Disruption of any of the overlapping steps that contribute to this process can result in profound and stereotypical cortical malformations. In view of the multiple regulatory functions played by Rho GTPases during cerebral cortex development, it is thus not surprising that their forebrain specific suppression leads to cortical malformations. Accordingly, Cappello et al. found that *Emx1-Cre* mediated deletion of *RhoA* causes three types of malformations in the mouse cerebral cortex (Cappello et al., 2012; Cappello, 2013). First, the adult mutant cerebral cortex is about 1.3-fold thicker than control. This megalencephaly phenotype has been linked to the increased proliferation observed in the *RhoA* conditional mutant (see Section Rho GTPases and Regulation of Progenitor Cell Division, Proliferation and Cell Fate and **Table 1**). The second malformation observed in these mutants is a subcortical band heterotopia, which is characterized by a heterotopic cortex formed of ectopic neurons embedded within the white matter and underlying a normotopic cortex. Interestingly, the formation of this double cortex may not result from direct defects in migrating neurons, but rather from defective radial glia fibers that neurons use as a scaffold to migrate. Indeed, as mentioned previously (see Section Rho GTPases and Regulation of Radial Migration and **Table 1**), *RhoA* null neurons migrate normally when transplanted into wild-type cerebral cortex, whereas the converse is not the case, probably because of the strongly disorganized RG processes in the mutant cortices (Cappello et al., 2012). The last cortical abnormality observed in *Emx1-Cre RhoA* null mutants is the formation of cobblestones or neuronal ectopias at the basal side of the developing cerebral cortex. These neuronal protrusions beyond layer I may result from the increase speed of migrating nascent neurons and/or from the aberrant RG endfeets (Cappello, 2013). Interestingly, *FoxG1Cre RhoA-*deficient embryos also exhibit expansion of the neural progenitor pool and exencephaly-like protrusions (Katayama et al., 2011).

In contrast to *RhoA* deletion, which induces excessive proliferation, *Rac1* loss of function in the mouse cerebral cortex reduces progenitor cell proliferation and increases apoptosis, which may be the major causes of microcephaly observed in the mutants (Chen et al., 2009; Leone et al., 2010). The suppression of *Cdc42* in the cortex also affects the overall cortical morphology. Indeed, *Cdc42*-deficient telencephalon fails to bulge or separate into two cerebral hemispheres, resulting in holoprosencephaly (Chen et al., 2006). This phenotype may result from the essential role of Cdc42 in establishing the apico-basal polarity of RGs and subsequently of the telencephalic neuroepithelium, which is needed for the expansion and bifurcation of cerebral hemispheres (Chen et al., 2006).

Interestingly, *NOMA-GAP* deficiency also leads to a decrease of cortical thickness in different cortical areas (Rosario et al., 2012). This reduction was observed in adult mice, but also at early postnatal stages, suggesting that cortical thinning is due to a defective developmental process. Since the absence of *NOMA-GAP* does not impact the early stages of cortical development, including neuronal birth, survival, fate determination and migration, cortical thinning in these mutants may arise from defective formation of cortical dendritic trees (Rosario et al., 2012).

#### **CONCLUDING REMARKS**

Altogether the above studies highlight the crucial roles played by the Rho GTPase family in the regulation of cerebral cortex development and emphasize that a better understanding of these functions might help to clarify the etiology of several cortical malformations. Further work is required to fully characterize the contribution of the different Rho GTPases expressed in the developing cerebral cortex as well as the downstream signaling involved and the mechanisms regulating their expression and/or activity. Another major challenge for the future will be to understand how the signals through the different Rho GTPases as well as the other small GTPases of the Ras superfamily (Ras, Rab, Ran, Arf) are integrated in nascent cortical neurons and how they are spatially and temporally controlled during cortical development.

Abnormal signaling through Rho GTPases is associated with cognitive dysfunction (Newey et al., 2005; De Filippis et al., 2014) and recent findings have showed their involvement in the development and progression of neurodegenerative diseases (Antoine-Bertrand et al., 2011). Studies on Rho GTPase functions *in vivo* thus represent an underexplored territory that may hold therapeutic potential.

#### **ACKNOWLEDGMENTS**

This work was supported by INSERM and Marie Curie Actions (Intra European Fellowship). Roberta Azzarelli is supported by a Medical Research Council (MRC) post-doctoral fellowship and Thomas Kerloch is recipient of a studentship from the Ministère de l'Enseignement Supérieur et de la Recherche.

#### **REFERENCES**


Yoshizawa, M., Kawauchi, T., Sone, M., Nishimura, Y. V., Terao, M., Chihama, K., et al. (2005). Involvement of a Rac activator,P-Rex1, in neurotrophinderived signaling and neuronal migration. *J. Neurosci.* 25, 4406–4419. doi: 10.1523/JNEUROSCI.4955-04.2005

**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 22 October 2014; accepted: 11 December 2014; published online: 07 January 2015.*

*Citation: Azzarelli R, Kerloch T and Pacary E (2015) Regulation of cerebral cortex development by Rho GTPases: insights from in vivo studies. Front. Cell. Neurosci. 8:445. doi: 10.3389/fncel.2014.00445*

*This article was submitted to the journal Frontiers in Cellular Neuroscience.*

*Copyright © 2015 Azzarelli, Kerloch and Pacary. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Polarity transitions during neurogenesis and germinal zone exit in the developing central nervous system

# **Shalini Singh and David J. Solecki \***

Department of Developmental Neurobiology, St. Jude Children's Research Hospital, Memphis, TN, USA

#### **Edited by:**

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan

#### **Reviewed by:**

Orly Reiner, Weizmann Institute of Science, Israel Yasuhiro Itoh, Harvard University, USA

#### **\*Correspondence:**

David J. Solecki, Department of Developmental Neurobiology D2025C, St. Jude Children's Research Hospital, 262 Danny Thomas Place-MS 323, Memphis, TN 38105, USA e-mail: david.solecki@stjude.org

During neural development, billions of neurons differentiate, polarize, migrate and form synapses in a precisely choreographed sequence. These precise developmental events are accompanied by discreet transitions in cellular polarity. While radial glial neural stem cells are highly polarized, transiently amplifying neural progenitors are less polarized after delaminating from their parental stem cell. Moreover, preceding their radial migration to a final laminar position neural progenitors re-adopt a polarized morphology before they embarking on their journey along a glial guide to the destination where they will fully mature. In this review, we will compare and contrast the key polarity transitions of cells derived from a neuroepithelium to the well-characterized polarity transitions that occur in true epithelia. We will highlight recent advances in the field that shows that neuronal progenitor delamination from germinal zone (GZ) niche shares similarities to an epithelialmesenchymal transition. Moreover, studies in the cerebellum suggest the acquisition of radial migration and polarity in transiently amplifying neural progenitors share similarities to mesenchymal-epithelial transitions. Where applicable, we will compare and contrast the precise molecular mechanisms used by epithelial cells and neuronal progenitors to control plasticity in cell polarity during their distinct developmental programs.

**Keywords: neurogenesis, neuronal progenitor, neuroepithelial, neuronal polarity, cell junction, epithelial mesenchymal transition, delamination**

#### **INTRODUCTION**

The developing central nervous system, including the spinal cord, retina and brain, occupies a complex developmental landscape wherein neural stem cells are born and then proliferate, differentiate and migrate long distances to form intricate networks, all of which ensure proper CNS functioning. Notably, almost all of the diverse cell types comprising the CNS originate from the neuroepithelium lining the embryonic neural tube and neural plate. Self-renewing multipotent cells (neural stem cells) orient themselves along the apical-basal axis in a single layer, conferring a highly polarized structure on this germinal niche. Subsequent symmetric and asymmetric division of the neural stem cells imparts a pseudostratified appearance to neuroepithelium, in which the nuclei undergo interkinetic nuclear migration while the apical-basal surfaces of the cells remain anchored through intercellular junctions (**Figure 1A**; Haubensak et al., 2004; Götz and Huttner, 2005). Alteration of polarity signaling cascade or cell adhesion dynamics leading to improper neural development substantiates the architectural organization of the neuroepithelium (Ayala et al., 2007; Métin et al., 2008; Roussel and Hatten, 2011). As development proceeds these progenitors must commit to a specific neuronal fate and migrate to their final destinations. This step requires them to sever ties with the ventricular zone (VZ), undergo a transition in polarity, change their adhesive preference and delaminate. Our understanding of key components and signaling

cascades, such as the Par polarity complex and its interplay with adhesion molecules such as cadherins, nectins, claudins and junctional adhesion molecules (JAMs), has advanced considerably (Tsukita and Furuse, 1999; Mizoguchi et al., 2002; Costa et al., 2008; Ishiuchi et al., 2009; Famulski et al., 2010), but some key questions remain, including: What are the specific biological processes that precede delamination? What initiates and controls the switch in polarity and how is this linked to adherens junction (AJ) disassembly? Epithelial cells frequently display polar plasticity through processes known as epithelial-to-mesenchymal transition (EMT) and its reverse, mesenchymal-to-epithelial transition (MET), which developing neurons appear to mirror. By highlighting recent work addressing these specific challenges in the developing cortex, spinal cord, retina and cerebellum, we will highlight the emerging idea that akin to epithelial cells, progenitors in the GZs of the CNS require the EMT-MET machinery to undergo a change in polarity that leads to their delamination, differentiation and maturation.

#### **DEVELOPING NEOCORTEX**

Development of the neocortex, with its unique lamination process, provides an attractive model to investigate the cellular remodeling essential for establishing the CNS (Rakic, 1971, 2009). Neural stem cells at the transient embryonic zone switch from a proliferative symmetric cell division phase to an asymmetric

phase, giving rise to radial glial cells (RGCs), the progenitor cells of the cortex (Bayer, 1986, 1990; Malatesta et al., 2000; Campbell and Götz, 2002; Nadarajah and Parnavelas, 2002; Götz, 2003). Successive waves of migration at the VZ form an insideout gradient of neurogenesis to establish the laminar cortical structure (**Figure 1B**; Angevine and Sidman, 1961). Stratification of the cortex requires radial and tangential migration of neurons, as shown by electron microscopy, lineage tracing and realtime imaging of brain slices (Mione et al., 1997; Wilson and Rubenstein, 2000; Marín and Rubenstein, 2001). Lamination of the cortex in CNS development requires precise spatial-temporal regulation of cortical migration (Métin et al., 2008). To guide the reader's appreciation and understanding of the morphogenetic changes that occur during corticogenesis, we will briefly discuss the cytoarchitecture of the newly formed RGCs.

Ultrastructure studies show that like epithelial cells, newly formed RGCs are morphologically polarized (Aaku-Saraste et al., 1996; Huttner and Brand, 1997; Chenn et al., 1998). They attach apically to the ventricular surface while extending long basal processes that span the entire cortical plate that affix at the overlying matrix produced by the pia. Apical anchoring is mediated by specialized intercellular adhesion complexes that involve cadherins, nectins, JAMs and β-catenin (Aaku-Saraste et al., 1996; Zhadanov et al., 1999; Manabe et al., 2002; Junghans et al., 2005; Kadowaki et al., 2007). These complexes link the cytoskeletal scaffolds and coordinate signaling pathways in neighboring cells while the pial attachment is established by integrins and possibly cadherins (Anton et al., 1999; Graus-Porta et al., 2001). Relevance of the cell-cell contact in establishing the radial glial scaffold at the apical surface and its role in

signaling networks such as reelin is highlighted by cortical defects observed in β1integrin and Dab1/Rap1/cadherin deficient mice (Graus-Porta et al., 2001; Franco et al., 2011). As in epithelial cells, Cdc42 and polarity complexes such as Par proteins and Crumbs complex (Crb, PALS, PATJ, Lin7) are essential in establishing and maintaining these AJs (**Figure 2**; Manabe et al., 2002; Cappello et al., 2006; Imai et al., 2006; Bultje et al., 2009). Nevertheless, as development proceeds the RGCs divide giving rise to a progeny that become committed to their fate and must migrate to occupy their respective positions in the cortical layers. It appears that the newly born neurons must first break their AJs to delaminate from the apical surface. Second, they must remodel their cytoskeletons to initiate movement out of the VZ. Remodeling of polarity is apparent from time-lapse imaging and electron microscopy that show migrating neurons adopting an intermediate multipolar state before reacquiring some of the features needed for glial– guided migration (Shoukimas and Hinds, 1978; Nadarajah et al., 2003). Remodeling of junctions and polarity in these newly born neurons closely resemble the sequence of events during EMT-MET in epithelial cells (**Figure 1B**; Nelson, 2009; Lamouille et al., 2014).

Given that apical anchoring and delamination are crucial to cortical arrangement there has been great focus on understanding the apical protein complexes that allow for this GZ exit; and have helped to define the roles of N-cadherin/Rap1, adaptor protein Afadin, Lis1, β-catenin, cdc42/GSK-3β and polarity protein Pard3/Notch in maintaining the apical cytoarchitecture (Bos et al., 2001; Ooshio et al., 2007; Bi et al., 2009; Severson et al., 2009; Yokota et al., 2010; Franco et al., 2011; Jossin, 2011; Jossin and Cooper, 2011). A recent report showed that cadherinbased adhesions also facilitate Notch signaling (Hatakeyama et al., 2014). While these studies have identified key components of the machinery involved in establishing the polarity and GZ occupancy, the overarching question of how these components might be regulated demands attention. Because fate commitment and delamination are sequential events, it seemed plausible that proneural genes might also have a role in regulating delamination. Actin reorganization by proneural factors such as Neurogenin 2 (Neurog2) and Ascl1, activating Rho GTPases Rnd2 and Rnd3, respectively, allows remodeling of the actin cytoskeleton by inhibiting Rho activity, thus linking neuronal commitment and migration (Ge et al., 2006; Heng et al., 2008; Pacary et al., 2011). Although they may provide the physical force needed for RGC exit from the GZ niche, they do not initiate delamination, as silencing of Rnd3 disrupts the distribution of β-catenin and N-cadherin at AJs, indicating that Rnd3 is essential for maintaining the integrity of the junctions (**Figure 2**; Pacary et al., 2011, 2013). Additionally, the physical forces can be effective only if the RGCs first attenuate their intercellular adhesion through dissolution of AJs, lose their apical-basal polarity and acquire a multipolar nature. In neural crest cells and epithelial cells, this phenomenon is regulated by EMT factors that promote loss of epithelial characteristics and acquisition of mesenchymal attributes (Thiery and Sleeman, 2006; Baum et al., 2008). Central to this process are the Snail, Slug and Zeb family of transcriptional repressors (Chaffer et al., 2007; Baum et al., 2008). From these systems and the fact that newly committed neurons express such regulators, one can extrapolate a possible molecular link between commitment and delamination.

In support of the above idea, Scratch 1 and Scratch 2, members of the Snail superfamily, are expressed upon neuronal fate commitment in neocortex, in which they trigger active disintegration of AJs by directly repressing expression of Ecadherin (**Figure 2**; Itoh et al., 2013). Additionally, classical genes in the EMT pathway, such as those encoding Occludin and Nephrins, also show downregulation accompanied by increased expression of Mmp19 (Thiery and Sleeman, 2006; Itoh et al., 2013). Notably, Scratch1 overexpression has no effect on neurogenesis, thus identifying it as a delamination-specific pathway. The existence of an EMT-like pathway in corticogenesis is further buttressed by the finding that another regulator of the pathway, Snail, is expressed in both radial precursors and newborn neurons during corticogenesis (Zander et al., 2014a,b). Snail's role in the neocortex is not limited to regulating AJs; it also helps promote the survival, proliferation and self-renewal of cortical progenitors.

#### **DEVELOPING SPINAL CORD**

The central question of what molecular mechanism controls AJ disassembly during migration has been addressed in developing motor neurons (MNs) of the spinal cord. Spinal cord development is a multistage process with distinct subtypes V0, V1, V2, V3, interneurons and MNs (**Figure 1C**; Alaynick et al., 2011). Convergent action of morphogens such as Shh and retinoic acid elicits a unique set of transcriptional networks and factors (e.g., Olig2) that specify different neuronal subtypes (Mukouyama et al., 2006; Briscoe and Novitch, 2008). Studies in Olig2-mutant mice showed that the forkhead proteins Foxp1, 2 and 4 are essential for specifying MN fate and for migration (Ferland et al., 2003; Dasen et al., 2008; Rousso et al., 2008; Palmesino et al., 2010). Rousso et al. investigated the role of Foxp proteins in chicken and mouse spinal cord, showing them to be components of a gene regulatory network that links and balances AJs with cell fate (Rousso et al., 2012). Increased expression of Foxp2 and Foxp4 inversely affected expression of N-cadherin and Sox2, leading to dissolution of AJs and ectopic neurogenesis in the VZ. Conversely, their loss maintained the precursors in the progenitor state (**Figure 2**). Such transcriptional regulation of N-cadherin and Sox2 might also regulate the size of the embryonic VZ niche by controlling delamination (Rousso et al., 2012). Decreased Sox2 and cadherin levels have also been demonstrated in neural crest cells before delamination (Zappone et al., 2000; Bylund et al., 2003; Graham et al., 2003; Bello et al., 2012). These findings, together with the documented role of forkhead proteins in maintaining AJs in other systems by regulating cadherins, reinforces the idea of an EMT-like signature in CNS development (**Figure 1C**; Mani et al., 2007; Rousso et al., 2012).

Besides transcriptional control of adhesion proteins, delamination of progenitors also involves a change in polarity and cellular architecture, recently addressed in a new light by Das and Storey. By using structured illumination microscopy in chick neural tube, they discovered a cellular mechanism called apical abscission that participates in actin-myosin–dependent remodeling of primary cilium (Das and Storey, 2014). Apical abscission at the VZ detaches progenitors and leads to loss of apical complex proteins, a process characterizing loss of apical polarity (Farkas and Huttner, 2008; Das and Storey, 2012). This study answered the question of how the VZ, unlike archetypal epithelium undergoing EMT, maintains its integrity. It seems plausible that apical abscission might provide the progenitors an efficient way to transiently change their polarity without substantially altering their transcriptional profiles. If basal progenitors remodel their primary cilia in preparation for delamination in the neocortex and for apical shedding of MNs that could explain how newly differentiated neurons lose and reacquire polarity during delamination and migration by remodeling their apical complex proteins.

#### **DEVELOPING RETINA**

Like other developing CNS structures, the retinal neuroepithelium is pseudostratified. During retinogenesis, polarized progenitors remain anchored to the apical-basal membrane by AJs and TJs while their fate is determined by interkinetic nuclear migration (**Figure 1D**; Frade, 2002; Baye and Link, 2007, 2008). Subsequent establishment of the retinal layers requires postmitotic detachment of retinal progenitors from the apical surface, reorganization of polarity and migration to the appropriate layers (Miyata, 2008).

In mouse retinal progenitors, polarity is imparted by two apical complexes, the partitioning defect (PARD) complex comprising Pard3, Pard6, aPKC and cdc42 and crumbshomologue (CRB) comprising Crb1-3, Pals1/MPP5 and Patj (Koike et al., 2005; Cappello et al., 2006; van de Pavert et al., 2007; Park et al., 2011; Alves et al., 2013; Dudok et al., 2013; Heynen et al., 2013). In mice lacking CRB2, early retinal progenitors show abnormal lamination due to greater proliferation of lastborn progenitors. Membrane-palmitoylated protein 3 (MPP3), a scaffolding cell-cell adhesion protein, is also localized to the retinal sub-apical region and interacts with Pals1 and Dlg1 (Pellikka et al., 2002; Alves et al., 2013). Perturbation of MPP3 in developing retina was found to cause focal disruption of AJs and ectopic cell localization (Dudok et al., 2013). The need to maintain polarity is further stressed by laminar defects in Pals1 conditional-knockout mice, in which mislocalization of neurons was found throughout the retina (Park et al., 2011). Additionally, dysregulation of aPKC and cdc42 lead to lamination defects (Koike et al., 2005; Heynen et al., 2013).

Structural and signaling components come together at neural junctions. In vertebrates, the polarized epithelium's apical junction proteins, such as cadherin, occludins, claudins and JAMs, interact on the cytoplasmic side of the cell with a variety of polarity and signaling molecules, including Pard3 and βcatenin, that connect them to the actin cytoskeleton (Tsukita and Furuse, 1999; Tsukita et al., 1999; Nishimura et al., 2004; Umeda et al., 2004; Perez-Moreno and Fuchs, 2006). In the retina, alteration of β-catenin causes loss of polarity and adhesion, cell detachment, ectopic migration and spatiotemporally perturbed retinogenesis (Brault et al., 2001; Fu et al., 2006). The role of apical complexes is exemplified in retinogenesis, in which they not only establish adhesion and polarity but act as important nodes for diverse signaling pathways (e.g., Notch, Wnt, Hippo, mTOR) essential for specifying retinogenesis. As supporting evidence, Ohata et al., showed that Crb complex and Notch form a positive feedback loop that maintains apicobasal polarity; disruption of feedback causes laminar defects (Ohata et al., 2011).

There are some interesting differences between retinal neuroepithelium and other developing CNS structures, one of which was highlighted by Ivanovitch et al. By combining mosaic labeling of single cells with 4D confocal imaging during optic evagination in zebrafish, they showed two discrete populations of cells: basally positioned cells and core cells involved in establishing the retinal neuroepithelium (Ivanovitch et al., 2013). Interestingly, live imaging showed that basally positioned cells undergo precocious polarization, while core cells remain mesenchymal until optic vesicle formation. The authors found that core cells that localize Pard3 in a polarized manner undergo a special MET using apical points for intercalation in CNS outpocketing. Furthermore, modulation of Pard6γb and Laminin leads to failure of optic vesicle evagination (Ivanovitch et al., 2013).

Clearly, the transition in polarity and morphology of retinal progenitors requires rearrangement of their cytoskeleton that generates physical forces required for movement. While this process is a quintessential requirement for determining the laminar structure of developing CNS, how are such processes regulated remains an open question. A chemical mutagenesis screen in medaka fish identified the guanine-nucleotide exchange factor ArhGEF18, a RhoA- and Rac-specific GEF factor, to be a key regulator of retinal architecture and function that also controls apicobasal polarity and proliferation (Herder et al., 2013; Loosli, 2013). In epithelial cells, small GTPases of the Rho family act as molecular switches that regulate and coordinate the actin cytoskeleton, cell junction assembly and polarity (Heasman and Ridley, 2008; Samarin and Nusrat, 2009). Additionally, Rho GTPases are regulated by transforming growth factor β (TGFβ), which degrades RhoA by phosphorylating Pard6 (Ozdamar et al., 2005). TGFβ also downregulates Pard3 expression while activating the EMT regulator Snail (**Figure 2**; Wang et al., 2008). As TGFβ is expressed during early retinal embryogenesis, it could plausibly play a role in regulating or guiding a similar process at that time.

#### **DEVELOPING CEREBELLUM**

The cerebellum is an attractive model for studies of CNS development, primarily because of its remarkably simple laminar organization, which consists of two principal neurons (cerebellar granule neurons (CGNs) and Purkinje cells) and minority interneuronal populations (Roussel and Hatten, 2011). Granule neuron progenitors (GNPs) arise along the rhombic lip, at the midbrain-hindbrain junction, and migrate dorsally over the cerebellar anlage (Wingate, 2001), while a stream of radially migrating progenitors arising from neuroepithelium form the other cerebellar cell types (Hallonet and Le Douarin, 1993; Gao

and Hatten, 1994). After GNPs migrate away from the rhombic lip, they proliferate under the guidance of external cues, such as Shh, to generate a second GZ called the external granular layer (EGL; Hatten, 1999). The remaining cerebellar histogenesis occurs mainly postnatally. This protracted development provides an excellent opportunity to study distinct intermediate stages of neurogenesis. After their proliferative phase, GNPs exit the cell cycle to form a layer of cells that migrate tangentially, then undergo a morphological transformation that initiates radial migration through the molecular layer (ML) along glial fibers to their final laminar position (Rakic, 1971; Edmondson and Hatten, 1987; Komuro et al., 2001). In the outer EGL, the GNPs have a round morphology with multiple short processes (Manzini et al., 2006). As they migrate to the inner EGL, their cytoskeleton is reorganized to form two short horizontal processes that elongate during their journey to the ML (**Figure 1E**). After a brief latency at the ML, they acquire an elongated spindle shape and a vertical process (Komuro et al., 2001).

The maturation of CGNs and their migration requires polarization, which was carefully examined in two studies delineating role of Par complex in this process. The first study showed that manipulation of Pard6α function, both *in vitro* and *ex vivo*, disrupts the actomyosin cytoskeleton and blocks radial migration of CGNs (Solecki et al., 2009). As expression of Pard3A polarity protein increases as the CGNs mature, the second study investigated its role in CGN migration (Famulski et al., 2010). Whereas loss of Pard3A impeded radial migration of CGNs, ectopic expression led to precocious migration. Further, Pard3A expression was shown to be regulated by E3 ubiquitin ligase seven in absentia homolog (Siah) 1 and 2 (**Figure 2**). Reciprocal expression of Siah and Pard3A in developing CGNs indicated that Siah negatively regulates Pard3A expression. This study also linked polarity proteins with adhesion of CGNs. In the epithelial cells, Pard3A binds to three members of the JAM family via its PDZ domain and recruits them to TJs to establish polarity (Ooshio et al., 2007). Utilizing live probe imaging, Famulski et al., demonstrated identical spatiotemporal expression of these proteins during CGN maturation and showed that JAM-C is necessary and sufficient for CGN exit from the EGL (Famulski et al., 2010). These results indicate that not only is cell adhesion crucial for guiding a neuron's migration path but the acquisition of adhesion by CGNs and loss of some of multipolar features reflects the MET-like process (**Figure 1E**). This is unique from other developing CNS structures where the newly born neurons lose polarity and require it upon reaching their final laminar position. A pertinent question that sequentially arises from these observations is how polarity is reorganized and how immature neurons initiate, implement and conclude this essential morphogenic event.

#### **SUMMARY**

Addressing the morphological changes observed in newly born neurons and its relevance in setting up the laminar structure of CNS raises some interesting questions. How the polar plasticity of newly born is regulated and what are the key factors that are involved in this process across the various developing CNS structures? Further in the developing cerebellum, how do the neurons born from the RGCs after adopting a multipolar migratory feature upon reaching the second GZ restore their polarity? Like the epithelial system is there a regulator whose expression temporally coincides with the CGNs reacquiring epithelial characteristics aka MET like processes. If so, what would be some of the targets of this MET like processes in developing neurons? Do these factors also play a role in differentiation and maturation of the neurons?

#### **ACKNOWLEDGMENTS**

We thank Sharon Naron for excellent editorial support in preparation of the manuscript and members of the Solecki lab for editing the manuscript. The Solecki Lab is funded by the American Lebanese Syrian Associated Charities (ALSAC), by grant #1-FY12-455 from the March of Dimes, and by grant 1R01NS066936 from the National Institute of Neurological Disorders and Stroke (NINDS). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NINDS or the National Institutes of Health.

#### **REFERENCES**


deletion results in dramatic brain malformation and failure of craniofacial development. *Development* 128, 1253–1264.


Wang, X., Nie, J., Zhou, Q., Liu, W., Zhu, F., Chen, W., et al. (2008). Downregulation of Par-3 expression and disruption of Par complex integrity by TGF-beta during the process of epithelial to mesenchymal transition in rat proximal epithelial cells. *Biochim. Biophys. Acta* 1782, 51–59. doi: 10.1016/j.bbadis.2007.11.002


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 17 December 2014; accepted: 10 February 2015; published online: 04 March 2015*.

*Citation: Singh S and Solecki DJ (2015) Polarity transitions during neurogenesis and germinal zone exit in the developing central nervous system. Front. Cell. Neurosci. 9:62. doi: 10.3389/fncel.2015.00062*

*This article was submitted to the journal Frontiers in Cellular Neuroscience*.

*Copyright © 2015 Singh and Solecki. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms*.

# Molecules and mechanisms that regulate multipolar migration in the intermediate zone

#### **Jonathan A. Cooper \***

Fred Hutchinson Cancer Research Center, Division of Basic Sciences, Seattle, Washington, USA

#### **Edited by:**

Takeshi Kawauchi, Keio University School of Medicine / Japan Science and Technology Agency, Japan

#### **Reviewed by:**

Eckart Förster, University of Hamburg, Germany Hidenori Tabata, Aichi Human Service Center, Japan

#### **\*Correspondence:**

Jonathan A. Cooper, Fred Hutchinson Cancer Research Center, Division of Basic Sciences, 1100 Fairview Ave N, Seattle Washington 98109, USA e-mail: jcooper@fhcrc.org

Most neurons migrate with an elongated, "bipolar" morphology, extending a long leading process that explores the environment. However, when immature projection neurons enter the intermediate zone (IZ) of the neocortex they become "multipolar". Multipolar cells extend and retract cytoplasmic processes in different directions and move erratically sideways, up and down. Multipolar cells extend axons while they are in the lower half of the IZ. Remarkably, the cells then resume radial migration: they reorient their centrosome and Golgi apparatus towards the pia, transform back to bipolar morphology, and commence locomotion along radial glia (RG) fibers. This reorientation implies the existence of directional signals in the IZ that are ignored during the multipolar stage but sensed after axonogenesis. In vivo genetic manipulation has implicated a variety of candidate directional signals, cell surface receptors, and signaling pathways, that may be involved in polarizing multipolar cells and stabilizing a pia-directed leading process for radial migration. Other signals are implicated in starting multipolar migration and triggering axon outgrowth. Here we review the molecules and mechanisms that regulate multipolar migration, and also discuss how multipolar migration affects the orderly arrangement of neurons in layers and columns in the developing neocortex.

**Keywords: neuron migration, axonogenesis, multipolar migration, neocortex development, radial migration, cortical lamination, neuron locomotion, mini-columns**

#### **INTRODUCTION AND SCOPE**

The neocortex develops by the coordinated migration of projection neurons from the neocortical ventricular zone (VZ) and interneurons from the ganglionic eminences (Hatten, 2002; Marín and Rubenstein, 2003). While projection neurons move generally outwards from the VZ to the top of the cortical plate (CP), live imaging of individual cells has revealed that their radial progress is interrupted by an extended period of random migration. During this time the neurons appear to be "stellate" or "multipolar" (MP), characterized by multiple (>3) cytoplasmic projections that point in different directions. The primary cilium and centrosome are oriented randomly relative to the pial surface. MP cells migrate with frequent changes of direction, moving sideways (tangentially), up (towards the pia) or down (towards the VZ). Axons are initiated during the MP phase. MP neurons are quite different from migrating neurons in other brain regions, which generally are elongated in the direction of travel and have a prominent leading process. MP migration can lead to horizontal dispersion of neurons, and may be functionally significant for forming cortical circuits. In addition, the duration of MP migration differs from neuron to neuron, which has implications for cortical layering. Finally, the signaling mechanisms that initially cause polarized cells from the VZ to become MP, and which cause MP cells to resume radial migration and become bipolar, remain mysterious despite intense study.

This review addresses new developments and continuing uncertainty regarding the external signals and intracellular mechanisms that regulate MP cells at three important times: when MP migration starts, when the axon starts to grow, and when MP cells repolarize towards the pia and resume radial migration with bipolar morphology. The review also discusses the implications of MP migration for the radial unit hypothesis of cortical wiring and for cortical lamination. The focus is on newer research from the last 5–10 years, with an emphasis on loss of function studies. Readers are referred to excellent reviews for earlier work (Bielas et al., 2004; LoTurco and Bai, 2006; Ayala et al., 2007).

#### **BACKGROUND: PHASES OF MIGRATION OF NEOCORTICAL PROJECTION NEURONS**

Four phases of projection neuron migration have been described through detailed histological and live imaging studies (Shoukimas and Hinds, 1978; O'Rourke et al., 1992; Nadarajah et al., 2001; Hatanaka and Murakami, 2002; Tabata and Nakajima, 2003; Hatanaka et al., 2004, 2009; Noctor et al., 2004; Ochiai et al., 2007; de Anda et al., 2010; Namba et al., 2014; **Figure 1**). In phase 1, asymmetric division of radial glia progenitors (RG) in the VZ creates new post-mitotic neurons and intermediate progenitors (IP). These cells exit the VZ with bipolar or "pin-like" morphology. Phase 2 starts when cells reach the subventricular zone (SVZ)/IZ and become multipolar (**Figure 1**, stage 2A). MP IP divide in the SVZ and their daughters resume MP migration (stage 2A'). After a day or more in the MP phase, a ventricleor horizontally-oriented process near the centrosome begins to extend and becomes the axon (stage 2B). Phase 2 ends when the MP cell reorients the Golgi and centrosome towards the pia, establishes a dominant pia-directed leading process, and starts radial migration, trailing the axon behind (Hatanaka et al., 2004; de Anda et al., 2010) (stage 2C). This is known at the multipolar to bipolar (MP-BP) transition, and requires the stabilization of a dominant leading process and the correct orientation of that process towards the pia. After the MP-BP transition, neurons rapidly exit the IZ by locomotion along RG (phase 3), followed by phase 4, terminal translocation to the top of the CP.

The principal signaling pathways implicated in these major transitions are summarized in **Figure 1** and described in more detail below. Many additional molecules are required for the multipolar to bipolar transition but their regulation by external signals is unclear. These molecules are very briefly reviewed in **Box 1**.

## **THE START OF MP MIGRATION**

Neurons become multipolar at the boundary between the VZ and SVZ/IZ (Tabata and Nakajima, 2003; Noctor et al., 2004; **Figure 2**, stage 2A). It is not clear whether this transformation is active or passive. The BP-MP transition could be actively induced by signals present in the SVZ/IZ, but such signals have not been identified. Alternatively, physical interactions could induce the morphological change. The IZ contains a dense neuropil of horizontally-packed axons that may be a barrier to radial migration of BP cells. This region is also the stiffest part of the developing brain (Iwashita et al., 2014). Matrix stiffness influences the morphology and migration behavior of mesenchymal cells, and may similarly affect neurons (Roca-Cusachs et al., 2013).

#### **BOX 1 | GTPases, protein kinases and cytoskeletal proteins implicated in the transition from multipolar to bipolar migration.**

In addition to the principal signaling pathways discussed in the main text, the following genes and proteins are implicated in regulating the transition from MP migration to BP migration in the CP. It is not clear whether they are regulated by external signals: they may be cell-intrinsic, or controlled by a transcription program set in motion at the time of neurogenesis.

The protein kinase **Mst3/Stk25** regulates MP migration, apparently by inhibiting RhoA (Tang et al., 2014). Mst3-deficient cells accumulate in the IZ with rounded morphology. Mst3 contains a site for phosphorylation by Cdk5. Cdk5 phosphorylates Mst3 in vitro, and Mst3 phosphorylation is reduced in Cdk5-/- brain. Kinase-defective and non-phosphorylated mutants of Mst3 fail to rescue migration. This suggests that Cdk5 phosphorylation and activation of Mst3 is required for migration out of the IZ. In vitro, Mst3 over-expression reduces RhoA GTP loading. Remarkably, Mst3 directly phosphorylates RhoA and a non-phosphorylated RhoA mutant has increased GTP loading. This suggests that Mst3 directly inhibits RhoA. Overexpressed RhoA causes IZ arrest, and RhoA knockdown rescues Mst3-inhibited neurons, consistent with Mst3 activation by Cdk5 inhibiting RhoA and permitting exit from the IZ. However, it is not known whether Mst3 activity and RhoA phosphorylation change during migration.

**Rnd2** is an unusual GTPase that is regulated primarily by expression level. Rnd2 expression is induced soon after neurons leave the VZ by proneural gene Neurog2 and maintained in MP cells by NeuroD1 (Heng et al., 2008). Neurog2 and Rnd2 are required for the MP-BP transition, and the requirement for Neurog2 can be by-passed if Rnd2 is artificially expressed (Hand et al., 2005; Heng et al., 2008). The main function of Rnd2 appears to be as an inhibitor of **RhoA**. Accordingly, neurons lacking Neurog2 can be rescued by inhibiting RhoA. The results imply that Rnd2 inhibits RhoA during MP migration and high RhoA activity delays the MP-BP transition.

A requirement for inhibition of RhoA by Rnd2 seems to be in conflict with the requirements for activation by Mst3 and by PlxB2 (see main text) (Azzarelli et al., 2014). How can we reconcile these findings? It is possible that cycling of RhoA between GTP and GDP states is required, or activity be increased in some parts of the cell and inhibited in others. Alternatively, it is possible that some intermediate level of RhoA activity is permissive for radial migration.

**GSK3** is important for neuron polarization and axon growth and branching in vitro (Hur and Zhou, 2010). In bipolar neurons, activated GSK3β localizes the microtubule plus-end binding protein adenomatous polyposis coli (APC) to the distal ends of microtubules in the tip, allowing centrosomal forward movement and neuronal migration (Asada and Sanada, 2010). Morgan-Smith et al examined cortical lamination by use of mice lacking GSK3α and deleting GSK3β in early postmitotic neurons with NeuroD6- Cre (Morgan-Smith et al., 2014). While lower neuron layers were normal, upper layers were dispersed, suggesting defects in MP migration or locomotion along RG. When Cre was electroporated into the VZ at E15, GSK3-deficient neurons became arrested in the IZ with MP morphology. Axons were present, but branched abnormally. Phosphorylations of DCX at Ser327 and CRMP2 at Thr514 were inhibited. GSK3 also regulates the canonical Wnt signaling pathway, but mutation of β-Catenin or triple mutation of all Disheveled family members with NeuroD6-Cre did not lead to gross layering defects.

#### **BOX 1 | Continued**

Rab GTPases regulate membrane traffic. When endocytosis is inhibited with dominant-interfering **Rab5** or Rab5 knockdown, some neurons accumulate in the IZ with abnormal, rounded morphology (Kawauchi et al., 2010). However, other neurons had bipolar morphology but were stalled at the bottom of the CP with an abnormally thick trailing process. Co-culture experiments suggested that Rab5-inhibited neurons bind more tightly to RG. Rab5-inhibited cells have a small increase in surface NCad, and NCad knockdown partially rescues their migration. This suggests that abnormally high NCad after the MP-BP transition may inhibit RG-dependent locomotion.

In contrast with Rab5's role in endocytosis, **Rab11** regulates membrane recycling to the cell surface. Dominant-interfering Rab11 reduces NCad on the cell surface and increases NCad in recycling endosomes (Kawauchi et al., 2010). Neuron migration is also slowed, although the exact stage was not determined. If the delay is in the IZ, then the results would be consistent with Rab11 mediating the Reelin-dependent NCad exocytosis in the MP stage, and Rab5 mediating NCad endocytosis to allow locomotion.

**Cdk5** is a protein kinase that is related to cell cycle kinases but is expressed and functions in non-mitotic cells, including neurons. Cdk5 is required for the MP-BP transition (Ohshima et al., 2007). Cdk5-deficient cells extend axons but remain multipolar and retarded in the IZ. When cells do enter the CP, their leading processes are often branched, suggesting problems with stabilizing a single leading process (Hatanaka et al., 2004; Ohshima et al., 2007). A similar phenotype was noted for cells lacking the Cdk5 activator, p35 (Gupta et al., 2003). Cdk5 phosphorylates many proteins involved in MP migration, including DCX, Ndel1 and p27kip1 (reviewed by Ayala et al., 2007) and axin (Fang et al., 2011). A key question for Cdk5 is whether it is dynamically regulated by external signals during MP migration or is constitutively active. The serotonin 6 receptor, **5HT6R**, was recently reported to regulate MP exit (Jacobshagen et al., 2014). The authors suggested that 5HT6R may work through Cdk5, because 5HT6R binds to Cdk5 and the migration of 5HT6R-deficient cells was partly rescued by over-expressing Cdk5 and its activator, p35 (Jacobshagen et al., 2014). Unfortunately, there is no evidence that serotonin or other extracellular 5HT6R ligands regulate migration.

The role of the **Jnk1** pathway is controversial. Jnk1 activity, measured using antibodies to phospho-Jnk, is high in the IZ, where the Jnk1 activating kinase, **DLK** is highly expressed (Hirai et al., 2002; Kawauchi et al., 2003). Deletion of the DLK gene decreases Jnk1 activity, inhibits axonogenesis, and slows neurons at the IZ-MZ boundary (Hirai et al., 2006). DLK is a member of the Mixed Lineage Kinase family, and like other family members can be activated by Rac1. This suggests that Jnk may be activated by Rac1 via DLK. Indeed, dominant-interfering **Rac1** inhibits Jnk activation in vivo (Kawauchi et al., 2003). In turn, Rac1 may be activated by Rac1 GEFs **STEF** and **Tiam1** which are highly expressed in the IZ and CP (Kawauchi et al., 2003). Accordingly dominantinterfering STEF/Tiam1, dominant-interfering Jnk1, or Jnk inhibitors all stall neurons in the IZ, suggesting that a STEF/Tiam1-Rac1- DLK-Jnk1 pathway is involved in IZ exit. The phenotypes of the inhibited neurons are not consistent with a simple, linear pathway, however. Rac1-inhibited neurons are rounded, with reduced processes, suggesting a general defect in process extension and a failure of radial polarization. In contrast, Jnk1-inhibited neurons have undergone the MP-BP transition but the leading process is twisted and irregular (Kawauchi et al., 2003). This suggests that Rac1 regulates MP migration through several effectors, and that Jnk1 is involved in locomotion of BP neurons out of the IZ.

#### **BOX 1 | Continued**

A different conclusion was drawn by Westerlund et al., who found that **Jnk1** is an inhibitor, not an activator, of the MP-BP transition (Westerlund et al., 2011). They found that Jnk1 knockdown or gene deletion stimulates axon outgrowth, IZ exit and migration through the CP. The mechanism appears to involve the neuron-specific Stathmin family member **SCG10**. SCG10 stabilizes microtubules when it is phosphorylated by Jnk1. In vivo, SCG10 knockdown or expression of non-phosphorylated mutant SCG10 stimulates axon outgrowth and IZ exit (Westerlund et al., 2011). Tyrosinated (unstable) tubulin is increased in Jnk1-/- cortex. Expressing phosphomimetic mutant SCG10 in Jnk1 knockdown neurons restores the normal, slow, exit from the MP zone. This leads to a model in which Jnk1 phosphorylates SCG10, stabilizes microtubules and inhibits the microtubule remodeling required for axon outgrowth and for the MP to BP transition. It is not clear how to reconcile this study with those of Kawauchi et al. (2003) and Hirai et al. (2006).

**srGAP2**: srGAP2 contains an F-BAR membrane-bending domain and a GAP domain specific for Rac1. Knockdown of srGAP2 decreases the number of neurons in the IZ and lower CP, suggesting that endogenous srGAP inhibits IZ exit and BP locomotion through the CP (Guerrier et al., 2009). Rapidly migrating srGAPdeficient neurons in the CP have a less-branched leading process than normal. Endogenous srGAP2 may thus inhibit formation of a single leading process. Accordingly, over-expressed srGAP2, or just the F-BAR domain, inhibits the MP-BP transition. However, it is not known if srGAP2 activity changes during migration.

**Kinesin6**: Kinesin6 is a plus-end directed microtubule motor that also binds actin. Kinesin6 knockdown inhibits the MP-BP transition (Falnikar et al., 2013). Kinesin6 crosslinks and slides antiparallel microtubules in the mitotic spindle. Similarly, in differentiating neurons, kinesin6 helps establish and maintain antiparallel microtubules in the dendrites (Lin et al., 2012). In polarized migrating neurons, kinesin6 concentrates near the centrosome in the base of the leading process, potentially helping maintain mixed orientation microtubules in the leading process and concentrating actin in this region (Falnikar et al., 2013).

**FGF13** is an FGF-homologous family (FHF) member that is not secreted, but that acts inside the cell through mechanisms that are receptor-independent. Wu et al showed that FGF13 localizes in growth cones of cultured neurons, binds to microtubules, and stimulates microtubule polymerization (Wu et al., 2012). FGF13 deleted neurons have defective axon outgrowth in vitro. In vivo, FGF13 knockout delays MP neurons in the IZ, and those neurons that do enter the CP have an excessively branched leading process. Knockdown of FGF13 in utero causes a similar MP delay, which is rescued by wildtype FGF13 but not by a multi-alanine mutant that does not bind microtubules. In many ways, FGF13 resembles DCX, which is also a + end microtubule stabilizer that is enriched at the leading edge. Indeed, DCX knockdown causes IZ delay, which is partly rescued by FGF13 over-expression. Reciprocally, FGF13 knockdown is partly rescued by DCX over-expression, suggesting that DCX and FGF13 function in parallel.

Alternatively, cells may transform to MP morphology in order to weave between the axons that cross the radial path.

A third possibility is that proneural genes, expressed at the last RG division, induce the expression of proteins that cause pin-like cells to detach from the VZ and become multipolar (Itoh et al., 2013b). For example, induction of **Scratch** down-regulates **E Cadherin** (ECad, Cdh1), allowing separation of apical junctional

on processes that appear to be dynamically regulated by intrinsic transcription changes or signals from the environment. Reelin and neurotrophins BDNF and NT3 are shown as non-directional signals, while Sema3A is thought to form a gradient and provide a direction signal. TAG1 is a protein displayed on the surface of axons in the lower IZ. Abbreviations: junction dis., junction disassembly; actin reorg., actin reorganization; up arrow, increase; down arrow, decrease.

complexes (Itoh et al., 2013a). **p27kip1**, **Dcx** and **Rnd2** are other examples. Dcx and Rnd2 seem to be regulated by cell-intrinsic factors and are considered further in **Box 1**. p27 is considered further here because its expression may be dynamically regulated by signaling through gap junctions (Liu et al., 2012).

**p27** is well known for regulating the cell cycle, but it also inhibits the small GTPase **RhoA** and regulates microtubules, thereby increasing cell motility (Besson et al., 2004, 2008; Godin et al., 2012). p27 knockout inhibits neuron migration from the VZ to the IZ, which is rescued by a mutant that lacks the N-terminal, cell cycle kinase regulatory function but retains its C-terminal, RhoA-binding region (Nguyen et al., 2006). Surprisingly, p27 expression in MP cells requires **Connexin 43** (Cx43): p27 expression is inhibited by Cx43 knockdown and increased by Cx43 overexpression (Liu et al., 2012). Knockdown of Cx43 inhibits IZ entry, with a consequent decrease in MP cells (Elias et al., 2007; Liu et al., 2012). Cx43 is best known as a component of gap junctions, aqueous channels for cell-cell communication. In the CP, however, Cx43 provides adhesion between BP neurons and RG, thereby enabling locomotion (Elias et al., 2007). Empirically, one would not expect that MP entry would require cell-cell adhesion. Indeed, MP entry requires Cx43's channel function and C terminus, suggesting that gap junctions are involved (Liu et al., 2012). Cx43-dependent cell-cell communication could prime the p27 gene for induction during cell migration. The communication could occur between cells in the SVZ or between RG progenitors. Indeed, Cx43 is also needed for gap junction communication between progenitors and for interkinetic nuclear movement in the VZ (Liu et al., 2010). Therefore, gap junction communication between RG progenitors could be permissive for p27 gene induction by proneural genes when neuroblasts reach the IZ.

**p27** is also required later, for MP migration and exit from the IZ. A detailed study from Kawauchi et al. (2006) showed that acute knockdown of p27 with shRNA arrests neurons in the IZ. Cells in the lower IZ arrest with rounded morphology and reduced processes but those in the upper IZ have a clear radial process, suggesting defects in MP migration and locomotion of BP cells out of the IZ. The authors found that p27 inhibits phosphorylation of **cofilin**, an actin-severing protein that is inhibited by phosphorylation (i.e., p27 stimulates actin severing) (Kawauchi et al., 2006). This function of p27 is critically dependent on phosphorylation of p27 by **Cdk5** at Ser10, which protects p27 from proteasomal degradation and increases p27 levels. p27 likely inhibits cofilin phosphorylation by inhibiting **RhoA**, a known activator of the cofilin kinase, **LIMK** (Kawauchi et al., 2006). Therefore, Cdk5 and p27 together promote IZ exit by activating cofilin. Curiously, however, excess active cofilin also inhibits IZ exit (Kawauchi et al., 2006), suggesting that the balance between active and inactive cofilin is critical.

#### **AXON INDUCTION**

MP cells extend axons during the random migration phase, before they reorient towards the pia and begin radial migration (Hatanaka et al., 2004; Barnes and Polleux, 2009; Hatanaka and Yamauchi, 2013; Namba et al., 2014; **Figure 2**, stage 2B). Axon extension is first detected in the lower IZ, which contains corticofugal efferents (Hatanaka et al., 2009; Namba et al., 2014). This suggests that signals from corticofugal axons may induce or stabilize axons. Corticofugal axons express the homophilic adhesion molecule **TAG1** (Cntn2), which is absent from thalamocortical axons in the upper IZ (Fukuda et al., 1997). Nascent axons produced by MP cells also express TAG1 and align with the corticofugal axons (Namba et al., 2014). TAG1 knockdown in the MP neurons inhibits axon formation and inhibited radial migration. The functional domains of TAG1 that are required for homophilic binding in trans *in vitro* are also required cellautonomously for axon outgrowth. These results suggest that signals from TAG1 in the environment induce and stabilize the nascent axon by homophilic interactions. TAG1 is linked to the cell surface by a C-terminal glycosyl-phosphatidyl-inositol (GPI) moiety and is localized to lipid rafts, where it potentially activates Src-family kinases. Indeed, Namba et al show that the Src family kinase (SFK) **Lyn** and small GTPase **Rac1** are likely involved in stabilizing axons downstream of TAG1 (Namba et al., 2014).

In addition to TAG1, several receptor tyrosine kinases (RTKs) regulate axon induction *in vivo*. The neurotrophin receptors **TrkB** (Ntrk2) and **TrkC** (Ntrk3) and ligands **BDNF** and **NT3** are expressed in the VZ/SVZ. Trk inhibitors inhibit axonogenesis and migration out of the IZ in slice cultures (Ip et al., 2012). Also, sequestering neurotrophin ligands by expressing dominant interfering mutants of both TrkB and TrkC by *in utero* electroporation inhibited axon outgrowth and increased the percentage of MP neurons in the IZ (Nakamuta et al., 2011). TrkB knockout causes cortical lamination defects but axon defects were not noted, perhaps due to compensation by TrkC (Medina et al., 2004). Injection of BDNF or anti-BDNF antibodies into the ventricle stimulates or inhibits neuron migration, respectively (Fukumitsu et al., 2006). However, injected BDNF accelerates the entire neurogenic and migration process, so neurons born on a given day arrive more quickly at the top of the CP, settle in a lower layer position, and express earlier fate markers than usual. Taken together, these experiments suggest that neurotrophins probably stimulate axonogenesis and radial migration independently of neurogenesis, but cause and effect are not completely clear. Another RTK, **Kit**, is also expressed in migrating neurons and knockdown inhibits axon outgrowth and delays radial migration, although the mechanism remains to be determined (Guijarro et al., 2013).

Axon induction by BDNF *in vitro* requires **phospholipase C**γ, **calcium**, and calcium-calmodulin regulated kinase kinase (**CaMKK**α**; Nakamuta et al., 2011**). Dual inhibition of TrkB and TrkC or knockdown of phospholipase Cγ or CaMKKα inhibited axon outgrowth and radial migration *in vivo*, suggesting that signaling through phospholipase Cγ and CaMKKα is important. An alternative mechanism involves α2-chimerin (Ip et al., 2012). α2 chimerin is a Rho GAP (GTPase activating protein) that contains an SH2 domain through which it binds to BDNF-stimulated TrkB (Ip et al., 2012). α2-chimerin knockdown at E14 stalls neurons in the MP stage. Rescue experiments showed that migration requires the SH2 domain but not GAP activity. α2-chimerininhibited neurons have abnormally high levels of dephosphorylated (active) **CRMP2**, a protein that stabilizes microtubules. This suggests that α2-chimerin-stimulation of CRMP2 phosphorylation may regulate migration. Consistent with this, migration of α2-chimerin-deficient cells was rescued by over-expressing wildtype or phospho-mimetic CRMP2 but not a non-phosphorylated mutant. Other studies have shown that CRMP2 is needed for the MP-BP transition, consistent with the idea that CRMP2 phosphorylation is needed to remodel the microtubule cytoskeleton (Sun et al., 2010). However, axon formation requires **Cdk5** and **axin** to inhibit **GSK3**β and thereby reduce phosphorylation of CRMP2 (Fang et al., 2011). This implies that CRMP2 is dephosphorylated during axon formation and re-phosphorylated when a radial leading process is stabilized. Nevertheless, there is no evidence that CRMP2 phosphorylation state changes during axon formation or migration. Indeed, the immunoreactivity of phospho-CRMP2 relative to total CRMP2 is constant through the IZ and CP (Ip et al., 2012). More study of specific phosphorylation sites and subcellular localization of CRMP2 may help resolve the mechanism.

## **THE COORDINATION OF AXON INDUCTION WITH THE TRANSITION FROM RANDOM TO RADIAL MIGRATION**

The appearance of an axon before random migration ends suggests the possibility that axon formation is a prerequisite for starting radial migration (**Figure 2**, stage 2C). The TAG1 results are interesting in this regard. As mentioned above, TAG1 knockdown inhibits both axonogenesis and radial migration (Namba et al., 2014). Since TAG1 is an axonal molecule, it is unlikely to regulate radial migration directly. Therefore, this study suggests that axon outgrowth may be required before radial migration starts. In addition, inhibiting TrkB, TrkC and Kit impaired radial migration as well as axon induction (Fukumitsu et al., 2006; Nakamuta et al., 2011; Ip et al., 2012; Guijarro et al., 2013), consistent with, but not proving, causality.

On the other hand, there is evidence that neurons can enter the CP without axons. Deleting the kinase **LKB1** (Stk11, Par4) in progenitors inhibited axon outgrowth but neurons still migrated out of the IZ and into the CP, albeit in reduced numbers and with abnormal, highly branched leading processes (Barnes et al., 2007). LKB1 mutation does not cause gross lamination defects (Morgan-Smith et al., 2014). However, different results were obtained when LKB1 was acutely knocked down with shRNA (Asada et al., 2007; Matsuki et al., 2013). Many LKB1-deficient neurons accumulated in the IZ with multiple axon-like processes, suggesting defects in polarization which impact the selection of a single axon and migration (Asada et al., 2007). Acute knockdown of two LKB1 co-activators, **Stk25** and **STRAD**α, also inhibited axon outgrowth and delayed cells in the IZ (Matsuki et al., 2010, 2013; Orlova et al., 2010). However, germline knockout of Stk25 did not affect layering or axonogenesis (Matsuki et al., 2013). The results are consistent with important roles for Stk25-STRADα-LKB1 in axon outgrowth and radial migration, but there may be bypass mechanisms that come into play when these proteins are absent. Overall the question of whether axonogenesis is required before a multipolar cell can stabilize a pia-directed leading process and start radial migration remains unanswered.

## **Unc5D REGULATES THE TRANSITION TO RADIAL MIGRATION**

**Unc5D** is a co-receptor for **Netrins**, forming a complex with **DCC**/Unc40 and modulating DCC signaling. Unc5D expression is highest in the IZ (Sasaki et al., 2008). Transcription factor **FoxG1** represses Unc5D expression (Miyoshi and Fishell, 2012). FoxG1 is expressed in the VZ/SVZ and CP but is reduced in the IZ. Thus Unc5D increases when cells start MP migration and decreases when cells enter the CP. This suggests that the decline in Unc5D may trigger radial migration. However, inhibiting Unc5D expression (by over-expressing FoxG1) delayed the MP-BP transition, and co-over-expression of FoxG1 and Unc5D rescued normal migration. This suggests that the MP-BP transition requires high levels of Unc5D but the disappearance of Unc5D does not define the timing. In other systems, Netrin can bind to DCC in the absence of Unc5D, but the responses to DCC and Unc5D-DCC are different. DCC is expressed continuously during neuron migration. This means that MP cells have Unc5D-DCC complexes and Unc5D-deficient cells (which arrest in the IZ) have only DCC. This suggests that free DCC might inhibit IZ exit. Indeed, over-expressing DCC delays MP exit, which is rescued by co-expressing Unc5D (Miyoshi and Fishell, 2012). This suggests that MP exit may require DCC repression by Unc5D as a permissive event.

# **REELIN AND N-CADHERIN AS PERMISSIVE SIGNALS FOR RADIAL POLARIZATION**

**Reelin** is a secreted protein that regulates neuron migration in many brain regions (Honda et al., 2011). Full-length Reelin is most abundant where it is made in the MZ, but it is also cleaved and active fragments diffuse through the CP to the IZ (Jossin et al., 2007; Uchida et al., 2009). Genetic deletion of Reelin interferes with migration at several steps (Honda et al., 2011). Overexpressing dominant-negative Reelin receptors induces delay in the MP phase, with little or no inhibition of axon outgrowth (Jossin and Cooper, 2011).

Signaling proteins that are activated by Reelin are also required for radial polarization of MP cells. *In vitro*, Reelin stimulates a pathway including tyrosine phosphorylation of **Dab1** by **Src** and **Fyn**, recruitment of **Crk/CrkL** adaptors and **C3G**, a guanine nucleotide exchange factor (GEF) for **Rap1**, and increased GTP binding to Rap1 (Cooper et al., 2008). Some of these signaling proteins are also important in the MP-BP transition. For example, Dab1 knockdown inhibits migration in the IZ (Young-Pearse et al., 2007). Src and Fyn, the Crk/CrkL binding sites on Dab1, and the Crk and CrkL genes, are all required for normal lamination, but MP migration was not analyzed (Park and Curran, 2008; Feng and Cooper, 2009). C3G mutants arrest in the IZ with MP morphology and decreased Rap1GTP, implicating C3G and Rap1 (Voss et al., 2008). *In utero* electroporation with dominantinterfering Reelin receptors or Rap1A inhibitors also delays radial polarization, and over-expressed Rap1A partly rescues Reelininhibited MP cells, suggesting that Rap1A activation is necessary and partly sufficient for polarization (Jossin and Cooper, 2011). Rap1GTP regulates GEFs for the Ras-family member **RalA**, which regulates exocytosis, as well as Rho-family members **Rac1** and **Cdc42**, which regulate the actin cytoskeleton. Inhibition of Rac1or Cdc42 causes arrest of MP cells in the IZ (Kawauchi et al., 2003; Konno et al., 2005; Jossin and Cooper, 2011). Inhibition and rescue experiments implicated RalA, RalB, Rac1 and Cdc42 in Rap1-dependent radial polarization, suggesting that exocytosis and the actin cytoskeleton are involved (Jossin and Cooper, 2011).

The importance of exocytosis for radial migration is reinforced by the finding that a primary role of Reelin in the IZ is to induce surface traffic of **N-Cadherin** (NCad, Cdh2) (Jossin and Cooper, 2011). Rap1- or RalA-inhibited neurons have decreased cell-surface NCad and their migration is rescued by NCad over-expression. Dominant-interfering NCad inhibits MP exit (Kawauchi et al., 2010; Jossin and Cooper, 2011). The results suggest that Reelin regulates Rap1, RalA and other small GTPases to upregulate NCad, and that surface NCad is required to sense direction signals for radial migration and the MP-BP transition. NCad is not needed for locomotion of BP neurons, but is required again later, when Reelin regulates the final positioning of neurons at the top of the CP (Franco et al., 2011; Gil-Sanz et al., 2013).

It is unclear whether Reelin induces radial migration or is permissive. *In vivo*, NCad levels are high throughout the IZ, dependent on Reelin (Jossin and Cooper, 2011), so Reelin may induce NCad translocation to the cell surface throughout the period of MP migration. Moreover, Reelin is unlikely to be a direction signal: Reelin produced in the VZ or soaked into a brain slice can substitute for Reelin from the MZ (Magdaleno et al., 2002; Jossin et al., 2004). Therefore, Reelin may prime cells to sense another molecule that provides a direction signal and triggers radial migration as an instructive cue (Jossin and Cooper, 2011). Reelin is also required for correct radial orientation of other types of neurons (Landrieu and Goffinet, 1981; Nichols and Olson, 2010; Schneider et al., 2011; O'Dell et al., 2012), but in these situations it is also not clear whether it is instructive or permissive.

#### **SEMA3A MAY INDUCE AND ORIENT THE LEADING PROCESS**

**Plexin** (Plx) and **Neuropilin** (Nrp) family proteins are coreceptors for secreted **Semaphorins**, such as Sema3A. Importantly, Sema3A is secreted at the top of the CP and forms a gradient through the cortex. It is thus ideally suited to be a chemotactic factor for radial migration.

Seminal studies from Polleux and Ghosh showed that Sema3A orients cortical neurons so their dendrites grow towards the pia and axons towards the ventricle (Polleux et al., 1998; Polleux and Ghosh, 2002). Sema3A acts directly as an instructive signal by inhibiting axons and promoting dendrites (Shelly et al., 2011). Since the leading process of radially migrating neurons later develops into the dendritic tree, Sema3A may similarly induce the radial leading process on MP neurons. Indeed, Azzarelli et al found that MP neurons accumulate in the IZ when PlxB2 is inhibited (Azzarelli et al., 2014). Similarly, knockdown or Cremediated deletion of Nrp1, PlxA2, PlxA4 or PlxD1 increased the number of neurons trapped in the white matter and lower layers after birth, consistent with inhibition of MP migration (Chen et al., 2008). *In vitro*, Sema3A attracts neurons when added to one side of a slice culture or to one side of a porous membrane (Chen et al., 2008). This suggests that the gradient of Sema3A could provide a direction signal for radial migration, perhaps by stabilizing the leading process. However, it should be noted that cortical lamination is normal in Nrp1 mutant mice, at least up to E14.5 when embryonic lethality precludes analysis (Hatanaka et al., 2009). *In utero* expression of Nrp1/2 inhibitors at E12.5 or E15.5 also had no major effects on migration, although axon pathfinding was impaired (Hatanaka et al., 2009). It is not clear how to reconcile these findings.

The downstream signaling from Plx and Nrp receptors is currently unclear. However, PlxB2 may induce radial migration by stimulating **RhoA**. RhoA has a general role in stabilizing filamentous actin and inducing actomyosin contraction. PlxB2 binds to a RhoA activator, **PDZ-RhoGEF**/LARG. Removing PlxB2 lowered RhoA activity and over-expressing RhoA partly rescued the migration of PlxB2-deficient neurons. The C terminus of PlxB2 is required to bind PDZ-RhoGEF and rescue migration of PlxB2-deficient cells. This suggests that PlxB2 activates RhoA through PDZ-RhoGEF and that RhoA activity is needed for radial migration. However, there is currently no evidence that RhoA activity increases when radial migration starts. Cyclic nucleotides may also be involved (Liu et al., 2004). One attractive possibility is that Sema3A, coming from the top of the cortex, contacts the upper side of a multipolar cell, activates PlxB2, Nrp1 and RhoA locally, inducing and stabilizing a pia-directed process. However, testing this model will require improved RhoA activity reporters and high-resolution live-imaging of MP neurons as they reorient radially in the IZ.

## **OTHER SIGNALING MOLECULES REQUIRED FOR RADIAL MIGRATION: INSTRUCTIVE OR PERMISSIVE?**

The last decade has seen the discovery of a plethora of proteins that are required for or inhibit the MP-BP transition. In most cases, authors have used loss of function (knockdown or knockout) and gain of function (in most cases, over-expression) experiments to demonstrate necessity and sufficiency of specific proteins for radial migration. Some of the proteins, including LIS1 (Pafah1b1), Ndel1, dynamin, FilaminA and Doublecortin (DCX), were reviewed thoroughly previously, and will not be considered here (LoTurco and Bai, 2006; Ayala et al., 2007). Progress in understanding other genes that regulate radial migration is summarized in **Box 1**. However, in nearly every case, it remains unclear whether they are permissive or whether their activation (or inhibition) sets the start time for radial migration and transformation to BP morphology.

# **LATERAL MIGRATION OF MP CELLS AND THE RADIAL UNIT HYPOTHESIS**

Once MP migration starts, neurons migrate laterally as well as up and down. The horizontal movement needs to be reconciled with the radial unit hypothesis (Rakic, 2000). This postulates that cells move radially from the VZ to CP, allowing the protomap of functional areas in the VZ to be relayed into the CP. Radial migration means that sister neurons are arranged in "mini-column" units (reviewed by Gao et al., 2013). Consistent with strictly radial migration, locomoting neurons are generally detected within 1 or 2 cell diameters (10–20 µm) of their sister RG (Noctor et al., 2001, 2002). After migration is complete, neurons preferentially form electrical and chemical synapses with their sisters in other layers (Yu et al., 2009, 2012). If neurons wander laterally during MP migration, how do they remain close to their parental RG and how do they find their sisters when they are forming circuits?

One way that the MP cells could remain close to their parent RG would be to follow a random walk. Given a MP velocity of 2–4 µm/h (Tabata and Nakajima, 2003; Jossin and Cooper, 2011; Dimidschstein et al., 2013) and travel time of 24 h, a neuron could travel 50–100 µm in a straight line but just a few cell diameters (10–20 µm) if it changes direction every hour. However, the distance would be expected to increase with the time spent migrating. The relationship between horizontal spread and time spent migrating has not been established, but Sanada et al. (2004) found that the lateral dispersion of neurons was reduced from 16 µm to 4–8 µm when Reelin signaling was inhibited, even though the cells had presumably spent more time in the IZ and had progressed less far into the CP. This is the opposite of what would be expected if lateral dispersion is passive.

An alternative hypothesis is that the direction of MP migration is not random, but is controlled by signaling molecules. Indeed, there is now excellent evidence that horizontal movement of MP cells is regulated by Ephrin (Efn) and Eph family proteins (**Figure 3**).

Efns are cell surface proteins that activate Eph family RTKs on other cells and generate "forward signals". Reciprocally, an Eph can activate "reverse signaling" into an Efn-expressing cell. EfnA molecules are glycosylphosphatidylinositol-linked cell surface proteins that bind EphA family members. EfnB molecules are transmembrane proteins that bind EphB family members. Efns and Ephs regulate cell sorting in many developmental systems (Fagotto, 2014).

Several **EfnA** and **EphA** genes are expressed in the developing cortex (Torii et al., 2009). In the IZ, EfnAs are expressed more medially and less laterally. Their ligands, EphA RTKs, are expressed in the IZ in a counter gradient, higher laterally than medially. This expression pattern suggests that EfnA or EphA activity could regulate lateral dispersion, e.g., if cells of the same EfnA:EphA ratio sort together. This hypothesis was tested by studying a EfnA2/3/5 triple mutant (Torii et al., 2009). At birth, there were no gross effects on cortical development or layer order, but the cortical layers contained undulations. When clones were labeled at E12.5 and examined at E14.5, neurons in wildtype clones averaged 20 µm from the RG while they were only 10 µm distant in the triple knockout. Live imaging confirmed that tangential migration is inhibited when EfnAs are absent.

Somewhat paradoxically, over-expression of an EfnA ligand, EphA7 or EphA4, caused over-expressing neurons to segregate from non-expressing neurons in the IZ (Torii et al., 2009). Each cluster contained neurons from several clones. Later, these clusters gave rise to cortical columns of over-expressing cells separated from other columns by normal cells. Column formation required EphA cytoplasmic domains, suggesting forward signaling from EfnA to EphA. Live imaging showed that this increased clumping is actually due to greater dispersion of EphA over-expressing neurons while they are migrating in the MP zone. Thus, the live imaging is consistent with the EfnA triple knockout, in that decreased EfnA (and decreased EphA forward signaling) decreases

on the strengths of signaling from EphA-EfnA and EphB-EfnB signals. See

text for details and references.

lateral movement in the IZ and increased EphA expression (and presumably EphA forward signaling) increases lateral movement of MP cells in the IZ. The results are consistent with MP cells choosing their lateral position based on the strength of EphA forward signaling (**Figure 3**, compare blue and red routes in absence and presence of EphA signaling).

In contrast with EfnAs stimulating lateral movement by forward signaling, endogenous EfnBs may inhibit lateral movement by reverse signaling. Over-expression of **EfnB1** at E13 in post-mitotic neurons caused over-expressing neurons to segregate away from non-expressing neurons (Dimidschstein et al., 2013). Despite the marked change in lateral distribution, layering was normal. There was no evidence for increased homotypic adhesion. Rather, the MP cells had shorter processes, perhaps indicating reduced migration in the IZ. This effect required the extracellular and cytoplasmic domains of EfnB1, consistent with reverse signaling from EphBs. The C terminus of EfnB1 binds a Rho-family GEF, **P-Rex1**. The abnormal clustering induced by over-expressing EfnB1 required the EfnB1 C terminus, P-Rex1 and **Rac3**, a little studied Rho-family member that is a primary substrate for P-Rex1 and is implicated in opposing Rac1 and reducing neurite number (Dimidschstein et al., 2013). This suggests that high levels of EfnB1 activity reduce MP migration in the IZ, reducing lateral spread (**Figure 3**). Consistently, deletion of EfnB1 increased the number of MP cell neurites and the speed of movement in the IZ. However, lateral spread of deleted clones was only increased by ∼20%. It is possible that combined disruption of EfnB1 with EfnB2 and EfnB3 would give a more severe dispersion and confirm the importance of endogenous EfnB family members in regulating horizontal movement.

#### **THE VARIED DURATION OF MP MIGRATION MAY AFFECT LAMINATION**

Time-lapse recordings and birthdating studies indicate that MP migration lasts for at least a day in the mouse, but the timing is very heterogeneous (Noctor et al., 2004; Westerlund et al., 2011). Neurons that spend longer in the MP phase are expected to arrive later at the top of the CP than neurons that travel more rapidly. This means they will enter higher cortical layers. Since neuron fate is thought to be fixed at the last division of the RG progenitor, different transit times are expected to blur the discrete layering of neuronal subtypes that characterizes the mature cortex (Molyneaux et al., 2007; **Figure 4**). It remains unclear whether or how the varied timing of MP migration is compensated at other stages of development.

One major source of variability is that post-mitotic neurons and IP enter and leave the IZ at different rates (Tabata et al., 2009). Radial glia daughters (RGDs) that are post-mitotic move slowly (>10 h) from the VZ to the SVZ/IZ but have a relatively short MP phase, entering the CP between 36 and 60 h (**Figure 4**, see RGD1). In contrast, IPs move quickly (<10 h) to the SVZ/IZ but then divide. The intermediate progenitor daughters (IPDs, **Figure 3**) become multipolar and continue migrating. They enter the CP >60 h after the RG division (Tabata et al., 2009). IP daughters are thus expected to layer above their "aunts" (RG daughters) at the top of the CP (**Figure 3**). Does this mean that IP daughters switch

**FIGURE 4 | Different timing of intermediate zone exit influences cortical lamination** Neurons spend variable times migrating in the intermediate zone (IZ). Postmitotic neurons derived directly from radial glia (i.e., radial glia daughters, RGD) spend approximately 2 days in the IZ before entering the cortical plate (CP). Intermediate progenitors (IP) divide to create intermediate progenitor daughters (IPDs), for a total time of approximately 5 days in the IZ. As a result RGD born on day 0 (RGD1) arrive at the top of the CP 2–3 days before IPDs whose mother IP also left the VZ on day 0. As a result the IPDs layer above RGD1, and co-layer with RGD2, born 2 days later. If neuron fate is fixed at the last RG division, then this would cause mixing of neurons of different fates in the same layer.

to a later fate than RG daughters or does this lead to broadening and mixing of cortical layers?

One idea is that fates are not fixed at the last division of the RG progenitor but remain plastic (Fishell and Hanashima, 2008). Cell fate may be determined at the last division, whether it occurs in the VZ or SVZ/IZ. In this way, the fates of IP daughters and RG daughters entering the same lamina could be equivalent. Indeed, manipulation of FoxG1 not only delays cell exit from the IZ, so neurons shift to a higher lamina, but also adjusts the neuron fate to match the new position (Miyoshi and Fishell, 2012; Toma et al., 2014). Further testing of this hypothesis during normal development requires approaches for distinguishing IP daughters from RG daughters, and testing whether they co-layer and express the same markers (Tabata et al., 2009).

#### **CONCLUSIONS**

Amazing technical refinements in mouse genetics, imaging, *in utero* manipulation, and slice culture methods have led to the discovery of a large number of genes that regulate MP migration, axonogenesis, and direction of movement. Many of the genes have been subjected to epistasis analysis, implying causal relationships and sequences of contingent events. Some protein activities are regulated by external signals, while others may be "hard-wired" by cell intrinsic mechanisms or transcription changes. Despite dramatic progress, it is still unclear how these different processes fit together. We have limited understanding of how different pathways interact and exactly when and where in the cell signaling occurs. Analysis is extremely challenging given the asynchrony of the cell population, the small size of the cells, and the difficulties of imaging signaling events within living tissue. However, given the continuous development of new technology, we can be optimistic that these obstacles will be overcome.

#### **ACKNOWLEDGMENTS**

I am grateful to Drs. Goichi Miyoshi, Yves Jossin, Gord Fishell and Nancy Ip for discussion during preparation of this review. The selection of papers and interpretation is my own, however, and I apologize to authors for papers omitted due to lack of space. Research in the author's laboratory is supported by research grants R01-NS080194 and R21-NS089888 from the U.S. Public Health Service.

### **REFERENCES**


cortex development. *J. Neurosci.* 31, 13613–13624. doi: 10.1523/jneurosci.3120- 11.2011


**Conflict of Interest Statement**: The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 06 October 2014; accepted: 29 October 2014; published online: 14 November 2014*.

*Citation: Cooper JA (2014) Molecules and mechanisms that regulate multipolar migration in the intermediate zone. Front. Cell. Neurosci. 8:386. doi: 10.3389/fncel.2014.00386*

*This article was submitted to the journal Frontiers in Cellular Neuroscience*.

*Copyright © 2014 Cooper. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms*.

# Cellullar insights into cerebral cortical development: focusing on the locomotion mode of neuronal migration

#### Takeshi Kawauchi 1, 2, 3 \*

<sup>1</sup> Department of Physiology, Keio University School of Medicine, Tokyo, Japan, <sup>2</sup> Precursory Research for Embryonic Science and Technology (PRESTO), Japan Science and Technology Agency, Saitama, Japan, <sup>3</sup> Laboratory of Molecular Life Science, Institute of Biomedical Research and Innovation, Foundation for Biomedical Research and Innovation, Kobe, Japan

The mammalian brain consists of numerous compartments that are closely connected with each other via neural networks, comprising the basis of higher order brain functions. The highly specialized structure originates from simple pseudostratified neuroepithelium-derived neural progenitors located near the ventricle. A long journey by neurons from the ventricular side is essential for the formation of a sophisticated brain structure, including a mammalian-specific six-layered cerebral cortex. Neuronal migration consists of several contiguous steps, but the locomotion mode comprises a large part of the migration. The locomoting neurons exhibit unique features; a radial glial fiber-dependent migration requiring the endocytic recycling of N-cadherin and a neuron-specific migration mode with dilation/swelling formation that requires the actin and microtubule organization possibly regulated by cyclin-dependent kinase 5 (Cdk5), Dcx, p27kip1, Rac1, and POSH. Here I will introduce the roles of various cellular events, such as cytoskeletal organization, cell adhesion, and membrane trafficking, in the regulation of the neuronal migration, with particular focus on the locomotion mode.

#### Edited by:

Thomas Knöpfel, Imperial College London, UK

#### Reviewed by:

Zoltan Molnar, University of Oxford, UK Gunnar P. H. Dietz, Schwabe Pharma Deutschland, Germany

#### \*Correspondence:

Takeshi Kawauchi, Laboratory of Molecular Life Science, Institute of Biomedical Research and Innovation, Foundation for Biomedical Research and Innovation, 2-2 Minatojima-Minamimachi Chuo-ku, Kobe 650-0047, Japan takeshi-kawauchi@umin.ac.jp

> Received: 08 May 2015 Accepted: 22 September 2015 Published: 07 October 2015

#### Citation:

Kawauchi T (2015) Cellullar insights into cerebral cortical development: focusing on the locomotion mode of neuronal migration. Front. Cell. Neurosci. 9:394. doi: 10.3389/fncel.2015.00394 Keywords: microtubule, actin cytoskeleton, endocytosis, JNK, Rab5, Rab11, Rab7, Rap1

# Introduction

The brain is divided into many compartments, such as nuclei, layered structures, and cortical areas, allowing highly organized role allocations. The systematically allocated neuronal populations are generated from spatially restricted regions, the ventricular, and subventricular zones. Therefore, a long-distance migration from the ventricular side to the final destination is essential for constructing a functional brain. In line with this, defects in neuronal migration are associated with various neurological disorders (Gleeson and Walsh, 2000; Kawauchi and Hoshino, 2008). Several types of cortical malformations, including lissencephaly, double cortex syndrome (subcortical band heterotopia) and periventricular heterotopia (PVH), are thought to result from neuronal migration defects. These cortical malformations are frequently associated with intellectual disability and intractable epilepsy (Francis et al., 2006; Moon and Wynshaw-Boris, 2013; Reiner and Sapir, 2013; Lian and Sheen, 2015). Lis1, Dcx, Filamin A, ArfGEF2, Arx, Reelin, and several Tubulin genes (TUBA1A, TUBA8, TUBB2B, TUBB3, TUBB5, and TUBG1) are identified as causative genes for these cortical malformations (des Portes et al., 1998; Fox et al., 1998; Gleeson et al., 1998; Hong et al., 2000; Kitamura et al., 2002; Kato et al., 2004; Sheen et al., 2004; Keays et al., 2007; Abdollahi et al., 2009; Reiner and Sapir, 2013; Bahi-Buisson et al., 2014; Magen et al., 2015) (**Figures 1A,B**). Furthermore, suppression of genes related to dyslexia (e.g., DCDC2, KIAA0319), autism spectrum disorder (ASD) (e.g., Auts2, CNTNAP2) and schizophrenia (e.g., SDCCAG8) disturbs neuronal migration, although it is unclear whether the neuronal migration defect is the main cause of the pathogenesis of these neurological and psychiatric disorders (Hannula-Jouppi et al., 2005; Kamiya et al., 2005; Meng et al., 2005; Paracchini et al., 2006; Wang et al., 2006; Kähler et al., 2008; Peñagarikano et al., 2011; O'roak et al., 2012; Zhang et al., 2013a; Hori et al., 2014; Insolera et al., 2014; La Fata et al., 2014) (**Figure 1B**). Interestingly, in addition to these genes, several environmental factors, such as stress and inflammation, are also associated with cortical development, including neuronal migration (Stolp et al., 2012; Hashimoto-Torii et al., 2014; Ishii and Hashimoto-Torii, 2015).

In the developing cerebral cortex, neuronal migration consists of several contiguous steps (Nadarajah and Parnavelas, 2002; Cooper, 2014; Takano et al., 2015) (**Figure 1A**). Newly generated neurons exhibit multipolar morphology in the lower part of the intermediate zone. The multipolar neurons form an axon and a leading process almost coincidentally with retraction of other processes. The bipolar-shaped neurons are called "locomoting neurons," which migrate along radial glial fibers with unique morphological changes (the locomotion mode of neuronal migration) (**Figure 2A**). At the final phase of neuronal migration, they switch over from the "locomotion" into a "terminal translocation."

Most studies so far have focused on the mechanisms of the morphological changes at the early phase of neuronal migration, including the multipolar-to-bipolar transition (Kawauchi and Hoshino, 2008; Heng et al., 2010). One reason is that many cortical malformation-related gene products are involved in the multipolar-to-bipolar transition. Second, the acquirement of neuronal polarity, a key step of neuronal maturation, occurs nearly simultaneously with the multipolar-bipolar transition. Third, suppression of cytoskeletal proteins or kinases often leads to defects in the early phase of migration due to the occurrence of various morphological changes at this stage.

In contrast to the early phase of migration, our knowledge of the locomotion mode is relatively poor. However, recent morphological and cell biological analyses have uncovered unique features of locomoting neurons. In this review, I will introduce recent advances in the molecular and cellular biology of neuronal migration with particular focus on the locomotion mode.

# The Early Phase of Neuronal Migration

Several steps of the early phase of neuronal migration, including the multipolar-to-bipolar transition, are required for the formation of the morphologies of the locomoting neurons. The formation of a leading process requires c-jun N-terminal kinase (JNK) (Kawauchi et al., 2003). Filamin A (FLNA), an actin-binding protein, is also involved in the early phase of migration, possibly including the leading process formation (Nagano et al., 2004). Knockdown of Lis1, a regulator of the dynein complex, a microtubule minus end-directed motor, suppresses the multipolar-bipolar transition (Tsai et al., 2005). FLNA and Lis1 have been identified as causative genes for PVH and lissencephaly, respectively, and both knockout of FLNA and Lis1 heterodeficiency show neuronal migration defects (Hirotsune et al., 1998; Zhang et al., 2013b).

In addition, many other molecules are reported to regulate the formation of multipolar morphology (e.g., Cdk5, p27kip1 , Arx, Rab5) (Kawauchi et al., 2006, 2010; Friocourt et al., 2008; Friocourt and Parnavelas, 2010) and multipolar-to-bipolar transition (e.g., Cdk5, PHF6, FMRP) (Ohshima et al., 2007; Zhang et al., 2013a; La Fata et al., 2014; Franzoni et al., 2015) at the early phase of migration (**Figure 1A**).

# The Locomotion Mode of Neuronal Migration

The locomotion mode of neuronal migration covers the largest part of the neuronal journey, and is therefore a main contributor to proper neuronal positioning (Rakic, 2006; Nishimura et al., 2010). As described above, however, analysis of molecular mechanisms underlying the locomotion mode is difficult, because in many cases, neurons with defects in cytoskeletal proteins or kinases also show abnormalities early in neuronal migration prior to starting the locomotion (or no phenotypes). However, recent advances in in vivo cell biological approaches and novel technologies have uncovered several molecules regulating the unique features of the locomotion mode of neuronal migration. For example, a novel method, the ex vivo chemical inhibitor technique, that allows us to directly analyze molecules involved in the locomotion mode, has recently been established (Nishimura et al., 2010). Using this technique, Cdk5 and Src family kinases were shown to regulate the locomotion mode (Nishimura et al., 2010).

The locomotion mode of neuronal migration displays two major characteristics, a radial glial fiber-dependent migration and a neuron-specific unique migration mode with dilation/swelling formation and nuclear elongation (Rakic, 1972; Bellion et al., 2005; Schaar and McConnell, 2005) (**Figure 2**). In the next subsections, I will introduce the morphological, molecular, and cellular mechanisms of these unique characters of the locomotion mode.

#### A Unique Migration Mode with Dilation/Swelling Formation

Locomoting neurons exhibit distinct migration features (Bellion et al., 2005; Schaar and McConnell, 2005; Nishimura et al., 2014). (1) Locomoting neurons extend a leading process and form a cytoplasmic dilation (also referred as to "swelling" especially in tangentially migrating interneurons) at the proximal region of a leading process. (2) The nucleus in the locomoting neurons becomes elongated to enter the cytoplasmic dilation (**Figure 2**).


FIGURE 1 | Molecules involved in the multi-step modes of neuronal migration and its related neurological disorders. (A) Immature neurons (light yellow cells), generated from radial glial progenitors (light blue cells) near the ventricle, migrate toward the pial surface. At the early phase of neuronal migration (the left three migrating neurons), many cytoskeletal-regulatory proteins (e.g., Lis1 and Filamin A), kinases (e.g., Cdk5 and JNK), and other proteins (e.g., p27kip<sup>1</sup> and N-cadherin) (Continued)

#### FIGURE 1 | Continued

are required for proper morphological changes of migrating neurons (Kawauchi and Hoshino, 2008; Lickiss et al., 2012; Kawauchi, 2014). Interestingly, some of these molecules are also involved in neurological disorders, such as Börjeson-Forssman-Lehmann syndrome (for PHF6) (Zhang et al., 2013a; Franzoni et al., 2015), Fragile X syndrome (for FMRP) (La Fata et al., 2014) and lissencephaly (for Lis1, Cdk5, and Arx) (Friocourt and Parnavelas, 2010; Reiner and Sapir, 2013; Magen et al., 2015). Subsequently, neurons undergo the locomotion mode of neuronal migration (the middle three migrating neurons). Cdk5 and its substrates, Dcx, and p27kip1, control the dilation/swelling formation during the locomotion. POSH and Rac1 are also required for the dilation/swelling formation. N-cadherin and its regulation by Rab5- and Rab11-dependent endocytic recycling play important roles in radial glial fiber-dependent migration of the locomoting neurons. At the final phase of migration, neurons undergo the terminal translocation mode (the right two migrating neurons). The Reelin-Dab1-C3G-Rap1-Talin-Integrin pathway regulates the terminal translocation. PKCδ is also required for the terminal translocation. (B) Molecules involved in neuronal migration-related brain diseases.

The cytoplasmic dilation or swelling was first identified in 2005 as a migrating neuron-specific subcellular domain, because not only other migrating cells, such as neutrophils, keratocytes, and fibroblasts, but also static neurons do not form a cytoplasmic dilation/swelling (Bellion et al., 2005; Schaar and McConnell, 2005). Electron microscopy studies show that the cytoplasmic dilation/swelling contains the centrosome, Golgi apparatus, and microtubules. Although the centrosome frequently is a part of the cytoplasmic dilation/swelling (Bellion et al., 2005; Schaar and McConnell, 2005), suppression of dynein heavy chain or Lis1, both of which are known to regulate centrosomal positioning and nuclear forward movement in radially migrating neurons, does not disrupt cytoplasmic dilation/swelling (Tsai et al., 2007). Furthermore, mDia, an actin nucleator that acts as a downstream effector of RhoA, regulates centrosomal positioning, and nuclear translocation in tangentially migrating GABAergic interneurons. However, mDia deficiency does not impair the cytoplasmic dilation/swelling formation (Shinohara et al., 2012).

In contrast to RhoA, another Rho family small GTPase, Rac1 and its binding protein, POSH, are required for the formation of cytoplasmic dilation/swelling in cortical excitatory neurons (Yang et al., 2012). Suppression of Rac1 by the expression of the dominant negative mutant, shRNA-mediated knockdown, or gene targeting, disturbs neuronal migration (Kawauchi et al., 2003; Chen et al., 2007; Govek et al., 2011; Yang et al., 2012). Although Rac1 promotes the activity of JNK, which is known to regulate leading process morphology and neuronal migration (Kawauchi et al., 2003), JNK1-suppressing neurons are able to form the cytoplasmic dilation/swelling. The area of the cytoplasmic dilation/swelling is not significantly different between control and JNK1-knockdown neurons, although the morphologies of the cytoplasmic dilation/swelling in the JNK1 knockdown neurons are rough and irregular in part (Nishimura et al., 2014). Therefore, Rac1 and POSH are believed to control the formation of cytoplasmic dilation/swelling mainly in a JNK1 independent manner.

Considering that abundant microtubules are observed in the cytoplasmic dilation/swelling, microtubule-regulatory proteins may be involved in the formation of this subcellular domain. In fact, it has been reported that Dcx (previously known as Doublecortin) and its upstream kinase, Cdk5, are required for the formation of the cytoplasmic dilation/swelling (Nishimura et al., 2014). Interestingly, both Dcx and Cdk5 are known as causative genes for lissencephaly (Gleeson et al., 1998; des Portes et al., 1998; Magen et al., 2015).

Dcx controls not only microtubule polymerization but also endocytic trafficking (Francis et al., 1999; Gleeson et al., 1999; Yap et al., 2012; Yap and Winckler, 2015), and clathrincoated pits are observed in cytoplasmic dilation/swelling (Shieh et al., 2011). Pharmacological inhibition of microtubule polymerization (nocodazole treatment) or endocytosis (dynasore treatment) by using the ex vivo chemical inhibitor technique, disturbs the formation of the cytoplasmic dilation/swelling (Nishimura et al., 2014). Consistently, knockdown of Rab5, a regulator for endocytosis and trafficking to early endosomes, shows defects similar to the dynasore treatment.

Furthermore, p27kip1, another Cdk5 substrate, is required for the cytoplasmic dilation/swelling formation (Nishimura et al., 2014). Although p27kip1 controls the G1 length in the cell cycle and promotes cell cycle exit, p27kip1 also plays a role in actin reorganization through suppression of RhoA activity and activation of an actin-binding protein, Cofilin, in the multipolar processes of neurons at the early phase of migration (Kawauchi et al., 2006, 2013). In addition, p27kip1 is required for the tangential migration of cortical GABAergic interneurons via microtubule organization (Godin et al., 2012). However, it is still unclear which downstream event(s) (the regulation of actin or microtubule or something else) is important for the cytoplasmic dilation/swelling formation in the locomoting neurons.

#### Nuclear Elongation and Forward Movement

After the formation of the cytoplasmic dilation/swelling, the nucleus elongates, and moves into the newly formed dilation. This nuclear elongation is closely coupled with the cytoplasmic dilation/swelling formation. In fact, suppression of Cdk5, Dcx, p27kip1, Rab5, microtubule polymerization, or endocytosis perturbs the nuclear elongation as well as dilation/swelling formation during the locomotion mode (Nishimura et al., 2014). Interestingly, however, knockdown of JNK, which does not affect the area of the cytoplasmic dilation/swelling, suppresses the nuclear elongation, suggesting that Cdk5 and JNK, both of which promote microtubule dynamics (Kawauchi et al., 2003, 2005), have different roles in the locomotion mode of migration (Nishimura et al., 2014).

The elongated nuclei are surrounded by perinuclear cagelike microtubules, which contain abundant tyrosinated tubulins, components of dynamic microtubules (Rivas and Hatten, 1995; Schaar and McConnell, 2005; Umeshima et al., 2007). The regulation of microtubule dynamics is known to require Cdk5 and JNK activities (Kawauchi et al., 2005). Cdk5 phosphorylates focal adhesion kinase (FAK) at Ser732, and

nucleus shows an elongated morphology and moves into the dilation/swelling (left). Locomoting neurons also show a scaffold cell-dependent migration. They attach to and migrate along the neural progenitor-derived radial glial fibers (right). (B) Cellular events regulating two unique features of the locomotion mode of neuronal migration. N-cadherin is involved in the adhesion to radial glial fibers. Endocytic recycling of N-cadherin is required for the forward movement of neurons. JNK controls the leading process formation possibly through the regulation of microtubule dynamics. Par6α is localized at centrosomes and regulates perinuclear cage-like microtubules. Both POSH-Rac1 and Cdk5-Dcx pathways are required for the formation of cytoplasmic dilation/swelling. Another Cdk5 substrate, p27kip1 also plays an important role in cytoplasmic dilation/swelling formation. Furthermore, Cdk5, Dcx, p27kip1 as well as JNK are involved in the nuclear elongation in the locomoting neurons. Lis, Dynein, SUN1/2, Nesprin-1/2, and Myosin II regulate the nuclear forward movement. See the main text for more details.

Ser732-phosphorylated FAK is localized on the perinuclear cagelike microtubules (Xie et al., 2003). Cdk5 deficiency or expression of the Ser732-nonphosphorylatable mutant of FAK (S732A) disturbs the nuclear elongation in migrating neurons. It is also known that overexpression of Par6α, which is localized at the centrosome, disrupts the perinuclear cage (Solecki et al., 2004).

The forward movement of the nuclei (nucleokinesis) requires Lis1- and dynein-mediated motor activity (see the following excellent reviews: Tsai and Gleeson, 2005; Marín et al., 2010). SUN1/2 and Nesprin-1/2, which are localized at the inner and outer membranes of the nuclear envelope, respectively, connect the nucleus to the dynein complex on microtubules in the locomoting neurons (Zhang et al., 2009). In addition, actomyosin-mediated contractility at the posterior end of the cell is known to play an important role in the nuclear forward movement (Schaar and McConnell, 2005; Martini and Valdeolmillos, 2010). Myosin II is also observed at the proximal region of the leading process and controls the coordinated movement of the centrosome and soma in cerebellar granule neurons (Solecki et al., 2009).

#### A Radial Glial Fiber-dependent Migration

Another feature of the locomotion mode of neuronal migration is migration on other cells, called a scaffold cell-dependent migration (Kawauchi, 2012) (**Figure 2A**). It has been suggested that Astrotactin (Astn1) is involved in the interaction between migrating neurons and Bergmann glial fibers (Adams et al., 2002). Treatment with antibodies against Astn1, but not Ncadherin and L1-CAM, inhibits the attachment of cultured cerebellar granule neurons to astroglia and glia-guided neuronal migration (Stitt and Hatten, 1990; Fishell and Hatten, 1991).

The discovery that the locomoting neurons migrate along radial glial fibers in the developing cerebral cortex was reported in 1972 (Rakic, 1972). Unlike the cerebellar granule neurons, suppression of a cell-cell adhesion molecule, N-cadherin, in the developing cerebral cortex perturbs the attachment of migrating neurons to the radial glial fibers and neuronal migration (Kawauchi et al., 2010; Shikanai et al., 2011). Importantly, a portion of N-cadherin is internalized by Rab5-dependent endocytic pathways, and subsequently transported to the plasma membrane via Rab11-dependent recycling pathways. This active transport of N-cadherin is essential for the radial glial fiberdependent migration of locomoting neurons in the developing cerebral cortex (Kawauchi et al., 2010).

N-cadherin is involved in other modes of neuronal migration and adhesions between radial glial neural progenitors (Solecki, 2012). Under the control of Reelin and Rap1, N-cadherin regulates the transition from multipolar to bipolar neurons (Jossin and Cooper, 2011; Gärtner et al., 2012). N-cadherin is also required for the somal translocation mode of neuronal migration, which is applied to the early-born neurons (Franco et al., 2011). Cajal-Retzius cells in the marginal zone and somal translocating neurons express immunoglobulin-like adhesion molecules, Nectin-1 and Nectin-3, respectively, and Nectin-3 upregulates N-cadherin to promote the somal translocation mode in early corticogenesis (Gil-Sanz et al., 2013). Interestingly, N-cadherin has also been implicated for a role in the tangential migration of cortical interneurons (Luccardini et al., 2013, 2015).

In addition to N-cadherin, other cell adhesion molecules, such as Connexin 43 (Cx43), Cx26, and JAM-C, have been shown to control neuronal migration. Cx43 and Cx26, gap junction proteins, stabilize a leading process on the radial glial fibers via enhancement of cell-cell adhesion, rather than formation of an aqueous channel (Elias et al., 2007). FAK promotes the assembly of Cx26 at contact sites between the locomoting neurons and radial glial fibers (Valiente et al., 2011). Interestingly, Cx43 is also involved in the formation of multipolar morphologies at the early phase of neuronal migration (Liu et al., 2012). Cx43 upregulates p27kip1, which controls the multipolar morphologies through actin reorganization (Kawauchi et al., 2006). JAM-C and its binding adaptor protein, Pard3, are localized at the tight junctions in epithelial cells. However, in migrating cerebellar granule neurons, Pard3A promotes the recruitment of JAM-C to neuron-neuron or neuron-glial cell contacts (Famulski et al., 2010).

Integrin heterodimers are mainly involved in cell-to-extra cellular matrix adhesion (Kawauchi, 2012). It has been reported that treatment with antibodies against β1-integrin suppresses radial glial fiber-dependent neuronal migration in vitro (Anton et al., 1999). However, NEX promoter-mediated conditional knockout of β1-integrin in neurons revealed no migration defects in the cortical six-layered structures, while Nestin promotermediated disruption of β1-integrin in both neurons and radial glial progenitors resulted in disorganization of radial glial fibers and cortical laminae, similar to type II-lissencephaly (Graus-Porta et al., 2001; Belvindrah et al., 2007).

# The Final Phase of Neuronal Migration After the Locomotion: A Terminal Translocation Mode

At the final phase of neuronal migration when the leading process reaches the marginal zone, neurons undergo a shortdistance migration in a radial glial fiber-independent manner (**Figure 1A**). N-cadherin expression is decreased at the cell soma of neurons undergoing terminal translocation (Kawauchi et al., 2010). Suppression of Rab7, a regulator for lysosomal degradation pathways, leads to a defect in the terminal translocation. Taken together with the involvement of Rab7 in the degradation of N-cadherin in vitro, it suggests that Rab7 dependent lysosomal degradation of N-cadherin at the cell soma is required for the terminal translocation (Kawauchi et al., 2010). N-cadherin is still expressed in the distal region of the leading processes (immature dendrites) in terminal translocating neurons (Kawauchi et al., 2010), and therefore might play a role in the terminal translocation as it has been reported to control somal translocation during early corticogenesis (Franco et al., 2011).

Additionally, involvement of other cell adhesion molecules, such as α5β1-integrin, a receptor for fibronectin, and L1- CAM, and Protein kinase C delta (PKCδ) has been reported (Nishimura et al., 2010; Sekine et al., 2012; Tonosaki et al., 2014) (**Figure 1A**). Suppression of either α5-integrin or β1 integrin perturbs terminal translocation (Sekine et al., 2012). Reelin-mediated activation of Rap1 promotes the recruitment of Talin to the plasma membrane, which activates the Integrin heterodimers possibly through direct binding to the cytoplasmic region of β1-integrin (Sekine et al., 2012). As described above, Reelin also enhances the activation of Rap1 during the early phase of neuronal migration (Jossin and Cooper, 2011). Recent studies have revealed that two guanine-nucleotide exchange factors (GEFs), C3G, and RapGEF2, differentially activate Rap1 at the final or early phases of migration, respectively (Ye et al., 2014). However, defects in the multipolar-to-bipolar transition have been reported in C3G-knockout brains (Voss et al., 2008), suggesting that C3G may also be required for the early phase of neuronal migration.

# Conclusion

From the 1990's, several key molecules involved in neuronal migration, such as Lis1, Dcx, FLNA, and Reelin, have been identified mainly by the use of molecular genetics. Furthermore, recent in vivo and ex vivo cell biological techniques, including in vivo electroporation, slice culture methods, time-lapse imaging and electron microscopy analyses, have uncovered essential roles for dynamic regulation of cytoskeleton and cell adhesion in neuronal migration. In the locomoting neurons, the formation of dilation/swelling requires proper regulation of microtubules, actin cytoskeleton, and endocytic pathways (**Figure 2**). Another feature of locomotion, a radial glial fiber-dependent migration, depends on the membrane trafficking-mediated remodeling of

# References


the cell adhesion complex (**Figure 2**). Thus, molecular pieces, identified from molecular genetics and in vivo electroporation, begin to take shape. However, the spatio-temporal regulation of these cellular events remains unclear. Furthermore, the dynamic behavior of each endosome in migrating neurons in cortical slices remains to be observed. Continual technological advances in in vivo cell biology and related research fields will shed light on unsolved questions to help us better understand the whole picture of cerebral cortical development.

# Acknowledgments

The author thanks Dr. Ruth T. Yu for critical reading of the manuscript. Research in the author's group is supported by research grants from the JST PRESTO and the JSPS KAKENHI (26290015 and 26110718).


cytoplasmic dilation formation and nuclear elongation in migrating neurons. Development 141, 3540–3550. doi: 10.1242/dev.111294


**Conflict of Interest Statement:** The author declares that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Kawauchi. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Adapting for endocytosis: roles for endocytic sorting adaptors in directing neural development

Chan Choo Yap \* and Bettina Winckler\*

Department of Neuroscience, University of Virginia, Charlottesville, VA, USA

Proper cortical development depends on the orchestrated actions of a multitude of guidance receptors and adhesion molecules and their downstream signaling. The levels of these receptors on the surface and their precise locations can greatly affect guidance outcomes. Trafficking of receptors to a particular surface locale and removal by endocytosis thus feed crucially into the final guidance outcomes. In addition, endocytosis of receptors can affect downstream signaling (both quantitatively and qualitatively) and regulated endocytosis of guidance receptors is thus an important component of ensuring proper neural development. We will discuss the cell biology of regulated endocytosis and the impact on neural development. We focus our discussion on endocytic accessory proteins (EAPs) (such as numb and disabled) and how they regulate endocytosis and subsequent post-endocytic trafficking of their cognate receptors (such as Notch, TrkB, β-APP, VLDLR, and ApoER2).

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan

#### Reviewed by:

John R. Henley, Mayo Clinic, USA Brian W. Howell, SUNY Upstate Medical University, USA

#### \*Correspondence:

Chan Choo Yap and Bettina Winckler, Department of Neuroscience, University of Virginia, 409 Lane Rd. Extd., MR-4 6116, Charlottesville, VA 22936, USA cy5x@virginia.edu; bwinckler@virginia.edu

> Received: 01 January 2015 Accepted: 16 March 2015 Published: 08 April 2015

#### Citation:

Yap CC and Winckler B (2015) Adapting for endocytosis: roles for endocytic sorting adaptors in directing neural development. Front. Cell. Neurosci. 9:119. doi: 10.3389/fncel.2015.00119 Keywords: clathrin, signaling endosomes, TrkB, numb, disabled, doublecortin, migration

# Introduction

The development of the mammalian brain involves a multitude of precisely coordinated processes, including proliferation and differentiation of neural stem cells followed by migration of newborn neurons from their birthplace to their final destination, elaboration of axons and dendrites, and finally synapse formation (LoTurco and Bai, 2006; Kawauchi and Hoshino, 2008; Liu, 2011; Lewis et al., 2013; Wu et al., 2014). During cortical development, neurons born in the ventricular zone (VZ) migrate radially by moving along radial glia fibers to form proper cortical layers. Interneurons also migrate long distances, but follow a tangential migratory route that does not involve migration along radial glia processes (Guo and Anton, 2014). Defective neuronal migration causes neurodevelopmental disorders such as mental retardation, Lissencephaly, epilepsy, and others (Jamuar and Walsh, 2014). Other cell types in the brain also migrate to reach their final distribution, including oligodendrocyte precursor cells (Choe et al., 2014) and microglia (Arnò et al., 2014), but less is known about the cues and receptors that guide their movements.

Directional migration in any cell type requires cell polarization and regulated cycles of adhesion and de-adhesion (Vicente-Manzanares and Horwitz, 2011). Multiple extracellular ligands and their receptors mediate the elaboration of polarized cell morphology and of directional protrusions by dynamically regulating linkages to cytoskeletal elements. Cadherins are often implicated in cell-cell contact mediated guidance (such as might occur for radial migration along radial glial fibers) whereas integrins are well known to mediate migratory behavior for cell-extracellular matrix interactions (such as migrating along basement membranes). In addition, polarized membrane addition and removal contributes to directional motility (Vitriol and Zheng, 2012). Because migratory behavior is mediated by membrane receptors, their intracellular trafficking is essential for regulating the responsiveness of migratory cells to extracellular guidance cues. This is also true for neuronal migration. Precise sorting of membrane guidance receptors to specific regions in migratory neurons, regulating surface levels via endocytic trafficking of the receptors, as well as decisions to recycle or degrade receptors post-endocytically are crucial for brain development, but most of the detailed mechanisms are still poorly understood. Since ligand-receptor systems continue to play important roles past the migratory stage for axon and dendrite growth and for synaptogenesis, understanding membrane traffic in brain development will shed light on many normal brain processes as well as on dysfunctions of the nervous system (Yap and Winckler, 2012).

Not surprisingly, multiple genes important for proper brain development have molecular roles in endocytic membrane traffic. One prominent example is numb, an evolutionary conserved protein originally identified as a cell fate determinant during peripheral and CNS development in Drosophila (Uemura et al., 1989). The molecular role of Numb is as an endocytic coadaptor that interacts with cargos and with endocytic machinery, such as the clathrin adaptor AP-2 and the endocytic accessory proteins Eps15. It thus associates with the clathrin complex for clathrin-dependent endocytosis of membrane proteins (Salcini et al., 1997; Santolini et al., 2000; Smith et al., 2004; McGill et al., 2009). How exactly is neural development regulated by an endocytic adaptor, such as numb? The mechanisms involve regulating the endocytosis, trafficking, and signaling of developmentally relevant receptors. Which other endocytic adaptors play roles in receptor endocytosis during neural development? Which receptors require which co-adaptor (numb or another co-adaptor) for endocytosis and how is their endocytosis important for normal development? Are endocytic co-adaptors also involved in post-endocytic trafficking of the receptor? These are some of the question we will explore in this review. In particular, we will discuss co-adaptor proteins that are involved in clathrin-mediated pathways via AP-2, such as Numb, Dab family proteins, and DCX, and explore their respective roles in brain development.

# Clathrin-Mediated Endocytosis: The Basic Process

Many excellent reviews have been published about clathrinmediated endocytosis, and we only want to briefly summarize the most important facts here. Most endocytosis of receptors in mammalian cells is mediated by assembling a multimeric complex between membrane receptors (i.e., cargos), adaptor protein complexes (AP-2, in particular) and a coat lattice composed of many trimers (''triskelions'') of clathrin (McMahon, 1999; McMahon and Boucrot, 2011; Merrifield and Kaksonen, 2014). Due to better and better imaging, the sequence of events in assembling this multimeric complex at the plasma membrane (called a ''clathrin-coated pit'', or CCP) is becoming better understood on a mechanistic level: A nucleation complex (including FCHO proteins, Eps15, and intersectins) associates with the plasma membrane via PIP(4, 5)P<sup>2</sup> binding and initiates CCP formation. The tetrameric AP-2 adaptor is recruited to the nucleation sites and serves as a major hub for recruiting other accessory proteins as well as cargos to the forming CCP. As clathrin coat assembly progresses, the membrane is increasingly deformed into a larger clathrincoated invagination that accumulates cargos. Ultimately, the invaginating membrane cup fissions to bud a clathrin-coated vesicle into the cell interior. The clathrin coat is rapidly removed by uncoating enzymes, and the endocytic vesicle delivers its cargos into the endosomal system by fusion with early endosomes. In addition to the essential core components of cargo, AP-2 adaptor, and clathrin, many other endocytic accessory proteins (EAPs) or ''co-adaptors'' associate with CCPs and aid in cargo selection, in the efficiency of cargo enrichment at the CCP, and in execution of subsequent membrane deformation, fission, uncoating, and endosomal fusion events. In addition to clathrin-mediated endocytosis, there are nonclathrin endocytosis pathways that are less well understood. It was found recently that in a mammalian cell line 95% of cargos enter the cells via clathrin-mediated endocytosis (Bitsikas et al., 2014). Surprisingly, this includes GPI-linked proteins, which had been thought to enter primarily via non-clathrin pathways. If this overwhelming preponderance of clathrin- vs. non-clathrin pathways also holds in the nervous system remains to be seen.

### Tetrameric Adaptor Complexes

To date, five tetrameric adaptor complexes (AP-1 through AP-5) have been identified in mammals. Each adaptor complex consists of two large (α, γ, δ, ε, ζ and β1–5), one medium (µ1–5), and one small subunit (σ1–5) (Fölsch, 2008; Hirst et al., 2011). The large subunits mediate binding to the target membrane (via specific phosphoinsoitide lipids), to clathrin, and to other accessory proteins (Brodsky et al., 2001). The medium subunit (µ) directly interacts with the ''tyrosine-based'' sorting signals (YXXΦ) in the cytoplasmic tail of cargo molecules [''X'' means any amino acid can be at that position whereas ''Φ'' means an amino acid with a bulky hydrophobic side chain can be at that position.]. The small σ subunits are involved in the recognition of ''dileucine-based'' sorting motifs ([DE]XXXL[LI]) [square brackets mean that either of the bracketed amino acids can be at that position], such as the one found in the cation-independent mannose 6-phosphate receptor (CI-M6PR). AP-1 and AP-2 are both components of clathrin-coated vesicles, either for sorting at the trans-Golgi network (TGN)/endosomes (AP-1) or for endocytosis from the plasma membrane (AP-2). Both AP-3 and AP-4 are associated with TGN/endosomal membranes, and either mediate endosome/lysosome delivery (AP3 and AP4) or the biogenesis of specialized secretory vesicles in neurons (AP3) from the TGN (Faúndez et al., 1998; Mullins and Bonifacino, 2001; Robinson and Bonifacino, 2001). AP-3 also binds clathrin, whereas AP-4 is not significantly enriched in clathrin-coated vesicles and has no clathrin box consensus motif in its β4 subunit, suggesting it acts independently of clathrin (Simpson et al., 1996; Dell'Angelica et al., 1997, 1998; Hirst and Robinson, 1998; Hirst et al., 1999; Barois and Bakke, 2005). AP-5 was only recently identified and shown to associate with late endosomal/lysosomal compartments and might regulate the sorting from early endosome to lysosomes (Hirst et al., 2011). Importantly, mutations in adaptor protein complexes have been implicated in several neurodevelopmental disorders, such as Mednik syndrome (which manifests with abnormal Copper metabolism due to a copper transporter defect) and hereditary spastic paraplegia (HSP; Hirst et al., 2013).

#### Clathrin-Mediated Endocytosis Via AP-2: Constitutive or Regulated?

Much of the regulation of endocytosis likely involves regulating when and where to form a coated pit and which cargo proteins to recruit into it. Developmental processes that rely on endocytosis of specific receptors are thus subject to regulation of cargo/adaptor/coat interactions. Some of these regulatory mechanisms have been worked out, and many others will certainly be discovered in the future (Traub and Bonifacino, 2013). For example, AP-2 recruitment to the plasma membrane requires PI(4, 5)P<sup>2</sup> clustering, and can be inhibited by phosphorylation of µ2 by the kinase AAK (Conner and Schmid, 2002; Ricotta et al., 2002). Upon binding to PI(4, 5)P2, AP-2 undergoes a large conformational change that exposes the binding sites on the µ2 subunit for tyrosinebased endocytic signals on cargo receptors. Phosphorylation of tyrosine-based signals on the cargos similarly inhibits their binding to AP-2. Endocytosis of receptors is thus regulated from two ends: regulating receptor affinity for endocytic adaptors, and regulating affinity of the adaptor for the sorting signal on the receptor. Endocytosis of receptors is thus far from constitutive, but rather highly regulated.

## AP-2 Function in Neurons

AP-2 has been implicated in clathrin-mediated endocytosis of membrane proteins important for synaptic plasticity and neurotransmission (Kononenko et al., 2014). In addition, clathrin-mediated endocytosis is needed during development for axon and dendrite outgrowth and for pathfinding. Clathrin/AP-2 mediated endocytosis has been associated to play an important role in adhesion disassembly during cell migration. For example, clathrin coated pits are enriched at adhesive contacts with matrix substrates and co-localized with adhesion receptors in migratory neurons. Inhibition of dynamin or clathrin function impaired neuronal migration both in vivo and in vitro, and coincided with a shift in the distribution of adhesion proteins from the original region proximal to the cell body to the rear of the cells (Shieh et al., 2011). In addition, AP-2 also interacts with the tyrosine-based motif of L1-CAM to mediate its endocytosis at axonal growth cones, a process crucial for its local recycling and detachment from adhesive environment to promote axon elongation and outgrowth (Kamiguchi et al., 1998). Additionally, endocytosis of L1-CAM in dendrites is also essential for targeting L1-CAM to axons via an indirect transcytosis pathway (Wisco et al., 2003; Yap et al., 2008).

# Monomeric Endocytic Co-Adaptors

Different endocytic cargos have different endocytic signals and can be endocytosed via different molecular assemblies. Since AP-2 is the best studied of the clathrin adaptors, we also understand those endocytic signals best that directly bind to AP-2 itself. Among those signals are short linear signals, in particular the tyrosine-based motifs that bind µ2 as well as the di-leucine motifs that bind to the interface of σ2 and α-subunits of AP-2 (Bonifacino and Traub, 2003; Traub and Bonifacino, 2013). Other endocytic signals do not bind directly to AP-2, but require accessory proteins or ''co-adaptors''. For instance, the ''NPXY'' motif ([YF]XNPX[YF]) binds to the PTB domain [''PTB'' stands for ''phospho-tyrosine binding'' but PTB domains can bind other motifs as well] of a number of EAPs, including ARH, numb, and Dab2. These additional accessory proteins are often referred to as ''CLASPs'' (clathrin-associated sorting proteins), ''EAPs'', or ''co-adaptors''. [Note to reader: There is also a class of microtubule-associated proteins named CLASP (cytoplasmic linker associated protein) which are not clathrin-associated proteins. To avoid confusion, we will use the terms ''EAP'' or ''co-adaptor'' in this review]. EAPs can be subdivided into several categories depending on their binding interactions. A full adaptor has four direct interactions: with lipids, with cargo, with clathrin, and with additional accessory endocytic proteins (Reider and Wendland, 2011). The AP-2 complex itself falls into this category. Other EAPs/co-adaptors contain only a subset of these binding sites and require additional binding partners to supply the rest of the interactions. For example, Eps15 has been shown to bind ubiquitinated cargos (via its UIM domains) and accessory proteins (via its EH domain), but to date direct binding to lipids or clathrin is not known (Reider and Wendland, 2011).

# A Large Arsenal of EAPs Enables Regulated Cargo- and Cell Type-Specific Endocytosis

What might be some of the physiological consequences of having multiple ''full'' adaptors and/or various ''partial'' co-adaptors expressed in the same cell? In fact, up to 60 different proteins can participate in the various steps of clathrin-mediated endocytosis (Merrifield and Kaksonen, 2014). Deletion of many of the EAPs does not completely disrupt clathrin-mediated endocytosis. In fact, clathrin-mediated endocytosis is notoriously robust, and continues or compensates rapidly for the loss of a single component. This likely represents the cooperative nature of building a large multimeric assembly from many components that are partially redundant (McMahon, 1999). Surprisingly, under some circumstances even AP-2 itself is dispensable. A large number of EAPs thus endows cells with endocytic robustness. This robustness creates experimental challenges in terms of testing the involvement of specific EAPs in the endocytosis of a cargo of interest (see McMahon and Boucrot, 2011) for a good discussion). In fact, global interference approaches might not be optimally suited to uncover the detailed molecular roles of individual EAPs. New technological advances have given rise to approaches that allow the analysis of the formation of single coated pits and their maturation in relation to the recruitment levels of individual EAPs. These approaches have shown that the kinetics of CCP formation and the concerted maturation to a productive coated pit are regulated by a multi-step cascade. In addition, compensatory mechanisms can be detected, allowing the analysis of phenotypes even when bulk transferrin uptake is not significantly impaired. This has led to a new framework of a concerted molecular checkpoint cascade that regulates CCP maturation. These new approaches with better signal-to-noise detection demonstrate that EAPs binding to the ear domain of α-adaptin are required for the concerted and regulated maturation of CCPs (Aguet et al., 2013).

Another consequence of having cargo-specific EAPs is that it allows cells to endocytose cargos with greatly differing cell surface abundance without competing for potentially limiting cargo binding sites on AP-2 itself (McMahon, 1999; McMahon and Boucrot, 2011). Increasing the abundance of cargo-specific EAPs, thus, leads to increased endocytosis of its specific cognate receptor cargos (containing the specific binding motif) without affecting overall endocytic levels of non-cognate cargos (not containing the specific binding motif).

Another physiological consequence of the large number of EAPs it that many are expressed in a cell-type restricted fashion, resulting in a versatile arsenal for regulating cargo selectivity. For some of them, multiple isoforms exist that can further expand the repertoire of combinatorial players. For the tetrameric AP adaptors themselves, several subunits have multiple isoforms. Some of these are expressed in a cell-type restricted fashion, including in neurons. For example, both the ubiquitously expressed AP-3A (subunit composition δ–β3A–µ3A–σ3) and the neuron-specific AP-3B (subunit composition δ–β3B–µ3B–σ3) complexes are expressed in the brain. AP-3 has been implicated in sorting synaptic vesicle membrane proteins to synaptic terminals (Danglot and Galli, 2007). Loss of the neuronal-specific AP-3B in µ3B mutant mice results in spontaneous epileptic seizure and impairment of GABA release due to a reduction in vesicular GABA transporters (Nakatsu et al., 2004). In humans, Hermansky–Pudlak syndrome (HPS) is associated with mutations in the gene encoding AP3-β3A and is characterized by albinism and defects in lysosome-related organelles (Starcevic et al., 2002). The lack of neurological defects reported in human HPS is most likely due to the presence of functional AP-3B (using the β3B subunit) in neurons. It thus appears that expression of cell type-restricted EAPs provides tissues with a versatile arsenal from which to choose in order to achieve exquisitely precise control over which receptor is endocytosed when and where. Our own attention was directed towards exploring the roles that the ''EAP arsenal'' might play in neuronal development after we made some surprising observations, first with regard to the neuronal- and cargo-specific roles that the ubiquitous EHD proteins play (Yap et al., 2010), and second with regard to the previously unknown role that doublecortin (DCX) plays in the regulated endocytosis of neurofascin (Yap et al., 2012).

Many genes have been identified over the past decades for their roles in various aspects of neural development. Many of them were first identified in Drosophila and C. elegans, but mammalian homologs exist and play crucial roles. The inventory list of genes involved in any particular developmental process is ever growing and sheds light on the genetic networks that regulate development. In addition, creating this inventory is a crucial steppingstone to proceed towards a molecular and mechanistic understanding of the identified gene products and their function in a complex network of proteins. Identifying the exact molecular role of each protein in the genetic network is an arduous and slow process, but exciting new discoveries are continuously being reported. The need to regulate endocytosis for proper neural development is highlighted by the fact that ''famous'' neurodevelopmental genes, such as numb, are now known to be EAPs. We will first discuss the functions of two EAPs with known neurodevelopmental roles, numb and disabled (Dab), and subsequently explore what is currently known about some other EAPs in regulating neuronal endocytosis for a specific subset of cargos.

# Numb—A Monomeric EAP Regulating Neural Development

In vertebrates, there are two genes encoding for the related Numb and numb-like (numbl) proteins (Zhong et al., 1996, 1997; Salcini et al., 1997). Numb is a peripheral membraneassociated protein with an amino-terminal phosphotyrosinebinding domain (PTB) and a C-terminal proline rich region (PRR), as well as EH-domain-binding motifs (two DPF and one NPF motif) (**Figure 1**). Numb binds its cargos via its PTB domain, binds the α-adaptin subunit of AP-2 via the C-terminal DPF motif, and binds other accessory proteins, such as Eps15 and EHD1 and 4, via the NPF motif. Numb thus meets the criteria for being an endocytic EAP. There are four spliced isoforms of numb (Dho et al., 1999; Verdi et al., 1999) which differ in the length of the PTB (either including or lacking a 11 aa insert) and PRR domains (including or lacking). Numb-like shares high sequence homology with Numb (**Figure 1**): it contains a PTB domain, the EH-binding motifs but lacks the PRR region and has an additional specific poly-glutamine repeat which is absent in Numb (Zhong et al., 1997). This isoform diversity adds additional variety into the cellular toolbox. The PTB region and the EH-binding motifs are conserved in vertebrate and Drosophila numb, suggesting these two regions are functionally relevant and that the roles of numb may be similar across vertebrate and invertebrate species.

# What are the Roles for Numb in Neural Development?

Numb has multiple functions in the nervous system and has been implicated in a dazzling number of processes, including cell fate determination, proliferation, neurogenesis, cell migration, cell adhesion, and axon outgrowth (Gulino et al., 2010). In the mouse embryo, Numb is expressed in all layers of the cortical plate, as well as the progenitor cells of the VZ (Zhong et al., 1997). During cortical neurogenesis and cell division, numb is asymmetrically localized to the apical membrane of the dividing cells, and subsequently segregated to the apical daughter cells that remain progenitors. A numb-knockout mouse displays premature neuronal differentiation in the forebrain, implying numb functions in maintaining progenitor cell numbers (Verdi et al., 1999; Zhong et al., 2000; Shen et al., 2002). In other contexts, Numb appears to be involved in promoting neurogenesis and differentiation instead. In fact, whereas at earlier stages (E10) the daughter cell inheriting numb remains a progenitor, later on in corticogenesis (E13) the daughter inheriting numb becomes a neuron (Shen et al., 2002). Studies from different numb knockout mouse models have given rise to conflicting results pointing to the diverse functions and complex regulation of Numb. For instance, in a second numb knockout mouse model, no premature differentiation phenotype was observed in forebrain, while impaired differentiation was detected in hindbrain and in cerebellum (Zilian et al., 2001; Klein et al., 2004), further suggesting that numb might also play a role in neurogenesis.

Numb has also been reported to play a critical role in cerebellar granule cell polarization during migration (Zhou et al., 2011): Conditional ablation of numb/numbl using the transcription factor math1-cre system in cerebellar granule cell precursors (GCPs) impairs BDNF-induced GCP migration both in vitro and in vivo. How can numb play so many, sometimes contradictory roles? Since numb is an endocytic adaptor, its particular role depends on the cargos whose trafficking numb regulates in different cellular contexts. Among the known numb cargos are receptors with well established roles in regulating developmental processes: Notch1, βAPP, β1 integrin, and TrkB. numb binds to the conserved NPXY sorting motifs in the cytoplasmic tail of these cargos and mediates their endocytosis. What is the current evidence that the brain phenotypes of numb knockouts are due to impaired endocytosis of a numb cargo?

#### Numb-Mediated Endocytosis of Receptors Controls Neural Development Notch

Numb was originally identified as an antagonist of Notch. Notch signaling regulates numerous developmental decisions and patterning events from worms to human. Notch signaling promotes radial glial identity and controls cell fate specification during development of the neocortex (Gaiano et al., 2000). It is likely that endocytosis of Notch receptor per se is regulated by Numb because the PTB domain of Numb binds the ram23 and the PEST regions of the Notch cytoplasmic tail directly, and numb-mediated inhibition of Notch requires the AP-2 specific subunit α-adaptin (Guo et al., 1996; Berdnik et al., 2002). The endocytic removal of Notch via numb reduces Notch receptor levels on the surface and reduces Notch signaling. Reduced Notch signaling then affects cell fate.

Additional molecular roles of numb are becoming apparent as well. Numb was found to have additional interacting partners, such as Par3 and atypical protein kinase C (aPKC), protein complexes which function in polarized cell migration. Knockdown of mPar-3 reduces Notch signaling activity and causes premature depletion of progenitor cells from the VZ (Bultje et al., 2009). Conversely, depletion of Numb/numblike abolishes this effect and increases Notch signaling activity. Numb and Numblike are thus required for mPar-3 function in regulating Notch signaling and neocortical neurogenesis.

# TrkB

Granule cell precursors in the cerebellum migrate towards a source of the neurotrophin BDNF via activation of the BDNF receptor TrkB at the leading process of the cell (Zhou et al., 2007). Neurotrophin signaling via Trk receptors is well established to require endocytosis into endosomes. These endosomes contain activated Trk receptors and recruit a variety of signaling components, including PI3K, Ras-MAPK, and PLC-γ pathway components. Other cascades, such as PKA via elevated cAMP, have also been implicated in BDNF signaling. High levels of signaling takes place on these so-called ''signaling endosomes'' (Cosker and Segal, 2014). For some forms of signaling (such as long-range survival signaling by NGF), the signaling endosomes travel long distances back to the cell soma where they continue to signal and regulate transcription. There are also more locally restricted signaling events downstream of Trk receptor activation close to the sites of initial endocytosis that regulate axon growth and polarized mobility of migratory neuronal precursors. Neurotrophin signaling thus is one of the best-documented examples of an endocytosis-dependent signaling event.

How is endocytosis of activated TrkB and the subsequent signaling from signaling endosomes regulated in time and space? Numb is an endocytic co-adaptor for activated TrkB (**Figure 2**): Numb colocalizes with α-adaptin and TrkB in the leading processes, it interacts with the NPXY motif on TrkB after activation (via the numb PTB domain), and it promotes TrkB endocytosis and polarized localization to the leading process of migrating GCPs. Depletion of Numb either by shRNA or conditional knockout impairs BDNF-induced TrkB endocytosis and polarized localization of TrkB to leading processes of migrating GCPs. The polarized localization of numb leads to enhanced endocytosis of activated TrkB at the leading process, increased generation and accumulation of signaling endosomes locally in the leading process of GCPs, and consequently increased activation of downstream signaling cascades that promote local membrane trafficking (such as increased BDNF exocytosis) and cytoskeletal rearrangements that drive directional protrusions and mobility (Zhou et al., 2011; see **Figure 3**).

In addition to promoting local endocytosis of activated TrkB, numb also serves as an adaptor to recruit signaling components, such as aPKC, a well known regulator of polarized cellular events upstream of cytoskeletal rearrangements. Genetic deletion

of numb/numbl impairs BDNF-induced aPKC activation in GCPs, suggesting BDNF stimulation of TrkB recruits Numb, which in turn acts as an adaptor to recruit and activate aPKC via the PTB region. In BDNF knockout mice, polarization of Numb at the front of GCPs and its interaction with aPKC are significantly decreased, suggesting BDNF regulates the polarization of Numb and its interaction with aPKC in migrating GCPs. Interestingly, Numb is also itself a target of aPKC. BDNFinduced phosphorylation of numb by aPKC increases its binding to TrkB and promotes the chemotactic response to BDNF. Taken together, Numb acts as an adaptor linking BDNF, an extracellular cue, to intrinsic cellular polarity machinery, including aPKC, via TrkB endocytosis and serves as a feed-forward loop to promote BDNF-induced directed GCP migration (Zhou et al., 2011; see **Figure 3**).

## L1

Numb is also expressed in postmitotic neurons and functions in axonal outgrowth. Numb accumulates at the tip of growing axons in cultured hippocampal neurons and mediates endocytosis of L1/NgCAM for axon growth (Nishimura et al., 2003). Internalization of L1/NgCAM occurs preferentially in the central domain of migrating growth cones, followed by anterograde transport of L1/NgCAM-containing vesicles and subsequent recycling near the leading edge. Numb has been reported to co-localize with L1/NgCAM at the central region of growth cones, suggesting numb plays a role in regulating local L1/NgCAM internalization and maybe recycling for growth cone advance (Nishimura et al., 2003). Numb co-immunoprecipitates in a complex with L1 and AP-2 and is important for efficient endocytosis of L1/NgCAM. The PTB domain of numb is required for L1/NgCAM internalization, but L1/NgCAM does not itself contain an NPXY motif. It is currently not known if numb binds directly to L1/NgCAM and if so, via what binding motif. The regulation of L1/NgCAM endocytosis on numb is surprising since L1/NgCAM has its own tyrosine-based motif that binds directly to µ2 of the AP-2 tetrameric adaptor complex (Kamiguchi et al., 1998). It will be interesting to determine what aspects of L1/NgCAM function and signaling might be additionally influenced by numb.

# How is the Polarized Distribution of Numb Regulated?

In many cell types, numb localization is highly polarized to one side of the cell. This polarized distribution of numb is ideally suited to promote non-uniform endocytosis of numb cargos (see also Section TrkB). How is the polarized distribution of numb regulated? Some evidence points to important roles of the numb interacting polarity proteins par3 and aPKC. For example, numb mediates the endocytosis of integrins via binding of the PTB domain of numb to the NPXY motif in the cytoplasmic tail of integrins (Calderwood et al., 2003; Nishimura and Kaibuchi, 2007). In migrating ECV304 and HeLa cells, Numb polarizes toward the leading edge, accumulates around focal adhesions, and localizes to clathrin-coated structures at the substratumfacing surface of the leading edge. Its localization to clathrincoated structures is α-adaptin dependent. Depletion of Numb by RNAi impairs both integrin endocytosis and cell migration.

Par3-dependent phosphorylation by aPKC regulates the polarized localization of numb and its association with clathrin coated structures. Phosphorylated numb is released from the clathrin-coated structures and no longer binds integrin, suggesting aPKC negatively regulated functions of numb with respect to integrins. However, aPKC is required for polarized subcellular localization of Numb as knockdown of aPKC mislocalizes Numb to both the apical and basal surface leading to less enrichment in leading processes of the cells. Taken together, these results suggested that the polarized numb phosphorylation regulated by aPKC/Par3 complex is important for integrin endocytosis and integrin-substrate based cell migration (Nishimura and Kaibuchi, 2007). This is in contrast to the effects of aPKC phosphorylation on numb-TrkB binding which is increased by phosphorylation of numb (see Section TrkB).

Similarly in dividing sensory organ precursor (SOP) cells in Drosophila, a numb mutant with five PKC phosphorylation–deficient sites is mislocalized to both anterior and posterior sides of the cell. The results suggested that aPKCmediated phosphorylation of numb regulates the asymmetric localization of numb in dividing cells (Smith et al., 2007), and that overall Numb functions relies on its phosphorylation status which determines its subcellular localization. Similarly, hyperphosphorylated Numb after treatment with the phosphatase inhibitor calyculin-A dissociates from the AP-2 and Eps15 complex, indicating that phosphorylation at certain sites of numb negatively regulates its endocytic functions (Krieger et al., 2013). Interestingly, phosphorylation of Numb by Ca2+/calmodulin-dependent protein kinase I at positions S264 and S283 prevents Numb binding to AP-2, but promotes its interaction with 14-3-3 proteins (Tokumitsu et al., 2006), pointing to the participation of numb phospho-isoforms in distinct molecular complexes with presumably distinct functions in the cell. In addition, the recruitment of numb to endosomes is also regulated: phosphorylation of numb by AAK1 at position T102 redistributed numb from the plasma membrane to perinuclear endosomes (Sorensen and Conner, 2008).

There is thus strong evidence that numb participates in endocytosis and subsequent endosomal trafficking of NPXYmotif containing cargos during multiple processes in neural development, such as maintaining proper neurogenesis via regulated Notch trafficking, cerebellar migration via regulated TrkB trafficking, and axon growth via regulated L1 endocytosis. Since numb is localized in a polarized manner to only a part of the cell, endocytosis would be preferentially occurring on one side leading to polarized surface distribution of the numb cargo. This leads to changed signaling (for Notch or TrkB) or reduced adhesion (for integrin) or both, and changed cellular behavior. Since most of the numb cargos are signaling receptors that continue to signal after internalization, numb-mediated endocytosis of such a receptor would not stop signaling immediately, but allow signaling from endosomes. Furthermore, numb itself provides a scaffold for signaling components and thus participates directly in determining signaling output from its cargos. Subsequently, these receptors can be recycled to the surface for another round of ligand-binding and signaling (i.e., continued signaling) or be trafficked for degradation (i.e., termination of signaling).

## Multiple Splice Isoforms with Different Domain Structure Play Different Roles

In mammalian cells, numb isoforms are expressed in cell type-specific manner and localized differentially subcellularly dependent on the insert in the PTB domain. For instance, the absence of the 11 aa in the PTB domain results in more cytosolic and less membrane-associated Numb (Dho et al., 1999). The Numb isoform with long PRR domain is expressed transiently during early brain development and disappears prior to cell differentiation in P19 embryonic carcinoma cells, suggesting the isoform may function in promoting proliferation of progenitor cells. Conversely, the isoform with short PRR domain is expressed throughout neurogenesis in the developing brain and in adult brain, and during the course of retinoic acid-induced P19 cell differentiation (Dho et al., 1999; Verdi et al., 1999; Toriya et al., 2006). Expression of the short PRR Numb isoform in the outer optic anlage of the Drosophila larvae brain also promotes neuronal differentiation and reduces the levels of nuclear-localized Notch (Toriya et al., 2006), implying this short PRR isoform might mediate neuronal differentiation by regulating the endocytic sorting of Notch. Thus, different Numb isoforms might function in different developmental stages dependent on the time of expression and their respective subcellular localization.

# Other Co-Adaptor Proteins for NPXY-Motifs Involved in Signaling During Brain Development: Disabled Proteins

The NPXY motif is not only recognized by numb, but also by another set of co-adaptors, the Disabled (Dab) family of proteins. In addition, NPXY motifs bind to other adaptors, such as sorting nexins (SNX), Fe65, and X11/Mint (Uhlik et al., 2005; Stolt and Bock, 2006). Is there specificity of endocytic adaptors to a subset of NPXY-containing cargos, or do multiple endocytic adaptors play roles (either redundant or distinct) in trafficking the same cargos? Participation of a particular adaptor in cargo sorting would largely depend on cell-type expression, and on subcellular localization of the cargo.

## The Discovery of Disabled (Dab)

One of the best studied ''full'' EAPs (i.e., binding sites for cargo, lipid, AP-2, and clathrin) is disabled (Dab), in particular the mammalian isoform Dab2 (**Figure 4**). Dab was first isolated from Drosophila in screens for mutations in genes that enhanced phenotypes of Abl-/- (Abelson tyrosine kinase, a non receptor tyrosine kinase) (Gertler et al., 1989, 1993). Although a later study showed that Dab is not the bona fide enhancer of Abl

(Liebl et al., 2003), Dab is still a positive regulator of the Abl signaling pathway. Null mutations of Dab in Drosophila causes defects in motor axon patterning and epithelial morphogenesis, phenotypes resembling those of Abl mutants. Genetically, Dab acts in conjunction with Abl for proper growth and guidance of motor axons (Song et al., 2010). While it is not well established which cargos might undergo Dab-mediated endocytosis during axon guidance, a clear role for a clathrin-associated role for Drosophila Dab was shown by Kawasaki et al. in mature neurons: a novel Dab mutant that has a nonsense mutation within the conserved N-terminal PTB domain exhibits impaired synaptic function. The experiments indicated that Dab is involved in clathrin-mediated endocytosis for rapid clearance of neurotransmitter release sites, which is important for subsequent vesicle priming and refilling of the readily releasable pool (Kawasaki et al., 2011).

In mammals, the situation is more complicated. There are two members of the Dab family identified thus far, Dab1 and Dab2. Dab is most closely homologous to Dab2. Dab1 was originally isolated as a Src-binding protein (Howell et al., 1997a), whereas Dab2 was identified as a phosphoprotein regulated by colony-stimulating factor CSF1 (Xu et al., 1995). Dab1 is highly neuron-enriched and functions in cell positioning during brain development, whereas Dab2 regulates endodermal cell organization during embryogenesis (Howell et al., 1997b, 1999a; Morris et al., 2002; Yang et al., 2002) and is widely expressed. Dab1 and Dab2 are multi-domain cytoplasmic adaptor proteins and function in mediating several signaling pathways. Structurally, Dab1 contains a PTB-domain, clathrin adaptor AP-2-, SH3 domain-, and myosin-binding sites at its unique C-terminus (**Figure 4**). Similarly to Dab1, Dab2 is composed of an N-terminal PTB domain which is 63% identical to that of Dab1. The Dab2 C-terminus contains clathrin- and EH-domain-binding sites, which are absent in Dab1, in addition to AP-2-, SH3-domain-, and myosin VI-binding sites, which are shared with Dab1. The Dab N-terminal PTB domain is known to bind preferably to non-phosphorylated NPXY internalization motifs present in the cytoplasmic tails of βAPP and LDLrelated family proteins, such as LDLR/VLDLR, LRP, ApoER2, and megalin (Trommsdorff et al., 1998, 1999; Howell et al., 1999b; Oleinikov et al., 2000). Additionally, the PTB domain is required for recruitment to the plasma membrane by binding to phospholipids.

#### Dab2 as an Endocytic Adaptor

Most studies of Dab2 highlight its roles in clathrin-mediated endocytosis. For instance, loss of Dab2 results in decreased endocytosis of transferrin within visceral endoderm and fewer early endosomes in conditional Dab2 knockout mice. Other tissues, such as the kidney, are also affected. In NIH3T3 cells, Dab2 localizes very close to the plasma membrane at CCPs that are also positive for AP-2, but is absent from early endosomal and lysosomal compartments. Its localization to clathrin-coated vesicles is dependent on the clathrin- and AP2-binding sites located in the C-terminus as the Dab2 p67 isoform lacking the binding sites (**Figure 4**) do not associate with CCPs and is defective for receptor endocytosis (Morris and Cooper, 2001; Mishra et al., 2002; Maurer and Cooper, 2006). Mishra at al. demonstrated that Dab2 interacts directly with clathrin independently of AP-2 association, and engages soluble clathrin trimers via its multiple clathrin binding sites to assemble complete polyhedral clathrin cages. Importantly, clathrin recruitment by Dab2 does not affect its interaction with AP-2 (Mishra et al., 2002). Many lines of evidence thus demonstrate that Dab2 is a bona fide endocytic co-adaptor in clathrin-mediated endocytosis of NPXY-containing cargos, such as members of the LDL receptor family.

# Disabled in the Mammalian Nervous System: Dab1 Regulates Neural Development

There is great interest in Dab1 among developmental neuroscientists since it is highly expressed in neurons, and it has been linked to signaling of major developmental receptors pathways, including reelin (via the reelin receptors ApoER2 and VLDLR), and βAPP. Two mutant mice arising from spontaneous mutations in Dab1, Scrambler and yotari (Sweet et al., 1996; Yoneshima et al., 1997), display inverted cortical lamination, abnormal positioning of neurons, and aberrant orientation of cell bodies and fibers, similar to the reeler mouse which lacks the ligand reelin itself (Falconer, 1951; Goffinet, 1979; Sheldon et al., 1997; Ware et al., 1997). A targeted disruption of the PTB domain of Dab1 resulted in a Dab1 null with no Dab1 mRNA detectable. This Dab1 null mouse also exhibits identical phenotypes to the reeler mouse (Howell et al., 1997b). The Dab1 null mouse is ataxic and dies prematurely between P20 to P30. No clear lamination into layers could be distinguished in either the cerebral cortex or hippocampus. Similarly, the development of the cerebellum is severely affected, unfoliated and small in size. The mutant Purkinje cells are present in the central mass with their dendrites oriented randomly. This suggested that interaction of Dab1 with reelin receptors is essential for reelin-dab1-mediated neuronal migration. Since the phenotypes of the Dab1 null mouse resembles those of the reeler mouse, Dab1 and reelin are part of a single genetic signaling pathway (Rice and Curran, 2001).

# The Mechanism of Dab1 Regulation of Reelin Signaling

The PTB domain of Dab1 directly interacts with the NPXY motifs located in the cytoplasmic tails of the reelin receptors VLDLR and ApoER2 (Trommsdorff et al., 1999). Reelin-signaling is initiated by direct binding of reelin to the extracellular domains of VLDLR and ApoER2 (D'Arcangelo et al., 1999; Hiesberger et al., 1999; Rice and Curran, 2001) and subsequently leads to increased levels of phosphorylated Dab1. Reelin-induced phosphorylation of Dab1 leads Src family of non-receptor tyrosine kinases and recruitment of Crk family adaptor proteins (Park and Curran, 2008) which trigger downstream intracellular signaling cascades, such as activation of AKT, PI3K and Erk1/2. Tyrosine phosphorylation of Dab1 at baseline level prior to reelin stimulation is essential for the subsequent reelin-induced activation of Src (Arnaud et al., 2003; Bock and Herz, 2003). VLDLR and ApoER2 are two essential components of the Reelinsignaling pathway. The VLDLR–knockout mouse has smaller cerebellum, whereas the ApoER2-deficient mouse displays two layers of CA1 region. Importantly, the VLDLR and ApoER2 double knockout mouse exhibits identical phenotypes to that observed in reeler/scrambler mice (Trommsdorff et al., 1999). Howell et al. (2000) identified several tyrosine phosphorylation sites clustered close to the PTB domain of Dab1 that are important for cell positioning during brain development. They demonstrated that a mutant mouse with all the identified potential phosphorylation sites mutated display phenotypes identical to those observed in reeler/Dab1-null mice (Howell et al., 2000).

Surprisingly, despite the fact that Dab1 binds the NPXY motif and contains an AP-2 binding site, evidence for a direct role in endocytosis is lacking, and no detailed study demonstrating its association with clathrin-AP2 complex in the regulation of reelin/ApoER2/VLDLR endocytosis exists. Similarly, the endocytic trafficking routes of reelin/receptorbound Dab1 signaling complex remain poorly understood. Whether phosphorylation affects the function of Dab1 in the endocytic trafficking of reelin-ApoER2 complex, such as dissociation of phosphorylated Dab1 from the complex, remains to be answered. So far there are only a handful of papers showing Dab1-signaling complex in endosomal-like structures. Brian Howell was the first to show Dab1 diffusely in the cell soma and enriched in the axon where it co-localizes with βAPP in small vesicular-like structures (Howell et al., 1999b). Subsequently, Curran's group showed Dab1 co-localization with APLP1 in membrane ruffles and vesicular structures (Homayouni et al., 2001). Although Hiesberger and D'Arcangelo both reported that binding of reelin to VLDLR/ApoER2 on the cell surface mediates its internalization into vesicles (D'Arcangelo et al., 1999; Hiesberger et al., 1999), it is not clear whether vesicles that are positive for internalized reelin also contain Dab1. Similarly, Leeb et al. did not show the presence of Dab1 in EEA1 positive early endosomes colocalizing with clusterin ligandbound ApoER2/VLDLR complexes (Leeb et al., 2014).

In 2002, it was reported that deletion of the C-terminal region of Dab1 (corresponding to the p45 isoform) (**Figure 4**) has no affect on cortical neuronal migration in vivo (Herrick and Cooper, 2002), and the isoform still appears to be phosphorylated, indicating that the AP-2 and SH-domain binding sites are not crucial for activation of reelin-Dab1 signaling during migration. This result argued strongly that Dab1 works in endocytosis-independent pathways during corticogenesis downstream of reelin signaling. However, a mutant mouse expressing only a single copy of the p45-Dab1 isoform (Dab1p45/−, i.e., p45 hemizygote) displays disrupted neocortex and hippocampus development, where the marginal zone of the neocortex shows the presence of late-born neurons, while the CA1 region of hippocampus is separated into two layers. The same paper showed that a mouse carrying one copy of full length p80-Dab1 (Dab1p80/−, i.e., p80 hemizygote) shows no migration defect. This observation suggested that the C-terminus of Dab1 might have important functions required for signaling in specific neurons during later stages of brain development.

Is there evidence for or against a role of Dab1 in clathrinmediated endocytosis? A report by Morimura et al. revealed that tyrosine phosphorylated Dab1 is recruited to the plasma membrane and co-localizes with reelin/receptor complexes in puncta on the cell periphery of cortical neurons of reeler mice after addition of reelin (Morimura et al., 2005). They observed that two minutes after reelin wash-out, phosphorylated Dab1 still co-localizes with reelin in puncta. However, the phosphorylated Dab1 appears to dissociate from the receptor complex 20 min after removal of unbound reelin, the time point when endocytosis is presumably completed, suggesting phosphorylated Dab1 does not traffic extensively with the endocytosed reelin/receptor complex after internalization. Importantly, inhibition of Dab1 phosphorylation by the src kinase inhibitor PP2 prevents internalization of reelin, leaving reelin co-localized with Dab1 near the plasma membrane for a prolonged period. The study concluded that Dab1 regulates cell surface expression of reelin receptors by promoting translocation of the receptors to the plasma membrane, and that phosphorylation of Dab1 initiates intracellular trafficking of reelin/receptor complex in neurons. It is not known whether the puncta positive for reelin and Dab1 at the two-minute time point are on the cell surface or whether they are endocytosed vesicles as the immunostaining for reelin and Dab1 was done after permeabilization. Since endocytosis can be extremely fast, the two-minute time point might include Dab1 on endosomes. This remains to be established more firmly.

In addition, Dab1 has been implicated in regulating the processing and trafficking of βAPP and ApoER2, but this might be during post-endocytic trafficking. A study by Rebeck's group demonstrated that interaction with Dab1 increases the cell surface levels of ApoER2 and βAPP, increases cleavage and secretion of the extracellular domain of the proteins and decreases levels of βAPP C-terminal fragments (Hoe et al., 2006). The effects are NPXY motif- and PTB domaindependent. Treatment with reelin increases the interaction between Dab1 and βAPP or ApoER2 and significantly lowers the levels of secreted Aβ (Hoe et al., 2006). Notably, Dab1 KO mice have higher levels of Aβ compared to littermate controls, implicating the Dab1-βAPP interaction in regulating Aβ production. Overall the data indicated that Dab1 is involved in regulating the intracellular trafficking and processing of βAPP and of ApoER2 facilitated by reelin. The question of whether Dab1 acts as a scaffold/stabilizer for βAPP/ApoER2 on the cell surface to prevent the proteins from being sorted into endosomal compartments that facilitate Aβ production, or as an adaptor to ensure correct endocytic sorting remains to be investigated.

In addition to direct binding of Dab1 to endocytic machinery, there might also be Dab1 interactions with additional co-adaptors that mediate endocytosis of Dab1 cargos. For example, Fuchigami et al. demonstrated that Dab1 co-localizes with ApoER2 and CIN85 in vesicle-like structures at the plasma membrane (Fuchigami et al., 2013). CIN85 is an adaptor protein involved in endocytosis of receptors including EGFR and dopamine receptor (Dikic, 2003; Shimokawa et al., 2010) by binding to the cbl-EGFR complex via its SH3 domains and to endophilins via its PRR domain. In addition, CIN85 interacts with the PRR region of Dab1. Treatment with a reelin fragment containing the ApoER2/VLDLR binding sites (called the ''reelin-repeats'') in the presence of Dab1 induced CIN85 localization to early endosomes which contained EEA1, the reelin-repeat fragment and ApoER2 in neurons. Since CIN85 binds phosphorylated Dab1, it is highly possible that interaction of Dab1 with Cin85 might facilitate Dab1-mediated internalization of the receptor complex and possibly trafficking to early endosomal compartments.

# Endocytosis Regulation in Neurons—What Other Players are Known?

Numb plays crucial role in the nervous system but is expressed widely outside the nervous system and plays important roles in many places. Are there also neuronal-specific EAPs that might play neuronal-specific roles? And how are cargos with different endocytosis signals endocytosed? What other EAPs and co-adaptors have functions in nervous system development? There are likely many still undiscovered and their roles are largely unknown. We will discuss below the identification of one putative neuronal endocytic co-adaptor, namely doublecortin.

# Doublecortin (DCX): A Critical Regulator of Cell Migration and Axon Tract Formation

Doublecortin (DCX) was first identified as the major gene causing X-linked subcortical laminar heterotopia in female and lissencephaly syndrome in male patients (des Portes et al., 1998a,b). These defects manifest as cortical and hippocampal layering defects, leading to phenotypes including epilepsy and mental retardation. The layering defects are due to problems with proper neuronal migration. DCX has two closely related homologs, DCLK1 and DCLK2, both of which contain a kinase-domain at their C-termini, absent in DCX (**Figure 5**). DCX is prominently expressed in young postmitotic neurons, with transient expression detected during adult neurogenesis. Unlike DCX, DCLKs are expressed across the nervous system throughout adulthood. Although DCX is the causative gene for X-linked Lissencephaly in humans, DCX knockout mice display only mild neurodevelopmental delays in the cortex, and cortical lamination in the adult DCX KO mouse is indistinguishable from wild type. In contrast, severe lamination defects are observed in the CA3 region of the hippocampus (Corbo et al., 2002). However, DCX/DCLK1 double KO mice exhibit severely disrupted lamination in the cortical plate, suggesting

compensation from DCLK1 might have contributed to the mild cortical phenotypes observed in the DCX KO mouse. Since DCX and DCLK1 are highly related in protein sequence (**Figure 5**), functional redundancy between the two proteins is very likely. In addition, cooperative functions of both proteins in axon outgrowth and in mediating fiber tract decussation have been reported (Deuel et al., 2006; Koizumi et al., 2006; Tanaka et al., 2006). Study on an allelic series of DCX/DCLK1 KO mice with cresyl violet staining reveals a dosage-dependent effect on cortical lamination. Disorganized cytoarchitetonics with dispersion of layer2/3 in the cortex was observed in DCX mutant mice with one copy of DCLK removed. Similar phenotypic defects were detected in DCLK1 mutant mice lacking a copy of DCX. In addition, Koizumi et al. (2006) observed a dosage-dependent interaction between DCX and DCLK1 in commissural fiber tract formation. Removal of one copy of DCLK1 in DCXnull mice causes hypoplastic corpus callosum and thin anterior commissures, whereas anterior commissures are hypoplastic in DCLK1 mutant mice lacking a copy of DCX. On the other hand, deletion of all four copies of DCX and DCLK1 results in disappearance of the corpus callosum, anterior commissures and hippocampus commissures.

#### Microtubule-Based Roles for DCX

On the molecular level, DCX binds microtubules (MTs) via its two DC repeats to stabilize MTs and promote MT polymerization (**Figure 5A**). DCX has been reported to function with Lis1 and dynein to mediate nucleus-centrosome coupling in neuronal migration. Depletion of DCX in neurons results in delayed centrosomal-nuclear movement, whereas DCX overexpression is able to rescue the nucleus-centrosome coupling defect caused by dynein inhibition and increases the migration rates (Tanaka et al., 2004). These defects are largely attributed to the microtubule-binding activity of DCX. Similar to Dabs, DCX is a phosphoprotein and is known to be a substrate for kinases such as PKA, Cdk5, MAPK8 (aka JNK1), and others. Phosphorylation of DCX regulates its MT binding activity and its localization at the leading processes of migrating neurons: Cdk5-mediated phosphorylation of DCX at Ser297 increases DCX binding to MTs, leading to MT stabilization and promotion of tubulin polymerization, events crucial for neuronal migration (Tanaka et al., 2004).

In addition to microtubule binding, DCX also binds several other proteins (Caspi et al., 2000; Friocourt et al., 2001; Kizhatil et al., 2002; Tanaka et al., 2004; Liu et al., 2012). The roles of these binding interactions are still under investigation. Analysis of double knockout mice for DCX and DCLK1 have uncovered a previously unsuspected role for the DCX proteins in axon outgrowth and dendrite branching and demonstrated aberrant trafficking of synaptic vesicle proteins (Deuel et al., 2006). In neurons cultured from DCLK1 KO with additional DCX knockdown, VAMP2 and synaptophysin accumulate in the cell body, but are completely absent from axons, a drastic contrast from the WT where both proteins are detected in the cell body and the axons. In a follow-up study by Liu et al., the VAMP2 transport defect found in DCX/DCLK1-deficient neurons is caused by the mislocalization of Kif1a. Interestingly, a diseaseassociated DCX mutation impairs Kif1a motility and disrupts Kif1a-mediated VAMP2 transport from the soma to the neurites. The finding suggested that DCX/DCLK1 is involved in neuronal migration and axonal outgrowth via its interaction with Kif1a on the microtubules for transport regulation (Liu et al., 2012).

## DCX as an Endocytic Co-Adaptor for Cell Adhesion Molecules

Several of the other reported defects in DCX deficiencies have led to suggestions that DCX is involved in vesicle trafficking. Defects in synaptic vesicles and synaptic vesicle proteins (such as VAMP2) have all been found (Friocourt et al., 2003; Deuel et al., 2006). The exact molecular mechanisms of such putative trafficking effects of DCX, though, are not known, nor are the cargos. We recently found that DCX might act as an endocytic adaptor and modulate the surface distribution of the cell adhesion molecule neurofascin in developing cultured rat neurons (Yap et al., 2012). DCX binds directly to the cytoplasmic tail of neurofascin via a binding site dependent on residue G253 in DCX (Kizhatil et al., 2002). In contrast to numb and Dab2 which interact with the α subunit of AP-2, DCX associates with the AP2-µ2 subunit via a YLPL motif located in the DCX C-terminus (Friocourt et al., 2001; **Figure 5A**). The YLPL motif conforms to the YXXΦ class of tyrosine-based sorting motifs and is absent in DCLK1/2. Since DCX had been reported to bind to neurofascin (Kizhatil et al., 2002), we tested if DCX mediated the endocytosis of neurofascin. In fact, depletion of DCX diminishes endocytosis of endogenous neurofascin in cultured neurons (Yap et al., 2012). It was previously shown that a cytoplasmic motif in neurofascin (FIGQY) binds to ankyrin when unphosphorylated. In contrast, the phosphorylated FIGQY motif has low affinity for ankyrin, but high affinity for DCX (Kizhatil et al., 2002). Signaling of neurotrophins leads to phosphorylation of the FIGQY motif in the cytoplasmic tail of neurofascin, thereby increasing its affinity for DCX. We proposed (Yap et al., 2012) that local signaling via NGF leads to phosphorylation of the FIGQY motif in neurofascin, which can then bind to DCX locally. Neurofascin is then endocytosed in a DCX-dependent manner from the neuronal plasma membrane, especially in the soma and dendrites (**Figure 5B**). The endocytic machinery required for this event is not yet clear but may involve AP-2 clathrin adaptors. Since we observed the most robust endocytosis of neurofascin in young neurons, DCX-mediated modulation of neurofascin levels could play a role during development. Whether DCX interacts with and modulates sorting of other cargos via clathrin/AP-2 mediated endocytic trafficking remains to be investigated.

# Postendocytic Trafficking—Same Signals, Different Machinery?

Given the fact that the µ-subunits of tetrameric AP complexes recognize and interact with similar cytoplasmic motifs in cargos, it is believed that endocytic trafficking of a cargo involves several AP complexes to act in conjunction with each other in a sequential manner for sorting a cargo to its final destination. Yuzaki's group showed that NMDAinduced AMPAR trafficking to the late endosome requires sequential interactions of stargazin, a transmembrane AMPA receptor regulatory protein (aka TARP-γ2), with AP-2 followed by AP-3 during NMDA-dependent long-term depression. Stargazin promotes association of AMPAR with AP-2 and AP-3, and mediates formation of ternary complexes containing AMPAR and the AP complexes. Inhibition of the stargazin interaction with AP-2 impairs NMDA-induced AMPAR endocytosis, whereas inhibition of the stargazin interaction with AP-3 disrupts late endosomal/lysosomal trafficking of AMPAR, thereby leading to recycling of AMPAR back to the cell surface (Matsuda et al., 2013). Similarly, Schachner's group showed that NCAM promotes switching of synaptic vesicles recycling from an AP-3 to an AP-2–dependent mechanism during synapse maturation (Shetty et al., 2013). Both studies suggested that the sequential interaction of a cargo with different AP complexes for its sorting from one endosomal compartment to the subsequent compartment along the endocytic pathway is essential for proper functions in neurons.

# Post-Endocytic Roles of Numb

Numb localizes not only to the plasma membrane where it can aid endocytosis, but also to endocytic organelles as well as the TGN, and co-traffics to endosomes with endocytosed receptors (Santolini et al., 2000). In addition to its function in clathrin/AP2-mediated internalization, growing evidence thus implicates Numb in regulating endosomal sorting of receptors post-endocytosis. This is not surprising since many adaptor binding proteins bind more than one of the 5 tetrameric adaptors, such as AP-1 and AP-2. In fact, Drosophila numb physically interacts with the AP-1 complex by co-immunoprecipitation (Cotton et al., 2013).

#### Notch

Numb has been implicated in regulating the post-endocytic trafficking of Notch1. Endocytosis is required to maintain the steady state level of Notch receptors (Le Borgne et al., 2005). Mammalian Notch1 is known to constitutively internalize and traffic to recycling and late endosomal compartments. Despite the fact that numb-mediated inhibition of Notch signaling requires α-adaptin, it is still unknown whether Numb directly regulates the endocytosis or it regulates the endocytic trafficking of Notch after internalization via its association with Eps15 and α-adaptin (Le Borgne et al., 2005). What is known is that interaction of numb with the intracellular domain of Notch (Notch ICD) recruits the E3-ubiquitin ligase itch to the membrane-tethered Notch and leads to polyubiquitination and degradation of Notch ICD. In a study using mammalian cell lines, overexpression of Numb promotes trafficking and degradation of Notch 1, whereas depletion of numb facilitates recycling of Notch1. Numb mutants defective for binding to Itch, Eps15 or adaptin, fail to promote Notch 1 degradation, suggesting Numb suppresses Notch activity by regulating post-endocytic sorting pathways that lead to the degradation of Notch (McGill and McGlade, 2003; McGill et al., 2009).

### Sanpodo

In Drosophila, interaction of Numb with α-adaptin is required for numb-mediated asymmetric cell division. During asymmetric cell division of SOPs, numb distributes asymmetrically between two daughter cells (called pIIa and pIIb), which is important for the subsequent binary cell fate decision. Hutterer et al reported that internalization of Sanpodo, a transmembrane protein required for Notch signaling in Drosophila, is mediated by α-adaptin via interaction with Numb (Hutterer and Knoblich, 2005). However, two recent studies demonstrated that Numb is not essential for the internalization of Sanpodo. The bulk of AP-2-dependent Sanpodo endocytosis still occurs in Numb mutant SOPs (Cotton et al., 2013; Couturier et al., 2013a,b). The studies proposed that Numb interacts with the sanpodo-Notch complex at early endosomes in concert with AP-1 to regulate the endosomal trafficking of the complex. The interaction of numb with sanpodo blocks the recycling of Notch back to the plasma membrane, leading to the asymmetric distribution of Notch along the pIIa-pIIb cell contact interface. A recent finding by Couturier et al. (2014) using dual GFP/cherry-tagged sensors in live SOP cells further confirmed that sanpodo is internalized into early/sorting endosomes in the Numb-inheriting daughter cell and sorted toward late endosomes, a process which is dependent on Numb. Similar to mammalian cells then, numb in flies mediates inhibition of Notch via its regulation of cargo sorting towards late endosomes and inhibition of recycling (Couturier et al., 2014). Some of the seemingly disparate roles of numb on Notch signaling might thus be due to differential post-endocytic sorting of numb cargos towards either recycling or degradative pathways. Depending on the cell type or developmental timing, numb-dependent Notch trafficking could either lead to increased signaling (via recycling) or decreased signaling (via degradation).

#### Cadherins

Cadherins are required for adhesion and polarity of radial glia cells (RGCs) in the cortex as depletion of cadherins by shRNA results in loss of end-feet and of bipolar morphology. Numb and Numbl are required for the maintenance of cadherinmediated radial glial adherens junctions (Rasin et al., 2007). EM analysis demonstrated that Numb was enriched just basally of the apical end-feet of interphase RGCs and localized to internalized cadherin-containing Rab11-positive endosomes. Numb physically interacts with the cadherin/catenin complex via its PTB and C-terminal domains, and proper localization of cadherins requires Numb/numbl. Inactivation of Numb and Numbl in conditional double knockout mice decreases basolateral insertion of cadherins and mislocalizes cadherins to the cytoplasm and to apical membrane regions of RGCs. The changed cadherin distribution is likely the reason for the disrupted adherens junctions and loss of polarity observed in the mice. This results in progenitor cells dispersion and disorganized cortical lamination. In contrast, overexpression of Numb and numbl prolongs cadherin-dependent RGC apical attachment and polarization. Thus, Numb plays a critical role in ensuring correct trafficking of cadherins, a process required for the maintenance of adherens junctions during neurogenesis (Rasin et al., 2007).

#### Sorting Nexin 17 (SNX17): Regulating Post-Endocytic Trafficking of ApoER2

In addition to binding to Dab1, ApoER2 has recently been reported to interact with SNX17 via its cytoplasmic NPXY motif for its endocytic trafficking and receptor signaling (Sotelo et al., 2014). SNX17 is a cytosolic protein that is highly expressed in mouse brain and localized to early endosomes. SNX17 is involved in the endocytic trafficking of membrane proteins, including LRP1, integrins, and P-selectin. For instance, SNX17 has been reported to mediate recycling of integrins. SNX17 binds integrin and prevents degradation in lysosomes (Böttcher et al., 2012; Steinberg et al., 2012). Similarly, SNX17 co-localizes with endocytosed ApoER2 and facilitates the trafficking from early endosomes to recycling endosomes (Sotelo et al., 2014). Depletion of SNX17 causes retention of ApoER2 in Rab5 positive early endosome compartments, leading to a decrease of ApoER2 in Rab11 recycling endosomes. The defect in ApoER2 recycling causes low surface level of ApoER2 and leads to dendritic outgrowth defects. Interestingly, accumulation of ApoER2 in the early endosome of SNX17-deficient neurons promotes proteolytic cleavage of its C-terminal domain. Furthermore, downregulation of SNX17 also promotes reelininduced degradation of ApoER2, and dampens the activation of downstream reelin effectors, such as the phosphorylation of Dab1. The data suggested that SNX17 is an endosomal adaptor

# References

Aguet, F., Antonescu, C. N., Mettlen, M., Schmid, S. L., and Danuser, G. (2013). Advances in analysis of low signal-to-noise images link dynamin and AP2 to the functions of an endocytic checkpoint. Dev. Cell 26, 279–291. doi: 10.1016/j. devcel.2013.06.019

that regulates the post-endocytic trafficking and recycling of ApoER2. It is via this post-endocytic role that SNX17 participates in reelin-receptor signaling (Sotelo et al., 2014). Therefore, in neurons some co-adaptors such as Numb (and possibly Dab1) regulate endocytosis of NPXY-containing cargos at the plasma membrane as well as their postendocytic sorting. Other co-adaptors, such as SNX and likely other still unidentified co-adaptors, might be additionally involved in the subsequent steps of sorting following endocytosis.

# Conclusions

In this review, we laid out some of the emerging themes of how regulating the endocytosis and endosomal trafficking of critical receptors contributes to orchestrating developmental decisions in the nervous system. We highlighted some of the better understood endocytic co-adaptors, but many other pathways are similarly coordinated by endocytic regulation. For instance, Wnt signaling via disshevelled is dependent on endocytosis (Onishi et al., 2013). In particular, we discussed examples highlighting the following concepts:


Given the large number of receptor pathways, we look forward to many more exciting discoveries of the diverse roles of EAPs in neural development.


ApoE receptor 2 induces tyrosine phosphorylation of disabled-1 and modulates tau phosphorylation. Neuron 24, 481–489. doi: 10.1016/s0896-6273(00) 80861-2


(SRM) quantification of endocytosis factors associated with Numb. Mol. Cell. Proteomics 12, 499–514. doi: 10.1074/mcp.M112.020768


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Yap and Winckler. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Emerging roles of Axin in cerebral cortical development

#### Tao Ye , Amy K. Y. Fu and Nancy Y. Ip\*

Division of Life Science, Molecular Neuroscience Center and State Key Laboratory of Molecular Neuroscience, The Hong Kong University of Science and Technology, Hong Kong, China

Proper functioning of the cerebral cortex depends on the appropriate production and positioning of neurons, establishment of axon–dendrite polarity, and formation of proper neuronal connectivity. Deficits in any of these processes greatly impair neural functions and are associated with various human neurodevelopmental disorders including microcephaly, cortical heterotopias, and autism. The application of in vivo manipulation techniques such as in utero electroporation has resulted in significant advances in our understanding of the cellular and molecular mechanisms that underlie neural development in vivo. Axin is a scaffold protein that regulates neuronal differentiation and morphogenesis in vitro. Recent studies provide novel insights into the emerging roles of Axin in gene expression and cytoskeletal regulation during neurogenesis, neuronal polarization, and axon formation. This review summarizes current knowledge on Axin as a key molecular controller of cerebral cortical development.

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/ PRESTO, JST, Japan

#### Reviewed by:

Phillip R. Gordon-Weeks, King's College London, UK Harish Pant, NINDS, National Institutes of Health, USA

#### \*Correspondence:

Nancy Y. Ip, Division of Life Science, Molecular Neuroscience Center and State Key Laboratory of Molecular Neuroscience, The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, China boip@ust.hk

> Received: 28 February 2015 Accepted: 21 May 2015 Published: 08 June 2015

#### Citation:

Ye T, Fu AKY and Ip NY (2015) Emerging roles of Axin in cerebral cortical development. Front. Cell. Neurosci. 9:217. doi: 10.3389/fncel.2015.00217 Keywords: Axin, cerebral cortex, neurogenesis, polarization, axon formation, cytoskeletal regulation

# Introduction

The mammalian cerebral cortex is characterized by a six-layered laminar structure, which forms the structural basis for higher cognitive function. Each cortical layer contains distinct distributions of neuron types with specific dendritic morphology, electrophysiological properties, and axonal connection with other brain regions. Remarkably, the characteristic distribution and connectivity of cortical neurons originate from a single layer of progenitor cells called the neuroepithelium. During cortical development, neural progenitor cells located in the ventricular zone undergo symmetric cell division to proliferate and maintain a proper progenitor pool; moreover, they utilize asymmetric cell division to differentiate and generate successive waves of neurons. After neuronal differentiation, the postmitotic neurons go through a multipolar stage; they subsequently polarize, taking on a bipolar morphology with a leading process towards the cortical plate and a nascent axon toward the opposite direction. The cell body continues to migrate along radial glial fibers towards the cortical plate and past the existing layers of neurons. Thus, cortical layers are created in an inside-out manner, with the early- and later-born neurons occupying the deeper and superficial layers, respectively.

When the neurons mature at their destination, the leading process spawns the apical dendrite, followed by dendritic arborization and synaptic formation. The nascent axon elongates tangentially, projecting to form synaptic connections with other neurons. Defects in any of these coordinated events can greatly impair neural functions and are implicated in various neurological and psychiatric disorders including lissencephaly, cortical heterotopias, and autism (Ayala et al., 2007). Thus, these steps require the orchestration of various signaling cascades; accordingly, scaffold proteins have emerged as key molecular controls of cortical development owing to their abilities to interact with and regulate a myriad of signaling proteins. This review summarizes the emerging roles of the scaffold protein Axin in the orchestration of the transcriptional program of neurogenesis and cytoskeletal regulation of neural development in the cerebral cortex.

# Identification and Regulation of Axin

Axin was initially identified from the analysis of the Fused locus, mutations of which cause defects in axis formation in mouse embryos (Zeng et al., 1997). Because of its inhibition of axis formation, the Fused gene was named Axis inhibitor (Axin). Genetic analyses identified Axin as a negative regulator of the canonical Wnt signaling pathway, although its precise roles remained unclear (Kikuchi, 1999). Biochemical studies demonstrated that Axin acts as a scaffold and associates with various components of the canonical Wnt signaling pathway, forming the β-catenin destruction complex, which leads to the GSK3β–dependent phosphorylation of β-catenin (Luo and Lin, 2004).

Axin possesses several functional domains including the Regulators of G protein signaling domain near its N-terminus and the C-terminal DIX domain, which is also found in Disheveled and Dixdc1 (also called Ccd1). In addition, Axin contains domains for interacting with other proteins such as GSK3β and β-catenin (**Figure 1**). In brief, Axin interacts with adenomatosis polyposis coli (APC), GSK3β, and β-catenin via distinct domains to form the β-catenin destruction complex (Furuhashi et al., 2001). Without Wnt ligand stimulation, Axin facilitates GSK3β-mediated phosphorylation of β-catenin, and thus triggers the proteasome-dependent ubiquitination and degradation of β-catenin. Hence, although β-catenin is constitutively expressed in the cytoplasm, its protein level remains relatively low because of destruction complex activity. When the Wnt ligands bind to the receptors Frizzled and LRP5/6, they activate several signaling components such as Disheveled. Disheveled associates with Axin, translocating Axin to the membrane, which leads to Axin degradation. Loss of Axin results in the disassembly of the destruction complex as well as the suppression of β-catenin phosphorylation by GSK3β. Subsequently, β-catenin accumulates and translocates into the nucleus, where it forms a complex with TCF/Lef transcription factors and turns on Wnt-responsive gene transcription (Furuhashi et al., 2001). It should be noted that most of these findings are from studies using different types of cultured cells. Thus, the precise mechanisms in vivo are only beginning to be elucidated.

Axin has also emerged as a master scaffold for multiple signaling pathways (Kikuchi, 1999). Axin interacts with and activates MEKK, which activates JNK, suggesting that Axin regulates the JNK pathway (Liu et al., 2006). Axin also functions as a scaffold in the transforming growth factor β (TGFβ) pathway by facilitating the ubiquitin E3 ligase Arkadiadependent degradation of Smad7 (Zeng et al., 1997). It also directly interacts with Smad3 and promotes its phosphorylation by TGF-β receptors (Furuhashi et al., 2001; Guo et al., 2008).

# Coordination of Neurogenesis by Axin-Mediated Signaling

Neurons are generated from neural progenitor cells. Cortical neural progenitors include neuroepithelial progenitors, radial glial progenitors (RGPs; Ever and Gaiano, 2005), and intermediate progenitors (IPs; Farkas and Huttner, 2008). Among them, RGPs in the ventricular zone are responsible for generating most or all neurons during embryonic development (Kriegstein et al., 2002; Götz et al., 2003; Heintz et al., 2004). RGPs enlarge the progenitor pool through symmetric proliferative divisions in order to enable the subsequent rapid increase in neurons (Götz and Huttner, 2005; Huttner and Kosodo, 2005). As development proceeds, RGPs switch to neuronal differentiation, dividing asymmetrically and producing several neurons. RGPs adopt two modes of asymmetric divisions to generate neurons: producing one RGP for selfrenewal and either one neuron (direct neurogenesis) or one neurogenic IP (indirect neurogenesis) to enlarge the population of neurons (Noctor et al., 2004; Götz and Huttner, 2005). Consequently, asymmetric cell division of RGPs plays a critical role in generating a proper number of neurons, while at the same time maintaining an adequate pool of neural

and Cdk5-dependent phosphorylation site. Axin possesses functional domains including the Regulators of G protein signaling (RGS which interacts with adenomatosis polyposis coli, APC) and DIX domains. Axin contains domains for interacting with other proteins such as MEKK1 (MID), GSK3β (GID), poly-ADP-ribosylating enzyme tankyrase which stimulates Axin degradation through the ubiquitin-proteasome pathway. It is worth noting that the Cdk5-dependent phosphorylation site (Thr485) is located close to the nuclear export signal (NES) of Axin (amino acids 413–423).

progenitors for self-renewal. IPs that preferentially reside in the subventricular zone are transient neurogenic progenitors with limited amplifying capability (1–3 mitotic cycles). IPs undergo symmetric divisions to produce pairs of IPs that subsequently differentiate into neurons (Pontious et al., 2008; Kowalczyk et al., 2009). Thus, the birthdates and numbers of neurons are dependent on the balance between neural progenitor proliferation and differentiation.

Recent evidence suggests that Axin serves as a master scaffold for coordinating the proliferation and differentiation of neural progenitors during cerebral cortical development (Fang et al., 2013). Our laboratory has demonstrated that the level and subcellular localization of Axin in neural progenitors determine their fate, either self-proliferation or neuronal differentiation. In the presence of proliferating cues such as SHH, the Axin-GSK3β interaction in the cytoplasm is crucial for the proliferation of the intermediate progenitor cells (**Figure 2**). Upon stimulation by neurogenic cues such as WNT, RA, and TGFβ, the Axin-βcatenin interaction in the nucleus promotes neuronal production through the activation of neurogenic transcription factors (**Figure 3**). Of note, Axin phosphorylation at Thr485 (**Figure 1**) by cyclin-dependent kinase-5 (Cdk5) determines the subcellular localization of Axin, which translocates from the cytoplasm to the nucleus upon neural progenitor differentiation, thus serving as a molecular switch that causes IPs to switch from proliferation to differentiation (Fang et al., 2013).

factors, such as Ngn1 and NeuroD1.

# Axin-GSK3β Signaling in Neuronal Polarization and Migration

Newborn neurons exit the ventricular and subventricular zones, and migrate into the intermediate zone. They initially exhibit a multipolar morphology characterized by several thin processes in various directions but subsequently adopt a bipolar morphology with a leading process pointing toward the apical pia and a trailing process extending toward the ventricle. Directed by their leading processes, neurons migrate toward the cortical (i.e., pial) surface in an orderly progression to occupy their proper positions (Tsai and Gleeson, 2005). Each wave of migrating neurons travels past their predecessors, forming cortical layers of the cortical plate in an inside-out manner, in which the older and younger neurons reside in deeper layers closer to the ventricle and outer layers of the cerebral cortex closer to the pia, respectively (Noctor et al., 2004). Projection neurons generally migrate in two modes: radial glial fiber-independent somal translocation and radial glial fiber-dependent locomotion. Somal translocating neurons attach their long leading processes to the pia through Reelin-mediated cell-matrix adhesion (Sekine et al., 2012) and subsequently shorten them to move their cell bodies to their final positions and detach from the pia to terminate migration (Dulabon et al., 2000). Notably, as the cortical plate is sufficiently thin during the early stage of cortical development, early-generated neurons can extend leading processes to the pia and migrate through somal translocation alone (Nadarajah et al., 2001). On the other hand, locomoting neurons adhere to the radial glial fibers and extend their short leading processes to wrap around the fibers, forming a temporary adhesion that facilitates their migration along the fiber (Noctor et al., 2004; Ayala et al., 2007). Thus, various factors including the polarity and morphology of migrating neurons as well as adhesion with radial glial fibers contribute to the appropriate laminar positioning of neurons.

The first line of evidence suggesting that Axin plays a role in neuronal migration came from the effect of Axin overexpression in migrating neurons (Fang et al., 2013). Notably, a substantial number of Axin-overexpressing neurons are stacked in the intermediate zone, suggesting that Axin has an alternative function in neuronal migration probably through the regulation of GSK3 signaling (Fang et al., 2013). A recent study using mutant mice lacking both GSK3α and GSK3β confirms the in vivo role of GSK3 signaling in neuronal migration (Morgan-Smith et al., 2014). Conditional Gsk3 deletion in cortical neurons under neuron-specific Neurod6 promoter resulted in dramatic mislocalization of layer 2/3 neurons in deeper layers. In particular, Gsk3 deletion disrupts the transition of cortical neurons from multipolar migration phase to bipolar migration phase (Morgan-Smith et al., 2014). Since GSK3β but not GSK3α interacts with Axin during neural development (Fang et al., 2011, 2013), these findings collectively suggest that Axin-GSK3β interaction can target the neuronal polarization process to regulate neuronal migration during cerebral cortical development.

How does Axin-GSK3β signaling specify the neuronal polarity? Increasing evidence suggests that GSK3β functions as a critical regulator of neuronal polarization and migration by controlling microtubule dynamics (Hur and Zhou, 2010). Initial studies on neuronal polarization came from cultured dissociated hippocampal neurons. Before polarity establishment, Ser9-phosphorylated GSK3β, which is the inactive form, is universally expressed at the tip of each neurite. Upon neuronal polarization, one of the neurites gives rise to the axon, where phosphorylated GSK3β becomes enriched at the tip of the nascent axon. Therefore, axon specification depends on the local inhibition of GSK3β in one neurite and activation of GSK3β in other neurites. Accordingly, multiple axons are induced when GSK3β activity is globally suppressed by pharmaceutical inhibitors or GSK3β-specific knockdown. In contrast, axon specification is impaired when the constitutive GSK3β-Ser9Ala mutant is overexpressed, and thus neuronal polarization is inhibited. Besides, the instructive role of GSK3β inactivation in axon specification is further demonstrated by the observation that the differentiated dendrite can be transformed into axons after axon-dendrite polarity specification. This evidence collectively indicates that local inhibition of GSK3β is essential for the establishment and maintenance of neuronal polarity (Hur and Zhou, 2010). Notably, we have suggested that the specific Axin-GSK3β interaction ensures the precise localization and inactivation of GSK3β in one neurite during neuronal polarization (Fang et al., 2011).

Cytoskeletal reorganization is required for neurons to undergo morphological changes during polarization and migration. Several substrates of GSK3β have been identified as important downstream effectors in regulating neuronal polarization through microtubule rearrangement (**Figure 4**). These substrates includes microtubule plus-end binding proteins CRMP-2 (Yoshimura et al., 2005), APC (Zumbrunn et al., 2001), and doublecortin DCX (Bilimoria et al., 2010), as well as structural microtubule associated proteins (MAPs) such as MAP1B (Trivedi et al., 2005) and Tau (Johnson and Stoothoff, 2004). In polarizing neurons, plus-end binding proteins CRMP-2 and APC localizes at the tips of immature neurites and later becomes concentrated in the growth cone of the nascent axon (Inagaki et al., 2001; Zumbrunn et al., 2001). Binding of unphosphorylated active forms of CRMP2, DCX, APC, and Tau to microtubules promotes their assembly and stabilization. The microtubule-binding activity of these proteins is abrogated by GSK3β phosphorylation (Hur and Zhou, 2010). Therefore, GSK3β inactivation in the growth cone of the nascent axon facilitates the interaction between MAPs and microtubules, and thus promotes axon initiation by specifically stabilizing one neurite. Similar to the effect of global GSK3β inhibition, overexpression of phosphorylation-deficient CRMP2-Ser514Ala is sufficient to induce multiple axons during neuronal polarization. In addition, GSK3β-mediated phosphorylation of MAP1B may also contribute to microtubule dynamics (Trivedi et al., 2005).

# Roles of Axin in Axon Formation and Outgrowth

After neurons reach their final destinations, the leading process develops into an apical dendrite and extends up to the pia,

while the trailing process differentiates into an axon and grows toward the intermediate zone. Over time, the apical dendrite branches extensively and is accompanied by basal dendrites arborizing radially from the cell soma (Whitford et al., 2002). The axons pass vertically through the cortical plate and intermediate zone, sprout collaterals that arborize in specific intracortical layers, and bundle to form axonal tracts that project to intraand subcortical areas (Hatanaka and Murakami, 2002). The projection targets determine the three basic subtypes of neurons: associative, commissural, and corticofugal, in which axons form connections within the cortex in the same hemisphere, opposite hemisphere, or away from the cortex, respectively (Molyneaux et al., 2007). Neurons in distinctive cortical and subcortical areas are functionally coordinated and integrated through their axonal projections to allow the proper cortical information processing. The patterns of axon projection and connection involve axon formation and extension, guidance, recognition of and targeting to specific areas, and elimination of inappropriate axon segments and branches. Among these processes, axon formation and extension are the most basic and are regulated by cytoskeletal reorganization.

Axon is initiated and facilitated by the extracellular cues. For example, TGF-β and its receptors specify the axon during brain development. TGF-β receptors are expressed in axons during embryonic development, and their receptor kinase activity is required for axon formation (Yi et al., 2010). The effect of TGF-β signaling on axon specification is mediated by the phosphorylation of Par6, a component of the Par3/Par6/aPKC complex. Neurotrophic factors such as brain-derived neurotrophic factor (BDNF) can also direct axon specification, because the first neurite contacting a BDNF stripe becomes the axon (Shelly et al., 2007). The effect of BDNF on axon specification requires the activation of the polarityinducing kinase LKB1 via a cAMP-dependent protein kinase A pathway (Shelly et al., 2007). In another case, both BDNF and neurotrophin 3 stimulate the inhibition of GSK3β, resulting in the dephosphorylation of CRMP2, which consequently promotes axon outgrowth in cultured hippocampal neurons (Yoshimura et al., 2005).

Interestingly, recent evidence demonstrates that Axin plays an instrumental role in microtubule assembly and axonal transport in supporting axon formation and growth. Through Axin knockdown in utero, our laboratory demonstrated that the Cdk5-mediated phosphorylation of Axin contributes to axon formation through the inhibition of GSK3β in vivo (Fang et al., 2011). In particular, Cdk5 is activated through p35 to phosphorylate Axin at Thr485 in response to neurotrophins such as BDNF and neurotrophin 3 (NT-3). The phosphorylation of Axin enhances its interaction with GSK3β, which inhibits GSK3β activity, thereby increasing nonphosphorylated CRMP-2 and Tau in the growth cone (Fang et al., 2011). Nonphosphorylated CRMP-2 and Tau, which represents the active form, promotes microtubule assembly and stabilization to support the elongation of growing axons during development. Therefore, Axin expression and phosphorylation by Cdk5 are essential for axon formation and outgrowth through the regulation of microtubule dynamics (**Figure 5**). In addition to microtubule dynamics, efficient axonal transport requires Cdk5 mediated suppression of GSK3β, thereby preventing premature GSK3β-mediated cargo release (Morfini et al., 2004). Whether Cdk5 inhibits GSK3β by facilitating Axin-GSK3β interaction awaits further investigation.

# Implications of Axin in Neurodevelopmental Disorders

As Axin is a multifaceted scaffold protein involved in diverse signaling pathways that regulate neural progenitor proliferation and differentiation, it is not surprising that its deregulation is associated with brain malformations such as micro- and

macrocephaly. Notably, a previous study demonstrates that a mutation in the GSK3-binding domain of Axin results in the formation of small brains, mimicking human microcephaly (Heisenberg et al., 2001). Additional evidence in a human genetic study shows that the Axin gene is located in the same chromosomal region (16p13.3–12.1) as genetic mutations found in microcephaly patients (Kavaslar et al., 2000). Despite the lack of an Axin genetic mouse model, the recent development of in utero electroporation techniques have enabled the manipulation of Axin protein in neural progenitor cells during cerebral cortical development in vivo (Saito and Nakatsuji, 2001; Tabata and Nakajima, 2001). Accordingly, depletion or overexpression of Axin in the mouse neocortex results in premature and suppressed neuronal differentiation, which eventually lead to reduced and excessive numbers of cortical neurons at birth, respectively (Fang et al., 2013).

Axin is also implicated in the etiology of autism, a childhood-onset neurodevelopmental disorder characterized by disabilities in social interaction and communication as well as repetitive and compulsive behaviors. A postmortem study suggests that patients with autism often have increased neuronal density and number in the prefrontal cortex (McCaffery and Deutsch, 2005; Courchesne et al., 2011). We have recently generated a mouse model with enhanced neurogenesis by intraventricular injection of a tankyrase inhibitor, XAV939 (**Figure 1**; Huang et al., 2009). Injection of XAV939 transiently elevates the protein level of Axin in neural progenitors by preventing tankyrase-mediated protein degradation and consequently increases the number of upper-layer neurons in the developing mouse cortex without affecting astrogenesis or microglial reactivity (Fang et al., 2014). We demonstrated that enhanced neurogenesis leads to the overproduction of excitatory neurons, and impairs excitatory and inhibitory synaptic connection and balance. More importantly, these mice exhibit autism-like features, namely social interaction deficits in the three-chamber sociability test and compulsive behaviors such as increased self-grooming and marble burying (Fang et al., 2014). These results collectively illustrate a neurodevelopmental mechanism wherein enhanced neurogenesis and increased neuronal production result in autism-like features, hinting at the etiology of autism.

# Concluding Remarks and Perspectives

Current evidence from several studies clearly indicates that Axin acts as a scaffold protein to integrate upstream signals while regulating various downstream interacting proteins involved in a wide array of cellular activities. Specifically, cytosolic Axin modulates downstream signaling of extracellular signals, including neurotrophins, WNT, Notch, and TGFβ. In addition, Axin enables efficient cytoskeletal reorganization and axon transport through GSK3β signaling and various MAPs. Moreover, nuclear Axin regulates gene expression by interacting with β-catenin and activating various neurogenic transcriptional factors. Therefore, Axin appears to be a master scaffold that modulates various developmental steps in the development of cerebral cortex.

After embryonic development, lifelong neurogenesis occurs in specific regions of the adult brain; various Axin-regulated signaling proteins such as GSK3β and β-catenin continue to play important roles in adult neurogenesis (Lie et al., 2005; Mao et al., 2009; Morales-Garcia et al., 2012). Therefore, it will be of interest to investigate how Axin regulates the proliferation and differentiation of adult neural stem cells and if enhancing Axin protein level improves adult neurogenesis, which is implicated in depression and neurodegenerative diseases such as Alzheimer's disease.

# Acknowledgments

We apologize to the researchers whose studies could not be discussed or cited because of space limitations. We would like to thank Ka-Chun Lok for his help with preparing the figure and members of the Ip Laboratory for helpful discussions. This study was supported in part by the Hong Kong Research Grants Council Theme-based Research Scheme (T13-607/12R), the National Key Basic Research Program of China (2013CB530900), the Research Grants Council of Hong

# References


Kong SAR (HKUST660110, HKUST660610, HKUST660810, HKUST660110, HKUST661111, HKUST661212, and HKUST661013), the Innovation and Technology Fund for State Key Laboratory (ITCPT/17-9), and the SH Ho Foundation.


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Ye, Fu and Ip. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Crosstalk between intracellular and extracellular signals regulating interneuron production, migration and integration into the cortex

#### Elise Peyre1, 2 †, Carla G. Silva1, 2 † and Laurent Nguyen1, 2, 3 \*

<sup>1</sup> GIGA-Neurosciences, University of Liège, Liège, Belgium, <sup>2</sup> Interdisciplinary Cluster for Applied Genoproteomics (GIGA-R), University of Liège, Liège, Belgium, <sup>3</sup> Wallon Excellence in Lifesciences and Biotechnology, University of Liège, Liège, Belgium

During embryogenesis, cortical interneurons are generated by ventral progenitors located in the ganglionic eminences of the telencephalon. They travel along multiple tangential paths to populate the cortical wall. As they reach this structure they undergo intracortical dispersion to settle in their final destination. At the cellular level, migrating interneurons are highly polarized cells that extend and retract processes using dynamic remodeling of microtubule and actin cytoskeleton. Different levels of molecular regulation contribute to interneuron migration. These include: (1) Extrinsic guidance cues distributed along migratory streams that are sensed and integrated by migrating interneurons; (2) Intrinsic genetic programs driven by specific transcription factors that grant specification and set the timing of migration for different subtypes of interneurons; (3) Adhesion molecules and cytoskeletal elements/regulators that transduce molecular signalings into coherent movement. These levels of molecular regulation must be properly integrated by interneurons to allow their migration in the cortex. The aim of this review is to summarize our current knowledge of the interplay between microenvironmental signals and cell autonomous programs that drive cortical interneuron porduction, tangential migration, and intergration in the developing cerebral cortex.

#### Keywords: interneurons, cortex, migration, progenitors, nucleokinesis, branching

# Introduction

During mouse embryogenesis, cortical interneurons (cINs) are generated in the ventral subpallium. Distinct proliferative regions can be identified in this area, including the lateral ganglionic eminence (LGE), the medial ganglionic eminence (MGE), the caudal ganglionic eminence (CGE) and the preoptic area (POA). While the MGE, CGE and POA contribute to the generation of cortical interneurons (cINs) (Flames et al., 2007; Rubin et al., 2010), the LGE is mostly involved in striatal and olfactory bulb histogenesis (Yun et al., 2003). These distinct subpallial regions differ

#### Edited by:

Yoko Arai, Université Paris Diderot, France

#### Reviewed by:

Christine Métin, Institut National de la Santé et de la Recherche Médicale, France Diego Matias Gelman, Instituto de Biologia y Medicina Experimental, Argentina

#### \*Correspondence:

Laurent Nguyen, GIGA-Neurosciences, Quartier Hôpital, University of Liège, Tour B36, Avenue Hippocrate 15, 4000 Liège, Belgium lnguyen@ulg.ac.be †

These authors have contributed equally to this work.

Received: 19 December 2014 Accepted: 19 March 2015 Published: 14 April 2015

#### Citation:

Peyre E, Silva CG and Nguyen L (2015) Crosstalk between intracellular and extracellular signals regulating interneuron production, migration and integration into the cortex. Front. Cell. Neurosci. 9:129. doi: 10.3389/fncel.2015.00129

**Abbreviations:** CGE, caudal ganglionic eminence; cIN, cortical interneuron; CP, cortical plate; CR, calretinin; dMGE, dorsal medial ganglionic eminence; F-actine, fibrilar actine; G-actine, globular actine; IP, intermediate progenitor; IZ, intermediate zone; LGE, lateral ganglionic eminence; MGE, medial ganglionic eminence; MT, microtubule; MZ, marginal zone; NPY, neuropeptide Y; POA, preoptic areo; PTM, post-translational modification; PV, parvalbumin; SAP, subapical progenitor; SNP, short neuroal precursor; SP, subplate; SST, somatostatin; SVZ, subventricular zone; VIP, vasointestinal peptide; vMGE, ventral medial ganglionic eminence; VZ, ventricular zone.

in progenitor domain composition and in the ability to generate IN subtypes characterized by specific networks of transcription factors. In addition to genetic programs, diffusing molecules also participate in shaping the timing, space and specificity of cIN subtype production. Indeed, at the earlier stages of corticogenesis, molecules acting as mitogens and morphogens induce genetic programs eventually leading to the expansion of proliferative regions, specification and maturation of cINs. In rodents, cINs migrate long distances to reach their final destination. Again, they are under control of genetic programs and signaling cascades triggered by extracellular cues that work together to produce a synchronized, harmonious and directed movement toward the cortex. At the cellular level, these informations are integrated and translated by the cytoskeleton into appropriate cellular behavior. After settling at their final location, cINs integrate and organize in coherent networks. Here we will review the current understanding of how genetic programs intermingle with extracellular signaling pathways to achieve the production, migration and network integration of INs in the cortex.

The exact number of cIN subtypes remains debated, mainly due to the diversity in their morphological, molecular and functional properties (Petilla Interneuron Nomenclature et al., 2008). In this review, a simplified nomenclature combining molecular and physiological properties of cINs will be used. According to this nomenclature (Gelman and Marin, 2010), the large variety of cIN subtypes will fall in one of four major groups: (a) fast spiking, parvalbumin (PV)-expressing cINs; (b) burst spiking or adapting non-fast spiking somatostatin (SST)-expressing cINs; (c) non-fast spiking and fast adapting calretinin (CR)- and/or vasointestinal peptide (VIP)-expressing cINs; (d) rapidly adapting neuropeptide Y (NPY)- and/or reelin-expressing cINs. Most studies discussed here used rodents as experimental model. They have been useful in the understanding of the physiopathology underlying cortical interneuron development. We will finalize by giving example of how these findings fit with what starts being known about the production, migration and cortical integration of GABAergic neurons in primate and human.

# Generation and Specification of Cortical Interneurons

#### Role of Morphogens in Establishing Ventral Identity

MGE histogenesis starts at around embryonic day (E) 9, followed by the generation of LGE at E10 and CGE at E11 (Smart, 1976). GEs histogenesis requires a complex interplay between morphogens and transcription factors to ventralize the structure and promote IN production. Sonic hedgehog (SHH) and fibroblast growth factors (FGFs) contribute to the dorso-ventral patterning and subpallium development (Jessell, 2000; Briscoe and Ericson, 2001; Ingham and McMahon, 2001). Shh is widely expressed in the prospective MGE by E9 and by E12 its expression spreads to the mantle zone of the MGE and POA (Echelard et al., 1993). Cell responsiveness to SHH greatly depends on the action of the transcription factors of the GLI family that can, in their constitutively cleaved forms, act as transcriptional repressors in the absence of SHH signaling. Conversely if uncleaved, they function as transcriptional activators in the presence of SHH (Bai et al., 2004; Pan et al., 2006). The current understanding postulates that SHH signaling is required to counteract GLI3 repressor activity therefore contributing to the positioning of the dorsoventral boundary (**Figure 1**). At early stages of brain development, SHH prevents dorsalization of the ventral telencephalon, allowing the subsequent formation of the GEs. This is in contrast with the role of GLI proteins in the spinal cord where these factors act as activators by directly promoting ventral patterning (Rallu et al., 2002; Bai et al., 2004). Downstream effectors of SHH signaling are also required for ventral development. They include smoothened, the low-density lipoprotein receptor-related protein 2 (LRP2) or megalin and the multifunctional transmembrane protein Cdo (Fuccillo et al., 2004; Spoelgen et al., 2005; Zhang et al., 2006). In Shh knockout mouse models, rescue of the ventral telencephalon by compound Gli3 removal gives the indication that other genes might act independently or downstream of Shh (Rallu et al., 2002). Several studies provide evidence that FGFs act downstream of SHH and can directly induce ventral gene expression in dorsal telencephalic explants when SHH signaling is inhibited (Aoto et al., 2002; Ohkubo et al., 2002; Kuschel et al., 2003; Gutin et al., 2006; Rash and Grove, 2007). Since SHH promotes Fgf expression (Martynoga et al., 2005), FGFs are considered mandatory effectors of SHH signaling (**Figure 1**). The forkhead G1 factor (FOXG1) is the main generator of direct ventralization within the forming ventral telencephalon as it induces expression of Fgf8 (Martynoga et al., 2005) in the MGE or FGF15 in the CGE (Borello et al., 2008). Foxg1 acts in concert with FGF signaling, forming a positive feedback loop (Shimamura et al., 1995; Martynoga et al., 2005). Foxg1 expression is independent of the direct action of SHH but impaired in the absence of Shh due to the increased repressor activity of GLI3 (Rash and Grove, 2007).

## Transcription Factors acting in Concert with Morphogens in the MGE

The MGE contributes to the production of 50–60% of the total population of cINs in the mouse (Pleasure et al., 2000; Butt et al., 2005; Wonders and Anderson, 2006). On top of the hierarchical MGE SHH- and FGF-dependent organizers is Nkx2-1 transcription factor (Sussel et al., 1999; Gutin et al., 2006; Storm et al., 2006; Fogarty et al., 2007; Xu et al., 2008, 2010). Nkx2-1 itself maintains Shh expression within the early MGE, a process depending on FoxA2/HNF-3β transcription factor (Sussel et al., 1999) (**Figure 1**). Moreover, early removal of Nkx2.1 from MGE progenitors re-specifies INs into early LGE medium spiny neuron identity, while its late removal leads to acquisition of CGE IN profile (Butt et al., 2008). Nkx2-1 is no longer detected in mice that lack expression of both Fgfr1 and Fgfr2 (Gutin et al., 2006). The analysis of the expression of several transcription factors within the ventricular zone (VZ) of the MGE have led to the proposal that this region can be compartmentalized into five different progenitor domains (Flames et al., 2007). The dorsal region of the MGE (dMGE) preferentially gives rise to somatostatin (SST) expressing cINs. In contrast, the ventral part of the MGE (vMGE) was shown to mostly generate paravalbulin (PV)-expressing cINs.

Since Nkx2-1 expression promotes the specification of both SST and PV cIN subtypes, it was suggested that the SHH gradient determines the final fate of interneurons. High levels of SHH signaling would favor the generation of SST-expressing cINs at the expenses of PV-expressing cINs (Xu et al., 2010). Consistently, high expression levels of SHH effectors were found in the dMGE (Wonders et al., 2008). More recently, it was demonstrated that neurons from the mantle zone are an additional source of SHH (Flandin et al., 2011). They play a relevant role in maintaining a high SHH gradient in the dMGE, far away from the VZ. SHH production by mantle neurons was shown to require expression of Lhx6 (Flandin et al., 2011), a direct target of Nkx2-1, also implicated in PV or SST fate acquisition. In the absence of Lhx6, NPY fate was promoted at the expenses of PV or SST (Liodis et al., 2007; Zhao et al., 2008). It was also shown that disruption of the gene encoding FGF receptor 1 (Fgfr1) lead to a loss of Lhx6 and Lhx8(7) expression, both necessary for the formation of MGE-derived INs (Fragkouli et al., 2005; Gutin et al., 2006; Liodis et al., 2007) and maturation of PV-cINs(Smith et al., 2014). Other genes act downstream of or in concert with Nkx2-1. For example, high levels of Lhx8(7) expression shifts the fate of cINs toward globus pallidus GABAergic neurons and into cholinergic INs from the striatum (Zhao et al., 2003; Fragkouli et al., 2005). Sox6 expression is required for the generation of the appropriate number of PV and SST INs as demonstrated in studies using germline and conditional knockout mice (Azim et al., 2009; Batista-Brito et al., 2009). In these mice models, a concomitant increase of NPY cINs was observed (Azim et al., 2009; Batista-Brito et al., 2009). Sox6 acts downstream Lhx6 and is expressed continuously within the MGE from postmitotic progenitor stage until adulthood and is implicated in the placement and maturation of PV and ST cINs (Batista-Brito et al., 2009). Dlx genes Peyre et al. Cortical interneuron development

also contribute to cINs specification and maturation. Dlx genes expression is temporally regulated, following the sequence: Dlx2, Dlx1, Dlx5, and Dlx6 (Liu et al., 1997; Eisenstat et al., 1999). Dlx1/2 gene seems to be particularly important for the acquisition of SST, calretinin (CR), NPY and reelin fates (Cobos et al., 2005) as its absence leads to an abnormal expression of cortical markers in the ventral telencephalon (Long et al., 2009a,b). The expression of a wide range of transcription factors in the MGE progenitors from the VZ and subventricular zone (SVZ) as well as non-transcription factor proteins involved in migration and cortical integration are also under the control of Dlx1/2 (Long et al., 2009a). For example, Dlx1/2 has a repressor activity over Arx transcription factor, required for MGE differentiation. In their absence, cIN migration is blocked, resulting in accumulation of cells in the GEs and reduced numbers of INs in the cortex (Colombo et al., 2007). In addition, Dlx1/2 genes are also required for the delayed expression of Dlx5/6 genes (Anderson et al., 1997; Yun et al., 2002; Long et al., 2009b), particularly important for the establishment of PV subtype identity (Wang et al., 2010). Furthermore, Dlx1/2 tightly control the generation of oligodendrocytes in the forebrain by repressing Olig2 (Petryniak et al., 2007). More precisely, the choice between neuronal and glial fate involves cross-regulation between Mash1 or Acsl1 and Dlx genes. Mash1 binds to and represses the regulatory DNA elements in the intergenic region of Dlx1/2 (Parras et al., 2007). In Mash1 mutants, Dlx1/2 expression is expanded in the VZ/SVZ (Casarosa et al., 1999; Horton et al., 1999; Yun et al., 2002; Poitras et al., 2007). Conversely Mash1 expression is increased in the VZ/SVZ of Dlx1/2 mutants (Yun et al., 2002). Thus, the role of Mash1 consists in restricting the number of Dlx-expressing progenitors (Petryniak et al., 2007).

# Transcription Factors and Morphogens Shaping the Generation of cINs in the CGE

CGE contributes to the generation of 30–40% of all cortical interneurons. Several studies have demonstrated that CGE derived interneurons acquire either a CR and/or VIP (Pleasure et al., 2000; Butt et al., 2005) or reelin identity (Miyoshi et al., 2010). Gsh or Gsx homeobox TFs act at the top of the genetic network involved in CGE cell specification (**Figure 1**). Gsh2 is particularly relevant for the generation of CR bipolar cINs (Xu et al., 2010). Interestingly, Gsh1 and Gsh2 are co-expressed but have antagonist functions within the CGE, Gsh2 promoting progenitor state and Gsh1 promoting neuronal differentiation (Pei et al., 2011). Interestingly, the control of the choice between proliferation and differentiation by Gsh genes seems to involve the downstream target Mash1 (Fode et al., 2000). In Mash1 loss of function there is premature differentiation of progenitors located in the SVZ and precocious expression of Dlx genes (Casarosa et al., 1999; Yun et al., 2002), downstream effectors. On the other hand, overexpression of Mash1 contributes to cell type specification (Fode et al., 2000). Dlx1 and Dlx2 are co-expressed in subsets of progenitor cells and contribute to cell maturation by downregulating Gsh2/Mash1 (Yun et al., 2002). Other CGE transcription factors include Nrf2f1and Nrf2f2 or Couptf1 and Couptf2, respectively, as well as SP8. These genes are however not exclusive from CGE, as they have been identified in the dMGE and POA (Lodato et al., 2011a). Nrf2f1 is required for proper progenitor proliferation and necessary for generation of interneuron diversity in the cortex (Lodato et al., 2011a). Nrf2f2 is important for directing interneurons through a caudal migratory path (Cai et al., 2013). SP8 function in the hierarchy of CGE specification/maturation is yet unknown (Ma et al., 2012).

# The POA Produces a Reduced Number of Diverse cIN Subtypes

The POA is the most ventral region of the developing subpallium and it has been shown to generate around 10% of GABAergic INs. Using a Cre line driven by Nkx5-1 or Hmx3, a gene exclusive from POA, gives rise to a small population (around 4%) of multipolar GABAergic cells (Gelman et al., 2009). Another 5% of total INs is also produced by progenitors present in this region, characterized by their expression of the transcription factor Dbx1. Fate mapping and in utero transplantation demonstrated that POA generates diverse cINs subtypes (Gelman et al., 2011). In terms of molecular markers expression, the cells generated by the POA resemble the ones originating from the CGE (Gelman et al., 2009). Shh and Nkx2-1 but not Lhx6 are also expressed in the POA (Flames et al., 2007). Dbx1 and Nkx6-2 are respectively markers of the dorsal and ventral POA (see **Figure 1**). The function of these genes remains, however, elusive.

# Progenitors and Proliferation in the Ventral Subpallium

Ventral telencephalon expansion and generation of a great diversity of cINs relies on the proliferation of pools of progenitors. The molecular rules governing cell proliferation in the ventral telencephalon as well as the characterization of the distinct cIN progenitor behavior has just started to be unveiled. For some time it was anticipated that GE progenitors would display a proliferative behavior similar to progenitors in the cerebral cortex (Ross, 2011). This view relied on anatomical and cumulative bromodeoxyuridine (BrdU) experiments. These studies were important as they served identifying both VZ and SVZ as two distinct proliferative compartments (Sheth and Bhide, 1997). The lack of selective markers for the SVZ and the superposition of proliferating and migrating cINs hampered for some time detailed studies aiming at characterizing the cellular biology of cell division of ventral progenitors. Improvement of molecular tools and imaging techniques overcame these limitations. For example, an elegant study by Brown and colleagues used a clonal approach to understand how cINs were generated within these regions. Low concentration of retrovirus expressing GFP was injected in the ventricle of E11 embryos. In order to specifically infect INs progenitor cells, the virus entry receptor was expressed under the control of Nkx2-1 (Brown et al., 2011). They found that cINs are produced as spatially organized clonal units and clonally related INs form spatially isolated cluster in the neocortex. They identified the presence of radial glia (RG) in the VZ of MGE and POA that undergo interkinetic nuclear migration and divide asymmetrically in the VZ to self-renew and produce intermediate progenitors (IPs) or differentiating cINs. Using time-lapse microscopy, Pilz et al. (2013) proposed a more complex hierarchical classification for ventral progenitors. RG cells sit at the base of this classification and divide asymmetrically to generate both an amplifying and a self-renewal branch. These cells give rise to short neural precursors (SNPs). Both RG and SNPs generate subapical progenitors (SAPs) which in turn divide to produce basal radial glia (bRG) or basal progenitors (BPs). Basal radial glia and BPs contribute to the great SVZ expansion. Mash1 levels were shown to control the numbers of SAPs (Pilz et al., 2013). Although this study was performed in the LGE, such hierarchical complexity might be expected for the entirety of the ventral telencephalon.

# Molecular Regulation of Ventral Proliferation

Studies performed by Vidaki et al. (2012) showed that classical proteins displaying a role in dorsal proliferation, such as Rasrelated C3 botulinum toxin substrate 1 (Rac1), are also important regulators of Nkx2-1-expressing MGE progenitor division. In the absence of Rac1, cyclin D proteins levels are reduced and similarly low levels of Retinoblastoma (Rb) phosphorylation is detected. Cortical interneuron progenitors are thus blocked from completing cell cycle (halted in G1 phase) and accumulate in the GEs. Interestingly, the lack of cyclin D2 in SVZ progenitors lead to the production of lower number of PV but not SST cIN subtypes (Glickstein et al., 2007a,b, 2009), suggesting that cell division and cell fate acquisition are linked events (Glickstein et al., 2007b; Ross, 2011). Rb family proteins and the closely related protein p107, play a role in cell proliferation by regulating the activity of E2F transcription factors, notably E2F4 a transcription repressor (Trimarchi and Lees, 2002). Deficiency of E2F4 expression impairs the selfrenewal of neuronal precursor cells (Ruzhynsky et al., 2007) and results in loss of ventral telencephalic structures. The underlying mechanism involves a dramatic loss of Shh, Nkx2-1, and Dlx2 expression.

Acting extracellularly, morphogens such as SHH and FGFs can also potentially act as mitogens (Hebert and Fishell, 2008). SHH-mediated proliferation is regulated in space and time (Blaess et al., 2006). If progenitors are exposed to SHH during the peak of neurogenesis, it will enhance proliferation, whereas exposure during the post-neurogenic period maintains cells in the undifferentiated state (Rowitch et al., 1999). Downstream effectors of SHH regulating the cell cycle are N-myc (Kenney et al., 2003), cyclin D1 (Kenney and Rowitch, 2000), E2f1 and E2f2 (Oliver et al., 2003). The induction of N-myc occurs through GLI proteins and its stabilization depends on phosphatidylinositol-3-kinase (PI3-K) (Kenney et al., 2004; Sjostrom et al., 2005). FGFs control cell cycle length mainly during G1 phase. In a cell culture model, addition of FGF2 results in G1 phase shortening and an increase in the number of proliferative divisions by E14-E16 (Lukaszewicz et al., 2002). In vivo, Fgf8 controls ventral telencephalon size mainly in rostral regions (Storm et al., 2006). This effect mainly relied on the control over progenitor survival (Storm et al., 2006). Deletion of Fgf3 in addition to Fgf8 further decreased the telencephalic size, indicating that both genes act in synergy (Theil et al., 2008). Fgf15 controls progenitor differentiation at earlier developmental stages by promoting cell cycle shortening and exit, an effect opposite of what was observed at later stages (Borello et al., 2008). Neurotransmitter receptors are another class of diffusible molecules that control cell proliferation in the developing telencephalon (Cameron et al., 1998; Nguyen et al., 2001; Owens and Kriegstein, 2002a,b). cIN precursors and progenitors appear to be sources of gamma aminobutyric acid (GABA) (Bellion and Metin, 2005). Indeed, the extracellular GABA concentrations in the GEs may be as high as 0.5µM (Cuzon et al., 2006). Proliferating cIN precursors also display detectable levels of GABA synthetizing enzymes and functional GABA<sup>B</sup> receptors (Maric et al., 2001) as well as GABA and chloride transporters (Laurie et al., 1992; Ma and Barker, 1995). Glutamate is also present in the telencephalic germinal zones where it acts as mitogen. In the cortex, both GABA and glutamate where shown to decrease proliferation in the SVZ probably by reducing DNA synthesis as a consequence of membrane depolarization and Ca2<sup>+</sup> increase (LoTurco et al., 1995; Haydar et al., 2000), in opposition to what was observed on VZ progenitors (Haydar et al., 2000). Glutamate actions were found to be diverse and depend on the glutamate receptor subtype involved in the signaling. For example, DNA synthesis inhibition occurs when α-amino-3-hydroxy-5-méthylisoazol-4-propionate (AMPA)/ kainate (KA) receptors are activated (LoTurco et al., 1995). N-methyl-D-aspartate (NMDA) receptor-dependent signaling instead promotes proliferation of striatal neural progenitors (Sadikot et al., 1998; Luk et al., 2003). The ERK (Extracellular Signal-Regulated Kinase)-PI3K pathway is triggered downstream NMDA receptor activation to control proliferation of striatal progenitors (Luk et al., 2003). Variation in glutamate concentration (Haydar et al., 2000) and interaction between glutamate receptormediated and growth factors-mediates signaling pathways (Dobbertin et al., 2000) might further contribute for a differential responsiveness of distinct progenitors. The action of morphogens and/or mitogens can be disrupted by many environmental agents and by epigenetic modifications during the period of corticogenesis. It is thus of utmost interest to fully characterize the signaling cascades triggering ventral proliferation.

# Migration of Cortical Interneurons

During development, cINs migrate over long distances to reach the cortex and settle within cortical layers. Migrating cINs are highly polarized cells harboring a branched and dynamic leading process that terminates in a growth cone-like structure. They also possess a membrane protrusion at the rear of the cell called trailing process. While migrating, cINs display a stereotyped cyclic movement (**Figure 2A**). First, there is an extension of branches emanating from the leading process and as one of the branches stabilizes, a transient swelling forms close to the cell body where the centrosome and Golgi apparatus are displaced (Bellion et al., 2005). Then, cINs move forward by sudden and fast nuclear translocation into the swelling, an event called nucleokinesis. The jumping behavior of cINs characterizes its migration pattern and distinguishes it from the treadmill-like movement observed in a large range of cells. Finally, the trailing process is retracted and the cycle repeats. cINs can significantly change the direction of migration by inverting polarity, the trailing process extending and becoming the new leading process while the older leading process undergoes retraction (Nadarajah

et al., 2002) (**Figure 2A**). All these dynamic phases heavily rely on cytoskeleton remodeling.

# Molecular Regulation of the Cytoskeleton in Migrating cINs

#### cINs Migration and Nucleokinesis

In cINs, a microtubule (MT) "cage" surrounds the nucleus and a large array of MTs connects this structure to the centrosome (Tanaka et al., 2004; Higginbotham and Gleeson, 2007; Godin et al., 2012) (**Figure 2B**). MTs are nucleated ahead of the nucleus to guide centrosome movement and docking to the cell membrane. At the membrane, the mother centriole is then able to grow a cilium allowing the cell to sense extracellular signals such as SHH (Baudoin et al., 2012). It was previously considered that the MT network was responsible for generating forces required for nucleokinesis, through dynein-dynactin directed motor movement. This type of MT-generated forces has been described for migration of projection neurons and cerebellar granule cells where MT-associated proteins Lissencephaly-1 (Lis-1) and Doublecortine (DCX) couple MT to the nucleus (Tanaka et al., 2004; Nasrallah et al., 2006; Tsai et al., 2007). Similarly, cINs depleted for Lis-1 show a reduced rate of migration (McManus et al., 2004). It was however demonstrated in vitro that the chemical destabilization of the MT network does not completely abolish nucleokinesis (Schaar and McConnell, 2005; Baudoin et al., 2008; Martini and Valdeolmillos, 2010). Instead, acto-myosin contractibility is necessary for nuclear movement, as shown by experiments in which non-muscle myosin-II-mediated contraction was blocked (Bellion et al., 2005; Schaar and McConnell, 2005; Baudoin et al., 2008; Martini and Valdeolmillos, 2010). Myosin-II is enriched behind the nucleus during nuclear translocation, where F-actin also shows a strong accumulation (Bellion et al., 2005; Schaar and McConnell, 2005; Martini and Valdeolmillos, 2010) (**Figure 2B**). Acto-myosin cytoskeleton is also necessary to promote centrosome separation from the nucleus upon swelling formation and for the dynamic remodeling of growth cones (Metin et al., 2006). Altogether, this strongly suggests that the forward movement occurring during nucleokinesis arises from pushing forces generated by acto-myosin contraction at the rear of the nucleus, together with limited MT- generated pulling forces (He et al., 2010; Steinecke et al., 2014a). Although well studied in the context of radial migration of projection neurons, the molecular cascade regulating acto-myosin cytoskeleton during tangential migration is not as well described. In many cell types the Rho family GTPases including Rho, Rac, and cell division control protein 42 homolog (CDC42) have been implicated in the regulation of acto-myosin contractility (reviewed in Heasman and Ridley, 2008; Govek et al., 2011). In cINs, modification of RhoA activation levels by loss of function of its inhibitor p27Kip1 leads to migration defects (Besson et al., 2004) due to myosin-II hyperactivation (Godin et al., 2012). Active RhoA, in its GTP bound state, regulates migration by activating the downstream kinase Rho-associated protein kinase (ROCK) and mDia but also by inhibiting cofilin, an actin severing enzyme (Kawauchi et al., 2006; Nguyen et al., 2006; Godin et al., 2012). ROCK promotes acto-myosin contraction using different mechanisms. It can directly phosphorylate Myosin Light Chain (MLC), inhibit Myosin Light Chain Phosphatase (MLCP) or activate Myosin Light Chain Kinase (MLCK) (Amano et al., 1996; Chrzanowska-Wodnicka and Burridge, 1996; Kimura et al., 1996; Ishizaki et al., 1997). MLCK activation leads to acto-myosin contraction by phosphorylating MLC. MLCK activity is modulated by its cofactor calmodulin, a calcium-activated protein (Gallagher et al., 1997). mDia is an actin nucleator that helps producing long filaments of actin fibers (Higashida et al., 2004) (**Figure 2B**). Interestingly, mutant mice for mDia1 show impaired tangential migration, but no defects in radial cortical dispersion. In this animal model the anterograde actin flow that moves the centrosome forward as well as F-actin accumulation at the rear of the nucleus are impaired. In contrast, actin dynamics in the growth cone is normal (Shinohara et al., 2012). This argues for a differential regulation of cytoskeleton-generated forces in different subcellular compartments where ROCK and mDia cooperate to grant proper acto-myosin dynamics during cIN migration. F-actin turnover is a crucial parameter in acto-myosin dynamics as inhibition of F-actin severing was shown to stabilize the leading process and can lead to cell migration arrest (Chai et al., 2009a,b). In INs, partial inactivation of cofilin in mice lacking p27 does not result in strong accumulation of F-actin, suggesting that F-actin severing can be mediated by redundant mechanisms, for example by the action of gelsolin (Godin et al., 2012) (**Figure 2B**).

#### cINs Leading Process, Branching, and Growth Cone Regulation

cINs navigate in their environment using a branched leadingprocess bearing dynamic growth cones. When a chemo-repulsive cue is sensed, the growth cone collapses and the branch is consequently retracted. In parallel, another branch is stabilized and it determines the new direction of migration (Martini et al., 2009). Leading process branching requires the coordination of both acto-myosin and MT networks. During this process a new membrane protrusion is formed thanks to the underlying ramified F-actin meshwork organized by cortactin and Actin-related proteins 2/3 (Arp2/3) (Spillane et al., 2011; Lysko et al., 2014). Then, invasion by unbundled and freely spreading MTs stabilizes the protrusion and allow the emergence of the branch. Recent evidence shows that fine-tuning of cIN branching is negatively regulated by CXCL12 or SDF-1 (Lysko et al., 2014). Binding of CXCL12 to its receptor CXCR4, results in decreased levels of cAMP and de-repression of calpain and DCX expression. This intracellular signaling has a double effect on the cytoskeleton: generation of straight F-actin fibers by calpain-mediated proteolysis of cortactin and MT bundling by DCX. Membrane protrusions are thus less likely to form in the absence of a branched actin meshwork and be stabilized by bundled MTs. Accordingly, hyperbranching is observed in cINs lacking DCX (Kappeler et al., 2006; Friocourt et al., 2007; Lysko et al., 2014) (**Figure 2B**). Proper regulation of MT dynamics is also essential for cIN tangential migration as neurite growth is depending on the establishment of new MT-networks. For example, in p27kip1 mutant, neurite growth defects are observed and cannot be fully rescued by modulating the acto-myosin contraction mediated by the RhoA pathway (Godin et al., 2012). Interestingly it was shown that p27kip1 acts also as a MT-associated protein (MAP) thanks to its prolinerich domain that promotes MT polymerization both in vivo and in vitro. This indicates that MT polymerization is an essential parameter in modulating neurite growth and migration properties of cINs (Godin et al., 2012). Similarly, the establishment of a long leading process also requires MT stabilization. In Rac1/Rac3 mutant mice, migrating cINs show a hyper-branched phenotype similar to the one observed in Dcx knockout mice as well as a reduced leading process length. This is also accompanied by MTs harboring less post-translational modifications (PTMs), frequent on stable and long-lived MTs. The phenotype was partially rescued by treating the cells with taxol, a MT stabilizing agent, indicating that establishment of a correct migration needs a minimal amount of stable MTs (Tivodar et al., 2014) (**Figure 2B**). Growth cone shape is modulated by both MT polymerization that generates pushing forces on the plasma membrane and by actomyosin network underlying pulling forces on the leading edge.

Actin undergoes polymerization and depolymerisation activity and acto-myosin contraction leads to the generation of an actin retrograde flow. The balance between these two networks will either allow growth or retraction of the leading process (Martini et al., 2009). Behind the leading edge, interactions between cell surface and migration substrate together with actin retrograde flow generate traction forces. Importantly, sectioning the leading tip or locally inhibiting acto-myosin contraction halts nucleokinesis, highlighting the role of this region in force generation (He et al., 2010). Actin cytoskeleton is regulated at the leading edge by Disc1 and cINs deficient for this protein accumulate less F-actin together with less phosphorylated Girders of Actin filaments (girdin) and Protein kinase B (Akt), crosslinkers of actin filaments (Steinecke et al., 2014a). Girdin targeting to to the leading tip requires interaction with Disc1 (Steinecke et al., 2014a). Mice mutant for Disc1 or Lis1, display decreased level of acetylation, a marker of stable MTs in the growth cone. This suggests that stable bundles of MTs are not properly entering this structure and are important to grant the proper shape to the growth cone (Gopal et al., 2010; Steinecke et al., 2014a) (**Figure 2B**).

#### cINs Migration Substrate

Adhesion to a migration substrate allows cINs to generate actomyosin forces as well as to organize cellular polarization and directionality. cINs seem to use cell-cell interaction to migrate along bundles of fibers of the corticofugal system invading the GEs (Metin and Godement, 1996). TAG-1 or contactin-2, a member of the immunoglobulin superfamily expressed by the corticofugal axons was proposed to mediate cell-cell contact between migrating cINs and corticofugal fibers as inhibition of TAG-1 leads to a strong reduction of cIN migration in vitro (Denaxa et al., 2001). However, in vivo TAG-1 does not play a role during migration cINs (Denaxa et al., 2005). In cINs, proteins displaying a role in adhesion such as talin, paxilin or Focal Adhesion Kinases (FAKs) are also calpain substrates (Franco and Huttenlocher, 2005). Calpain inhibition reduces migration speed despite the increased levels of F-actin (Lysko et al., 2014), implying that local adhesion turnover by cleavage is primordial for correct migration. Finally, N-cadherin, a homophilic cell adhesion molecule has been shown to be necessary during tangential migration (Luccardini et al., 2013). N-cadherin not only promotes cINs motility in vitro and in vivo, likely by promoting adhesion through the acto-myosin cytoskeleton (Giannone et al., 2009) but also contributes to the polarity maintenance. N-cadherin inhibition unable the centrosome/Golgi apparatus to enter the swelling and contributes to local defects of Myosin II contraction at the nuclear rear (Luccardini et al., 2013). Cell adhesion molecules play a relevantrole for cINs migration as they are at the interface between the extracellular environment and the cytoskeleton.

## Migration of cINs: Extracellular Cues Guide cINs Movement

To allow proper directionality, integration of extracellular signaling is paramount. The extracellular cues are received by exploring the local environment thanks to stochastic branching of the leading process, and relevant cues are transduced at the level of cytoskeleton to grant the proper response of the cell: either extension of a new leading process is the right direction and/or retraction of existing branches (Britto et al., 2009). The first stage in which cINs are challenged by extracellular cues occurs within the GEs. During the differentiation process, newborn cINs need to be displaced from the VZ and SVZ and be driven toward the exit of the GEs. Two types of signaling drive the movement of cINs away from GEs. One has a chemo-repulsive effect to steer the cells in the good direction and the other stimulates cellular motility to enhance migration.

#### Chemo-Repulsion within the GEs

Regarding the mechanisms mediating repulsion of INs from GEs, similarities were found with general mechanisms implicated in axonal guidance. Using a paradigm of in vitro brain slice preparation, Ephrin-A5/EphA4R interaction was shown to be necessary to control the repulsion response of cINs. The guidance molecule Ephrin-A5 is abundant in the VZ and cINs express the Ephrin receptor EphA4 (EphA4R) (Zimmer et al., 2008). In absence of Ephrin-A5, cINs were found ectopically invading the VZ, a phenotype rescued when the slices were treated with recombinant Ephrin-A5 (Zimmer et al., 2008). EphA4R-mediated forward signaling is also used by cINs to avoid migration toward the ventral region of the subpallium (Zimmer et al., 2011) as it also binds Ephrin-B3 present in the MGE and POA. It is noteworthy that EphA4R also promotes cIN migration trough EphrinA2 reverse signaling (Steinecke et al., 2014b). EphA signaling is integrated at the cellular level to remodel the actin cytoskeleton and steer cells in the right direction. Although little is known about the intracellular cascades downstream EphA4 in cINs, it generally acts in neurons through regulation of RhoA to stimulate growth cone collapse (Wahl et al., 2000). Moreover, in cINs, Src family kinases (SFKs) have been implicated in this process. Indeed, SFK inhibition results in the loss of Ephrin-A5/EphA4 repulsion (Zimmer et al., 2008). In mouse, four redundant SFKs have been identified (Thomas and Brugge, 1997) and they have been linked to the phosphorylation of numerous targets. They control the Rho family GTPase (Kullander and Klein, 2002) and inactivate cortactin (Huang et al., 1997; Weaver et al., 2001) resulting in decreased activity of Arp 2/3, actin filament breakdown and growth cone collapse (Weed and Parsons, 2001). Slit/Robo is another signaling pathway mediating chemo-repulsion in GEs and cINs express Roundabout homolog 1 (Robo1) (Bagri et al., 2002; Marillat et al., 2002), a receptor recognizing the ligands Slit homolog 1 and 2 (Slit1 and 2). These ligands and Robo1 are found in a complementary expression pattern in the VZ (Marin et al., 2002), and it was first thought that Robo/Slit signaling was at play to push cINs away from the VZ. However, mice deficient for Slit1/2 do not show cIN migration defects (Marin et al., 2003). Instead, the lack of Slit ligands or removal of Robo1 leads to aberrant striatal invasion by cINs (Andrews et al., 2006; Hernandez-Miranda et al., 2011). This indicates that on the way toward the GEs exit, cINs face a second set of signaling molecules that refine their migratory routes through the LGE, and around the forming striatum. It was previously shown that the striatum is a strong repulsive structure for cINs thanks to the expression of class 3 semaphorins: Sema3A and Sema3F (Marin et al., 2001). The effect of semaphorins is mediated in cINs by neuropilin (Nrps) and Plexin receptors. Sema3A transduces signal specifically trough Nrp1 and PlexinA1 receptors (McKinsey et al., 2013) and Robo1 modulate semaphorin-neuropilin/plexin expression levels (Hernandez-Miranda et al., 2011). It is noteworthy that Nrp1 is present exclusively in cINs so that they respond to the repulsive signals secreted by the striatum. Nrp1 expression depends on the transcription factor Sip1 involved in Nkx2-1 down-regulation and Nkx2-1 is a Nrp1 repressor (McKinsey et al., 2013). On the contrary, striatal INs generated by the same progenitors as cINs, do not express Nrp1 as they maintain highs levels of Nkx2.1 (Marin et al., 2000; Butt et al., 2008; Nobrega-Pereira et al., 2008). Since the signaling pathway downstream of Sema3A is highly conserved, although not directly studied in cINs, it is anticipated that similar mechanisms are at play in these cells. Sema3A signaling regulates actin-dependent growth cone collapse trough RhoA, ROCK and LIM Kinase (LIMK) activation to eventually phosphorylate cofilin and inhibits Factin turnover (Aizawa et al., 2001; Causeret et al., 2004). RhoA activation also leads to increased actin contractibility (Zhang et al., 2003). In the axon growth cone, a crosstalk between actin and MT cytoskeletons has been observed as retrograde F-actin flow on filopodia can displace distal position of MTs (Schaefer et al., 2002). Sema3A signaling could also directly regulate MT dynamics by inducing double phosphorylation of Collapsin Response Mediator Protein 2 (CRMP2) by Glycogen Synthase Kinase 3B (GSK3B) and Cyclin-dependent Kinase 5 (CdK5). The double phosphorylated state of CRMP2 reduces its tubulin affinity and overexpression of the phospho- mutant CRMP2 decreases Sema3A-induced growth cone collapse (Uchida et al., 2005).

#### Migration of cINs: Motogenic Factors

Newly produced cINs are stimulated by diffusible molecules to enhance movement and migration. A wide range of factors, including neurotrophins and neurotransmitters (NTs), were found to increase cINs motility in vitro (Heng et al., 2007). Demonstration came from experiments in which recombinant Brain Derived Neurotrophic Factor (BDNF) or Neurotrophin-4 (NT4) where applied to organotypic slice cultures (Polleux et al., 2002). BDNF- or NT4 effects are mediated by the Tyrosin Kinase B Receptor (TrkBR) (Polleux et al., 2002). Downstream signaling likely involves PI3K and the modulation of actin cytoskeleton. However this effect was observed in vitro but not in mouse mutant for TrkB where number and position of cINs are unchanged (Jones et al., 1994; Polleux et al., 2002; Sanchez-Huertas and Rico, 2011). In vivo, neurotrophin-mediated signaling promoting cINs motility was also linked to the action of Glial Cell-Derived Neurotrophic Factor (GDNF), binding and activating GDNF Family Receptor α1 (GFRα1). GDNF-mediated effects did not result from expression of Rearranged During Transfection (RET) or Neural Cell Adhesion Molecule (NCAM), two classical co-receptor signaling molecules (Pozas and Ibanez, 2005; Canty et al., 2009). The downstream signaling involves a matrix-bound form of GDNF and syndecan-3. This interaction then activates SFK to promote neurite outgrowth. GDNF may thus promote cell migration by acting on the actin cytoskeleton via SFK and cortactin pathway (Yoneda and Couchman, 2003; Bespalov et al., 2011). Since GFRα1 mutant mice shows perturbations in cINs regionalization and subtype differentiation, GDNF might have additional functions beyond modulating cINs motility (Pozas and Ibanez, 2005; Canty et al., 2009). The hepatocyte growth factor/scatter factor (HGF/SF) and its receptor MET were also reported to have a potent motogenic action on cINs in vitro. Mutant mouse for the urokinase-type plasminogen activator receptor (uPAR) that cleaves and releases the active form of HGF/SF (Powell et al., 2001) or mutant mice for MET (Eagleson et al., 2011) shows a decreased number of cINs in the cortex. The cell-autonomous effect of HGF/SF-mediated signaling was later questioned since MET is not found to be expressed in cINs in vivo but is rather found in projection neurons and their axonal fibers (Eagleson et al., 2011). Several NTs/neuromodulators have also been implicated in cINs migration. Ambient GABA is found in high concentration in the MGE and in the cortical migration streams (Cuzon et al., 2006). cINs express GABA<sup>A</sup> and GABA<sup>B</sup> receptors and as a result of an inverted chloride gradient, they respond to GABA by membrane depolarization (Owens et al., 1999) that triggers opening of L-type voltage-sensitive Ca2<sup>+</sup> channels and induces Ca2<sup>+</sup> transients (Bortone and Polleux, 2009). Antagonizing GABA<sup>A</sup> receptor function prevents cINs from crossing the cortico-striatal barrier, leading to their accumulation at the pallial/subpallial boundary. Conversely, in experiments where exogenous GABA or diazepam are added to brain organotypic slices, a higher number of cINs were found exiting the GEs (Cuzon et al., 2006). Similarly, Glycine α2 receptor is also regulating cIN migration and in particular nucleokinesis by fine tuning acto-myosin II contraction (Avila et al., 2013). Activation of GlyRs by glycine leads to Ca2<sup>+</sup> transients due to opening of voltage-gated Ca2<sup>+</sup> channels and GlyRs α2 loss of function impairs cINs migration in the cortex. Glutamate signaling through AMPARs, located on the plasma membrane of migrating cINs also induces membrane depolarization and sodium influx (Metin et al., 2000; Manent et al., 2006). Blockade of AMPARs decreases cortical and hippocampal invasion by INs (Manent et al., 2006; Bortone and Polleux, 2009). Membrane depolarization and Ca2<sup>+</sup> transients are thought to stimulate cINs motility as calmodulin bound to Ca2<sup>+</sup> ions can interact and activate MLCK. In turn, MLCK phosphorylates Myosin Light chain II on serine 19 and promotes acto-myosin contraction (Metin et al., 2000; Bortone and Polleux, 2009). Another example is the motogenic effect mediated by dopamine. Ambient dopamine is secreted in the GEs and close to the lateral VZ by projecting thalamo-striatal axons in the neo-striatum. Dopamine receptors D<sup>1</sup> and D<sup>2</sup> are expressed by cINs and have opposite effects on migration (Ohtani et al., 2003). When selectively blocked, D1R induces activation of D2R and impedes migration. This indicates that D1R has migration promoting action and conversely D2R is rather a migration stop signal. Other ubiquitous signaling molecules such as adenosine have also been implicated in the modulation of cINs migration. However the underlying mechanisms are still unknown (Silva et al., 2013).

# Migration of cINs: Chemo-Attraction Toward the Cortex

As cortical INs are steered away from the VZ/SVZ and around the striatum, they are simultaneously attracted toward the cortex by other molecules to cross the cortico-striatal junction and invade the pallium. Neuregulin-1 (Nrg1) is a strong attractant for cINs and can have both short and long-range effects depending on the isoform recruited. CRD-Nrg1, is membrane bound and acts on neighboring cells. It is mainly expressed by LGE cells and contributes to the formation of a permissive migration corridor for cINs. It is sensed by cINs as they probe their environment and inhibition of their branching behavior leads to a decreased ability to follow Nrg-1 gradient (Martini et al., 2009). The downstream signaling involves recruitment and activation of associated tyrosine kinases related to the epidermal growth factor (EGF) receptor ErbB4. Lg-Nrg1 is a diffusible form expressed by the cortex that can act as a long-range attractant to guide cells toward the cortical regions as they exit the LGE. In vivo, the loss of either Nrg-1 or ErbB4 causes migration defects and reduces the number of INs in the cortex (Yau et al., 2003; Flames et al., 2004; Neddens and Buonanno, 2010). At the cellular level, the presence of a Nrg-1 gradient does not prompt cINs to reorient an existing branch but rather to sprout a new leading process in the direction and at an angle corresponding to highest concentration of the attractant. The signaling events downstream ErbB4 have not been completely characterized in cINs, but they seem to be ROCK-independent (Martini et al., 2009). Thanks to its tyrosine kinase activity, ErbB4 can activate several distinct signaling pathways including the ras/raf/MAPK or the PI3 kinase pathway (Scaltriti and Baselga, 2006). In ErbB4 can be cleaved in its intracellular domain and be targeted to the nucleus (Ni et al., 2001). Interestingly, ErbBs can also phosphorylate β-catenin and modulate Cadherin signaling pathways involved in actin cytoskeleton remodeling (Hazan and Norton, 1998; Behrens, 1999; Al Moustafa et al., 2002). Moreover, other signaling molecules participate to the attraction of cINs toward the cortex. For example, CXCL12 is secreted by proliferating cortical cells from SVZ and IZ and is an attractant molecule for migrating cINs (Tiveron et al., 2006; Stumm et al., 2007).

# Migration of cINs: The Choice of the Migratory Route

Cortical invasion does not occur in a stochastic manner as cINs integrate and move in migratory streams. At earlier stages of cortical development, two streams can be identified in the pallium: one called intermediate zone (IZ) stream and located above the VZ/SVZ surface (Nadarajah et al., 2002) and the other positioned at the level of the marginal zone (MZ) and called MZ stream (Lavdas et al., 1999). Between E15 and E16, a third subplate (SP) stream forms between the IZ and MZ streams. Other routes have also been identified, for example a caudally directed stream arising from the CGE (Yozu et al., 2005). Although not depending on their birth place (MGE, CGE or POA), the choice of the migratory route by cINs does not seem to be random (Miyoshi and Fishell, 2011). Evidence comes from the observation that populations of cINs migrating along the different cortical streams do not show the same gene expression profile as revealed by microarray analysis on micro-dissected Gad67+ positive cells isolated from either IZ or MZ streams (Antypa et al., 2011). Some population-specific genes, such as Cdh8, Plxnd1, Sema5a, Robo 1 and 2 or the reelin receptor Dab1, play key roles in neuronal migration.

Moreover in vivo experimental approaches have also shed light on intrinsic cellular component that are implicated in cINs sorting between the different migration routes. For example, it was found that mutations in the Rb gene unable cINs to enter the MZ stream and redirects them to the IZ stream (Ferguson et al., 2005). This effect is cell autonomous as suggested by cINs transplantation experiments made on wild-type organotypic brain slices. A similar phenotype was found to be linked to glycine receptor α2 subunit (Glyα2R), present on the plasma membrane of migrating cINs. Activation of Glyα2R opens voltage-gated Ca2<sup>+</sup> channel (Avila et al., 2013). Extracellular cues seem to indeed play a relevant role in sorting cINs trough the different migration stream. Mutation of both Netrin1 and α3β1 integrin, in the cells serving as migration substrate for cINs, gives rise to major migration defects in the MZ stream (Stanco et al., 2009). Finally GABA signaling, in addition to its migration promoting role, seems to be also implicated in the choice of the migratory route during cortical migration. Upon selective blockade of metabotropic GABABR, cINs where found to be displaced from the MZ and CP stream toward the VZ/SVZ compartments (Lopez-Bendito et al., 2003).

# Migration of cINs: Molecular Regulation Controlling the Timing of Cortical Invasion

In the streams, tangentially migrating cINs do not invade the cortical plate (CP) and this process is coordinated with projection neurons migrating radially to form the cortical layers. Avoidance of the CP do not require repulsive cues coming from the projection neurons but rather the formation of a permissive corridor in the MZ and IZ thanks to chemokine signaling. CXCL12 is secreted by proliferating cortical cells from SVZ and IZ and it is a strong attractant molecule for migrating cINs (Daniel et al., 2005; Tiveron et al., 2006; Lopez-Bendito et al., 2008). Two receptors for this chemokine were identified in interneurons, CXCR4 and CXCR7 (Tiveron et al., 2006; Lopez-Bendito et al., 2008; Wang et al., 2011) and they signal through different downstream cascades. CXCR4 signals through Gα(i/o) while CXCR7 transduces independently on heterodymeric G proteins (Wang et al., 2011). In immature MGE neurons, CXCR7 acts as potent activator of MAP kinase signaling required for ERK1/2 phosphorylation (Wang et al., 2011). Cxcr7 expression in the cortical plate expression follows a dorsoventral gradient, opposite to Cxcl12 gradient in the SVZ. It was proposed that CXCR7 may lower the concentration of CXCL12 in the CP, generating a gradient from MZ and SVZ to the CP. A gradient of CXCL12 would be important for the regulation of cortical invasion (Wang et al., 2011). Indeed, disruption of CXCR4 or CXCR7 function results in premature exit of cINs from their migratory streams and perturbs their laminar and regional distribution within the neocortex (Li et al., 2008; Lopez-Bendito et al., 2008; Tanaka et al., 2010).

# Integration OF cINs into Cortical Assemblies

Cortical neurogenesis generates a six-layered cortex and cortical neurons distribute within these layers in an age-dependent manner, deep layers being generated before upper layers. Neurons sharing the same layers exhibit similar patterns of connectivity (Dantzker and Callaway, 2000). The cortex is also organized vertically and cells sharing the same cortical column will be linked by extrinsic connectivity (Mountcastle, 1997). Interneurons have to integrate into cortical circuits in order to fit in these two patterns of organization. Two studies have shown that clonally related cINs are preferentially consigned to specific cortical layers or columns (Brown et al., 2011; Ciceri et al., 2013). cINs generated by clonally-related progenitors are however diverse as they express markers of distinct subtypes. Furthermore, cINs from the same cardinal class are able to form connections with a variety of synaptic partners (Kepecs and Fishell, 2014). From their birthdates, the first interneurons populating the cortex are generated in the MGE and express PV, SST or CR/SST (Butt et al., 2005; Wonders and Anderson, 2006). The second wave of cortical invasion generated by the CGE produces INs expressing VIP, CR/VIP, calbindin (CB) or choline acetyltransferase (ChAT) and NPY, (Yozu et al., 2004; Miyoshi et al., 2010). While pioneer cINs populate the cortex in an inside-out mode following the pattern of cortical integration of projection neurons (Miller, 1985; Fairen et al., 1986; Valcanis and Tan, 2003; Rymar and Sadikot, 2007), late-migrating cINs instead concentrate in supragranular layers, independently of their birthdate (Xu et al., 2004; Rymar and Sadikot, 2007; Miyoshi et al., 2010). Since the birthdate and birthplace are not general predictors of the final specification and lamination of cINs, the mechanisms determining the proper and site-specific integration of INs in the developing cortex need to be clarified. This raises the question whether intrinsic and/or extrinsic factors determine the cortical integration and lamination.

# Intrinsic Factors Regulating cINs Integration into Cortical Layers

There is evidence that a maturational program exists and the behavior of cINs, at a given developmental stage, depends on their cellular age. In the cortical streams, cINs generated at different developmental stages co-exist and they exit the migratory streams at different time points even if the signaling regulating the exit from the streams is the same. Thus, cINs born earlier invade the cortical place before late-born interneurons (Lopez-Bendito et al., 2008). Further evidence supports this concept of intrinsic regulation. For example, it was found that the motility of interneurons in cortical slices gradually decreases as development proceeds and is almost abolished by the end of the first postnatal week (Inamura et al., 2012). Accordingly, lateborn cINs transplanted in younger embryos settle in deep layers instead of occupying the expected superficial layers (Pla et al., 2006). The mechanisms explaining intrinsic regulation of neuronal migration remain elusive. It was suggested that the frequency of Ca2<sup>+</sup> transients is reduced as the neuron completes its migratory program (Kumada and Komuro, 2004). Other studies proposed that the intrinsic regulation of motility might be linked to the expression of the potassium-chloride transporter KCC2. KCC2 could modulate cINs motility by reverting the chloride potential and thus reducing membrane depolarization upon GABA<sup>A</sup> receptor activation to serve as a stop signal for migration (Bortone and Polleux, 2009; Inamura et al., 2012). This is in agreement with the observation that cINs up-regulate KCC2 chloride transporter as soon as they exit the tangential mode of migration and start their radial sorting in the cortex (Miyoshi and Fishell, 2011). Cell autonomous regulation also contributes to the survival of interneurons as they invade the cortical plate. Transplantation experiments revealed that many cINs undergo programmed cell death in vivo between postnatal day (P) 7 and P11. When transplanted in older cortices, younger cINs die by apoptosis later than resident cINs (Southwell et al., 2012).

# Extrinsic Factors Regulating cINs Integration into Cortical Layers

Regional distribution of cINs seems also depending on extrinsic signaling. Elegant studies suggested, in the last past years, that molecular cues released by projection neurons contribute to the establishment of the laminar distribution of cINs (Hevner et al., 2004; Pla et al., 2006; Yabut et al., 2007; Lodato et al., 2011b). The works from Hevner (Hevner et al., 2004), Pla (Pla et al., 2006), and Yabut (Yabut et al., 2007) using the reeler mouse model showed that cINs distribution within the cortex mostly results from the aberrant organization of cortical layers rather than the loss of reelin signaling transduction in cINs. The subsequent work from Lodato et al. (2011b) further support this hypothesis. Fezf2 mutant mice lack sub-cerebral projection neurons, while all other projection neurons are normally produced. They show that Fezf2 depletion does not impair cINs specification but rather the lack of subcerebral projection neurons nonautonomously impairs the proper distribution of SST, PV but not CR cINs subtype. Delayed overexpression of Fezf2 in Fezf2 null mice leads to the production of ectopically located clusters of subcerebral projection neurons. Many cINs invade these aggregates while the number of INs recruited depends on the size of the ectopic aggregates. Local excitatory and inhibitory signals may also influence the final positioning of INs (De Marco Garcia et al., 2011; Lyons et al., 2012; McKinsey et al., 2013). For example, it was shown that attenuating the activity of specific cIN populations affects the migration and morphologic development of cIN (De Marco Garcia et al., 2011). A number of activity-dependent genes specifically expressed by cINs have been identified. These include Dlx1, Elmo1, and Mef2c. Moreover the observation that voltage-gated Ca2<sup>+</sup> influx may induce de novo gene expression opens the possibility that local activity might regulate direct region-specific differentiation and maturation of INs (De Marco Garcia et al., 2011; West and Greenberg, 2011). Additional evidence came from studies showing that MGE-derived interneurons are able to integrate in pathological neural circuits (Martinez-Cerdeno et al., 2010; Braz et al., 2012).

#### Peyre et al. Cortical interneuron development

# Conclusion

The recognition that many neurologic disorders such as schizophrenia, epilepsy and autism have components related to cIN development greatly prompted advances in this field. Postmortem analysis of the human brain and studies performed in primates strongly support the idea that human and primate cINs are produced in both dorsal and ventral regions (Fertuzinhos et al., 2009; Hansen et al., 2013; Ma et al., 2013). These studies have been instrumental to highlight the differences and similarities between in cINs generation across species, acknowledging more similarities than initially expected. These findings have given more credit to studies performed in rodents, designed to understand the genetic and molecular regulation underlying human pathology. For example, in the case of schizophrenia, NRG1 and ERB4 as well as DISC1 have been identified as susceptibility genes in human (Millar et al., 2000a,b; Corvin et al., 2004; Marin, 2012). Studies performed in rodents established a coherent outline of the biological causes of this human disease (Flames et al., 2004; Fazzari et al., 2010). How developmental perturbations of cINs lead to a specific brain disorder remains unclear although it has been proposed to depend, to some extent, on the cIN subtype affected (Marin, 2012). Thus, understanding the role of different classes of cINs and the neuro-circuitry they modulate will be relevant for unraveling the genetic causes of human diseases and to propose effective therapeutic approaches.

In the last decades, important advancements were made in the understanding of how INs are generated and function into networks. However, the fine regulation of cIN development might not be explained merely by genetic programs and extracellular signaling. Indeed, a fully new dimension of regulation, including post-transcriptional and post-translational modifications (PTMs) might be at play during corticogenesis. Post-transcriptional and PTMs can oppose or reinforce genetically-encoded and/or extrinsically-mediated signaling. A recent study explored the role of miRNA in distinct aspects of cINs development (Tuncdemir et al., 2014). In this study, Dicer, an enzyme required for miRNA processing and maturation was genetically deleted from MGE-derived cINs. Interestingly, the loss of miRNAs had no effect on cell proliferation and initiation of tangential migration but affected the transition from tangential to radial migration as well as modified the survival and maturation of cINs. Upon Dicer knockdown there was a precocious expression of cINs markers such as SST, GAD65 and NPY and at a later stage. MGE-derived cells failed to express markers of their subtype identity. PV-expressing cINs seem particularly affected by the absence of miRNAs. Furthermore, a different

# References


profile of gene expression related to differentiation, cellular interaction and survival was found in Dicer knockout mice. The role of miRNA signaling was also recently tested in a mouse model of schizophrenia and bipolar disorder (Toritsuka et al., 2013). This study showed a direct link between the 22q11 micro deletion and defects in cortical and hippocampal interneuron migration, relying on functional abnormalities in CXCR4/CXCL12 signaling. Mechanistically, Toritsuka et al. (2013) demonstrate the pivotal role of DiGeorge syndrome critical region gene 8 (Dgcr8) mediated in miR-200a regulation necessary for the maintenance of CXCR4 levels. PTMs occur in most proteins and often contribute to their functions and subcellular behaviors. Not many studies were developed to investigate the contribution of PTMs on brain development but there is evidence that they can regulate different aspects of cIN development. For example, it was demonstrated that acylation of SHH N-terminus changes its efficacy as a signaling molecule and greatly enhances its ability to ventralize early LGE progenitors (Kohtz et al., 2001). A C-terminal cholesterol modification has also been identified on SHH relevant for SHH tethering to the cell surface (Porter et al., 1996). At the functional level, these PTMs were showed to be primordial in determining SHH "short range" or "long range" function (Burke et al., 1999). Another example is the polysialylation of the neural cell adhesion molecule (PSA-NCAM), important to control the timing of the perisomatic GABAergic synapse maturation in the mouse cortex (Di Cristo et al., 2007). Di Cristo and colleagues showed that premature removal of PSA in the visual cortex resulted in precocious maturation of perisomatic innervation by PV basket cINs. Interestingly, polysialytransferases have been implicated in schizophrenia (Arai et al., 2006; Tao et al., 2007; Isomura et al., 2011). In light of this, it seems highly pertinent to pursue on understanding how genes, signaling molecules and environment communicate to shape brain development and function.

# Acknowledgments

LN is a Research Associate and CG a postdoctoral fellow at the Belgian National Fund for Scientific Research (F.R.S-F.N.R.S.). EP is a postdoctoral fellow. LN is funded by grants from the F.R.S.-F.N.R.S., the Fonds Léon Fredericq, the Fondation Médicale Reine Elisabeth, and the Belgian Science Policy (IAP-VII network P7/20). LN and EP are funded by a grant from Actions de Recherche Concertée (ARC11/16-01). Some scientific projects in the Nguyen are funded by the Walloon excellence in lifesciences and biotechnology (WELBIO).

receptor modulation in human lung cancer cells. Lung Cancer 37, 49–56. doi: 10.1016/S0169-5002(02)00025-9


migration of cortical interneurons through the ventral forebrain. J. Neurosci. 31, 6174–6187. doi: 10.1523/JNEUROSCI.5464-10.2011


interneurons expresses the transcription factor Sp8. Cereb. Cortex 22, 2120–2130. doi: 10.1093/cercor/bhr296


co-segregating with schizophrenia. Hum. Mol. Genet. 9, 1415–1423. doi: 10.1093/hmg/9.9.1415


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Peyre, Silva and Nguyen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Control of cortical neuronal migration by glutamate and GABA

# **Heiko J. Luhmann<sup>1</sup>\*, A. Fukuda<sup>2</sup> and W. Kilb<sup>1</sup>**

1 Institute of Physiology, University Medical Center of the Johannes Gutenberg University, Mainz, Germany <sup>2</sup> Department of Neurophysiology, Hamamatsu University School of Medicine, Hamamatsu, Shizuoka, Japan

#### **Edited by:**

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, Japan

#### **Reviewed by:**

Rustem Khazipov, Institut National de la Santé et de la Recherche Médicale, France Laura Cancedda, Istituto Italiano di Tecnologia, Italy

#### **\*Correspondence:**

Heiko J. Luhmann, Institute of Physiology, University Medical Center of the Johannes Gutenberg University, Duesbergweg 6, D-55128 Mainz, Germany e-mail: luhmann@uni-mainz.de

Neuronal migration in the cortex is controlled by the paracrine action of the classical neurotransmitters glutamate and GABA. Glutamate controls radial migration of pyramidal neurons by acting primarily on NMDA receptors and regulates tangential migration of inhibitory interneurons by activating non-NMDA and NMDA receptors. GABA, acting on ionotropic GABAA-rho and GABA<sup>A</sup> receptors, has a dichotomic action on radially migrating neurons by acting as a GO signal in lower layers and as a STOP signal in upper cortical plate (CP), respectively. Metabotropic GABA<sup>B</sup> receptors promote radial migration into the CP and tangential migration of interneurons. Besides GABA, the endogenous GABAergic agonist taurine is a relevant agonist controlling radial migration. To a smaller extent glycine receptor activation can also influence radial and tangential migration. Activation of glutamate and GABA receptors causes increases in intracellular Ca2<sup>+</sup> transients, which promote neuronal migration by acting on the cytoskeleton. Pharmacological or genetic manipulation of glutamate or GABA receptors during early corticogenesis induce heterotopic cell clusters in upper layers and loss of cortical lamination, i.e., neuronal migration disorders which can be associated with neurological or neuropsychiatric diseases. The pivotal role of NMDA and ionotropic GABA receptors in cortical neuronal migration is of major clinical relevance, since a number of drugs acting on these receptors (e.g., anti-epileptics, anesthetics, alcohol) may disturb the normal migration pattern when present during early corticogenesis.

**Keywords: neuronal migration, cerebral cortex, GABA, glutamate, neuronal migration disorders**

#### **INTRODUCTION**

During early brain development, mostly during embryonic phases and in some species also during early postnatal periods, newly generated neurons must migrate from their site of origin to their final target in a distinct brain area and a certain subregion, e.g., a specific layer at a certain site of the cerebral cortex. The distances that neurons must travel depend on the cell type and on the species and range from a few hundred micrometers to many millimeters. One exciting question in neuroscience is how newly generated neurons find their correct way to their final position. Over the last two decades we learned that the neuronal migration process is controlled by a number of different mechanisms. Transcription factors control the identity and the laminar position of developing neurons (for review, Kwan et al., 2012). Chemical cues, e.g., semaphorines and ephrins, are expressed as gradients in the brain and serve as attracting or repelling signals for migrating cells (Bagnard et al., 2001; Holmberg et al., 2006; Zimmer et al., 2008; Sentürk et al., 2011). In some brain regions the vertical fibers of radial glial cells act as chemico-mechanical guiding structures for migrating neurons (for review, Huang, 2009). Receptors activated by neurotransmitters or certain molecules, e.g., the extracellular matrix protein reelin, act as GO or STOP signals in neuronal migration (Huang, 2009).

This review will focus on the role of the two classical neurotransmitter systems glutamate and GABA in neuronal migration of cortical neurons. After briefly describing the different modes of neuronal migration and differences in the migration process between glutamatergic and GABAergic neurons, our current knowledge on the function of glutamate and GABA receptors in neuronal migration will be reviewed. Finally, we will also shortly address the putative role of glycine receptors in neuronal migration. Since the receptors of both transmitters are the target of numerous drugs acting for example as anesthetics or anti-epileptics, pathophysiological perturbations of the migration process by unwanted side effects of these drugs acting on glutamate and GABA receptors during early brain development will be also discussed.

For comprehensive overviews on the molecular and cellular mechanisms of neuronal migration the interested reader is referred to reviews by Ayala et al. (2007), Valiente and Marín (2010) and the review by David. J. Price on "Neuronal migration in the cerebral cortex" in this issue. A summary on neocortical layer formation, the timing of projection and interneuron migration and a comparison between rats and mice is given in the reviews by Kwan et al. (2012) and Tanaka and Nakajima (2012). An update on the early development of the human cerebral cortex is given by Bystron et al. (2008).

#### **MODES OF NEURONAL MIGRATION**

Different modes of neuronal migration have been described (**Figure 1**). In 1972, Pasko Rakic published a seminal paper on the "Mode of cell migration to the superficial layers of fetal monkey neocortex" and described the **radial migration** of immature neocortical neurons along the vertical fibers of radial glial cells (Rakic, 1972). Rakic postulated the existence of a "strong surface affinity" between the radial glial fibers and the migrating neuron and suggested that this "developmental mechanism . . . would allow for the vertical cell columns of adult neocortex" (Rakic, 1972). The radial unit hypothesis of the cerebral cortex was born. The radial glia-dependent locomotion is the dominant migration mode of newborn pyramidal, glutamatergic neurons in the hippocampus and cerebral cortex and also represents the central mechanism for the "inside first—outside last" developmental pattern of the cerebral cortex (neurons marked in red in **Figure 1B**; Nadarajah et al., 2003). Like building a house, the oldest neurons form the lowest layer 6 and subsequently generated neurons form layers 5, 4, 3 and finally layer 2. This inside-out layering also means that radially migrating neurons must pass beyond their predecessors before reaching their final position in the newly generated cortical layer, which they form (for review, Cooper, 2008). Recently Le Magueresse et al. (2012) described with time-lapse 2-photon microscopy in acute brain slice preparations of the neonatal mouse a new type of radial migration of subventricular zone (SVZ)-generated neurons along astrocytes lining blood vessels, which does not depend on radial glial cells.

A different mode of neuronal migration, which is independent of glial guiding fibers, is the **somal translocation** (Nadarajah et al., 2003; for review, Cooper, 2008). Somal translocation is smoother and faster than glia-guided radial migration. Here a leading coiled process extends into the marginal zone (MZ) and is anchored to the basement membrane or to the extracellular matrix. The soma moves upward in a spring-like manner by rapidly shortening the leading process. It seems likely that gliaindependent somal translocation and glia-dependent locomotion depend on different cytoskeletal machinery and motors and thereby are regulated by different processes.

In contrast to the radial migration of pyramidal cells, neocortical GABAergic interneurons show a **tangential migration** pattern throughout the developing telencephalon (de Carlos et al., 1996; for review, Marín, 2013). Inhibitory interneurons migrate tangentially over long distances by generating a leading process, which detects chemical cues in the extracellular environment, and subsequent movement of the nucleus towards to the branching point (nucleokinesis). Recent observations in slice cultures of the mouse embryonic brain indicate that endothelial cells may guide tangential migration (Won et al., 2013) and that tangential migration in the MZ is controlled by meningeal vessels (Borrell and Marín, 2006). The molecular mechanism of this blood vessel-guided migration to the cortex are not known, but neurotrophic factors such as brainderived neurotrophic factor (BDNF) and glial cell line-derived neurotrophic factor (GDNF) may be involved (Le Magueresse et al., 2012). Meninges affect tangential migration in the MZ via secretion of the chemokine CXCL12 which activates CXCR4 receptors (Borrell and Marín, 2006). This type of migration may become reactivated in the adult brain under pathophysiological conditions, e.g., stroke, when SVZ-generated neuroblasts are guided to the peri-infarct zone by blood vessels (Kojima et al., 2010).

Finally, so-called **random walk migration** has been described for medial ganglionic eminence (MGE)-derived cortical interneurons in the MZ of flat-mount cortices (Tanaka et al., 2009). Interneurons migrated tangentially over periods of up to

**GABAergic neurons. (A)** Schematic diagram illustrating migration pathway of the majority of glutamatergic neurons, originating in the ventricular zone (VZ) of the pallium and radially migrating into the developing cerebral cortex (red arrows). The majority of GABAergic neurons are generated in the medial (MGE) and lateral gangionic eminence (LGE) and reach their final position by tangential migration via deep pathways and superficial cortical layers. **(B)** Glutamatergic neurons (marked in different shades of red) are generated in

the VZ and migrate radially either by somal translocation or, at later phases, by locomotion along radial glial cells (light gray). Upon reaching the marginal zone (MZ) they detach and align on top of previously generated neurons of the cortical plate (CP), generating the "inside first—outside last" pattern of the cerebral cortex. The majority of GABAergic neurons (marked in different shades of blue) reach the cortex via tangential migration in the deep pathway within the subventricular zone (SVZ) or the superficial pathway in the MZ. Some GABAergic interneurons travel also within the subplate (SP).

2 days in an unpredictable manner, often changing the rate and direction of migration. These results suggest that MGE-derived cortical interneurons, once arriving at the MZ, are released from regulation by guidance cues and initiate random walk movement (Tanaka et al., 2009).

In summary, radial migration, somal translocation and tangential migration are the dominant forms of neuronal migration in the developing cerebral cortex. It is not surprising that mutations affecting genes, which control these forms of migration may cause severe brain malformations, which are generally categorized as neuronal migration disorders and which are often associated with a spectrum of neurological and/or neuropsychiatric diseases (for review, Guerrini et al., 2008; Guerrini and Parrini, 2010).

#### **MIGRATION OF GLUTAMATERGIC NEURONS**

Neocortical glutamatergic neurons mostly follow a pure radial migration pattern and for them radial glial cells in the ventricular zone (VZ) fulfill two important and different functions in the embryonic cortex (**Figure 1**). On the one hand radial glial cells serve as progenitors and produce by asymmetric cell division neurons and astrocytes, on the other hand radial glial cells serve as migratory guides for the newly generated glutamatergic neurons. Radial glial cells produces neocortical pyramidal and layer 4 spiny stellate cells, which migrate to the cortical plate (CP), thereby forming in the "inside first outside last" pattern the usually six-layered cerebral cortex. Sister glutamatergic neurons, which derive from the same mother cell, take up radially aligned positions in the cerebral cortex across layers and have a higher propensity to form unidirectional chemical synaptic connections with each other rather than with neighboring non-siblings (Yu et al., 2009). These data indicate that the columnar organization of the cerebral cortex may be determined to some extent by lineage (Noctor et al., 2001).

During early embryonic development, these glutamatergic neurons initially use somal translocation to migrate radially and then follow the vertical track along radial glial fibers for locomotion (Rakic, 1972). The extracellular matrix protein *reelin*, which is secreted from early born Cajal-Retzius neurons located in the MZ (for review, Kirischuk et al., 2014), controls this radial migration (for review, Valiente and Marín, 2010). Radial migration of single glutamatergic neurons does not occur continuously following a straightforward route, but rather shows phases of transient migratory arrest and even retrograde migration (Noctor et al., 2004). Gap junctions play important roles in the regulation of both proliferation and neuronal migration. Hemichannels formed by gap junctions mediate the spread of spontaneous intracellular Ca2<sup>+</sup> waves across progenitor cells and provide dynamic adhesive contacts between migrating neurons and radial glial fibers (for review, Elias and Kriegstein, 2008). For glia-guided neuronal migration the connexins Cx26 and Cx43 are essential and in the mouse their deletion disrupts migration to the CP (Elias et al., 2007). For Cx43 it has been demonstrated that deletion of the C-terminal domain modifies neuronal migration (Cina et al., 2009).

Recent immunohistochemical data obtained in embryonic mice demonstrated one population of transient glutamatergic neurons, which is generated early (at embryonic day (E) 12.5) and migrates tangentially over long distances from their generation site at the pallial-subpallial boundary to the CP (Teissier et al., 2010). At birth, these early glutamatergic neurons homogeneously populate all neocortical areas, but subsequently die massively by apoptosis. At birth, about 50% of the dying neocortical neurons belong to this population of tangential migrating glutamatergic neurons (Teissier et al., 2010).

In summary, glutamatergic neurons use mostly radial migration along radial glial fibers and somal translocation to move from their site of generation in the VZ into the developing cerebral cortex.

#### **MIGRATION OF GABAergic NEURONS**

In contrast to the large majority of the glutamatergic neurons, cortical GABAergic interneurons are at least in rodents generated in the subcortical telencephalon; in the lateral, medial, caudal and septal ganglionic eminence (LGE, MGE, CGE, and SGE, respectively; **Figure 1A**), to a minor extent also in the endopeduncular and preoptic area and also in the cortical SVZ (for review, Gelman and Marin, 2010) see also review by Wieland B. Huttner on "Neurogenesis in the developing cerebral cortex" in this issue. A subset of GABAergic neurons, which are 5-HT<sup>3</sup> positive, are generated postnatally in the SVZ and migrate into numerous forebrain regions, including the cerebral cortex, striatum, and nucleus accumbens (Inta et al., 2008). The origin of GABAergic neocortical interneurons in higher mammals, including humans, remains controversial, although a recent publication indicate that also in these species a substantial proportion of interneurons originate from subcortical telencephalic eminences (Letinic et al., 2002; Ma et al., 2013).

The spatio-temporal expression of various transcription factors control the generation and identity of different types of cortical GABAergic interneurons at different developmental periods (for review, Butt et al., 2007; Jovanovic and Thomson, 2011). *Dlx1/2* and *Mash1* are extensively expressed in the ganglionic eminence and determine the GABAergic lineage. *Lhx6*, which is under the control of *Nkx2.1* and *Dlx5/6*, control the generation of parvalbumin- and somatostatinimmunoreactive interneurons, which are generated first in the ventral and dorsal area of the MGE, respectively (Wang et al., 2010). The later generation of vasoactive intestinal polypeptide (VIP) and cholecystokinine (CCK) expressing GABAergic interneurons in the CGE is controlled by the transcription factors *Nkx6.2* and *CoupTF1/2*. The spatio-temporal developmental profile of cortical GABAergic interneurons predicts their intrinsic electrophysiological properties and firing patterns in the mature cortex (Butt et al., 2005). Rapidly adapting firing properties can be observed in mature neuropeptide Y (NPY), reelin, calretinin and/or vasointestinal peptide expressing cortical interneurons, which are generated in the CGE. Rapidly adapting NPY-containing interneurons are also produced in the preoptic area (for review, Marín, 2013).

From their birth place in the ganglionic eminence forebrain GABAergic interneurons migrate tangentially in the MZ, SVZ or intermediate zone (IZ) to the developing cerebral cortex (for review, Marín, 2013). Tangential migration is controlled by the spatio-temporal expression of a number of chemical cues, acting as attracting or repelling signals. Semaphorines, expressed in the LGE, prevent the entry of migrating interneurons into this region and Ephrin EphA5/EphA4 receptors, expressed in the VZ, repel MGE-generated interneurons (for review, Marín, 2013). Tangential migration of cortical GABAergic interneurons is enhanced by the neurotrophic factors BDNF, NT-4, hepatocyte growth factor, and GDNF. On their way to the cortex, interneurons use specific routes or migratory streams (marked in blue in **Figure 1B**): (i) a superficial route in the MZ; (ii) a deep route in the IZ/SVZ; and (iii) a route in the subplate (SP). Using an *in situ* migration assay, Tanaka et al. (2003) observed that neocortical GABAergic interneurons initially migrate predominantly in the IZ/SVZ and then invade the CP and MZ by departing from the major migratory stream in the IZ/SVZ. Once arriving in the MZ GABAergic interneurons show random walk migration and disperse throughout the cortex (Tanaka et al., 2009). A subpopulation of GABAergic interneurons descend from the MZ to be distributed in the CP.

During their tangential migration process, neocortical GABAergic interneurons progressively acquire responsiveness to GABA. Combining *in vitro* patch-clamp recordings, neuropharmacological experiments and single-cell PCR in E14.5 mouse acute slices, Carlson and Yeh (2011) characterized the functional expression of GABA<sup>A</sup> receptor subunits in tangentially migrating interneurons derived from the MGE. At this age, synapses have not yet formed and responsiveness to GABA reflect the functional expression of synaptic and extrasynaptic GABA<sup>A</sup> receptors. Early migrating interneurons located close to the corticostriate juncture showed a robust expression of the alpha2 and alpha3 subunits. When entering the developing cortex, both subunits were still highly expressed and in addition alpha1 and gamma1-3 subunits were upregulated (Carlson and Yeh, 2011). The functional implications of the simultaneous activation of multiple GABA<sup>A</sup> receptor isoforms and the upregulation of receptor isoforms with higher affinity to GABA in the migration process are not known and need to be elucidated.

Some experimental data indicate that migrating interneurons on their way to the cortex may move from one substrate to another, e.g., following specific axonal projections. Once they have reached their final cortical region, cortical GABAergic interneurons migrate radially to their final layer, which has been already formed by the radial migration of glutamatergic neurons. Thus, GABAergic interneurons invade their target layers after glutamatergic projection neurons have reached their final position. The mechanisms underlying this switch from tangential to radial migration are not completely understood. It may be that an intrinsic developmental program or connexins trigger the tangential-to-radial switch (for review, Marín, 2013). Elias et al. (2010) have demonstrated in embryonic rat brain slices including the MGE that this switch is controlled by Cx43 and depends on the adhesive properties and the C terminus of Cx43, but not on the Cx43 channel. These data indicate that the switch from tangential to radial migration depends on a gap junction-mediated interaction between migrating GABAergic interneurons and radial glia cells, similarly to the glia-dependent migration of glutamatergic neurons. In contrast, whereas reelin signaling is essential for proper radial migration of pyramidal neurons, layer acquisition of neocortical GABAergic interneurons does not depend on reelin, but rather on cues provided by projection neurons (Pla et al., 2006).

In summary, GABAergic interneurons migrate tangentially along specific streams from their site of origin in the subcortical telencephalon to their final neocortical site, where they then migrate radially to their final cortical layer.

# **ROLE OF GLUTAMATE IN NEURONAL MIGRATION**

The classical excitatory transmitter glutamate influences neuronal migration mainly by acting on two ionotropic receptors: (i) the NMDA receptor, a Ca2+-permeable subclass of glutamate receptor; (ii) the AMPA/kainate receptor, a usually Ca2+ impermeable glutamate receptor. Three (GluR1-3) of the four known subunits for AMPA receptors are expressed at prenatal stages in the developing cortex, while the GluR4 subunit appears only postnatally (Luján et al., 2005). Of the four subunits assembling kainate receptors, KA-2 and GluR5 and GluR6 are already expressed in the embryonic neocortex around E14 (Bahn et al., 1994). Functional NMDA receptors are composed from two NR1 and two NR2 subunits. NR1 and the highly Ca2<sup>+</sup> permeable NR2B subunits are already expressed at early postnatal stages, while expression of NR2A emerges at postnatal stages in the neocortex (Luján et al., 2005). Functional NMDA receptors have been found on migrating glutamatergic and GABAergic interneurons (Behar et al., 1999; Soria and Valdeolmillos, 2002). Metabotropic glutamate receptors, in particular mGlu1 and mGlu5, are also already expressed in the immature neocortex (López-Bendito et al., 2002a). A direct modulation of neuronal migration by NMDA receptors has been initially described by Komuro and Rakic for granule cells of the developing mouse cerebellum *in vitro*. Here, blockade of NMDA receptors by specific antagonists caused a slow-down of neuronal migration, whereas enhanced activation of NMDA receptors by removal of magnesium from the extracellular milieu or by application of the cotransmitter glycine accelerated cell movement (Komuro and Rakic, 1993).

Various *in vitro* studies using different models of cortical neuronal migration indicate that NMDA receptors also control radial neuronal migration in the cerebral cortex. In cell dissociates of murine embryonic cortical cells and cortical slice cultures, Behar et al. (1999) demonstrated that glutamate is a potent chemoattractant. Only activation of NMDA receptors, but not other ionotropic glutamate receptors, stimulated radial migration of immature neurons out of the cortical VZ/SVZ and application of NMDA antagonists blocked migration (**Figure 2B**; Behar et al., 1999). Inhibition of NMDA receptors using either MK801 or APV, attenuated radial migration in rat tissue explants *in vitro*

(Hirai et al., 1999). In contrast to these observations, which suggest a promigratory effect of NMDA receptors, a massive stimulation of NMDA receptors led to migratory arrest in cultured cerebral neurons (Kihara et al., 2002), indicating that only physiological levels of NMDA receptor activation may be a prerequisite for a promigratory stimulus. The observations that (i) Mg2<sup>+</sup> depletion enhances migration (Behar et al., 1999); (ii) overexpression of the Mg2+-sensitive NR2B subunit increases migration (Tárnok et al., 2008); and (iii) the NR2B subtype specific antagonist ifenprodil hinder migration of cerebellar neurons (Mancini and Atchison, 2007), all indicate that Mg2<sup>+</sup> sensitive NMDA receptors are involved in regulating neuronal migration. It has been also suggested that depolarized membrane potentials of migrating neurons contribute to the relative Mg2<sup>+</sup> insensitivity of the NMDA receptor-mediated effects (Gerber et al., 2010).

Also in tangentially migrating neocortical interneurons an inhibition of NMDA and AMPA receptors impedes migration (Bortone and Polleux, 2009), in accordance with the functional expression of NMDA and non-NMDA ionotropic glutamate receptors in migrating interneurons (Soria and Valdeolmillos, 2002). Unfortunately this study does not allow to discriminate whether AMPA and/or NMDA receptors affect migration. However, at least for mouse hippocampal interneurons it has been demonstrated that AMPA, but not NMDA receptors, influence radial migration (Manent et al., 2006). Therefore further analysis whether AMPA receptors are involved in the tangential migration of neocortical interneurons is required to elucidate if this is a common feature of interneuronal tangential migration.

In neurospheres it has been demonstrated that the early phases of neural progenitor cell migration strictly depend on AMPA receptors (Jansson et al., 2013). However, it is currently unclear whether AMPA receptors also contribute early phases of radial and/or tangential neuronal migration under *in vivo* conditions.

Further evidences for a role of glutamate in migration of neocortical neurons came from *in vivo* studies. Using intracerebral injections of ibotenate, an agonist of NMDA receptors and glutamatergic metabotropic receptors, Marret et al. (1996) demonstrated in the hamster by neuropharmacological experiments that activation of NMDA receptors caused a wide spectrum of abnormal neuronal migration patterns in the cerebral cortex *in vivo*. Golden hamsters were chosen for these studies because compared to mice and rats the cortex in hamsters is very immature at birth. While low doses of ibotenate produced mainly intracortical heterotopias and molecular layer ectopias, indicating an disturbed termination of migration, high ibotenate doses led mainly to periventricular and subcortical heterotopias, suggesting that they affected migratory onset (Marret et al., 1996). These migration defects could be attributed to both migration arrest and unsufficient termination of migration (Takano et al., 2004). Using sustained-release polymer Elvax implants (Smith et al., 1995) containing MK801 to deliver this NMDA antagonist focally to the cortical surface, Reiprich et al. (2005) could demonstrate that a local and transient NMDA receptor blockade in the somatosensory cortex of newborn rats *in vivo* produces structural and functional alterations in the cortical region underlying the implant (**Figure 2A**). MK801-treated animals showed disturbances in the cortical lamination and heterotopic cell clusters in the upper layers.

Complete knockout of NR1, an essential subunit of NMDA receptors, has no effect on the early migration pattern of neocortical neurons in the fetal mouse brain, but mice die at birth due to respiratory problems (Messersmith et al., 1997). A restricted knockout of NR1 in excitatory neocortical neurons (CxNR1KO) led to only slight changes in the neocortical organization, like a disordered barrel cortex, without gross anatomical disturbances reminiscent of cortical migration disorders (Iwasato et al., 2000), but in these animals residual amounts of functional NMDA receptors may be present during prenatal development. The function of NR1 in neuronal migration may be also compensated by other mechanisms in CxNR1KO. On the other hand, in chimeric mice transfected with NR1-deficient stem cells, neurons without functional NMDA receptors show a normal distribution within the hippocampus, indicating that NMDA receptors on neuronal membranes itself may be dispensable for correct radial migration (Maskos and McKay, 2003).

Thus no final conclusion on the role of NMDA receptors for migration can currently be given. While pharmacological *in vivo* and *in vitro* experiments strongly suggest an important role of NMDA receptors for radial migration, the observation that neurons lacking functional NMDA receptors show adequate migration questions this conclusion. These conflicting results may either indicate that the NMDA receptor dependent effects are mediated by non-neuronal target structures like glial cells or that compensatory mechanisms may counteract the lack of functional NMDA receptors.

The source of extracellular glutamate controlling neuronal migration is not completely known. *In vitro* studies on hippocampal organotypic slice co-culture assays from munc18-1 knockout mice, in which vesicular transmitter release is deleted, indicate that glutamate and also GABA is released in a SNAREindependent manner and both transmitters control neuronal migration via a paracrine action (Manent et al., 2005). Another mechanism of extracellular transmitter control are transporters. Glutamate uptake by transporters expressed in astrocytes set extracellular glutamate levels. The expression of glutamate transporters is relatively low in immature rodent hippocampus and increases during early postnatal development (Thomas et al., 2011), suggesting that extracellular concentrations of glutamate may be higher during early corticogenesis when neuronal migration occurs. However, extracellular space is also larger during early development (for review, Syková, 2004), therefore overall extracellular transmitter concentrations in the young brain may be not so much higher than in adult. Furthermore, inhibition of glutamate uptake enhances migration (Komuro and Rakic, 1993), which indicates that glutamate is sequestered rather than released in the vicinity of migration neurons. Related to the glutamatergic system, it has been demonstrated in the cerebellum that glutamate activates Bergmann glial cells to produce and release d-serine, which potentiates glutamate actions on NMDA receptors and enhances neuronal migration of cerebellar granule neurons (Kim et al., 2005).

The downstream molecular mechanisms how glutamate controls neuronal migration are not completely understood, but an appropriate increase in the intracellular Ca2<sup>+</sup> level is pivotal (for review, Komuro and Kumada, 2005; Zheng and Poo, 2007). Elegant experiments performed on migrating cerebellar neurons *in vitro* demonstrated that migratory and resting phases were directly correlated to elevated and resting Ca2<sup>+</sup> concentrations, respectively (**Figure 3**; Komuro and Rakic, 1996). In addition, this study demonstrates that the amplitude of Ca2<sup>+</sup> transients is directly correlated to the rate of saltatory cell movements. Disappearance of these Ca2<sup>+</sup> transients triggered the completion of cerebellar granule cell migration (Kumada and Komuro, 2004). In an interesting experiment Fahrion et al. (2012) were able to rescue methylmercury-induced migratory arrest of murine cerebellar neurons by restoring the frequency of Ca2<sup>+</sup> transients to control levels. Further support for a pivotal role of intracellular Ca2<sup>+</sup> in controlling neuronal migration comes from

**with migration speed and direction. (A)** Granule cells in cerebellar microexplant cultures were loaded with a mixture of the two calcium indicators Fluo-3 and Fura-Red. Upward deflections in Fluo-3/Fura-Red ratio indicate intracellular calcium rise and downward deflections represent calcium decrease. **(B)** Distance and direction of the same cell as in A. During a recording period of 30 min the migrating neuron exhibited 5 cycles of saltatory movements, which closely correlated with transient intracellular calcium changes. Modified and reproduced with permission from Komuro and Rakic (1996).

experiments in which the Ca2<sup>+</sup> chelator BAPTA inhibited radial migration in murine cerebellar (Komuro and Rakic, 1993) and murine neocortical cells (Hirai et al., 1999). Interestingly, soma translocation in migrating GABAergic interneurons depend on the occurrence of non-symmetrical Ca2<sup>+</sup> signals, with larger Ca2<sup>+</sup> transients observed toward the direction of migration (Moya and Valdeolmillos, 2004). On the other hand, a tonic Ca2<sup>+</sup> increase arrested motility in the absence of Ca2<sup>+</sup> transients (Komuro and Rakic, 1996). These data demonstrate that fluctuations in the intracellular Ca2<sup>+</sup> concentration within a physiological range control normal neuronal migration.

The Ca2<sup>+</sup> transients can interfere with the organization of the cytoskeleton via an activation of Ca2<sup>+</sup> dependent kinases, like Ca2+-calmodulin kinases II or doublecortin (DCX)-like kinases (Kumada and Komuro, 2004; Koizumi et al., 2006). Accordingly, inactivation of either DCX or the DCX-like kinase by shRNA slowed radial and tangential neuronal migration (Friocourt et al., 2007, see also review by Fiona Francis on "*The roles of DCX in cortical development*" in this issue). In addition, a Ca2<sup>+</sup> increase also activates Lis1-dependet rho-kinases, which are involved in connecting the microtubules in a Clip170 dependent manner to the actin cytoskeleton and dynein motor complexes (Kholmanskikh et al., 2006, see also review by Emilie Pacary on "*Role of RhoGTPases in cerebral cortex development*" in this issue). Interestingly, mutations in Lis1 and DCX have been directly linked to human neocortical migration disorders (Gleeson and Walsh, 2000).

In summary, there is compelling evidence that glutamate controls radial migration of glutamatergic neurons, most probably by acting on NMDA receptors. The mechanisms of the glutamate effect on tangential migration of GABAergic interneurons is less established and here AMPA receptors are more relevant.

#### **ROLE OF GABA AND TAURINE IN NEURONAL MIGRATION**

The classical inhibitory neurotransmitter GABA is important in controlling neuronal migration via ionotropic GABA<sup>A</sup> and metabotropic GABA<sup>B</sup> receptors (Manent and Represa, 2007). GABA<sup>A</sup> receptors are heteropentamers compiled from in total 19 subunits, divided into eight groups, while GABA<sup>B</sup> receptors are heterodimers co-assembled from the GABAB2 subunit with one of the two isoforms of the GABAB1 subunit (for a detailed review, see Farrant and Kaila, 2007; Ulrich and Bettler, 2007). Several GABA receptor subunits are abundantly expressed during early cortical development. At E14 the GABA<sup>A</sup> receptor subunits α2, α3, α4, β<sup>1</sup> and γ<sup>1</sup> are expressed, with α<sup>3</sup> expressed at particular high levels during prenatal development (Laurie et al., 1992). Accordingly, GABA<sup>A</sup> receptor mediated currents are observed already in proliferative neuroblasts and early postmitotic neurons (LoTurco et al., 1995; Owens et al., 1999). In line with the paucity of α<sup>1</sup> and γ<sup>2</sup> expression, immature cortical neurons show GABA<sup>A</sup> receptor mediated currents with slow kinetics and little desensitization, high GABA affinity and lack of synaptic GABAergic currents before they terminate migration in the CP (Owens et al., 1999). In addition to this classical GABA<sup>A</sup> receptor, ρ subunit containing GABAA-rho receptors, characterized by an exceptionally high GABA affinity and little desensitization, are found in the SVZ, while they are lacking in CP neurons (Denter et al., 2010). GABAB1 and GABAB2 subunits are expressed throughout all neocortical lamia during early stages of cortical development (López-Bendito et al., 2002b). Interestingly tangentially migrating neurons express only GABAB1 subunits and should thus lack functional GABA<sup>B</sup> receptors (López-Bendito et al., 2002b). Finally, it is important to consider that immature neocortical neurons show a high ratio in the expression of NKCC1 to KCC2, which renders GABA<sup>A</sup> mediated responses depolarizing (Yamada et al., 2004).

The implication of GABA receptors in the control of neuronal migration was first demonstrated by Behar et al. (1996), who could show by the use of a microchemotaxis chamber that neuronal migration of dissociated cortical neurons of embryonic rats is stimulated by low concentrations of GABA acting on GABAA/GABAA-rho and GABA<sup>B</sup> receptors. Femtomolar concentrations of GABA induced chemotaxis (migration along a chemical gradient) and micromolar GABA initiated chemokinesis (increased random movement). In a subsequent study Behar et al. (1998) showed that 1–5 µM GABA stimulated the migration of GAD-expressing neurons in the CP, whereas 500 fM stimulated motility of GAD-expressing neurons in the VZ. In this study the authors also postulate that GABA can promote migration via G-protein activation, mediated by GABA<sup>B</sup> receptors, and arrest migration via GABA<sup>A</sup> receptor-mediated depolarization (Behar et al., 1998). Using organotypic neocortical slice cultures Behar et al. demonstrated that specific activation of the different GABA receptors modified different parts of neural migration to the CP (Behar et al., 2000): (i) GABAA/GABAA-rho receptor activation promoted neuronal migration from the VZ/SVZ to the IZ; (ii) GABA<sup>B</sup> receptor controlled the migration from the IZ into the CP; and finally (iii) GABA<sup>A</sup> receptor activation provided the stop signal to terminate migration in the upper CP (Behar et al., 2000). The two ionotropic GABA receptor subtypes, GABA<sup>A</sup> and GABAA-rho receptors, play different roles in the control of radial migration (**Figure 4A**). In organotypic murine neocortical slice cultures application of bicuculline methiodide (BMI) facilitated neuronal migration, while the GABAA-rho specific antagonist TPMPA attenuated migration (Denter et al., 2010). These data add important information to the model on the role of GABA in neuronal migration as initially introduced by Behar et al. (2000). GABAA-rho receptors have a high affinity to GABA and are transiently expressed on migrating neurons in the IZ, where the extracellular GABA concentration at late embryonic stages is relatively low (**Figure 4A**). In the IZ, GABA acting primarily on GABAA-rho receptors acts as a GO signal for migrating neurons coming from the VZ/SVZ and passing through the IZ to the CP. Reaching the CP, migrating neurons no longer express functional GABAA-rho receptors. Due to the intracortical outside directed GABA gradient lowaffinity GABA<sup>A</sup> receptors are now activated and GABA serves as a STOP signal in the upper CP (Denter et al., 2010). Although this model is supported by a number of experimental studies, data on the existence of the outside directed GABA gradient are scare. Regional differences in the amplitudes of GABA<sup>A</sup> receptor-mediated tonic currents suggest a gradient of endogenous GABA<sup>A</sup> receptor agonists in embryonic murine cortex (Furukawa et al., 2014). However, detailed information about spatial gradients in extracellular GABA concentrations are currently not available (Bolteus et al., 2005). GABA imaging in brain slices using immobilized enzyme-linked photoanalysis may provide a possibility to demonstrate such a gradient (Morishima et al., 2010).

Blockade of GABA<sup>B</sup> receptors in organotypic cultures of the rat brain led to an accumulation of tangentially migrating neurons in the VZ/SVZ of the rat neocortex (López-Bendito et al., 2003), suggesting a role of GABA<sup>B</sup> receptors for final apposition of GABAergic interneurons. However since no functional GABA<sup>B</sup> receptors are expressed on these neurons (López-Bendito et al., 2002b), this effect may be caused by GABA<sup>B</sup> receptor-dependent effect on targets upstream from the migration tangential neurons itself (López-Bendito et al., 2002b).

A direct influence of GABA on the migration of cortical neurons has also be shown *in vivo*. Furukawa et al. (2014) demonstrated *in vivo* that continuous blockade of GABA<sup>A</sup> receptors with the GABA<sup>A</sup> antagonist gabazine (SR95531) during late embryonic stages accelerated radial migration in the murine neocortex (**Figure 4B**). In line with this, Elvax implants loaded with the GABA<sup>A</sup> antagonist BMI or the agonist muscimol placed on the neocortical surface of newborn rats induced heterotopic cell clusters in upper layers and a loss of neocortical lamination, probably because of an overmigration from the loss of a stop signal (**Figure 4C**). Whereas BMI caused this effect by blocking GABA<sup>A</sup> receptors, long-term application of muscimol induced

**(A)** Model of GABA<sup>A</sup> and GABAA-rho receptor dependent radial migration in the neonatal cerebral cortex, which shows outside directed GABA gradient (gray colored gradient). In the IZ migrating neurons express functional GABA<sup>A</sup> receptors (blue discs) and GABAA-rho receptors (orange discs), whereas in the CP migrating neurons express only functional GABA<sup>A</sup> receptors. Due to the outside directed GABA gradient the low-affinity GABA<sup>A</sup> receptors are only activated in the CP, while the lower GABA concentration in the IZ is sufficient to activate the high affinity GABAA-rho receptors. Activation of GABAA-rho receptors is necessary to support migration in the IZ (GO sign), while activation of GABA<sup>A</sup> receptors contributes to termination of migration (STOP sign). **(B)** Blockade of GABA<sup>A</sup>

a pronounced receptor desensitization, thereby also reducing GABA<sup>A</sup> receptor function on migrating rat neurons (Heck et al., 2007). In accordance with the results of *in vitro* studies identifying the role of GABA<sup>B</sup> receptors (Behar et al., 2001), *in utero* knockdown of GABA<sup>B</sup> receptors using RNA interference techniques impaired radial migration of the affected pyramidal neuron progenitors in the rat neocortex (Bony et al., 2013). The superficial stream of tangentially migrating GABAergic interneurons in the MZ of neonatal mice is also impaired after inhibition of GABA<sup>A</sup> receptors *in vivo* (Inada et al., 2011), demonstrating a direct influence of endogenous GABA also on tangential migration.

In summary, these reports demonstrate that ionotropic GABA<sup>A</sup> and metabotropic GABA<sup>B</sup> receptors are involved in the control of neuronal migration in the cortex.

One feature of the GABA receptors expressed on migrating neurons is their high GABA affinity, allowing them to sense even low ambient GABA concentrations. Microdialysis experiments in tangential neocortical slices revealed an extracellular GABA concentration of 25 nmol/l in the MZ of early postnatal rats (Qian et al., 2014), however, this experimental paradigm may underestimate the interstitial GABA concentration close to red fluorescent protein (RFP)-positive cells in control (left) and gabazine-treated (right) GAD67GFP/GFP fetuses at E17.5, which were injected with gabazine at E14.5 immediately after the electroporation of the RFP vectors. **(C)** Digital photographs of Nissl-stained coronal sections from a P7 rat that received at P0 on the cortical surface an Elvax implant containing DMSO (top, control), the GABA<sup>A</sup> antagonist bicuculline methiodide (middle, BMI) or the GABA<sup>A</sup> agonist muscimol (bottom). Note upper layer heterotopia due to increased radial migration in BMI- and muscimol-treated animals. Scale bar in B, C middle and C bottom corresponds to 200 µm, in C top to 500 µm. Modified with permission from Denter et al. (2010) **(A)**, Furukawa et al. (2014) **(B)**, and Heck et al. (2007) **(C)**.

migrating neurons. Using GABAergic modulation of glutamate release Dvorzhak et al. (2010) suggested a juxtasynaptic GABA concentration of 250 nmol/l during early postnatal stages in the mouse, with a substantial developmental decrease during the first postnatal week. A substantially higher ambient GABA concentration of ∼0.5 µmol/l was observed in the ganglionic eminence of mice using GABA<sup>A</sup> receptor expressing sniffer cells as GABA sensors (Cuzon et al., 2006), which may be relevant for the tangential migration of GABAergic interneurons from this region.

It was suggested that tangentially migrating GABAergic neurons are a source for GABA (Manent et al., 2005). *In vitro* assays indicated that CP neurons itself secrete promigratory signals acting on GABA receptors and suggested that these signals may include GABA and/or taurine (Behar et al., 2001). In line with this, tangentially migrating neurons in GAD67-GFP knockin mice had a substantially slower migration rate, which has been attributed to the lower extracellular GABA level in these animals (Inada et al., 2011). However, no obvious disorders in gross neocortical organization have been observed after complete blockade of GABA synthesis in GAD65/GAD67 knockout mice (Ji et al., 1999), indicating that other substances can act as GABAergic agonists during prenatal development.

A recent study identified taurine, released by volume-sensitive anion channels, as an important agonist of GABA<sup>A</sup> receptors directly influencing radial migration and its action was most apparent in the SP where taurine is most abundant (**Figure 5**). A decrease in ambient taurine, via pharmacological blockade of taurine synthesis, accelerated radial migration in the developing cerebral cortex. This effect was clearly mediated via GABA<sup>A</sup> receptors and is more substantial in GAD67 deficient mice with reduced extracellular GABA levels (Furukawa et al., 2014). Thus ambient GABA is not negligible, although ambient taurine is a main endogenous agonist. In addition, it was recently shown that taurine inhibits KCC-2 activity via activating the with-no-lysine protein kinase 1 (WNK1) and downstream STE20/SPS1-related proline/alanine-rich kinase (SPAK)/oxidative stress response 1 (OSR1) signaling pathway (Inoue et al., 2012). Thereby it may also play a role in maintaining the depolarizing GABAergic responses required for a promigratory action (see below). Microdialysis experiments in the MZ of early postnatal rats revealed a taurine concentration of 33 µmol/l, which was substantially higher than the GABA concentration (Qian et al., 2014). Thus taurine must also be considered as an important endogenous agonist influencing migration via GABA receptors.

Various mechanisms of GABA release at early stages of corticogenesis have been suggested. GABAergic precursor cells may release GABA tonically in a Ca2+- and SNARE-independent manner. Blocking GABA<sup>A</sup> receptors in hippocampal slice cultures from munc18-1-deficient mice, in which vesicular release is abolished, impairs neuronal migration, which supports the hypothesis that GABA is released in a non-canonical, paracrine manner (Manent et al., 2005). Candidates for a non-vesicular GABA release are GABA transporters (GATs), which take up GABA from the interstitial space. Whereas in the adult brain GAT-1 is mainly expressed in neurons and GAT-3 mainly in glial cells, in the immature cortex both GAT isoforms are expressed in astrocytes and neurons (for review, Kilb et al., 2013) and both transporters can release GABA by acting in reverse mode (for review, Kirischuk and Kilb, 2012). A nonvesicular release of GABA via reversal of GAT-1 has been demonstrated in tangentially migrating interneurons following glutamate induced activation of AMPA receptors and sodium influx (Poluch and König, 2002). In the MZ, GATs represent the major mechanism of GABA release and their operating mode is influenced by excitatory amino acid transporters (EAATs) via intracellular sodium signaling and/or cell depolarization (Unichenko et al., 2013), indicating that ambient glutamate and GABA levels are mutually dependent. As an alternative mechanism of GABA release, GABA could also be released via anion channels such as bestrophin-1 channel (Lee et al., 2010). However, no direct evidence of GABA release via volumesensitive anion channels has yet been reported. In contrast, a recent study identified taurine, released by volume-sensitive anion channels, as an important agonist of GABA<sup>A</sup> receptors (**Figure 5**; Furukawa et al., 2014).

In summary, non-synaptic release of GABA and taurine most probably is the main source of ambient GABAergic agonists that influence neuronal migration. However, the contribution

**(A)** Distribution of RFP-labeled radially migrating neurons in cortical sections from GAD67GFP/GFP fetuses at E17.5. Saline (control, **A1**) or D-CSA **(A2)** was administered to the mothers every12 h between E14 and E17. **(B)** Averaged proportions of radially migrating cells in the CP, SP, IZ, and VZ/VZ of cortical slices prepared from fetuses at E17.5 with (yellow bars) and without (blue bars, control) maternal D-CSA administration. Scale bar in A2: 200 µm. Modified with permission from Furukawa et al. (2014).

of different release mechanisms for GABA and taurine to their ambient levels have not been determined.

The down-stream pathways of migration control by GABA and taurine are not fully understood. As for glutamate, GABAinduced intracellular Ca2<sup>+</sup> transients are essential. Impairing GABA<sup>A</sup> receptor dependent Ca2<sup>+</sup> signals by Ca2<sup>+</sup> chelators inhibits the chemotropic effect of GABA in cortical cells (Behar et al., 1996) and impedes tangential migration (Inada et al., 2011). These GABA induced Ca2<sup>+</sup> transients are mediated by developmental changes in the expression of chloride transporters, leading to depolarizing GABAergic responses (for review, Ben-Ari et al., 2007; Kilb, 2012; Kaila, 2014; Luhmann et al., 2014). Whereas the chloride inward transporter NKCC1 is highly expressed in immature neurons, the efficacy of the chloride outward transporter KCC2 is low, and this imbalance in chloride transport causes a high intracellular concentration of chloride ions (Yamada et al., 2004). In accordance with this hypothesis, a pharmacological blockade of NKCC1 using bumetanide impairs tangential migration of murine GABAergic interneurons *in vivo* (Inada et al., 2011). This hypothesis was challenged by the observation, that premature expression of KCC2 by *in utero* expression at E17/18 causes no obvious migration deficits of rat neocortical neurons, while causing a hyperpolarizing shift in the chloride reversal potential of GABAinduced currents at early postnatal stages (Cancedda et al., 2007). This result is not too surprising, because ectopically expressed wild type KCC2 is not active in embryonic cerebral cortices and becomes functional only postnatally (Inoue et al., 2012). In addition, the *in utero* expression was performed at relatively late stages, so that a substantial part of radial migration to layer II/III was already accomplished until E21 (Cancedda et al., 2007). Indeed, ectopic expression of constitutive active KCC2 mutant at E15 lowered intracellular chloride concentrations, rendered hyperpolarizing GABA<sup>A</sup> receptor mediated responses in postmitotic neurons and perturbed their radial migration (Inoue et al., 2012). In migrating murine interneurons the chloride outward transporter KCC2 increases in expression and becomes functional after they enter the cerebral cortex (Bortone and Polleux, 2009), resulting in a reduced intracellular chloride concentration. The consequent shift in GABAergic action from excitation to inhibition leads to a decrease in the frequency of spontaneous intracellular Ca2<sup>+</sup> transients and terminates neuronal migration, thus turning GABA into a STOP signal for migrating interneurons (Bortone and Polleux, 2009). This scenario is supported by experimental data from Inoue et al. as mentioned above (Inoue et al., 2012).

In addition to a direct excitatory effect, depolarizing GABAergic responses are also involved in spontaneous activity patterns observed in neocortical networks during pre- and early postnatal development (for review, Khazipov and Luhmann, 2006; Allene and Cossart, 2010; Kilb et al., 2011). In a rat neocortical culture model de Lima et al. (2009) demonstrated a relationship between the expression of spontaneous synchronous network activity and neuronal migration. Although migrating interneurons did not participate in early cortical network activity, migration was terminated when interneurons became active in a synchronous network. These data indicate that synchronized GABA and also glutamate release during early network activity can terminate neuronal migration (de Lima et al., 2009).

In summary, GABA and the endogenous GABAergic agonist taurine have a strong impact on tangential and radial migration. These neurotransmitters have both, promigratory and migrationterminating actions, depending on the type of GABA receptor and the intracellular chloride concentration in the migrating neuron.

#### **ROLE OF GLYCINE IN NEURONAL MIGRATION**

Beside ionotropic GABA receptors, glycine receptors also have an influence on neuronal migration. As GABA<sup>A</sup> and GABAA-rho receptors, glycine receptors are also transmittergated chloride channels, which upon activation by glycine or taurine mediate a depolarizing or even excitatory action in the immature cortex (Flint et al., 1998; Kilb et al., 2002, 2008). A functional expression of heteromeric glycine receptors, compiled from α2/β subunits, has already been described in various types of immature neurons, including putative migratory neurons in the IZ (Flint et al., 1998; Kilb et al., 2002, 2008; Okabe et al., 2004), whereas tangentially migrating neurons express α2 homomeric glycine receptors (Avila et al., 2013). It is therefore not surprising that an activation of glycine receptors also promoted radial neuronal migration as demonstrated in organotypic slice cultures from embryonic mouse cerebral cortex (Nimmervoll et al., 2011). However, as pharmacological inhibition of glycine receptors did not interfere with radial migration, Nimmervoll et al. (2011) suggest that glycine receptors do not contribute substantially to radial migration in the neocortex (see also Furukawa et al., 2014). In contrast, tangential migration of cortical interneurons was effectively attenuated by genetic or pharmacological suppression of glycine receptor function in organotypic slice cultures from mouse cortex (Avila et al., 2013). In this study, the migration speed was not affected by addition of taurine, suggesting that glycine itself acts as endogenous neurotransmitter. In line with this suggestion, the estimated extracellular glycine levels of ∼150 nmol/l (Qian et al., 2014) would allow a partial activation of α2 subunit containing receptors with their EC<sup>50</sup> of ∼0.5 µmol/l (Flint et al., 1998; Okabe et al., 2004), while the estimated extracellular taurine concentration of 33 µmol/l (Qian et al., 2014) is most probably ineffective to activate glycine receptors (EC<sup>50</sup> for taurine ∼2.5 mM, Okabe et al., 2004).

In summary, these results indicate that glycine receptors can affect neuronal migration, although these receptors may be relevant only for tangential migration.

#### **ROLE OF GLUTAMATE AND GABA IN NEURONAL MIGRATION DISORDERS**

Given the pivotal role of glutamate and GABA in controlling neuronal migration in the developing cortex, it is not surprising that any modulation in the function of these neurotransmitters during pre- and early postnatal periods may have profound effects on the generation of the cortical architecture. Since neuronal migration is indirectly also controlled by spontaneous network activity, modulation of these transmitter systems may also cause disturbances in early neuronal activity patterns subsequently leading to migration deficits (for review, Kilb et al., 2011).

A direct role of GABA for migration disorders has recently been demonstrated in experimentally induced polymicrogyria. Wang et al. (2014) observed that the accumulation of neurons in the polymicrogyria forming below a neocortical freeze lesion was prevented by the administration of GABA<sup>A</sup> receptor antagonists *in vivo*. An altered migration caused by excitatory GABA<sup>A</sup> receptors has also been revealed as a cause for the hippocampal granule cell ectopia observed after febrile seizures (Koyama et al., 2012).

A number of drugs, which are taken by pregnant women for control of psychiatric or neurological disorders (e.g., epilepsy), anesthetics required for surgical operation of pregnant women, or drug abuse (e.g., ethanol consumption during pregnancy) may have profound effects on neuronal migration patterns in the cortex of the unborn child. These drugs often act on glutamate and/or GABA receptors and, when reaching the immature brain, may change the migration pattern of cortical pyramidal cells and GABA interneurons. Anti-epileptic drugs have a wide range of actions (for review Ikonomidou and Turski, 2010): (i) increasing GABAergic action by inhibiting degradation or uptake mechanisms of GABA; (ii) potentiating GABAergic function by acting on GABA<sup>A</sup> receptor subtypes; (iii) reducing presynaptic glutamate release; (iv) inhibiting glutamate receptor function; and (v) modulating neuronal activity by inhibiting voltage-gated sodium and Ca2<sup>+</sup> channels. The most potent anti-epileptic drugs are often unspecific and act via several of these mechanisms. In experimental studies, anti-epileptic drugs were administered in clinically relevant doses to pregnant rats during the last week of gestation (for review, Manent et al., 2011). Prenatal exposure to the antiepileptic drugs vigabatrin and valproate, which both increase extracellular GABA levels, induced neuronal migration disorders in the hippocampus and cerebral cortex (Manent et al., 2007). In line with this, the benzodiazepine diazepam, which augments the activation of GABA<sup>A</sup> receptors, substantially increased the motility rate of migrating GABAergic interneurons (Inada et al., 2011). In contrast the anticonvulsant carbamazepine, which mainly acts on voltage-dependent sodium channels and has only minor effects on GABA receptor function, did not cause major disturbances in neuronal migration (Manent et al., 2007). These data indicate that changes in the extracellular concentration of GABA have a stronger influence on neuronal migration than modifications in spontaneous neuronal activity.

Alcohol exposure during pregnancy is one of the leading causes of mental retardation and several neuroanatomical malformations, including migration disorders such as lissencephaly and cortical heterotopias (for review, Miller, 1992). Ethanol is a drug acting on ionotropic GABA receptors (for review, Lobo and Harris, 2008) and NMDA receptors (for review, Chandrasekar, 2013), which both are key players in the control of neuronal migration. Clinically relevant levels of ethanol substantially impaired migration of cerebellar neurons by attenuating intracellular Ca2<sup>+</sup> signals (Kumada et al., 2006). In contrast, exposure to relatively low levels of ethanol *in utero* elevated ambient GABA level, enhanced the sensitivity of MGE-derived interneurons to GABA and promoted premature tangential migration into the cortical anlage (Cuzon et al., 2008). A direct link between neuronal migration disorders and glutamate receptor dysfunction has been found in a mouse model for Zellweger disease. Here the migration defect results from a mutation in the NMDA receptor mediated Ca2<sup>+</sup> mobilization (Gressens et al., 2000).

In summary, exposure of the prenatal human brain to drugs and pharmacological agents acting on glutamate and/or GABA receptors, may have a profound influence on tangential and radial migration. The resulting neuronal migration disorders may be difficult to detect and may result in therapy-resistant neurological and neuropsychiatric disorders.

#### **OPEN QUESTIONS AND CONCLUSIONS**

Although the last two decades provided a large amount of experimental data on the role of glutamate and GABA on neuronal migration in the cerebral cortex, a number of questions need to be addressed in order to understand the function of these two classical neurotransmitters and other transmitters in more detail.


migration disorders in humans, may require further studies in primates.

In summary, a variety of studies provide substantial evidence that the classical neurotransmitters GABA and glutamate influence neuronal migration and may thus directly contribute to the pathogenesis of neuronal migration disorders. However, the effects of these neurotransmitters are not uniform, but depend on the brain region, identity and maturational state of the migrating neuron and the neurotransmitter receptor subtypes involved. Awareness of the complex interplay between neurotransmitter action and cellular migration processes may help to prevent migration disorders during fetal development.

#### **ACKNOWLEDGMENTS**

We thank our colleagues who contributed to the work reviewed in this paper. This work was supported by the Deutsche Forschungsgemeinschaft (SFB 1080 and KI835/2).

#### **REFERENCES**


interneuron migration. *J. Neurosci.* 27, 3875–3883. doi: 10.1523/jneurosci.4530- 06.2007


Letinic, K., Zoncu, R., and Rakic, P. (2002). Origin of GABAergic neurons in the human neocortex. *Nature* 417, 645–649. doi: 10.1038/nature00779


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Peyre, Silva and Nguyen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Cortical interneurons migrating on a pure substrate of N-cadherin exhibit fast synchronous centrosomal and nuclear movements and reduced ciliogenesis

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, Japan

#### Reviewed by:

Alfredo Cáceres, INEMEC-CONICET, Argentina Yves Jossin, University of Louvain Medical School, Belgium Masatoshi Takeichi, RIKEN, Japan

#### \*Correspondence:

Christine Métin, INSERM, UMR-S839, 17 Rue du Fer À Moulin, 75005 Paris, France christine.metin@inserm.fr

#### †Present address:

Camilla Luccardini, Université Claude Bernard Lyon 1, CGphiMC UMR CNRS 5534, Villeurbanne 69622, France camilla.luccardini@univ-lyon1.fr; Jean-Paul Rio, Institut du Cerveau et de la Moelle Epinière ICM-INSERM U 1127, Hôpital de la Pitié-Salpêtrière 47, Bd de L'Hôpital 75013 Paris, France jean-paul.rio@icm-institute.org

> Received: 01 April 2015 Accepted: 13 July 2015 Published: 03 August 2015

#### Citation:

Luccardini C, Leclech C, Viou L, Rio J-P and Métin C (2015) Cortical interneurons migrating on a pure substrate of N-cadherin exhibit fast synchronous centrosomal and nuclear movements and reduced ciliogenesis. Front. Cell. Neurosci. 9:286. doi: 10.3389/fncel.2015.00286

#### Camilla Luccardini 1,2,3† , Claire Leclech1,2,3 , Lucie Viou1,2,3 , Jean-Paul Rio1,2,3† and Christine Métin1,2,3 \*

1 INSERM, UMR-S839, Paris, France, <sup>2</sup> Sorbonne Universités, UPMC University Paris 06, UMR-S839, Paris, France, 3 Institut du Fer à Moulin, Paris, France

The embryonic development of the cortex involves a phase of long distance migration of interneurons born in the basal telencephalon. Interneurons first migrate tangentially and then reorient their trajectories radially to enter the developing cortex. We have shown that migrating interneurons can assemble a primary cilium, which maintains the centrosome to the plasma membrane and processes signals to control interneuron trajectory (Baudoin et al., 2012). In the developing cortex, N-cadherin is expressed by migrating interneurons and by cells in their migratory pathway. N-cadherin promotes the motility and maintains the polarity of tangentially migrating interneurons (Luccardini et al., 2013). Because N-cadherin is an important factor that regulates the migration of medial ganglionic eminence (MGE) cells in vivo, we further characterized the motility and polarity of MGE cells on a substrate that only comprises this protein. MGE cells migrating on a N-cadherin substrate were seven times faster than on a laminin substrate and two times faster than on a substrate of cortical cells. A primary cilium was much less frequently observed on MGE cells migrating on N-cadherin than on laminin. Nevertheless, the mature centriole (MC) frequently docked to the plasma membrane in MGE cells migrating on N-cadherin, suggesting that plasma membrane docking is a basic feature of the centrosome in migrating MGE cells. On the N-cadherin substrate, centrosomal and nuclear movements were remarkably synchronous and the centrosome remained near the nucleus. Interestingly, MGE cells with cadherin invalidation presented centrosomal movements no longer coordinated with nuclear movements. In summary, MGE cells migrating on a pure substrate of N-cadherin show fast, coordinated nuclear and centrosomal movements, and rarely present a primary cilium.

Keywords: migration, cortical interneuron, cytoskeleton, primary cilium, N-cadherin, electron microscopy, videomicroscopy

# Introduction

Cell migration, including neuronal migration, is described as a repetitive process comprising several steps (Ridley et al., 2003; Valiente and Marín, 2010): (i) cell polarization; (ii) leading process elongation; (iii) leading process stabilization through the formation of adherent junctions and/or adhesive contacts; (iv) forward displacement of organelles by forces anchored on adhesion sites; and (v) tail retraction. The whole process involves the capability of migrating cells to polarize and to interact with cues in their environment. Although adhesive interactions play a crucial role in cell migration, their role in the migration of embryonic neurons is not fully understood (Solecki, 2012). Recent studies have demonstrated a major role of N-cadherin mediated cell-cell adhesion in the radial migration of principal cortical neurons in the mouse embryo (Kawauchi et al., 2010; Franco et al., 2011; Jossin and Cooper, 2011). They showed that N-cadherin recycling controls the adhesion of migrating neurons to the radial glia, and that N-cadherinmediated adhesion functionally interacts with the reelin signaling pathway to control transition steps at the beginning and at the end of the radial migration. Another study in young postmitotic cortical neurons showed that N-cadherin enrichment at one pole determines the orientation of the polarity axis of the cell, on which the nucleus and centrosome thereafter position (Gärtner et al., 2012). Whether centrosome positioning organizes cell polarity or stabilizes a polarized organization resulting from interactions with extrinsic cues, is unclear in migrating cells of which the centrosome is most often positioned in front of the nucleus (Ueda et al., 1997; Solecki et al., 2004; Higginbotham and Gleeson, 2007; Yanagida et al., 2012).

In the developing forebrain, cortical GABAergic interneurons migrate long distances from the medial ganglionic eminence (MGE) and caudal ganglionic eminence (CGE) where they originate, to the cortex where they settle among principal cells and further differentiate (Anderson et al., 1997; Wichterle et al., 1999; Nery et al., 2002; Butt et al., 2005; Wonders and Anderson, 2006; Flames et al., 2007; Fogarty et al., 2007; Miyoshi et al., 2010). Maintaining the same direction of migration during a long journey toward their final target, is thus crucial for MGE and CGE cells (Marín et al., 2010). In a previous study, we examined whether N-cadherin contributed to the long distance migration of cortical interneurons, as previously demonstrated for precerebellar neurons (Taniguchi et al., 2006). We showed that N-cadherin ablation or invalidation in MGE cells delay the tangential migration of MGE cells and alter their capacity to colonize the cortical plate (Luccardini et al., 2013). Analyses in a co-culture model showed that MGE cells with N-cadherin invalidation exhibit reduced migration speed and abnormal cell directionality because nuclear movements frequently reversed.

N-cadherin is a homophilic cell-cell adhesion protein largely expressed by telencephalic cells on which MGE cells migrate in vivo (Kadowaki et al., 2007). To further understand how N-cadherin contributes to the motility and directionality of migrating MGE cells, we have compared the migration of MGE cells on a pure substrate of N-cadherin, on a pure substrate of laminin that also influences cell migration and polarization (Kawauchi, 2012; Solecki, 2012), and on a substrate of cortical cells on which MGE cells exhibit their ''in vivo-like'' migration cycle. MGE cells migrated faster on the N-cadherin substrate than on any other substrate, and did not aggregate to each other. We have recently shown that the centrosome of MGE cells migrating on cortical cells can dock to the plasma membrane and assemble a primary cilium able to collect extrinsic information (Baudoin et al., 2012; Métin and Pedraza, 2014). We show here that the proportion of MGE cells with a primary cilium was strongly decreased on the N-cadherin substrate as compared with the proportion of ciliated MGE cells on the laminin substrate or on the substrate of cortical cells. Ultrastructural analyses surprisingly revealed that the mature centriole (MC) that assembles and anchors the primary cilium at the cell surface, docked to the plasma membrane at the same frequency in MGE cells cultured on N-cadherin, which rarely assemble a primary cilium, and in MGE cells cultured on cortical cells, which frequently assemble a primary cilium. In MGE cells migrating on the N-cadherin substrate, centrosomal and nuclear movements were fast and synchronous, and the centrosome preferentially localized near the nucleus. Finally, because MGE cells with an acute invalidation of N-cadherin presented centrosomal movements no longer coordinated with nuclear movements, we conclude that a pure substrate of cadherin promotes synchronous nuclear and centrosomal movements. Results moreover suggest that increased centrosomal motility does not impair the capacity of the MC to dock to the plasma membrane.

# Materials and Methods

# Animals

Pregnant Swiss mouse were purchased from Janvier Labs, and mouse embryos were produced in the animal facility of the laboratory by crossing wild type adults (Swiss, Janvier Labs). Experiments and animal care followed the European guidelines and were approved by the Ethical Committee Charles Darwin N◦ 5 (E2CD5).

## Cell Cultures

#### Cell Cultures for Live Cell Imaging and Immunostaining

Cultures were performed on glass coverslips previously treated with nitric acid 67%, ethanol 100% and then sterilized in the oven and fixed with paraffin at the bottom of perforated petri dishes which were used as a culture chamber for the video microscopy experiments. For immunostaining, cultures were performed on glass coverslips placed in 4-well boxes.

#### **Co-cultures of MGE explants on dissociated cortical cells**

Cortices from wild type E13.5 mouse embryos were dissociated mechanically and the cortical cells were spread as a monolayer onto glass coverslips coated with poly-lysine and laminin (PLL/LN) as explained in Bellion et al. (2005). Cortical cells were cultured for 2 to 3 h in Dulbecco's Modified Eagle Medium/Nutrient Mixture F-12 (DMEM/F12) (1/1) supplemented with glucose, glutamax, penicillin/ streptomycin, 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES) buffer, N2 and B27 before placing the explants of MGE. MGEs dissected from E13.5 green fluorescent protein (GFP) expressing transgenic mouse embryos were divided into four to six explants and placed on the substrate of dissociated cortical cells. MGEs electroporated as explained in Baudoin et al. (2012) with a plasmid encoding the pericentrin-AKAP-450 centrosomal targeting (PACT) domain of pericentrin fused to the mKO1 fluorochrome (pCAG-PACT-mKO1) or with a plasmid encoding the LifeAct-RFP construct (pCAG-LifeAct-RFP, Ibidi) were cultured for 2–4 h for recovery before placing them in culture. To analyze centrosomal movements in cadherin invalidated MGE cells, MGE explants were co-electroporated with pCAG-EGFP-N-cad(t) and pCAG-PACT-mKO1 plasmids as explained in Luccardini et al. (2013). Co-cultures were maintained in the culture medium described above for 24 h before time-lapse imaging.

## **Cultures of MGE explants on a protein substrate**

Clean and sterile glass coverslips were coated with PLL/LN to prepare the laminin substrate. Coverslips were coated with the extracellular domain of N-cadherin fused to the human Fc receptor protein (Lambert et al., 2000) as explained in Luccardini et al. (2013) to prepare N-cadherin substrate. Briefly, coverslips were incubated overnight at 4◦C with 4 µg/ml poly-L-lysine (Sigma) and 4 µg/ml goat anti-human Fc antibody (Jackson ImmunoResearch). Coverslips were then washed in borate buffer and incubated with 1 µg/cm<sup>2</sup> purified N-cad-hFc chimera protein for 3 h at 37◦C. To prepare N-cadherin/laminin substrate, 4 µg/ml Laminin (Sigma) was added to the PLL and goat anti-human Fc antibody. GFP-expressing MGE explants dissected from E13.5 transgenic mouse embryos, electroporated or not with a plasmid encoding the PACT-mKO1 fusion protein, were placed on the protein substrate, and cultured 2–24 h before imaging.

# Cultures for Electron Microscopy **Co-cultures of MGE explants on cortical axons**

Co-cultures were performed on plastic coverslips coated with PLL/LN as explained in Baudoin et al. (2012). Because MGE cells cannot be identified by fluorescent markers in co-cultures destined to Electron Microscopy (EM) studies, they were cultured on cortical axons on which they were identified by their morphology. On this substrate, MGE cells exhibit the same migration cycle as on dissociated cortical cells (Bellion et al., 2005). Cortical explants dissected from E13.5 wild type mouse embryos were cultured for 3–5 days in DMEM/F12 (1/1) supplemented with glucose, glutamax, penicillin/streptomycin, HEPES buffer, N2 and B27. When long and numerous axons extended away from cortical explants and covered most of the surface surrounding explants, the MGE was then dissected from E12.5 wild type mouse embryos and cut equally into four to six explants. Explants were placed at the tip of corticofugal axons and cultured for 36–48 h, in order to observe the migration of numerous MGE cells on cortical axons. Co-cultures were then fixed in 1% paraformaldehyde, 1% glutaraldehyde in 0.12M PB/ 0.33 M sucrose at 4◦C.

# **Cultures of MGE explants on a N-cadherin-Fc biomimetic substrate**

Plastic coverslips (Thermanox) coated with the N-cadherin substrate were prepared as explained for glass coverslips (see above). The MGE dissected from E12.5 wild type mouse embryos were cut in half and deposited on a coverslip. After 24 h in culture in the DMEM/F12 culture medium (see above), cultures were fixed as described above in 1% paraformaldehyde, 1% glutaraldehyde in 0.12M PB/ 0.33 M sucrose at 4◦C.

## Electron Microscopy

#### Preparation of Ultra-Thin Sections

Cultures were post-fixed in 2% Osmium tetroxide (OsO4) and contrasted with 1% uranyl acetate in maleate buffer. After dehydration in graded series of ethanol, cultures were transferred in araldite/ethanol (1/1) for 2 h and then overnight in 100% araldite. Small blocks with individual MGE cells migrating on either cortical axons or the N-cadherin substrate were isolated from the rest of the culture under a binocular microscope. These small regions were then embedded in a capsule of araldite with the plastic coverslip oriented parallel to the surface of the capsule. Ultra-thin sections (60 nm), parallel to the plastic coverslips, were collected on copper grids (Maxtaform Finder type H6, 200 mesh, Ted Pella Inc., Redding, CA, USA). Sections were contrasted with lead citrate solution.

## Data Analysis

We selected several explants cultured on N-cadherin, which were surrounded by numerous migrating MGE cells (16 explants). An average of 6–11 grids were prepared with each culture. A total of 294 cells showing at least one centriole (30 grids, four independent cultures) were imaged, at minimum, at two different magnifications (×2000, ×20,000) with an ORIUS CCD camera (Gatan). In co-cultures performed on cortical axons, the density of MGE cells was strongly reduced compared to the density of MGE cells observed on the N-cadherin substrate. A total of 208 cells showing at least one centriole (98 grids, 10 independent cultures) were imaged and analyzed.

For each cell we measured the distance between the nuclear front and the centrosome, we characterized the plane of section of centriole(s) (transverse, longitudinal, oblique), and noted when centrioles were located within the leading process. The anchoring of the MC to the plasma membrane and the presence of ciliary structures (primary cilium, ciliary vesicle) at its distal end was also noted.

## Time-lapse Recording

Time-lapse recording started 2–24 h after placement of explants on the laminin or N-cadherin substrate, and 24–36 h after placement of explants on the dissociated cortical cells. For timelapse recording, the culture medium was replaced by a culture medium without phenol red and with an increased concentration of Hepes buffer (20 mM instead of 10 mM). Co-cultures and cultures were imaged on an inverted microscope (Leica CTR 4000) equipped with a spinning disk (Roper Scientific, Trenton, NJ, USA) and a QuantEM 512SC camera (Photometrics). Cells were recorded using an X63 immersion objective, every 3 min, for an average of 4–5 h. Acquisitions were controlled using Metamorph software (Roper Scientific, Trenton, NJ, USA).

#### Analyses of Recorded MGE Cells

Three independent culture experiments performed on each substrate of migration (dissociated cortical cells, laminin, N-cadherin, and N-cadherin/laminin) were analyzed in the present study. In experiments on cortical cells and N-cadherin, six to eight MGE cells exhibiting a correct expression pattern of PACT-mKO1 were selected for centrosomal and nuclear movement analysis in each experiment. In addition to its centrosome localization, the PACT construct was distributed throughout the cytoplasm and allowed for the identification of nuclear boundary. In experiments on laminin and N-cadherin/laminin, at least 15 cells were tracked in each experiment. Cell tracking was performed with either Metamorph or ImageJ (MTrackJ, NIH, USA).

#### Statistics

Statistical analyses were performed using XLSTAT software. Mean values for a parameter in two independent samples were compared using the Student t-test. Distributions into classes were compared by non-parametric statistical tests: the test of Mann-Whitney for two classes, and of Kolmogorov-Smirnov for five classes.

#### Immunostaining

Cultures were fixed with 4% paraformaldehyde in 0.12M phosphate buffer/0.33 M sucrose at 4◦C. Cells were incubated for 2 h with a blocking solution containing 0.25% Triton, 10% normal goat serum, 2% bovine serum albumin (BSA) in phosphate buffer saline (PBS). The following primary antibodies were used: mouse anti-Arl13b (1:1000; UC Davis/NIH NeuroMab Facility), rat anti-Tubulin YL1/2 (1:2000; abcam). Appropriate Alexa Fluor dye-conjugated secondary antibody (1:500; Invitrogen) was used to detect primary antibodies. Bisbenzimide (1:5000) was used for fluorescent nuclear counterstaining. Culture were mounted in Mowiol-Dabco and observed using an upright fluorescent microscope at X100 magnification (DM6000; Leica).

# Results

MGE cells contribute to 70% of cortical interneurons. We have previously shown that MGE cells exhibit the same migratory behavior in 3-dimensional organotypic cortical slices and on dissociated cortical cells (Bellion et al., 2005). When a piece of the MGE proliferative zone of a E12.5/E13.5 mouse embryo is placed on dissociated E13 cortical cells (see scheme in **Figure 1A1**), MGE cells exit the explant and migrate away with radially oriented trajectories. Cells present a polarized morphology with the nucleus facing the explant and the leading process extending at the opposite side (Bellion et al., 2005). After 24–36 h in culture, MGE cells distribute over a large area around their explant of origin (picture in **Figure 1A1**). In these co-cultures, MGE cells present a saltatory migration characterized by coordinated nuclear and centrosomal movements. The centrosome migrates first toward the cell front and then, the nucleus translocates toward the centrosome. During the nuclear resting phase, the centrosome can migrate far away from the nucleus in close association with the endoplasmic reticulum and Golgi apparatus (GA). We have recently shown that the MC of MGE cells can dock to the plasma membrane and organize a short primary cilium at the cell surface (Baudoin et al., 2012). About one third of MGE cells with the MC anchored to the plasma membrane have a primary cilium, making the MC a basal body. Plasma membrane anchoring could stabilize basal bodies and MCs, and thereby enhance the stabilization of MGE cell polarity.

We had previously shown that N-cadherin ablation and cadherin functional invalidation in MGE cells reduced cell motility and impaired cell polarization, identifying N-cadherin as an important cell-cell adhesion protein that controls the migration of cortical interneurons in vivo (Luccardini et al., 2013). Accordingly, MGE cells cultured on a biomimetic substrate of N-cad-Fc that consists of the extracellular domain of N-cadherin attached to a glass coverslips by the Fc receptor (Lambert et al., 2000) migrated actively. These observations confirmed that N-cadherin plays a unique role to control the migration of MGE cells as the extracellular domain of N-cadherin binds and activates N-cadherin receptors (Lambert et al., 2000). In particular, the motility of MGE cells was increased and the outgrowth of their leading process stimulated (Luccardini et al., 2013). Here, we first compared the motility and directionality of MGE cells cultured on a pure substrate of N-cadherin and on a pure substrate of laminin that binds integrin receptors and also controls cell polarity (Etienne-Manneville and Hall, 2003).

# Migratory Behavior of MGE Cells on Pure Substrates of N-Cadherin or Laminin

As already reported (Luccardini et al., 2013), MGE cells migrate much faster on the biomimetic substrate of N-cadherin than on the ''in vivo-like'' substrate of cortical cells (**Figure 1**). However, frequent directional changes and polarity reversals prevented MGE cells cultured on a pure N-cadherin substrate from colonizing larger areas than MGE cells cultured on cortical cells (**Figures 1B,D**). On a pure laminin substrate, in contrast, MGE cells migrated for very short distances around their explant of origin. After 24 h in culture, migrated MGE cells aggregated to each other to form short chains in which cells presented small amplitude forwards and backwards gliding movements (**Figure 2A**). When N-cad-Fc was added to the laminin substrate, MGE cells distributed over a significantly larger area that nevertheless remained inferior to the area covered by MGE cells on the pure N-cadherin substrate (**Figure 1**). On the mixed laminin/N-cadherin substrate, MGE cells showed fast migration speed and their trajectories were straighter than on

FIGURE 1 | Migratory behavior of medial ganglionic eminence (MGE) cells on various substrates of migration. (A1–B) Analysis of migration in fixed preparations. Schemes (A1–A4) illustrate co-culture and culture models compared in the present study. Co-culture of MGE explants on dissociated cortical cells (A1) mimics the in vivo situation. Biomimetic substrates are prepared on glass coverslips with the extracellular domain of N-cadherin fused to a human Fc fragment, N-cad-Fc (A2, see "Materials and Methods" Section), or with laminin (A3), or with a mixture laminin/N-cad-Fc (A4, see "Materials and Methods" Section). Pictures below schemes illustrate the typical distribution of green fluorescent protein (GFP)-expressing MGE cells after 24 h of migration. Scale bars, 200 µm. Histogram in (B) gives the average area of migration covered by MGE cells around their explant of origin. The largest

migration areas are observed on cortical cells (A1) and on N-cad-Fc substrate (A3). Values significantly different (\*\*p < 0.01; \*\*\*p < 0.001). (C,D) Dynamic analysis of MGE cell migration using time-lapse videomicroscopy. Cells were imaged every 3 min for at least 30 min. Histogram (C) shows the mean instantaneous migration speed of the cell body of wild type (gray bars) and N-cadherin invalidated (blue) MGE cells on various substrates. Graphs in (D) are examples of cell trajectories at the front of the migration area (MGE explant up). Each cell track is shown in a specific color and the recording duration (hours, minutes) is indicated with the same color on the left upper corner. Dots indicate the position of the nucleus at each time point. MGE cells trajectories on dissociated cortical cells are rather straight and dots indicate the saltatory progression of the cell body. (Continued)

#### FIGURE 1 | Continued

N-cad-Fc, alone or associated with laminin, strongly activates cell motility. On pure N-cadherin, MGE cells frequently turn or invert (green, yellow curves) their direction of migration. Cells movements on laminin are very slow. MGE cells are unbranched and form chains (see Figure 2A) in which cell bodies move alternatively forth and back. Trajectories are most often straight. On N-cadherin/laminin, trajectories are straighter than on N-cadherin (blue, orange curves).

N-cad-Fc (**Figure 1C**). On the N-cadherin/laminin substrate, neighboring MGE cells transiently migrated along each other and some of them exhibited polarity reversals (see yellow track in **Figure 1D**). As a consequence, the migration area of MGE cells was slightly but significantly reduced when compared to the migration area of MGE cells on a pure N-cadherin substrate (**Figure 1B**).

The morphologies of MGE cells and their spatial distribution strongly differed on the pure substrates of N-cadherin and of laminin (**Figures 2A,B**). Cells migrating on laminin exhibited bipolar shapes with short unbranched leading and trailing neurites whereas cells migrating on the pure N-cadherin substrate extended several long neurites at cell front and a thinner trailing process. On laminin, migrating MGE cells adhered to each other on their whole length. After 1 day in culture, they preferentially formed chains, with the exception of a few isolated cells. On N-cadherin, MGE cells moved on each other but never aggregated.

Cell dynamics and cell interactions appeared thus opposite on the pure substrates of N-cadherin and laminin. Cell motility was increased on the substrate of pure N-cadherin and decreased on the substrate of pure laminin. Cell-cell contacts were strengthened on laminin, and weakened on N-cadherin. Cell contacts moreover influenced MGE cell trajectories and the size of the area colonized by migrating MGE cells.

## Primary Cilium Frequency in MGE Cells Migrating on N-Cadherin and on Laminin Substrates

MGE cells migrating on a pure N-cadherin substrate or on a pure laminin substrate, frequently reversed their direction of migration. We have proposed that the primary cilium protruding at the cell surface could help stabilizing the centrosome (Baudoin et al., 2012; Métin and Pedraza, 2014). We thus immunostained the primary cilium in MGE cells migrating on those pure protein substrates. Immunostaining revealed that a primary cilium was much less frequent on MGE cells migrating on a N-cadherin substrate than on MGE cells migrating on a laminin substrate (**Figure 2**). In contrast, the primary cilium was observed at similar frequencies on MGE cells migrating on the laminin substrate and on the ''in vivo'' like substrate of cortical cells (**Figures 2C,D**). This result shows that adhesion proteins distributed on the substrate of migration of MGE cells directly influenced the formation and/or the stabilization of their primary cilium, and that N-cadherin alone was unable to promote the assembly and/or the stabilization of the primary cilium. In contrast, we did not correlate the presence of a primary cilium with the capacity of migrating MGE cells to maintain a fixed polarity.

The primary cilium is assembled at the cell surface by the MC that docks to the plasma membrane. In order to determine whether the low frequency of primary cilia on MGE cells migrating on the pure N-cadherin substrate could result from a change in the subcellular positioning of the MC, we analyzed by EM the localization of the centrioles, especially the MC, in MGE cells migrating on the pure N-cadherin substrate. As a reference, the subcellular positioning of centrioles was analyzed in MGE cells migrating on the ''in vivo-like'' substrate of cortical axons (**Figure 3**).

## Subcellular Localization of the Centrosome in MGE Cells Migrating on a N-Cadherin Substrate

Ultrathin sections were performed parallel to the surface of the coverslip in the plane of migration of MGE cells. Sections that comprised the cell body of MGE cells generally comprised a significant portion (up to 15 µm) of the neuritic processes extending from the cell body (**Figures 3A2,A2**<sup>0</sup> ). A trailing process, much thinner than the leading process, was sometimes observed at the rear of the nucleus (black star in **Figure 3A2**). Depending on the phase of the migration cycle, the nuclear region, the cytoplasm and the leading process were either in continuity (**Figure 3A2**) or separated by constricted regions that identified the rostral swelling comprising the centrosome and the GA (**Figure 3A2**; Métin et al., 2008; Valiente and Marín, 2010). Our previous analyses (Bellion et al., 2005; Luccardini et al., 2013) showed that the GA was positioned at cell front in almost all MGE cells cultured on cortical axons.

MGE cells cultured on N-cad-Fc showed either a polarized morphology with a leading and a trailing process that differed in thickness and length (**Figure 3B2**) or a complex multipolar morphology (not illustrated). Multipolar MGE cells were few in our sample. The size of the cell body was increased on N-cadherin and a clear constriction between the nuclear compartment and the rostral cytoplasmic compartment was less frequently observed. The GA was most often located in the same compartment as the nucleus. Our previous studies showed that the GA was located at the rear of the nucleus in 30% of MGE cells cultured on N-cad-Fc (see histogram of Figure 1G in Luccardini et al., 2013).

We analyzed the centrosomal-nuclear distance in MGE cells (n = 213) cultured on N-cad-Fc and in MGE cells (n = 208) cultured on cortical axons as illustrated in **Figures 3C1–C3**. We defined 5 classes of distances between the centrosome and the nuclear edge facing the leading process, or the GA when the leading process was difficult to identify in ultrathin section (**Figure 3E**). In the present study, the nuclear edge facing the leading process or the GA was considered as the nuclear front in 100% of MGE cells migrating in cortical axons, and in only 70% of MGE cells cultured on N-cadherin. It was considered as the nuclear rear in the remaining 30% of MGE cells cultured on N-cadherin.

Surprisingly, the centrosome of MGE cells cultured on Ncad-Fc was less frequently located close to the nucleus (less than 2 µm away) as compared to the centrosome of MGE cells cultured on cortical cells (24% compared to 40%, **Figure 3F**). On N-cad-Fc, the centrosome was most often located at a

distance between 2–5 µm from the nucleus, either at nuclear front or rear. Interestingly, in MGE cells cultured on cortical axons, the centrosome was able to escape the small perinuclear region and migrated into the leading process. Accordingly, the longest nuclear-centrosomal distances were observed in MGE cells cultured on cortical axons (mean value, 9.9 µm, table in **Figure 3G**) rather than in MGE cells cultured on N-cad-Fc (Table in **Figure 3G**, mean value 7.6 µm, significantly different by a t-test, p = 0.0053). These results show that the centrosome of MGE cells migrating on pure N-cadherin distributed in a perinuclear domain significantly larger than in MGE cells cultured on cortical axons. However, on N-cadherin,

the cytoplasm rarely formed a rostral swelling. We then examined whether the distribution of the centrosome in a rather large cytoplasmic compartment surrounding the nucleus affected the frequency to which the MC docked to the plasma membrane.

## Frequency of MC Docking to the Plasma Membrane in MGE Cells Migrating on N-Cadherin

The MC was detected in both MGE cells cultured on N-cad-Fc (n = 92) and MGE cells cultured on cortical axons (n = 103). In ultra-thin sections, the MC was sometimes accompanied by

#### FIGURE 3 | Continued

Comparison of centrosomal position in MGE cells migrating on cortical axons vs. N-cadherin substrate using electron microscopy (EM) on ultrathin sections. (A1–B2) MGE cells were cultured on either cortical axons (A2,A2<sup>0</sup> ) as schematized in (A1) or N-cad-Fc (B2) as schematized in (B1). On these low magnification views of ultrathin sections, migrating MGE cells are polarized. Nuclei (n) are elongated on the polarization axis. A large process, likely the leading process, extends at one pole of the nucleus. No process (A2<sup>0</sup> ) or a thin process resembling a trailing process (black stars in A2,B2) is observed at the opposite pole of the nucleus. Scale bars: 5 µm in (A2,B2). (C1–D3<sup>0</sup> ) Electron micrographs of MGE cells cultured on cortical axons (C1–C3) or on N-cad-Fc (D1–D3<sup>0</sup> ) illustrate cells with different nucleus-centrioles distances: centrioles less than 2 µm in front of the nucleus (C1,D1), more than 5 µm away from the nucleus (C3,D3) or between 2 and 5 µm (C2,D2). The centriole-nuclear distance was measured as shown in panels (C1–C3) (white dotted line). (D10–D3<sup>0</sup> ) are enlarged views of the encircled centriole pair in panels (D1–D3), respectively. Same scale in (D1–D3), and in (D10–D3<sup>0</sup> ), respectively. (E–G) Quantitative analysis. Scheme E defines the five classes of nuclear centrosomal distances used in the present study. The nuclear membrane facing the Golgi apparatus (GA) or the leading process defines the origin of measurements (0). Distances measured away from the nucleus, on the polarity axis of the cell, are positive. Negative centrosomal-nuclear distances were subdivided into two classes (1 and 2), whereas positive centrosomal-nuclear distances were divided into three classes (3–5). In cultures on N-cad, MGE cells with the centrosome located at the nuclear rear (about 30% of MGE cells, see text) were not distinguishable from cells with the centrosome located at nuclear front. Histograms in F show the proportion of MGE cells in each class of nuclear-centrosomal distance in co-cultures on cortical axons (C1) and in cultures on N-cad (C2). Orange and red bars identify cells in which the centriole was located in a process or in a cytoplasmic swelling, as illustrated in (E). (G) Table indicates the mean nuclear-centrosomal distance and the standard deviation in each class of distance.

a daughter centriole (**Figures 4A4,B1,B4**). Most often, the MC was isolated and identified by the presence of lateral appendages and/or distal structures such as a large vesicle or a primary cilium (**Figures 4A1,A3,A4,B1,B2,B3**). In some cells, the MC accumulated small vesicles at its distal end (**Figures 4A1,B2**). In MGE cells cultured on N-cad-Fc as well as on cortical axons, the MC was able to dock to the plasma membrane (**Figures 4A2,B3**). We counted very similar proportion of MC or basal bodies anchored to the plasma membrane in MGE cells migrating on either N-cad-Fc or cortical axons (26% and 32%, not significantly different by a test of Mann-Whitney, p = 0.67, **Figure 4C**).

We then examined whether the MC docked to the plasma membrane in a particular subcellular compartment. Analyses revealed that the centrosome could dock to the plasma membrane at any distance from the nucleus. Nevertheless, the MC most often docked to the plasma membrane in its preferred compartment of residency (see green bars in **Figure 4C2**: 50% of docked MCs were located 2–5 µm away from the nucleus). In MGE cells cultured on cortical axons, the proportion of MCs docked to the plasma membrane was slightly biased toward the leading process (**Figure 4C1**: 30% of docked MCs, and 20% of undocked MCs, the difference is not significant by a Kolmogorov-Smirnov test).

According to the results in MGE cells immunostained for Arl13b (**Figure 2**), the proportion of MCs docked to the plasma membrane and associated with a primary cilium was two folds higher in MGE cells cultured on cortical axons (55%) than in MGE cells cultured on N-cad-Fc (27%). However, in this small sample, the difference is not significant by a Mann-Whitney test. It is noteworthy that MCs located into the cytoplasm associated with a large distal vesicle or small vesicles attached to distal appendages at the same frequency regardless of the substrate of migration (69 and 71% respectively, Table **D** in **Figure 4**). Therefore, the ability to assemble a primary cilium at the distal end of the MC was similar in both samples, showing that the interaction of MGE cells with a N-cad-Fc substrate of migration rather did not promote primary cilium elongation on the cell surface.

# Dynamic Behavior of the Centrosome in MGE Cells Migrating on a N-Cadherin Substrate

Our EM analysis revealed that the MC of MGE cells migrating on a N-cad-Fc substrate docked to the plasma membrane at the same frequency as the MC of MGE cells migrating on cortical axons. However, the subcellular distribution of the centrosome in MGE cells migrating on a N-cadherin substrate and on cortical axons strongly differed. We thus analyzed the dynamic behavior of the centrosome in MGE cells cultured on a pure N-cadherin substrate. MGE cells were electroporated with a PACT-mKO1 construct to label the centrosome and identify nucleus position (**Figure 5**).

MGE cells migrating on N-cad-Fc alternated between straight and curved trajectories (**Figure 1D**) in correlation with their morphology. MGE cells with a bipolar shape followed straight trajectories (see panels 21–33 in **Figure 5A1**) whereas MGE cells with a multipolar shape continuously changed their direction of movements (see panels 3–15 and 39–51). The nucleus did not present resting phases (gray curves in **Figures 5A2,A3**) and the centrosome moved together with the nucleus (mean migration speed, 1.94 µm/min). The centrosome most often remained located at the same pole of the nucleus, even when the direction of migration reversed (pink arrow head in **Figure 5A1**, panels 39–45). The centrosome could also enter the leading process for very short duration. Nuclear centrosomal distance was maintained short (blue curves in **Figures 5A2,A3**, white bar in histogram **Figure 5D**) and the ratio between the mean instantaneous migration speed of the nucleus and of the centrosome was equal to 1 (1.05 ± 0.09, white bar in histogram of **Figure 5F**). In summary, the nucleus and centrosome presented fast and synchronous movements.

Synchronized centrosomal and nuclear movements are rarely observed in MGE cells migrating in vivo or on a complex substrate of cortical cells (Baudoin et al., 2012; **Figures 5B1,B2**). The nuclear progression is saltatory (see the distribution of nuclear positions in **Figure 1D**, and peaks of nuclear speed on gray curve in **Figure 5B2**), whereas centrosomal movements are progressive, with smaller peaks of migration speed and faster forward movements between

FIGURE 4 | Mature centriole (MC) subcellular localization and ciliogenesis in MGE cells cultured on cortical axons or N-cadherin. (A1–B4) Panels show examples of MC with cytoplasmic localization (A1,B1,B2) or MC docked to the plasma membrane (A2–A4,B3,B4) in MGE cells cultured on cortical axons (A1–A4) and N-cad-Fc (B1–B4). Inserts on the right side of panels are low magnification views of cells to localize the MC (circle). MC sectioned longitudinally show lateral appendages (black arrow heads) and/or distal appendages (gray arrow heads) which attach ciliary vesicles to


the MC (B1,B2) or the MC to the plasma membrane (A2–A4,B3,B4). The cilium is often attached to a basal body decorated with a long lateral appendage (A3,B4). Same scale for all high magnification pictures and inserts (see B1). Cil., primary cilium; c.v., ciliary vesicle DC, daughter centriole; n, nucleus. (C) Histograms compare in MGEs cultured on either cortical axons or N-cad-Fc, the distribution of cytoplasmic MC (blue bars) and of MC docked to the plasma membrane (green bars) in the different classes of nuclear centrosomal (Continued)

#### FIGURE 4 | Continued

distances (see Figure 3E). (D) Table gives the number of MC observed in the cytoplasm or docked to the plasma membrane, which exhibited a clear ciliary structure (primary cilium, cil.; ciliary vesicle, c.v.). Results confirm that primary cilia are less frequent on MGE cultured on N-cad-Fc than on cortical axons.

peaks (pink curve in **Figure 5B2**). The nuclear-centrosomal distance shows large amplitude variations, ranging from 0 to 20 µm (blue curve in **Figure 5B2**, and gray bar in histogram **Figure 5D**, significantly different from the white bar by a t-test, p = 0.009). Nevertheless, centrosomal movements are coordinated with nuclear movements even if both organelles do not move synchronously. Accordingly, the ratio between centrosomal and nuclear mean instantaneous migration speeds was close to 1 in the present sample (1.1 + 0.2, gray bar in histogram **Figure 5F**). To further understand the contribution of cadherin mediated cell-cell adhesion in those centrosomal and nuclear movements, we electroporated MGE cells with a dominant negative construct (pCAG-N-cad(t)) that sequesters endogenous catenins and interferes with signaling pathways downstream cadherin (Taniguchi et al., 2006). As previously reported (Luccardini et al., 2013), the migration speed of the nucleus was drastically reduced in MGE cells with cadherin invalidation and cell polarity frequently reversed (**Figures 5C1,C2**, and blue bar in histogram **Figure 5E**). Surprisingly, the centrosome remained more motile than the nucleus, moving from one nuclear pole to the other. Accordingly, the ratio between centrosomal and nuclear instantaneous migration speeds increased to 1.6 (±0.8), showing that cadherin invalidation in MGE cells partially disrupted the correlation between centrosomal and nuclear movements. This result suggests that cadherin activity promotes the synchronization of nuclear and centrosomal movements, which was indeed remarkably tight in MGE cells migrating on a pure N-cadherin substrate.

# Discussion

In the embryonic brain, N-cadherin is present with other adhesive and guidance cues in the migratory pathway of MGE cells. Because we observed that N-cadherin was the most efficient protein substrate to promote the migration of MGE cells in vitro, we decided to further characterize the migratory behavior of MGE cells on this permissive substrate. We moreover compared the migration of MGE cells on a pure N-cadherin substrate with their migration on a pure laminin substrate. The migration speed of MGE cells was very slow on the pure laminin substrate and cells preferentially adhered to each other. On N-cadherin in contrast, contacts between MGE cells were minimal. Centrosomal movements were highly synchronized with nuclear movements. The MC remained able to dock to the plasma membrane but a primary cilium rarely formed. Therefore, cadherin signaling not only influences nuclear motility in migrating MGE cells, but moreover influences centrosomal movements and ciliogenesis.

# Migratory Behavior of MGE Cells on a Pure Substrate of N-Cadherin or of Laminin

Two major types of cell adhesion regulate the cell migration: (1) cell-cell adhesion that is controlled, among other proteins, by classical cadherins that establish homophilic interactions; and (2) cell-extracellular matrix adhesion that is controlled, among others, by integrin receptors that link extra-cellular matrix (ECM) proteins as laminin or fibronectin to the actin cytoskeleton (Kawauchi, 2012). N-cadherin is expressed by MGE cells and by cells in their migratory pathway (Kadowaki et al., 2007). We previously showed that N-cadherin loss of function in migrating MGE cells altered cell motility and cell polarity (Luccardini et al., 2013). To further characterize the role of N-cadherin mediated cell-cell adhesion in the migration of cortical interneurons, we analyzed the migratory behavior of MGE cells on a biomimetic substrate of N-cadherin and on a pure laminin substrate that activates integrins, a class of cell surface receptors involved in neural migration and cell polarity (Lawson and Burridge, 2014). In contrast to laminin on which MGE cells moved very slowly, N-cadherin stimulated the outgrowth of leading processes and the motility of MGE cells. Additional differences distinguished MGE cells cultured on laminin or N-cadherin. First, reciprocal interactions between MGE cells differed. MGE cells migrating on N-cadherin only shortly interacted and never aggregated, even if their density was high. Adhesive interactions between MGE cells were thus minimal. On the laminin substrate in contrast, MGE cells exhibited longlasting and tight reciprocal interactions that quickly led to the formation of chains or cell aggregates. On this latter substrate of migration, MGE cells not only interacted with the protein substrate, but moreover with neighboring cells. These tight and long-lasting cell-cell interactions likely involve N-cadherin receptors, among other receptors, suggesting that the response of MGE cells to the laminin substrate was perturbated by additional and uncontrolled cell-cell interactions. By adding laminin to the N-cadherin substrate, we increased the frequency and duration of cell-cell contacts, suggesting functional interactions between integrin and N-cadherin receptors in migrating MGE cells. Second, MGE cells migrating on N-cadherin and on laminin differed by their morphologies and trajectories. On the laminin substrate, MGE cells were mainly bipolar and showed straight trajectories, whereas MGE cells migrating on N-cadherin alternated between bipolar and multipolar morphologies and showed complex trajectories. These differences in morphology suggest major differences in the organization of the cytoskeleton on the two substrates and will require further investigations.

# Influence of N-Cadherin Biomimetic Substrate on Centrosome Docking and on Primary Cilium Assembly

We have shown that the centrosome of GABAergic neurons born in the MGE of the basal telencephalon is a microtubuleorganizing center that can moreover assemble a primary cilium at the distal end of the MC (Baudoin et al., 2012). An important function of the primary cilium is to anchor the centrosome to the plasma membrane (Pedersen et al., 2008; Reiter et al., 2012). In GABAergic neurons migrating a long distance from the basal

(Continued)

Pericentrin expression is also used to track the nucleus (gray arrow head).

#### FIGURE 5 | Continued

(C1,C2) Migration on dissociated cortical cells of a cell electroporated with the dominant negative construct N-cad(t). (D–F) Histograms show the mean value and standard deviation of the average nuclear centrosomal distance (D), of the instantaneous migration speed of the nucleus (E), and of the ratio between centrosomal and nuclear instantaneous migration speeds (F) in wild type MGE cells monitored on N-cad-Fc (white bars), on cortical cells (gray bars) and in cadherin invalidated MGE cells monitored on cortical cells (blue bars). The nuclear speed is maximal in wild type MGE cells migrating on the pure N-cadherin substrate, and minimal in cadherin invalidated MGE cells. \*\*p < 0.01; \*\*\*p < 0.001. Centrosomal and nuclear movements are synchronized in MGE cells migrating on the pure N-cadherin substrate; cadherin invalidation in MGE cells migrating on cortical cells disrupts the coordination between nuclear and centrosomal movements observed in wild type MGE cells.

telencephalon to the cortex, the MC is able to dock to the plasma membrane. It can associate with a primary cilium protruding at the cell surface (Baudoin et al., 2012; Higginbotham et al., 2012). In the cell samples that we analyzed here, the primary cilium was less frequently observed on MGE cells cultured on N-cad-Fc than on MGE cells cultured on laminin or on cortical cells. However, the proportion of MGE cells with the MC docked to the plasma membrane did not decrease when MGE cells were migrating on a pure substrate of N-cadherin. Moreover, the MC was able to dock on the cell surface at any distance from the nucleus. We observed ciliary vesicles at the same frequency in MGE cells cultured on cortical axons or on N-cad-Fc. The large ciliary vesicle observed at the distal end of MCs likely contributes to the anchoring of the MC to the plasma membrane through a mechanism involving membrane fusion (Sorokin, 1962). Therefore, the ability of the MC to assemble a ciliary vesicle and to dock to the plasma membrane was not perturbed on the substrate of N-cadherin, suggesting that the N-cadherin mediated cell-cell interaction did not allow primary cilium elongation or stabilization. This response of the primary cilium to a cell adhesion protein distributed in the environment was specific to N-cadherin since MGE cells migrating on laminin exhibited a primary cilium at the same frequency as MGE migrating with a ''in vivo like'' substrate of cortical cells.

An attractive hypothesis is that MC docking could interfere with centrosome motility. Our results do not support this hypothesis. Indeed, the centrosome moved twice faster in MGE cells cultured on N-cad-Fc than in MGE cells cultured on cortical cells, although it docked to the plasma membrane at the same frequency in both culture conditions. Even if our results argue against a role of plasma membrane docking to reducing MC motility, we cannot totally exclude this possibility. The primary cilium itself could contribute to regulate MCs motility by interacting with the substrate of migration. In agreement with our present results, we previously showed that the genetic ablation of the primary cilium in Kif3a KOs MGE cells was correlated to a slight increase in the motility of the centrosome (Baudoin et al., 2012).

Integrin receptors have been described on the primary cilium (McGlashan et al., 2006; Seeger-Nukpezah and Golemis, 2012). Adhesive interactions with the extracellular matrix could transiently stabilize the primary cilium and the basal body. Accordingly, the apical abscission of the primary cilium helps centrioles to move away from the basal lamina in progenitors of the cortical neuroepithelium (Das and Storey, 2014).

The role(s) of the primary cilium in cell migration is far from understood (Albrecht-Buehler, 1977; Katsumoto et al., 1994; Schneider et al., 2010; Métin and Pedraza, 2014). In adherent cells like fibroblasts or smooth muscle cells, the cilium has been shown to control cell directionality. The primary cilium of cortical interneurons concentrates receptors to factors in the environment that likely influence the migration of cortical interneurons (Higginbotham et al., 2012). The simultaneous monitoring of centrosomal and primary cilium dynamics shall permit to examine whether the primary cilium interferes with centriole motility in addition to collecting extrinsic signals.

# In MGE Cells Migrating on N-Cadherin, the Centrosome is Maintained at One Pole of the Nucleus

The present study revealed an unexpected behavior of the centrosome in MGE cells migrating on N-cad-Fc. In control MGE cells migrating on cortical cells the centrosome showed a progressive forward migration in front of the nucleus (Métin et al., 2008). After each nuclear translocation, the centrosome starts moving toward the leading process where it can stabilize before the next nuclear translocation (Yanagida et al., 2012). On N-cad-Fc, centrosomal movements were most often restricted to a subcellular domain located at one pole of the nucleus. This domain, that contained the GA and endoplasmic reticulum (ER) vesicles, could be large and the centrosome only rarely escaped to enter a process emerging from the cell body. The centrosome remained there, even during the time period when the cell reversed its direction of migration and when the nucleus moved alternatively forward and backward. This behavior of the centrosome is opposite to the behavior observed in N-cadherin ablated or N-cadherin invalidated MGE cells, in which the centrosome exhibited fast and large amplitude movements in the whole nuclear compartment, moving from the front to the rear of the nucleus. N-cadherin KO MGE cells frequently reversed their direction of movement and showed alternating forward and backward nuclear movements, as control MGE cells cultured on N-cad-Fc (Luccardini et al., 2013). N-cadherin ablation or invalidation disrupted the coordination between nuclear and centrosomal movements in MGE cells. On the contrary, N-cadherin activation stabilized the centrosome in a sub-region of the nuclear compartment. This important role of N-cadherin to stabilize the positioning of the centrosome in migrating MGE cells recalls N-cadherin function to control cell polarity in neural cells (Dupin et al., 2009; Gärtner et al., 2012), and to control in vivo, the polarized organization of cortical progenitors in the proliferative neuroepithelium (Kadowaki et al., 2007). Accordingly, the expression level of N-cadherin has also been shown to control the polarity and migration speed of glial cells in a wound-healing test (Camand et al., 2012). In MGE cells migrating on N-cad-Fc, the nuclear and centrosomal movements were highly synchronized, which was not the case in MGE cells migrating on cortical cells. The regulatory mechanisms involved in this control remain to be explored.

In conclusion, we show here that N-cadherin-mediated cellcell interactions are efficient signals for stimulating MGE cell motility, including centrosome motility. The negative correlation between centrosomal movements and primary cilium formation observed in MGE cells migrating on N-cadherin, remembers results in MGE cells with a genetic ablation of the primary cilium (Baudoin et al., 2012). In addition, our results show that N-cadherin mediated cell-cell interactions and integrin mediated ECM-cell interactions have distinct influences on the ciliogenesis in migrating MGE cells.

#### References


#### Acknowledgments

We thank René-Marc Mège for help with the biomimetic substrates, Melissa Martin for english corrections and all the members of the team for helpful discussions. This work was supported by Institut National de la Santé et de la Recherche Médicale (INSERM), Agence Nationale pour la Recherche (ANR Grant MRGENE), Fondation J. Lejeune and Fondation pour la Recherche sur le Cerveau. CL was supported by a grant from Neuropole Ile de France and by the Grant MRGENE from ANR. C. Métin's team is affiliated with the Paris School of Neuroscience and the Bio-Psy Laboratory of Excellence. We thank Institut du Fer à Moulin for Imaging Facility and Animal Facility.


interneurons. J. Neurosci. 30, 1582–1594. doi: 10.1523/JNEUROSCI.4515-09. 2010


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Luccardini, Leclech, Viou, Rio and Métin. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Capture of microtubule plus-ends at the actin cortex promotes axophilic neuronal migration by enhancing microtubule tension in the leading process

#### *B. Ian Hutchins 1,2 and Susan Wray1 \**

*<sup>1</sup> Cellular and Developmental Neurobiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, MD, USA <sup>2</sup> Postdoctoral Research Associate Program, National Institute of General Medical Sciences, National Institutes of Health, Bethesda, MD, USA*

#### *Edited by:*

*Takeshi Kawauchi, Keio University School of Medicine/Japan Science and Technology Agency, Japan*

#### *Reviewed by:*

*Suresh Jesuthasan, Institute of Molecular and Cell Biology, Singapore Miguel Valdeolmillos, Universidad Miguel-Hernandez, Spain*

#### *\*Correspondence:*

*Susan Wray, Cellular and Developmental Neurobiology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, 35 Convent Dr., Bldg. 35, Rm. 3A1012, Bethesda, MD 20892, USA e-mail: wrays@ninds.nih.gov*

Microtubules are a critical part of neuronal polarity and leading process extension, thus microtubule movement plays an important role in neuronal migration. However, the dynamics of microtubules during the forward movement of the nucleus into the leading process (nucleokinesis) is unclear and may be dependent on the cell type and mode of migration used. In particular, little is known about cytoskeletal changes during axophilic migration, commonly used in anteroposterior neuronal migration. We recently showed that leading process actin flow in migrating GnRH neurons is controlled by a signaling cascade involving IP3 receptors, CaMKK, AMPK, and RhoA. In the present study, microtubule dynamics were examined in GnRH neurons. Failure of the migration of these cells leads to the neuroendocrine disorder Kallmann Syndrome. Microtubules translocated forward along the leading process shaft during migration, but reversed direction and moved toward the nucleus when migration stalled. Blocking calcium release through IP3 receptors halted migration and induced the same reversal of microtubule translocation, while blocking cortical actin flow prevented microtubules from translocating toward the distal leading process. Super-resolution imaging revealed that microtubule plus-end tips are captured at the actin cortex through calcium-dependent mechanisms. This work shows that cortical actin flow draws the microtubule network forward through calcium-dependent capture in order to promote nucleokinesis, revealing a novel mechanism engaged by migrating neurons to facilitate movement.

**Keywords: neuronal migration, neuronal migration disorders, microtubules, IP3 receptors, EB1, super resolution microscopy, actin cytoskeleton**

#### **INTRODUCTION**

For proper assembly of neural circuits, newly born neurons must migrate from their place of origin to their final location. Neuronal migration is commonly classified by the pathway the cells use, e.g., radial, tangential, or anteroposterior—anatomically indicating orientation to the cortex (Marín et al., 2010). However, neurons use many modes of migration within these categories. Some features are common to multiple populations of neurons such as saltatory locomotion occurring in radially migrating cortical neurons as well as in the axophilic migration of GnRH neurons (Nadarajah et al., 2001; Casoni et al., 2012). Similar features between different types of migrating neurons indicate that conserved movement mechanisms exist. Yet, certain aspects, such as the basic mechanisms underlying movement of cells during migration are clearly variable. These mechanisms include locomotion and nucleokinesis (Schaar and McConnell, 2005; Tsai and Gleeson, 2005), rapid spring-like somal translocation (Nadarajah et al., 2001), iterative extension and retraction of leading process branches (Martini et al., 2009), a highly branched "climbing mode" for pathfinding (Kitazawa et al., 2014) or multipolar migration (Tabata and Nakajima, 2003; Falnikar et al., 2013). These different mechanisms of migration often exhibit major alterations in the actin cytoskeleton (Solecki et al., 2009; Asada and Sanada, 2010; He et al., 2010; Martini and Valdeolmillos, 2010; Hutchins et al., 2013). Actin dynamics can promote neuronal migration by propulsive contractions at the cell rear (Martini and Valdeolmillos, 2010; Steinecke et al., 2014), through leading process actin dynamics away from the soma (Solecki et al., 2009; He et al., 2010), or via both mechanisms in tandem (Hutchins et al., 2013). However, microtubule forces, together with actin are most likely responsible for generating the sequential steps of nuclear translocation and neuronal cell migration (Pollard and Borisy, 2003; Tolic-Nørrelykke, 2010; Lysko et al., 2014 ´ ). Microtubules surround the nucleus. In the leading process, extended bundles of microtubules emanating from the centrosome define the direction of movement. Live-cell imaging data from mouse cerebellar granule cells showed that movement of nucleus and centrosome occur independently (Umeshima et al., 2007). These data suggest the existence of a pathway that may depend on a decentralized (i.e., away from the centrosome) microtubule organization and/or an interaction with actin cytoskeleton (Schaar and McConnell, 2005; Solecki et al., 2009).

The present study investigates the role of microtubules in neurons exhibiting axophilic anteroposterior migration, GnRH (gonadotropin releasing hormone 1-expressing) neurons. Recent work revealed that IP3 receptors promote nucleokinesis in these cells, signaling through CaMKK, AMPK, and RhoA, to engage cortical actin flow toward the distal leading process (Hutchins et al., 2013). Here, using the same model system, we show that microtubule linkage to the dynamic cortical actin in the leading process shaft transmit forces critical for nucleokinesis.

#### **MATERIALS AND METHODS**

#### **NASAL EXPLANTS**

All procedures were approved by NINDS ACUC and performed according to NIH guidelines. Explants were generated from E11.5 embryos of either gender as previously described (Klenke and Taylor-Burds, 2012). Explants were incubated at 37◦C in defined serum-free medium (SFM) in 5% CO2. Pharmacological treatments included 75µM 2-APB (Tocris Bioscience), 1µM nocodazole (Tocris Bioscience), and Concanavalin A (10µg/mL, Vector Labs).

#### **CONFOCAL MICROSCOPY**

Images were acquired using a Nikon TE200 microscope with a CSU10 spinning disk confocal (Yokogawa, Tokyo, Japan) and Hamumatsu ImagEM C9100-13 EMCCD camera (Hamumatsu, Hamumatsu, Japan) with a 60× objective (Nikon, Melville, NY) for microtubule imaging or Retiga SRV (Qimaging, Surrey, BC, Canada) with a 20× ELWD for DIC imaging.

#### **SUPER-RESOLUTION IMAGING**

Images were acquired using a Leica CW STED Confocal microscope (stimulated emission depletion) (Klar et al., 2000) with a 100× oil immersion objective (Leica). Images were over-sampled by a factor of ∼2.4 with a pixel size of 37.5 nm. Samples were fixed in 4% formaldehyde in PHEM buffer at 37◦C for 1 h and prepared for two-color STED microscopy with Atto 425 phalloidin (5 units/mL, equivalent to 165 nM, Sigma), and EB1 primary antibodies (1:100, BD) labeled with Oregon Green 488 secondary antibodies (1:1000, Life Technologies). Cells were also immunostained for GnRH (SW-1, 1:3000) (Hutchins et al., 2013) and labeled with Alexa Fluor 647 (1:1000, Life Technologies); this channel was imaged with conventional microscopy immediately prior to STED imaging, which bleached this fluorophore.

#### **MICROTUBULE IMAGING**

Microtubules were labeled by bath application of 150 nM TubulinTracker Green (Life Technologies) for up to 25 min. Six micrometer z-stacks at 1.5µm intervals were acquired every 30 s for imaging sessions lasting up to 20 min. During axophilic migration, GnRH neurons are closely apposed to olfactory sensory axons; measurements of microtubule dynamics were carefully taken to avoid fluorescence signal from intersecting pathway axons. Z-stacks were flattened for image analysis. Microtubules were manually tracked. This method was validated with automated cross-correlation measurement from the same cells (TRACKER ImageJ plugin, Olivier Cardoso, Paris Diderot

University, set to 9 × 9 pixel regions and 3-pixel correlation size) (Hutchins et al., 2013). In the 11 control cells that were cross-validated, manual and automated tracking of microtubules over the entire imaging session yielded a striking correspondence (*R*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*7840, simple linear regression). Nucleus centroids were tracked to calculate migration rates. GnRH neurons monitored by fluorescence imaging showed similar rates of movement (23.92 ± 6.46µm/h) as unlabeled cells monitored by DIC imaging (23*.*27 ± 1*.*42µm/h). To ensure that microtubule and nuclear movement occurred at the same time, movies were segmented into 2-min frames and movement compared within those frames. Microtubule/soma convergence was measured as the decrease over time of the distance between the edge of the soma and leading process microtubule bundles. Negative convergence indicates that microtubules are separating away from the edge of the soma.

#### **STATISTICS**

Statistics were performed in Prism 5 (GraphPad, La Jolla, CA) or R (R-Project; for 3D scatter plot and multiple regression). Unless otherwise noted, n is the number of cells and N is the number of explants. Model II linear regression was used to analyze cytoskeletal dynamics to account for measurement error in both the dependent and independent axes (Hutchins et al., 2013). This total least squares regression minimizes the sum of squared distances from the points to the regression line. Pearson's correlation coefficient *r* is given as a measure of effect size for these analyses.

Residuals analysis is performed to examine the contribution of a second parameter on a measured variable. This analysis was used to determine whether microtubule/soma convergence contributed to migration rate after removing the effects of forward microtubule movement. Residual soma speeds were evaluated with Model II linear regression as stated above.

# **RESULTS**

Different roles for microtubules in neuronal migration have been described including pulling the nucleus along the leading process toward the growth cone (Tsai et al., 2007; Asada and Sanada, 2010) and forming a barrier to nucleokinesis (He et al., 2010; Martini and Valdeolmillos, 2010; Falnikar et al., 2011). Observing microtubule dynamics with TubulinTracker Green showed that leading process microtubules translocated toward the growth cone during migration (**Figures 1A–C**). In addition, a strong relationship between the speed and direction of microtubule translocation with movement of the cell body was found (**Figure 1D**).

The relationship between simultaneous translocation of both microtubules and the nucleus toward the growth cone appears consistent with the proposed role of microtubules pulling the nucleus toward the growth cone (Asada and Sanada, 2010). However, in our experiments, single cell analysis revealed that the soma frequently advanced faster than the microtubules, corresponding to compression of the soma against microtubule bundles located immediately adjacent to the nucleus (**Figure 1C**). This soma/microtubule compression was measured as the speed at which the front edge of the soma and the microtubule bundle converged. This observation could be evidence of a microtubule

**FIGURE 1 | Microtubule dynamics during neuronal migration. (A)** Microtubules in a migrating GnRH neuron (raw fluorescence left, scale bar, 5µm; mid, nucleus outlined with dashed line; solid line indicates region for generating kymograph; image of the same cell at the end of the imaging session, right). **(B)** The microtubules translocate forward during neuronal migration (kymograph, duration 10 min) . (**C**, left) Lower magnification view of fluorescent microtubule staining in a migrating GnRH neuron. Inset, region of higher-magnification region shown in pseudocolored time-lapse images (right). Scale bar, 10µm. (**C**, right) Simultaneous forward microtubule translocation (arrow) and convergence with the soma (asterisk). Dotted lines show the distance between the

barrier in front of the nucleus. However, only a weak firstorder relationship between soma/microtubule compression and migration rates was found (**Figure 1E**).

One possibility is that forces causing soma/microtubule compression add to the influence of microtubule translocation described above. In this case compression should be compared to the residual migration rate (the migration rate left over after subtracting out the influence of microtubule translocation) to detect an additive contribution. To this end, a residuals analysis was performed (Hutchins et al., 2013). Residuals analysis subtracts the influence of the first independent variable (microtubule translocation rates) from the dependent variable (migration rate), giving a "residual" migration rate that can be compared to a new independent variable (soma/microtubule compression). This analysis revealed a strong correlation between soma/microtubule compression and the residual migration rate (**Figure 1F**). How well do these two measures combine to predict migration rates? The relationship between microtubule speed, soma/microtubule compression and soma speed are shown in a 3D scatterplot with a best-fit plane (**Movie 1** shows the 3D scatterplot). These data indicate that microtubule translocation and soma/microtubule compression strongly predict movement of the nucleus when taken together (multiple regression *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*7696).

To understand the mechanism(s) underlying these observations (in particular, the soma/microtubule convergence), two pertinent models of nucleus/microtubule interactions that have been previously reported were examined (see **Figures 2A,B**). The microtubule brake model (Falnikar et al., 2011) proposes soma edge and microtubule bundles at the beginning (blue) and end (green) of the imaging session; brackets (right) summarize the change in these distances from beginning (pre, 3.9µm) to end (post, 2.7µm). **(D–F)** Frame-by-frame analysis was performed (*n* = 65) on 2-min frames from 12 neurons (*N* = 9 explants). **(D)** Forward translocation of microtubules vs. soma speed within 2-min time frames (*p <* 0*.*0001, linear regression, *r* = 0*.*68). **(E)** Microtubule/soma convergence vs. soma speed shows only a weak relationship (*p* = 0*.*032, linear regression, *r* = 0*.*27). **(F)** Microtubule/soma convergence accounts for much of the residual soma speed after subtracting the effect of microtubule translocation rates (*p <* 0*.*0001, linear regression, *r* = 0*.*68).

that cross-linked microtubules in the leading process create a barrier for nucleokinesis (**Figure 2A**). Soma/microtubule compression could thus result from propulsive forces from the cell rear (Martini and Valdeolmillos, 2010) as the nucleus is forced through this microtubule lattice. In this case, nucleus/microtubule compression should occur *only* when the nucleus is propelled forward. Alternatively in the second model, microtubule motor proteins can pull the nucleus along microtubule bundles, drawing the two together (Tsai et al., 2007; Zhang et al., 2009). In this scenario, nucleus/microtubule convergence should also be observed when the nucleus pauses, as microtubules are drawn backward (schematic in **Figure 2B**). To test these possibilities, microtubule dynamics were monitored in neurons that were spontaneously pausing. In stalled neurons, microtubules displayed rapid movement backward toward the nucleus (**Figures 2C–E**), as if no longer coupled to movement of the soma as they are in forward migration (see **Figure 1**). Robust microtubule convergence with the cell body (in this case caused by backward movement of microtubules rather than forward movement of the soma) suggested that microtubules were actively drawn toward the soma and not acting as a brake (**Figure 2F**).

Some predictions made by these two models of nucleus/microtubule convergence were further tested with these live imaging experiments. A microtubule brake model would predict that the leading edge of the soma might be compressed backward toward the centroid of the nucleus as it pushed against microtubules located ahead of it (**Figure 2A**). In contrast, if the nucleus is drawn forward along microtubules

**FIGURE 2 | Models to explain convergence between nucleus and leading process microtubules. (A)** Testable model 1 ("Brake"): Microtubules act as a brake. Nucleokinesis is due to pushing forces from behind that cause the nucleus to "crash" against leading process microtubules. Leading process microtubules (green) form a barrier that slows (resistive force shown as arrow) the front edge of the nucleus (blue) as these compress together. In this model microtubule convergence and excess speed of the front edge are inversely correlated, and this convergence only occurs during nucleokinesis. **(B)** Testable model 2 ("Cable"): Microtubule motor proteins (black dots) draw the nucleus forward along the leading process microtubules, which can be thought of as cables or rails, as in other cell types (Zhang et al., 2009). The pulling force (arrow) from in front of the nucleus draws the front edge along microtubules faster than the center, causing elongation of the nucleus. In this model, microtubule/soma convergence and excess speed of the front edge are directly correlated, and convergence may also occur in GnRH neurons that have stalled. (**C**, left) Fluorescent staining of microtubules in a GnRH neuron that is not migrating. (**C**, right) Outlines indicate the border of the cell (solid) and nucleus (dotted), while the line shows the region measured for kymographs. Scale bar, 5µm. **(D)** Backward microtubule translocation (arrows) in a paused neuron. Dotted lines show the distance between the soma edge and microtubule bundles at the beginning (blue) and end (green) of the imaging session; dashed blue lines denote the nucleus; brackets (right) summarize the change in these distances from beginning (pre, 6.1 µm) to end (post, 2.7µm). Scale bar, 5µm. **(E)** Kymograph of the region shown in **(C)**, with an asterisk and arrow corresponding to the marked regions in **(D)**; the microtubule bundle in the leading process (arrow) and the front edge of the soma (asterisk). **(F)** Schematic: During stalling, microtubules (green) reverse direction and move toward the nucleus. **(G)** Measurements of microtubule/soma convergence vs. excess speed of the front edge show a direct relationship (*p* = 0*.*0325, linear regression, *r* = 0*.*27, *n* = 65 frames from 12 neurons, *N* = 9 explants), refuting the "brake" model in **(A)** and supporting the "cable" model in **(B)**. **(H)** Acute nocodazole (at microtubule depolymerizing concentrations) reduced neuronal migration rates in DIC-imaged GnRH neurons by 22% (*p* = 0*.*0051, Wilcoxon matched pairs signed rank test; *n* = 107 neurons from *N* = 4 explants).

by an active mechanism (rather than by a passive collision), the nucleus edge may elongate toward the leading process as it is pulled forward along leading process microtubules (**Figure 2B**). These models make opposite predictions about the relationship between elongation of the leading soma edge vs. microtubules and were tested by measuring the change in distance from the nucleus centroid to its edge facing the leading process. We found that microtubule/soma convergence was directly related to soma elongation (**Figure 2G**), supporting an active process drawing soma and microtubules together. Studies have reported unaltered or enhanced neuronal migration when microtubules are depolymerized, consistent with a braking mechanism (Schaar and McConnell, 2005; He et al., 2010; Martini and Valdeolmillos, 2010). Conversely, if convergence of the soma and microtubules is an active process in migrating GnRH neurons (e.g., caused by motor proteins drawing the two together, rather than a passive collision caused by propulsion from the cell rear), one would predict that depolymerization of microtubules should slow, rather than accelerate migration. In our system, acute nocodazole (at microtubule depolymerizing concentrations) reduced neuronal migration rates by 22% (**Figure 2H**, *p* = 0*.*0051, Wilcoxon matched pairs signed rank test; *n* = 107 neurons from *N* = 4 explants), further supporting an active process drawing soma and microtubules together during cell migration.

Cortical actin flow in the leading process promotes nucleokinesis, and thereby the migration of GnRH neurons, and is dependent on calcium release through IP3 receptors (Hutchins et al., 2013). To test whether microtubule dynamics during nucleokinesis operate using the same calcium release-dependent signaling pathway, calcium channels (IP3 receptors) were blocked with 2-APB (Li et al., 2009; Hutchins et al., 2011, 2013). TRP channels, also blocked by 2-APB, have been shown to have no effect on either spontaneous calcium activity or migration in GnRH neurons (Hutchins et al., 2013). Inhibiting calcium release through IP3 receptors reduced forward translocation of microtubules (**Figure 3** and **Movie 2**). Notably, after application of 2-APB, microtubules were observed to reverse direction–from moving toward the growth cone to instead moving toward the stalled soma (**Figures 3D,F**). These data indicate that, in contrast to microtubule translocation, soma/microtubule convergence rates were unaffected by 2-APB, i.e., not dependent on calcium release (*p* = 0*.*23, Two-Way ANOVA, *n* = 35 frames from *N* = 5 explants). Thus, these experiments showed that movement of leading process microtubules was uncoupled from movement of the soma during calcium channel inhibition. The fact that soma/microtubule convergence remained intact during 2-APB-induced stalling (**Figure 3**) suggested that another mechanism(s) was being utilized. Taken together, the results indicate that microtubules translocate forward along the leading process dependent on calcium release through IP3 receptors, while simultaneously and independently, the nucleus converges with microtubule bundles.

To determine whether the forward translocation of microtubules and cortical actin flow in the leading process (Hutchins et al., 2013) are (1) directly linked or (2) simultaneously regulated by calcium release, but not mechanically coupled, two experiments were performed. In the first experiment, Concanavalin A (ConA, a cortical actin flow inhibitor) (Canman and Bement, 1997; He et al., 2010; Hutchins et al., 2013) was applied during microtubule imaging. Forward microtubule translocation was nearly abolished in the presence of ConA, while reverse translocation toward the cell body was unaffected—45% of time-lapse frames showed forward microtubule translocation in control cells (*n* = 12) vs. only 13% of frames in ConA-treated cells (*n* = 7 cells from *N* = 3 explants; *p* = 0*.*0026, Fisher's exact test, **Figure 4**). These data indicate that microtubules require both calcium release and cortical actin flow to maintain their forward movement during axophilic neuronal migration. To determine whether these structures where mechanically coupled, microtubule plus-end tracking protein EB1 was examined. Microtubule plus-end tracking proteins (+TIPs) can be coupled to actin cortex in non-neuronal cells through a protein complex including the +TIP protein EB1 (Wen et al., 2004). If this protein complex also contributes to the phenomena observed here, then EB1 puncta should co-localize with the actin cortex in migratory GnRH neurons in a calcium release-dependent manner. To test for microtubule capture at cortical actin in GnRH neurons, microtubule plus-end locations were labeled with antibodies to EB1 (Jaworski et al., 2009). Conventional microscopy was unable to provide the resolution necessary to test this hypothesis, so super-resolution STED microscopy was used. Two-color super-resolution imaging of EB1 and phalloidin revealed many EB1 puncta embedded in the actin cortex in control neurons (**Figures 4A–F**). To determine whether manipulating calcium release through IP3 receptors altered microtubule capture at the actin cortex, 2-APB was applied. No differences in raw EB1 fluorescence signal were detected in the leading process shaft of vehicle and 2-APB treated cells (**Figure 4E**, 24.6 ± 2.6 arbitrary fluorescence units in vehicle treated cells vs. 30.0 ± 3.2 in 2-APB treated cells; *p* = 0*.*227, *t*-test, Cohen's *d* = 0*.*53), suggesting that treatment with 2-APB did not affect the total amount of EB1 in the shafts of GnRH neurons. However, EB1 localization to the actin cortex was significantly

brackets (right) summarize the change in these distances from beginning (pre, 3.4µm for **C** and 3.6µm for **D**) to end (post, 3.6µm for **C** and 2.3µm for **D**). Images in **(A–F)** are from the same cell. **(E,F)** Kymographs of the region shown in **(B)**, containing the microtubule bundle (arrow) and edge of the soma (asterisk) as shown in **(C,D)**. **(G)** Change in microtubule translocation rate corresponds to the change in soma speed (*n* = 35 frames from 7 neurons in *N* = 5 explants, *p <* 0*.*0001, *r* = 0*.*65, linear regression). Scale bars, 5µm.

reduced after blocking IP3 receptors (**Figure 4F**, 1.1 ± 0.13 EB1 puncta/µm of cortical actin in *n* = 10 control neurons vs. 0.59 ± 0.05 EB1 puncta/µm of cortical actin in *n* = 12*.*2-APB treated neurons, *p <* 0*.*0006, *t*-test; Cohen's *d* = 1*.*75, *N* = 3 explants for both conditions). This result indicates that microtubule capture at the actin cortex is a physical interaction that is attenuated in the absence of calcium release.

# **DISCUSSION**

The present results support a model for nuclear movement during neuronal migration whereby microtubules link to cortical actin draw the nucleus forward, flowing toward the growth cone, during calcium activity. Dissociation of microtubules from actin during calcium inhibition disrupts this process, resulting in microtubule translocation back toward the soma, possibly due to calcium-independent microtubule motor activity. In addition, our data show that during neuronal migration, microtubule capture at the moving actin cortex in the shaft transmits forces critical for nucleokinesis. These results reveal a fundamental mechanism of microtubule contribution to nucleokinesis as cells migrate to establish their proper neural circuits.

Genetic studies of tubulin subunits in neurological disorders have revealed that mutations in tubulin genes have severe effects on migration (Keays et al., 2007; Tischfield et al., 2011). However, mutations could affect any one of the many modes of migration used by the affected cells (neurite extension, multipolar movement, locomotion, or somal translocation) (Saillour et al., 2014). Thus, it is important to delineate cytoskeletal dynamics during neuronal migration in the unperturbed state, to better understand the etiology of the disease state.

Microtubules form a cage-like structure around the nucleus and extend into the leading process (Rivas and Hatten, 1995). As such, microtubules are well positioned to either pull or obstruct the nucleus. Microtubules are essential for leading process extension (Baudoin et al., 2008; Lysko et al., 2014). However, the contribution of microtubule dynamics to the movement of the soma during the nucleokinesis phase of neuronal migration is controversial. Movement of the centrosome, the structure where most microtubules are attached at their minus ends, precedes nucleokinesis in migrating cortical pyramidal neurons (Tsai et al., 2007). This observation is consistent with microtubules promoting migration by transmitting traction forces from the growth cone at the tip of the leading process (Asada and Sanada, 2010) and propulsion from actinomyosin contractions in the cell rear is halted when microtubules are artificially stabilized (Martini and Valdeolmillos, 2010). However, other work in cortical neurons suggests that leading process microtubules, combined with kinesin-5, form a molecular brake (Falnikar et al., 2011), consistent with data in migrating cerebellar granule neurons in which depolymerization of microtubules accelerated migration rates (He et al., 2010). Even in the same cell type, the centrosome sometimes leads the nucleus and sometimes trails (Yanagida et al., 2012). These studies suggest that mechanisms underlying migration are context-dependent and likely temporally modified. Thus, understanding neuronal migration will require discovering when and how microtubule mechanisms are engaged during neuronal migration.

Our results reveal a new mechanism engaged during the axophilic migration of GnRH neurons. We show that microtubule linkage to the dynamic cortical actin in the leading process shaft (Hutchins et al., 2013) promotes the forward movement of the nucleus. It is not known whether this is the same mechanism used to draw the centrosome forward in the radial migration of cortical neurons (Tsai et al., 2007), but appears to be independent of the mechanisms used in cortical interneurons and cerebellar granule cells that are either not affected or accelerate with microtubule depolymerization (He et al., 2010; Martini and Valdeolmillos, 2010; Falnikar et al., 2011). Although actin-dependent propulsion from the cell rear described in cortical interneurons and cerebellar granule cells (Martini and Valdeolmillos, 2010; Steinecke et al., 2014) does not discernably contribute to the microtubule/soma convergence observed in GnRH cells, it does correlate with ∼30% of the forward movement that is unexplained by leading process actin dynamics (Hutchins et al., 2013). Other mechanisms involving iterative branching and retraction of the leading (or multipolar) process (Martini et al., 2009; Kitazawa et al., 2014) seem not to be utilized in GnRH neurons, which instead form long, mostly unbranched leading processes. Since many cells exhibit simple morphology when undergoing migration, the cytoskeletal dynamics described here for GnRH neurons may be a common developmental mechanism. As such, microtubule linkage to the dynamic cortical actin in the leading process shaft promoting forward movement of the nucleus, adds to a growing repertoire of microtubule-based cellular tools used by neurons to accelerate or slow their migration, along with microtubule braking (He et al., 2010; Falnikar et al., 2011) and +TIP-dependent leading process protrusion (Kholmanskikh et al., 2006).

#### **ACKNOWLEDGMENTS**

This work was supported by the Intramural Research Program of NINDS, NIH (ZIA NS002824-21 to Susan Wray), and a postdoctoral fellowship from the Postdoctoral Research Associate Program of NIGMS, NIH (B. Ian Hutchins). We thank Dr. Carolyn Smith and the NINDS Light Imaging Facility for training and use of the STED microscope, and Dr. Ellen Flannery for thoughtful comments on the manuscript.

#### **SUPPLEMENTARY MATERIAL**

The Supplementary Material for this article can be found online at: http://www*.*frontiersin*.*org/journal/10*.*3389/fncel*.* 2014*.*00400/abstract

**Movie 1 | Microtubule translocation and compression strongly predict movement of the nucleus when taken together.** Three-dimensional scatter plot of microtubule translocation (MTspeed) and microtubule/soma convergence (labeled MTcompression) rates vs. migration speed (soma speed). **(A–C)** Different views of the 3D scatter plot of these parameters with best-fit plane. Multiple regression *<sup>R</sup>*<sup>2</sup> <sup>=</sup> <sup>0</sup>*.*7696 (*<sup>R</sup>* <sup>=</sup> 0.88).

**Movie 2 | Pseudocolored time-lapse video of labeled microtubules.** Arrows track microtubule bundles in each segment of the video (control vs. 2-APB treatment from the same cell). 1st half: microtubules advance as the soma moves forward (position of the cell border shown in an overlay). 2nd half: microtubules reverse toward the nucleus as movement stalls during 2-APB treatment.

#### **REFERENCES**


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 25 September 2014; accepted: 06 November 2014; published online: 27 November 2014.*

*Citation: Hutchins BI and Wray S (2014) Capture of microtubule plus-ends at the actin cortex promotes axophilic neuronal migration by enhancing microtubule tension in the leading process. Front. Cell. Neurosci. 8:400. doi: 10.3389/fncel.2014.00400 This article was submitted to the journal Frontiers in Cellular Neuroscience.*

*Copyright © 2014 Hutchins and Wray. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.*

# Morphological and functional aspects of progenitors perturbed in cortical malformations

# **Sara Bizzotto1,2,3 and Fiona Francis 1,2,3\***

1 INSERM UMRS 839, Paris, France

<sup>2</sup> Sorbonne Universités, Université Pierre et Marie Curie, Paris, France

3 Institut du Fer à Moulin, Paris, France

#### **Edited by:**

Yoko Arai, Université Paris Diderot, France

#### **Reviewed by:**

Silvia Cappello, Ludwig Maximilian University of Munich, Germany Veronique Marthiens, Institut National de la Santé et de la Recherche Médicale, France

#### **\*Correspondence:**

Fiona Francis, Inserm UMRS 839, Institut du Fer à Moulin, 17 rue du Fer à Moulin, 75005 Paris, France e-mail: fiona.francis@inserm.fr

In this review, we discuss molecular and cellular mechanisms important for the function of neuronal progenitors during development, revealed by their perturbation in different cortical malformations. We focus on a class of neuronal progenitors, radial glial cells (RGCs), which are renowned for their unique morphological and behavioral characteristics, constituting a key element during the development of the mammalian cerebral cortex. We describe how the particular morphology of these cells is related to their roles in the orchestration of cortical development and their influence on other progenitor types and post-mitotic neurons. Important for disease mechanisms, we overview what is currently known about RGC cellular components, cytoskeletal mechanisms, signaling pathways and cell cycle characteristics, focusing on how defects lead to abnormal development and cortical malformation phenotypes. The multiple recent entry points from human genetics and animal models are contributing to our understanding of this important cell type. Combining data from phenotypes in the mouse reveals molecules which potentially act in common pathways. Going beyond this, we discuss future directions that may provide new data in this expanding area.

**Keywords: neurodevelopment, mouse mutant, radial glial cells, proliferation, epilepsy, intellectual disability, lamination**

#### **INTRODUCTION**

Cortical malformations (**Figure 1**) are usually detected during pregnancy (fetal ultrasound), and are obvious after birth due to developmental delay, epilepsy and intellectual deficits. In human, magnetic resonance imaging (MRI) is used to classify the defects and if a genetic origin is suspected, this classification directs potential genetic screens. New variants of these disorders, unexplained by known genes, are currently the subject of exome sequencing projects. Studies in the mouse, as well as in other organisms, try to model these disorders. Knockdown or knockout of genes of interest reveals the cellular mechanisms. Alternatively, mouse mutants arise spontaneously and their characterization subsequently helps reveal both new genes and mechanisms. In general there are many different forms of cortical malformation, and many variants in each category. This review aims not to be exhaustive, but to resume general notions related to the abnormal functioning of progenitor cells. We start here by briefly describing the malformations of interest at the morphological level. We then group different gene mutations, classifying by similar phenotypes observed in mouse mutants, and in so-doing, dissect different aspects of progenitor cell function. Finally, we discuss and integrate all this information in order to have a more global current view of the cellular mechanisms related to malformations.

#### **CORTICAL MALFORMATIONS**

For so-called disorders of neuronal migration, neurons derived from zones of proliferation close to the ventricles, do not reach their correct destination in the cortical plate (CP), either because they are arrested in the white matter (subcortical band heterotopia, SBH or "double-cortex", online mendelian inheritance in man OMIM 300067), and/or form a disordered, often thickened, CP (Barkovich et al., 2012). A thicker cortex is often associated with abnormal cortical gyri, leading either to a simplified or abnormal gyral pattern or, in their absence, to a smooth appearance of the cortical surface (lissencephaly). The type 1 lissencephaly spectrum (e.g., OMIM 607432, 300067) hence includes a smooth, thickened and disorganized cortex (agyria), or simplified, thickened and abnormal gyri (pachygyria), or SBH. X-linked inheritance, different gene mutations and different genes (**Table 1;** Des Portes et al., 1998; Jaglin and Chelly, 2009) explain the spectrum of phenotypes. In general these latter disorders may involve intrinsic functions in migrating neurons which are not mentioned here, although some genes play multiple roles including in neural progenitors, which we discuss in Section Mouse and Human Mutations and Mechanisms Important for RGC Function.

Overmigration of neurons at the pial surface (so-called cobblestone or type II lissencephaly, e.g., OMIM 236670, 253800, Cormand et al., 2001; van Reeuwijk et al., 2005),

represents a different set of disorders involving perturbed progenitors (Section Basal Processes of RGCs, Shedding Light on Heterotopias, Polymicrogyria and Type II Lissencephaly). Walker Warburg syndrome (WWS) is a severe autosomal recessive disorder of this nature characterized by muscular dystrophy, eye and neuronal migration defects. Overmigration gives rise to disorganized cerebral and cerebellar cortices and multiple coarse gyri with agyric regions (cobblestone lissencephaly). As well as this, the structural brain anomalies include agenesis of the corpus callosum, cerebellar hypoplasia and hydrocephalus. WWS is grouped within a series of disorders which include Fukuyama congenital muscular dystrophy (FCMD; Kobayashi et al., 1998), and Muscle-Eye-Brain disease (MEB). Mutations in a number of related genes have been associated with the various types of cobblestone lissencephaly (Section Basal Processes of RGCs, Shedding Light on Heterotopias, Polymicrogyria and Type II Lissencephaly, **Table 1;** Godfrey et al., 2007).

Polymicrogyria (e.g., OMIM 615752, 610031, 606854, Leventer et al., 2010) is considered as a separate entity, although there can be overlapping features with cobblestone lissencephaly (Bahi-Buisson et al., 2010). This disorder is characterized by multiple small folds at the surface of the brain, either diffuse or restricted to one brain region. The mechanisms causing this disorder initially remained elusive and they have for some time been described as affecting the end stages of migration. With the identification of various mutant genes (**Table 1;** Squier and Jansen, 2014), and studies in the rodent, it has become clear that a number of genes play a role in progenitors (Section Basal Processes of RGCs, Shedding Light on Heterotopias, Polymicrogyria and Type II Lissencephaly), especially the end, basal attachment of radial glial cells (RGCs) forming the pial surface of the cortex.

Periventricular heterotopia (PH, e.g., OMIM 300049, Parrini et al., 2006) is associated with large clusters of neurons present at the ventricular surface. These are often observed on MRI as visible gray matter nodules extending into the ventricles. Since neurons are known to be generated in these regions during development, it is assumed that some neurons, after being produced, do not migrate at all. Mouse models reveal abnormalities at the ventricular lining, which is made up of apical RGC end-feet with intricate cell-cell junctions (Section Apical Adhesive Interactions and


<sup>1</sup>All genes code for proteins with centrosomal-related activities, MSGP, microcephaly with simplified gyral pattern; <sup>2</sup>Some genes appear in multiple categories depending on the patient and gene mutation; <sup>3</sup>Complex malformation of cortical development (MCD) associated with microcephaly; <sup>4</sup>PMG, Polymicrogyria; <sup>5</sup>Associated with macrocephaly and sometimes hydrocephaly.

#### **Table 1 | Genes and human malformations**.

Mechano-Transduction, Shedding Light on PH and Ciliopathies). Each of the PH genes (**Table 1;** Sheen, 2014), shows potential roles linking the plasma membrane to the cytoskeleton and some of these genes may also be important during neuronal migration.

Hydrocephaly (e.g., OMIM 307000) is associated with an abnormal quantity of cerebrospinal fluid in the ventricles, causing them to be larger than normal. Genetic causes related to specific human hydrocephaly phenotypes are still relatively unknown, with the notable exception of L1-CAM mutations (Adle-Biassette et al., 2013), involved in a syndrome including hydrocephalus due to aqueductal stenosis. Although causative mechanisms are indeed heterogeneous, hydrocephaly can arise because fluid movement is impaired. One defect is related to motile cilia on neuroependymal cells, as ciliary beating drives fluid flow (Tissir et al., 2010; Tong et al., 2014). Mouse mutations that affect motile ciliogenesis can thus lead to hydrocephalus, disruptions in neurogenesis and brain tumor formation (Han et al., 2009; Tissir et al., 2010; Hildebrandt et al., 2011). Primary non-motile cilia are known to act as mechano-transducers, transmitting signals to the developing tissue (Paridaen et al., 2013). Abnormalities in these processes may or may not be associated with hydrocephaly. Abnormal cilia, and the cycle where cilia components are disassembled to be reused during mitosis, can also be associated with other cortical malformations. Various mouse mutants with progenitor defects show hydrocephaly (Sections Apical Adhesive Interactions and Mechano-Transduction, Shedding Light on PH and Ciliopathies, **Table 2**), although often the exact causes of this remain unidentified.

Microcephaly (e.g., OMIM 251200, 605481, Gilmore and Walsh, 2013) in human refers to a disorder in which the brain at birth is found to be significantly (−2.5–3 standard deviations below the mean) smaller than control brains. This condition leads to intellectual disability. In microcephalia vera, or primary microcephaly, although the brain is proportionally smaller, brain architecture seems not to be dramatically changed and the brain exhibits cortical folds. Its small size is indicative of a highly reduced number of neurons, premature neurogenesis or excessive cell death is likely, and most of the genes identified suggest roles in centrosomal-associated activities during division (see **Table 1** and Section RGCs and Cell Division, Mechanisms Leading to Microcephaly, Gilmore and Walsh, 2013). There are a number of related disorders with microcephaly and additional cortical malformations, such as microcephaly with a simplified gyral pattern (MSGP, Adachi et al., 2011), or complex cortical malformations and polymicrogyria (e.g., OMIM 603802, 604317, e.g., Bilgüvar et al., 2010; Yu et al., 2010). These less "pure" forms show, as well as a reduced brain size, more noticeably affected gyrations (simplified or multiple small gyri), implying parallel changes in neuron production, organization and brain architecture. There is now known to be overlap between "pure" forms and those more obviously affecting gyri, as shown by unbiased genetic studies revealing mutations in genes previously identified mutated in other variants of the pathology (Poulton et al., 2014). Whole exome or genome sequencing is extremely useful in this respect, revealing unexpected genes associated with wider phenotypes than initially thought.

Macrocephaly (e.g., OMIM 600302) potentially has multiple origins related either to increased neuron number (inverse situation compared to microcephaly) but also to increased neuropil (e.g., dendritic arborizations), the latter linked to conditions such as autism spectrum disorder (OMIM 605309). Macrocephaly due to increased neuron number is not yet as clearly elucidated as conditions such as microcephaly, probably related to the multiple potential causes of this disorder, and different types of progenitors found in primate, which are not easily studied in the rodent (Hansen et al., 2010; Wang et al., 2011). Future studies with human genetics as a starting point (and see e.g., Keeney et al., 2014a,b) will almost certainly shed further light on this condition.

#### **MOUSE AND HUMAN MUTATIONS AND MECHANISMS IMPORTANT FOR RGC FUNCTION GENERAL CHARACTERISTICS OF RGCs**

One characteristic aspect of RGCs is their intrinsic highly polarized structure with the cell body confined to the ventricular zone (VZ), and two processes that depart from it: a long basal process reaching the pial surface, and a short apical process descending to the ventricular lining. RGCs need both processes to exert their function: the basal process constitutes the scaffold for migration of newly born neurons through the intermediate zone (IZ), while the apical process is responsible for attachment to the ventricular lining and contains key elements of signaling pathways. These are important to control the balance between proliferation and differentiation, and for cellular specification. RGCs, which are Pax6-positive, as well as selfrenewing, can give rise to basal progenitors in the subventricular zone (SVZ) which are Tbr2-positive, these then give rise to postmitotic neurons. These and further progenitor types are greatly expanded in the primate cortex (reviewed by LaMonica et al., 2012).

As RGCs present a very specialized morphology and dynamics, which are strictly linked with the function they exert during cortex development, every minor perturbation involving their structure or behavior is susceptible to lead to major developmental problems. Indeed, numerous genes coding for proteins influencing RGC morphology and function have been found mutated in mouse models and in cortical malformation patients. We try to bring together here mouse mutant data related to these genes (also resumed in **Table 2**), classifying these data by different RGC compartments and cellular mechanisms (resumed in **Figures 2**, **3**).

## **BASAL PROCESSES OF RGCs, SHEDDING LIGHT ON HETEROTOPIAS, POLYMICROGYRIA AND TYPE II LISSENCEPHALY**

Perturbations that affect basal process structure can cause subsequent problems of neuron migration and lead to cortical malformations such as heterotopia or cobblestone lissencephaly. Breaks of the cortical basement membrane (BM; **Figure 2**) have been associated with RGC basal process end-feet that are not well attached to the extracellular matrix (ECM). The

#### **Table 2 | Specific genes mouse mutants**.


#### **Table 2 | Continued**


#### **Table 2 | Continued**


#### **Table 2 | Continued**



meningeal BM is located immediately below the pia matter and serves as an anchor point for the end-feet of RGCs and as a physical barrier to migrating neurons. Through human genetics studies, it has been shown that cobblestone lissencephalies are associated with reduced glycosylation of alpha-dystroglycan, a basal process dystrophin-associated glycoprotein that is crucial to act as anchor between the dystrophin complex and the ECM (e.g., laminin, van Reeuwijk et al., 2005; Roscioli et al.,

phenotype, in this case some RGC basal processes are not well attached to the pial surface, possible breaks in the BM potentially cause neurons (burgundy nuclei at the surface of the brain) in some regions to overmigrate into the meningeal space. **(C)** Periventricular disorganization, some neurons (burgundy nuclei) remain stuck at the ventricular surface, most probably due to breaks in the ventricular lining where apical end-feet of RGCs normally

(e.g., HeCo mice), in this case a proportion of apical progenitors detach from the ventricular surface (represented by blue nuclei without apical attachment to the ventricular lining) and retain proliferation capacity, providing a local source of neurons in the IZ (burgundy nuclei). A subcortical heterotopia subsequently arises. VZ, ventricular zone; IZ, intermediate zone; CP, cortical plate; MZ, marginal zone; BM, basement membrane.

2012; Buysse et al., 2013). Six major genes have been identified encoding putative or demonstrated glycosyltransferases, POMT1, POMT2, FKTN, FKRP, LARGE and POMGnT1. Post-mortem studies have helped characterize this disorder, however, mouse models for these genes are often not viable, which has led to the difficulty of studying the exact mechanisms involved (Brockington et al., 2001; Yoshida et al., 2001; Beltrán-Valero de Bernabé et al., 2002; Longman et al., 2003; van Reeuwijk et al., 2005; Manzini et al., 2012; Roscioli et al., 2012; Willer et al., 2012). Fragmentation of the BM is however, also frequently seen after deletion of other ECM components or receptors. Laminins are major secreted glycoproteins found in the BM, where they influence cell proliferation, differentiation, migration, and adhesion. *LAMB1*, encoding the laminin subunit beta-1 is involved in basal process attachment to the pial surface and also

found mutated in a cobblestone brain malformation (Radmanesh et al., 2013). Thus, mutations in laminin subunit genes, as well as glycosyltransferases, may both lead to detachment of RGC processes from the pial surface leading to breaches of the BM, disintegration of the scaffold mediating neuronal migration, subcortical heterotopias and neuronal overmigration phenotypes. These studies hence reveal the progenitor origin of certain "neuronal migration" defects (but see also Moers et al., 2008, for intrinsic problems in migrating neurons, affecting potentially their ability to stop migrating).

Similarly, the *GPR56* gene encodes a heterotrimeric guanine nucleotide-binding adhesion protein (G-protein)-coupled receptor that is highly expressed in progenitors, is localized to the basal process, and binds ECM proteins at the pial surface (Li et al., 2008; Luo et al., 2011). Disruption of

*GPR56* was found to selectively and bilaterally perturb the human cortex surrounding the Sylvian fissure with a strikingly restricted polymicrogyria (Bae et al., 2014). Loss of *GPR56* disrupts RGCs pial anchorage and causes breaks in the BM, through which some neurons over-migrate (Li et al., 2008; Bahi-Buisson et al., 2010). Studies involving this gene hence link these phenotypes to mechanisms leading to polymicrogyria. Moreover, even when the pia is intact, as observed in *Gpr56* knockout mice, cortical thickness and organization are irregular with periodic thinner regions. Such defects suggest proliferation problems and indeed in these mice there are less mitotic progenitors in both the VZ and SVZ at embryonic day (E) 14.5. Conversely, in mice carrying a transgene that overexpresses human *GPR56*, the opposite effect was observed (Bae et al., 2014). These data show that disruption of basal processes and overmigration, can be intimately linked with a perturbation of proliferation.

RGC basal processes have been postulated to regulate progenitor proliferation via integrin signaling (Radakovits et al., 2009; Fietz et al., 2010, see Section Apical Adhesive Interactions and Mechano-Transduction, Shedding Light on PH and Ciliopathies). *GPR56* influences attachment to ECM proteins, such as collagen type III, and tetraspanins, which are known to also bind integrins expressed by basal endfeet (Xu and Hynes, 2007; Li et al., 2008; Luo et al., 2011). Studies in conditional β1-integrin knockout mice showed a wavy appearance of cortical layers at E15.5, indicative of defects in the organized laminar cytoarchitecture and abnormal positioning of cortical neurons. Neurons either invaded the marginal zone (MZ) or accumulated deep in the cortical wall, resembling cobblestone lissencephaly. RGC fibers in the mutants terminated at varying positions close to the BM and were highly irregular (Graus-Porta et al., 2001). All these observations, together with the finding, by conditional ablation specifically in neurons, that β1-integrins are not essential for neuron-glia interactions and neuronal migration *per se* (Belvindrah et al., 2007), indicated that they are likely to primarily be required for anchorage to the BM. Defects similar to those found in β1-integrin knockout mice were also found in mice with mutations in the genes encoding the integrin α6-subunit or both α3 and α6 (Georges-Labouesse et al., 1998; De Arcangelis et al., 1999; Hynes, 2002); other components of the BM (Miner et al., 1998; Halfter et al., 2002; Pöschl et al., 2004); and the integrin downstream effectors focal adhesion kinase (FAK; Beggs et al., 2003) and integrin-linked kinase (ILK; Niewmierzycka et al., 2005).

Studies involving β1-integrin, the small GTPase RhoA and the protein Marcks (Myristoylated alanine-rich substrate protein) also highlight the fact that a number of proteins are likely to have a role in both apical and basal processes. The conditional deletion of RhoA and Marcks in the developing mouse cortex leads to a prominent tissue mass (heterotopia) found underneath an apparently layered but thinner cortex. Moreover, in the case of RhoA deletion, there was a phenotype reminiscent of cobblestone lissencephaly. Analysis of progenitor morphology also revealed already at E12.5, that mitotic cells were scattered in the cortex instead of being neatly aligned at the ventricular lining. At E16 mitotic cells were assembled into a broad band located abnormally in the middle of the cortex between the pial and the ventricular surfaces. Moreover, RGCs had mis-oriented processes and had lost their apical anchoring. While RhoAdepleted neurons still migrated fairly normally in a wild-type

environment, they followed a largely non-radial path when the RGC scaffold was disturbed by RhoA depletion (Cappello et al., 2012). RhoA plays a role in polymerizing actin into fibers (Factin) (Etienne-Manneville and Hall, 2002). Thus, loss of RhoA destabilized the actin and microtubule (MT) cytoskeletons in both neurons and RGCs but the most severe consequences were in RGC positioning and in the proper formation of the basal scaffold (Cappello et al., 2012). Knockout mice for Marcks, an actincross-linking protein and subcellular substrate of protein kinase C (PKC), also presented a disorganized RGC scaffold, impaired cell polarity, disorganized VZ, and ectopic progenitors. Marcks is a potent upstream regulator of the localization and function of cell polarity complexes, and its mutation leads to disrupted VZ organization and mitotic orientation (Blackshear et al., 1997; Weimer et al., 2009). At E15.5, the RGC scaffold was severely disrupted and basal end-feet failed to branch appropriately and instead had a club-like, balled-up appearance as they reached the pial surface. There was a reduction in the thickness of both the VZ and SVZ, and perturbed migration. Phenotypes combining apical and basal perturbation thus show misplaced progenitors, perturbed proliferation and either inability of neurons to reach the cortex or overmigration. In the different models whether the basal detachment causes the detachment of the apical process, or *vice versa*, is not always clear. These studies also emphasize that the cytoskeleton is essential for the maintenance of RGC structure.

#### **APICAL ADHESIVE INTERACTIONS AND MECHANO-TRANSDUCTION, SHEDDING LIGHT ON PH AND CILIOPATHIES**

Perturbations of the apical domain of RGCs, a complex and relatively well-studied cell compartment, are becoming recognized as leading to neuroepithelial disorganization, different types of heterotopia (this section), or micro- and macrocephaly (Section RGCs and Cell Division, Mechanisms Leading to Microcephaly), depending on the affected cell mechanisms. The consequence of the perturbation can appear as obvious breaks in the ventricular lining and changes in VZ architecture, or be more subtle. To help explore these phenotypes, we further focus in this section on molecules which have been described to play a role in intercellular and ECM contacts in the VZ.

We re-mention here β1-integrins which have also been shown to be involved in end-feet anchoring to the ventricular surface, binding with laminins, located also in the apical ECM. These anchors are reinforced by cadherin–catenin-based adherens junctions, which help attach apical end-feet of adjacent RGCs to each other (Kadowaki et al., 2007). Blocking β1-integrin's function by injection of specific antibodies into the lateral ventricles of embryos at E12.5 and E15.5 showed a significant increase in dividing cells due to a larger abventricular (localized outside the VZ) dividing progenitor population (Loulier et al., 2009). This phenotype was linked to the detachment of apical processes from the ventricular surface and alterations in mitotic spindle orientation showing that β1-integrin plays a critical role in the adhesion that maintains the progenitor cells within their niche and preserves the architecture of the VZ. The same effect was observed in laminin-α2 deficient mice (Miyagoe et al., 1997; Loulier et al., 2009). Another link between apical VZ integrity and heterotopia formation is represented by the conditional deletion in the cortex of the apical junction molecule α-E-catenin. Due to the disruption of RGC morphology, caused by impaired actin cytoskeletal organization, progenitors were found disorganized in rosette-like structures, associated with a large heterotopia and a thin layered cortex (Schmid et al., 2014). Another component of adherens junctions is Afadin, which interacts with cadherins and stabilizes them (Sato et al., 2006). Its conditional inactivation in the developing cortex leads to disruption of adherens junctions, dispersion of dividing progenitors with a shorter cell cycle and reduced cell cycle exit, and formation of a double cortex-like phenotype (Gil-Sanz et al., 2014). In the same study, conditional knockout mice for a cadherin subunit, *Cdh2*, showed a very similar phenotype, close also to that described previously for RhoA-knockout mice (Cappello et al., 2012). Thus intercellular contacts and downstream pathways seem to be essential to maintain the integrity of the VZ and to regulate proliferation.

Similarly, the Eph/ephrin signaling pathway (Nievergall et al., 2012) activates signal transduction cascades, and exhibits extensive cross-talk with other receptors, including cadherins and integrins (Arvanitis and Davy, 2008). Ephrin B1 is expressed in apical progenitors from the neuroepithelial stage in a ventricular-high to pial-low gradient (Stuckmann et al., 2001). Prolonged Ephrin B1 activity was shown to prevent progenitor differentiation, while loss of function had an opposite effect promoting differentiation and leading to loss of progenitor cells (Qiu et al., 2008; Murai et al., 2010). Indeed, Ephrin B1 reverse signaling controls the switch between progenitor maintenance and neuronal differentiation (Arvanitis et al., 2010). In knockout mice, and more severely in heterozygous mice, the neuroepithelium had an irregular appearance with formation of micro-invaginations due to abnormal folding of the VZ without changes in apico-basal polarity of progenitors. Progenitor detachment was also observed. In absence of Ephrin B1, local alterations of the apical surface might weaken the rigidity and cohesion of the neuroepithelium. The reason why in heterozygous embryos the phenotype is more severe might be that sorting between Ephrin B1-positive and Ephrin B1-negative cells leads to discontinuous rigidity, which is more detrimental to the morphogenesis of this tissue than a homogeneous decrease in rigidity. Thus, the normal function of Ephrin B1, via an interaction with EphB2 on neighboring cells, is to maintain morphology and localization of progenitors in the VZ by promoting apical integrin-based adhesion (Arvanitis et al., 2013).

Mutations in the Filamin A (*FLNA)* gene were found in 100% of families with X-linked bilateral PH and in 26% of sporadic patients with PH (Fox et al., 1998; Parrini et al., 2006). β1 integrin mediated adhesion to the ECM was also found to be dependent on the binding of FLNA to vimentin and PKC epsilon (PKC1, Kim et al., 2010) allowing vimentin phosphorylation by PKC1. This step is crucial for the activation and trafficking of β1-integrin to the plasma membrane. *FLNA* encodes a large phosphoprotein that crosslinks actin filaments into orthogonal networks, reorganizing them by interacting with several proteins at the membrane (Stossel et al., 2001; Nakamura et al., 2007). It may play a role both in progenitors and migrating neurons. The ventricular surface has been shown to be disrupted in FlnA knockout mice (Feng et al., 2006). PH formation and alterations in the neuroepithelial lining were also shown in FlnA-knockdown brains where disruption of both the polarized RGC scaffold and the neuroepithelial lining were the likely cause of the PH (Carabalona et al., 2012). Also, loss of mitogen-activated protein kinase kinase kinase 4 (MEKK4) in mice, a regulator of FlnA phosphorylation, leads to a similar phenotype (Sarkisian et al., 2006). A second human PH gene, *ARFGEF2*, coding for brefeldin-A-inhibited guanine exchange factor-2 (BIG2) is likely to play a role in endocytosis, regulating levels of Arf1 at the plasma membrane, which is known to regulate cell-cell contacts (Zhang et al., 2013). PH was also induced by knockdown of C6orf70 in the developing rat cortex, a gene of unknown function, mutated in a PH patient, and coding for a protein with a vesicle-like subcellular localization (Conti et al., 2013). These combined data suggest a coordinated role for actin and vesicle trafficking in controlling cell adhesion in apical regions in the VZ (Sheen, 2014).

Related to this, protocadherins Dchs1 and Fat4 (Cappello et al., 2013) are important for an apically located adhesive complex (Ishiuchi et al., 2009). Van Maldergem syndrome, an autosomalrecessive multiple malformation syndrome, shows a partially penetrant PH phenotype caused by mutations in *FAT4* or *DCHS1*. Dchs1 is the ligand of the Fat4 receptor and the complex they constitute is situated apically, closer to the ventricle relative to adherens junctions. Fat4 and Dchs1 knockdown studies in mice also showed an increased cell proliferation in the VZ and SVZ, a block in differentiation between the Pax6+ and Tbr2+ states, and an accumulation of neuronal precursors, showing that this adhesive complex normally suppresses continued proliferation (Cappello et al., 2013). Adhesion and proliferation hence seem to be interlinked themes related to these phenotypes.

Another very important characteristic of the apical domain of RGCs is the presence of the primary cilium, an MTbased, slender projection from the cell that is thought to be important for sensing signaling factors present in the cerebrospinal fluid, and with a guiding role in the establishment of apical-basal polarity of the RGC scaffold. The importance of cilia function for cortical development is evident in developmental brain disorders such as Joubert, Meckel-Gruber, orofaciodigital and Bardet-Biedl syndromes (commonly referred to as ciliopathies), where disrupted cilia and the resulting changes in cortical formation may underlie cognitive deficits and intellectual disability (Cantagrel et al., 2008). Mutations in a gene encoding the centrosome-associated protein CEP290, important for ciliogenesis (Kim et al., 2008), have been found in both Meckel-Gruber and Joubert syndromes (Valente et al., 2006; Frank et al., 2008). Also, Arl13b, a small GTPase of the Arf/Arl family that is mutated in Joubert syndrome, is specifically localized to cilia and controls the MT-based, ciliary axoneme structure. Deletion of Arl13b impairs the cilium's ability to convey critical extracellular signals such as Shh (Caspary et al., 2007). In constitutive mutant mice, and in E9 conditional knockout mice, early neuroepithelial progenitors showed markedly perturbed polarity with the soma located near the pial surface and the basal end-feet located near the VZ. These cells divided ectopically at or near the pial surface, instead of adjacent to the ventricular surface (Higginbotham et al., 2013). These studies show that primary cilia play an important role in both signal transduction and polarity.

Indeed, RGC polarity is a crucial issue for cortical development. We resume this only briefly here (see details in **Table 2** and **Figure 3**). Conditional mutagenesis in the mouse or focal knockdown experiments, have often been necessary to reveal the role of a particular polarity protein, in this case it remains difficult to directly link these data with malformations. What clearly comes out of the different studies is the relationship between apical polarity complexes (Par-Complex and its regulators), maintenance of the structure of the ventricular lining and neuroepithelium, and regulation of cell proliferation/differentiation and fate. Thus, defects in the polarity complexes have been studied both prior to neurogenesis and during the neurogenic period. Changes in the balance between expansion of RGCs, production of basal progenitors, and differentiation of post-mitotic neurons have been identified but are still little-understood. This imbalance can be the cause of the incapacity of the brain to form an ordered layer structure and/or a brain of the correct size (see also Section RGCs and Cell Division, Mechanisms Leading to Microcephaly). Diverse mechanisms can be affected by the perturbation of different polarity molecules, related to the complex interactions that link the different players. The variable consequences are also likely to be due to the different importance these molecules have during neuroepithelial progenitor expansion and/or the neurogenic phase when RGCs have a major role. Also, cell adhesion complexes, strictly related to polarity components, change during the transition from early neuroepithelial cells to RGCs (Götz and Huttner, 2005), adding to the complexity. Mechanisms leading to hydrocephaly identified in some mutant mice remain complex, however, disruption of the early neuroepithelium and polarity changes are clearly associated.

#### **RGCs AND CELL DIVISION, MECHANISMS LEADING TO MICROCEPHALY**

We previously discussed how cell junctions and the integrity of apical polarity domains are important for regulating the structure of the ventricular lining and the balance between proliferation and differentiation. However, there are other cellular mechanisms that are more solely linked with proliferation/differentiation and cell fate. RGC centrosome behavior, mitosis, the regulation of spindle orientation (which has effects on symmetric and asymmetric division and cell localization), cytokinesis and interkinetic nuclear migration (INM) are finely regulated processes, and several cortical malformation genes or mouse mutants associated with these mechanisms have been studied. We have classified these phenotypes within separate sub-categories, but these can still often be considered as overlapping.

#### **Microcephaly genes and centrosome function**

The centrosome is important for correct spindle assembly and function during mitosis. Centrosomes influence the morphology of the MT cytoskeleton, function as the base for the primary cilium and integrate important signaling pathways.

At the core of a typical centrosome are two cylindrical MTbased structures termed centrioles, which recruit a matrix of associated pericentriolar material (Nigg and Stearns, 2011). RGC centrosomes are located at the extremity of the apical process and are aligned at the ventricular surface. This position influences cell polarity and anchoring in the VZ. Once a cell enters mitosis, centrosome duplication takes place and these move more basally to help form the spindle poles and the bipolar mitotic spindle. A set of proteins related to centrosome behavior has been identified, whose mutation was found to cause microcephaly. Mutations in ASPM (abnormal spindle-like microcephaly associated) are the most common cause of primary microcephaly in humans (Kumar et al., 2004; Pichon et al., 2004; Shen et al., 2005; Gul et al., 2006). Aspm has been shown to exert a critical role at the spindle poles of neuroepithelial cells, maintaining spindle position during mitosis and, consequently regulating the precise cleavage plane orientation required for symmetric, proliferative divisions (Fish et al., 2006). Microcephalin (*MCPH1*) mutations also cause primary microcephaly type 1 (Woods et al., 2005). *Mcph1* is expressed at high levels in the VZ and SVZ at E13.5 and E15.5 (Gruber et al., 2011) and *Mcph1*-deficient mice have a small brain. Characterization of the mutant cortex revealed premature production of neurons and exhaustion of progenitors. Mcph1 deficiency specifically caused a delayed and imbalanced centrosomal maturation, leading to a lengthening of the cell cycle due to abnormal spindles and chromosome misalignment (Gruber et al., 2011). Another example of a gene mutated in primary microcephaly is SCL-interrupting locus protein (*STIL*), encoding a centriole-duplication factor that localizes to the procentriolar cartwheel region, a key structure in procentriole assembly. STIL depletion was shown to completely block centriole formation, whereas its overexpression resulted in extensive centriole amplification (Arquint and Nigg, 2014).

Human primary microcephaly is also caused by mutations in *CDK5RAP2* (cyclin-dependent kinase 5 related activator protein 2, Bond et al., 2005). In somatic cells, CDK5RAP2 promotes centrosomal cohesion (Graser et al., 2007) and recruits the γ-tubulin ring complex (γ-TuRC)—the MT nucleator—to the centrosome (Fong et al., 2008). In a homozygous mouse model of *Hertwig's anemia (an)*, the disease is caused by a mutation in *Cdk5rap2* (Lizarraga et al., 2010). Brain size was reduced and an increased ventricular size and decreased cortical thickness were already detected at E13.5. Mutant animals had fewer total neurons and the last-born superficial neurons were particularly reduced. The premature decrease in progenitors was due to problems encountered during mitosis causing cell death affecting both progenitors and neurons and, possibly, changes in cell fate. Indeed, an increase in pro-metaphase and metaphase precursor cells with mono-, tri-, and other aneupolar spindle poles, together with defective spindle orientations, were detected (Lizarraga et al., 2010). Mutations in centromere protein J (CENPJ) also cause microcephaly (Bond et al., 2005). This gene is involved in the maintenance of centrosome and spindle integrity. A recent study described conditionally inactivated Cenpj also known as SAS-4 (Insolera et al., 2014). This led to mitotic delay, p53 activation and cell death of delocalized progenitors. Keeping cells alive by p53 inactivation showed many RGCs in the IZ, which were multipolar but could still divide, self-renewing and producing also basal progenitors and neurons. Under these conditions small heterotopias formed in the IZ. This study showed that cell death was not due to aneuploidy or other chromosomal abnormalities, unlike hypomorphic Cenpj mutants (McIntyre et al., 2012), instead delocalized RGCs were often remarkably deficient in centrioles and cell death was most probably due to mitotic delay.

Centrosome amplification may also cause microcephaly by affecting the correct formation of the spindle and continuation through mitosis. Polo-like kinase 4 (Plk4) is a centriole duplication protein whose overexpression leads to cells with supernumerary centrosomes. Conditional overexpression of Plk4 specifically in progenitors, led to reduced brain size, accompanied by a reduction of both apical and basal progenitors, and the neuronal population. In Plk4 overexpressing embryos, cells with extra centrosomes showed bipolar, as well as multipolar, spindle configurations, and spent more time in mitosis. This was at the origin of p53-dependent cell death and could be one of the main causes of brain reduction in this model. Deletion of p53 showed accumulation of aneuploid daughter progenitors, and these underwent premature neuronal differentiation, with subsequent depletion of the progenitor population (Marthiens et al., 2013). Related to this work, another two MCPH proteins, CEP63 and CEP152, form a complex that is an essential part of the molecular machinery controlling centrosome numbers, and defects in either component result in a diminished pool of precursors that cannot provide an adequate supply of neurons (Sir et al., 2011). Thus centrosome formation, numbers, maturation and function are all important for maintaining a correct progenitor number.

#### **Spindle genes and mitosis**

The cell cycle of RGCs is characterized by an oscillatory behavior called INM. Mitosis occurs apically close to the ventricular surface, nuclei then migrate basally during G1 to reach the most basal side of the VZ where they undergo S-phase, and migrate apically during G2 to reach the ventricular surface before undergoing mitosis again. This behavior of the nuclei of RGCs gives the VZ the appearance of a pseudo-stratified epithelium. A variety of molecules (Kif1a, Dynein (Tsai et al., 2010); Lis1 (Cappello et al., 2011); Tag-1 (Okamoto et al., 2013); Rnd3 (Pacary et al., 2013); Dock7, Tacc3 (Yang et al., 2012); SUN-KASH protein complex (Zhang et al., 2009); Tpx2 (Kosodo et al., 2011)) have been reported to play a role in this process, although since no cortical malformation in human has been shown to our knowledge to be caused directly by abnormal INM (but see discussion below for dynein and LIS1, and Asp in *Drosophila* (Rujano et al., 2013)), we do not mention them further here. We focus instead on mitosis itself and division occurring at the ventricular lining.

Even if the mechanisms of mitosis are still not clear, the orientation of the mitotic spindle was previously linked with symmetric or asymmetric modes of cell division and, consequently, also with the progenitor state or cell cycle exit. This remains a little-understood area. Mitotic division planes are coordinated with the polarized expression of cell fate determinants such as Numb, β-catenin, Par3 and Notch (Zhong et al., 1996; Chenn and Walsh, 2002; Bultje et al., 2009). In order to be RGCs, daughters of dividing progenitors need to inherit both the apical and the basal attachments, this is favored when the spindle is oriented parallel to the ventricular lining with a cleavage plane that bypasses both the apical and basal domains (Taverna et al., 2014). The molecular mechanisms that govern the mode of cell division in RGCs are still not clear (Knoblich, 2001; Lancaster and Knoblich, 2012). Orientations of the spindle other than parallel may favor asymmetric divisions and the generation of neurons or basal progenitors, which do not inherit the apical attachment, and migrate to the SVZ to undergo a symmetric final division and generate two neurons (Postiglione et al., 2011). However, other studies concern models in which ectopic RGC progenitors result from perturbations of spindle orientation and the unequal inheritance of apical attachment sites upon division, with the retention however, of the molecular signature of apical progenitors (Konno et al., 2008; review by Lancaster and Knoblich, 2012; Kielar et al., 2014). This suggests that the primary role of planar spindle orientation in apical divisions is to maintain daughter cells attached to the ventricular surface, but not directly to influence the choice between symmetric and asymmetric outcomes (Peyre and Morin, 2012). The size of the apical domain corresponds to only 1– 2% of the total membrane surface. This is why minor changes in spindle orientation may decide whether the cleavage plane would dissect or bypass the small apical domain and result in its equal or unequal repartition and the distribution of cell fate determinants between the daughter cells (Kosodo et al., 2004; Marthiens and ffrench-Constant, 2009; Peyre and Morin, 2012). Moreover, defects in mitotic spindle assembly, dynamics and function have often been linked with mitotic delay, changes in cell cycle length and, consequently, of daughter cell fate. The cell cycle length of wild-type progenitor cells increases from 8.1 h at E11 to 18.4 h at E17 in mouse embryos. In contrast, the period of the G2/M-phase, is very rigidly controlled and remains constant at 2 h throughout brain development (Takahashi et al., 1995; Sakai et al., 2012). Therefore, altering M-phase progression is likely to influence cell survival and fate determination. Thus, even if the role of spindle orientation in cell fate and mode of division are not clear, its timely and mechanistic regulation are finely controlled processes, and mutations have been found in several genes which severely perturb the formation of the cortex, often causing different versions of microcephaly.

Interestingly, Huntingtin (Htt), the protein whose mutation leads to Huntington's disease (HD), is one such gene. During mitosis, Htt was found specifically located at the spindle poles and at the spindle midzone (Godin et al., 2010). Htt was shown to control spindle orientation by ensuring the proper localization of several key components of the spindle and, as a consequence, its position. The MT-dependent transport of the dynein/dynactin complex to the spindle was reduced in Htt-depleted cells, and the localization of Protein Numa1 (NuMA) was modified. In mammalian cells, NuMA by assembling with dynein/dynactin is essential for the organization of MTs at the spindle pole (Merdes et al., 1996; Fant et al., 2004) and the regulation of astral MT interactions with the cell cortex (Du and Macara, 2004). Depletion of Htt by RNAi in progenitors*in vivo* led to spindle misorientation and promotion of premature neurogenesis (Godin et al., 2010; Molina-Calavita et al., 2014).

Spindle orientation is regulated by the interaction of astral MTs with the cellular membrane, and the polymerization of MTs directed toward the chromosomes assures their proper segregation. Related to this, mutations in the Treacher Collins Syndrome Treacle Protein (*TCOF1*) gene cause Treacher Collins Syndrome (TCS), which, amongst other defects, is associated with microcephaly. *TCOF1* codes for a nucleolar phosphoprotein known as Treacle (The Treacher Collins Syndrome Collaborative Group, 1996). *Tcof1* heterozygous mice exhibited considerable brain hypoplasia, with a reduced RGC population and cells already committed to neuronal fate. Vertical cleavage planes in dividing RGCs were found dramatically reduced showing that Treacle is important for correct spindle orientation. This was accompanied by an extension of M-phase and mitotic delay. Treacle was found to localize to centrosomes of RGCs during interphase and in mitotic cells, it co-localized with CENP-E at the kinetochore, and was also found at the midzone in anaphase cells and the midbody in telophase. In *Tcof1* knockdown cells, mitotic spindles were found disorganized, and chromosome assembly at the metaphase plate incomplete, suggesting roles for the Treacle protein in chromosome movement and spindle formation. Loss of Plk1 function, which phosphorylates Treacle, also resulted in perturbation of mitotic spindle orientation and mitotic delay (Sakai et al., 2012).

Thus, multiple human microcephaly proteins can take part in the assembly of the mitotic MT structure and its dynamics (Bond and Woods, 2006; Fish et al., 2006; Sun and Hevner, 2014; Valente et al., 2014). However, further similar function proteins seem also important for cortical layering. WD repeat-containing protein 62 (*WDR62*) encodes a centrosome- and spindle poleassociated protein in which mutations cause microcephaly with simplified gyri and abnormal cortical architecture (Bilgüvar et al., 2010; Yu et al., 2010). WDR62 accumulated strongly at the spindle poles during mitosis and the murine version, Wdr62 was found expressed in the neuroepithelium exclusively in apical precursors undergoing mitosis at the ventricular surface (Nicholas et al., 2010). Also, centromere-associated protein E (*CENPE*), the gene coding for centromere-associated protein E was found mutated in patients with microcephalic primordial dwarfism (MPD), featuring microcephaly and a simplified gyral pattern (OMIM 616051). Mutations in CENPE were shown to alter spindle dynamics and chromosome segregation leading to delayed mitotic progression (Mirzaa et al., 2014). CENPE is a core kinetochore component functioning initially to mediate the bringing together of misaligned chromosomes, and subsequently to capture spindle MTs during mitosis (Abrieu et al., 2000; Yao et al., 2000). Indeed, the stable propagation of genetic material during cell division depends on the congression of chromosomes to the spindle equator before the cell initiates anaphase (Kapoor et al., 2006). A replicated chromosome possesses two discrete, complex, dynamic, macromolecular assemblies, known as kinetochores that are positioned on opposite sides of the primary constriction of the chromosome. Proper chromosome congression depends on MT bundles (K fibers) that connect sister kinetochores of each chromosome to opposite spindle poles (Rieder and Salmon, 1998). CENPE is clearly involved in these processes, although the reason why layering is also affected with CENPE (or WDR62) mutations still remains unclear.

Similarly, *NDE1* is one of the known spindle-associated genes and mutations also cause a severe microlissencephaly syndrome that reflects both morphological and quantitative defects in RGCs. In apical cells, Nde1 was found enriched at the centrosome in interphase and early mitosis and then reduced during metaphase and telophase during which it was present at the mitotic spindle and at the level of kinetochores. NDE1 was shown to be important for normal mitotic spindle function (Alkuraya et al., 2011; Bakircioglu et al., 2011). Nde1 knockout mice showed a small-brain phenotype from birth (Feng and Walsh, 2004). The thinning of the cortex in these mice was much more pronounced in superficial cortical layers, which are formed near the end of neurogenesis. Mitotic spindle defects were described to result in mitotic delay/arrest and shifted orientation towards horizontal cleavage. Nde1 self-associates and has a scaffolding function in mitotic spindle assembly. Blocking its self-association induced defective centrosomal duplication, and this defect was at least partially responsible for observed spindle mis-assembly (Feng and Walsh, 2004). NDE1 is also a critical binding partner of LIS1 (Feng et al., 2000), a gene causative of neuronal migration defects and type I lissencephaly (Dobyns et al., 1993; Sicca et al., 2003). Nde1 null and Lis1 heterozygous double mutant mice showed not only a thinner but also a severely disorganized cortex where all the distinct cellular layers were lacking and reduced numbers of radial neuronal units were caused by loss of progenitors in early ages due to failed mitotic spindle function (Pawlisz et al., 2008). Lis1 is also a cytoplasmic scaffold protein that functions as an adaptor that controls the organization of the MT cytoskeleton and MT-associated motors, confirming their importance for spindle orientation (Faulkner et al., 2000; Yingling et al., 2008). Indeed, mutant RGCs were able to establish apical junctions and overall polarity, but failed to maintain apical cell shape and intimate association with the ventricular surface in particular during mitosis (Pawlisz et al., 2008). Thus, Nde1-Lis1 is essential for mitotic orientation determination, but also critically required for maintaining apical cell integrity and lateral contacts of RGCs during mitosis, showing that the polarity and morphology of metaphase progenitors must be co-regulated with mitotic spindle orientation for correct neuron number and organization (Pawlisz et al., 2008).

It is still unclear how LIS1 works in the human cortex, and if heterozygote gene dosage defects found in human lissencephaly patients affect more neuronal migration or progenitor proliferation. However, a role of Lis1 in spindle orientation was confirmed by studies in conditional knockout mice (Yingling et al., 2008). Complete Lis1 loss was found to have a deleterious effect early in development during symmetric divisions of neuroepithelial stem cells, however its loss specifically in RGCs was also shown to give rise to a thinner cortex with a less-organized structure. In the same study Lis1 was found to be important for localization of its binding partners Nde1-like (Nudel or NDEL1), dynein, and CLIP-170, and this localization was important for MT stability and capture at the cell cortex (Yingling et al., 2008; Moon et al., 2014). Related to this, protein phosphatase PP4c is required for proper asymmetric cell division in *Drosophila* neuroblasts, and conditional knockout mice were found to have thinner and disorganized cortical layers again partly related to spindle orientations that favor progenitor exhaustion. Indeed, PP4c can dephosphorylate Ndel1 and regulate its interaction with Lis1. Excessive phosphorylation of Ndel1 upon PP4c loss leads to disruption of the Ndel1/Lis1 complex and subsequent spindle orientation defects (Xie et al., 2013). As mentioned above, dynein, together with Kif1a and Lis1, has also been studied in progenitors in relation to mechanisms governing nuclear translocation during INM (Tsai et al., 2005, 2010). Recently mutations in *DYNC1H1*, encoding dynein heavy chain, together with *TUBG1*, *KIF5C* and *KIF2A*, have been associated with complex cortical malformations and microcephaly (Poirier et al., 2013), however, the mechanisms giving rise to these cortical malformations are still unclear. Even if they have not been found mutated in cortical malformation patients, other genes (e.g., AGS3, Vangl2, **Table 2**) have been described to play a role in mitotic regulation of RGCs and have been studied in mouse models with similar phenotypes (Montcouquiol et al., 2003; Sanada and Tsai, 2005; Lake and Sokol, 2009). These combined data underline a critical role of MTs and associated proteins during RGC mitosis, maintenance of RGC morphology, as well as during neuronal migration.

Another protein found to regulate spindle orientation, this time favoring basal progenitor production, is *Inscuteable* (Kraut et al., 1996). The mouse homolog, mInsc, is enriched at the spindle midzone in anaphase cells. Gain and loss of function experiments showed that when mInsc was knocked down the cortex was thinner, whilst mInsc overexpression led to a thicker cortex (Postiglione et al., 2011). A similar example is the Wnt3a overexpression mouse model (Munji et al., 2011). Loss of function mutations of Wnt3a lead to dramatic disruption of cortical development (Lee et al., 2000). Overexpression of Wnt3a in RGCs caused the formation of ectopic neuronal rosettes adjacent to the ventricle, and a dramatic increase in Tbr2 positive cells present in disorganized clumps or organized in rosettes in an expanded SVZ adjacent to a heterotopic neuronal mass (Munji et al., 2011, see also Schmid et al., 2014 for a similar phenotype). There was also RGC hyper-proliferation and unusual rosette organization, increasing the thickness of the VZ, similar to that observed in Lgl1 mutants (Klezovitch et al., 2004; **Table 2**), making a further link hence to polarity and adhesion complexes. Cdc42 deletion in RGCs also caused an increase in the number of basal progenitors (**Table 2**), but in this case the consequence was an increased neurogenesis and the formation of a thicker cerebral cortex, without the formation of rosettes or heterotopia (Cappello et al., 2006). Indeed, mechanisms determining the different progenitor outcomes in these cases are not yet elucidated.

Most genes mentioned so far in this section make the relationship between defects in spindle function and a depletion of RGC progenitors, associated with either premature differentiation, increased basal progenitors, or increased cell death. However, there are also models where spindle misorientation is found concomitant with misplaced progenitors that remain Pax6-positive and potentially maintain the ability to produce all cell types in an ectopic position. Several mutants with detached Pax6 progenitors, in which spindle orientation was not necessarily previously studied, were also mentioned in the previous sections. If detached Pax6-positive cells survive, this mechanism can either lead to a thinner, or a thicker cortex, or to subcortical heterotopia where ectopic masses of cells remain present in the white matter, but the overall apical-basal architecture of the cortex seems to be maintained. We discuss here certain mutants which can help us begin to understand such phenotypes. In Pax6 mutants, altered spindle orientation and cleavage planes in RGCs resulted in a markedly unequal inheritance of the ZO1-labeled adherens junction components and the apical membrane domain enclosed by these. Non-apical cell divisions were found increased in the mutant cortex, and most of the basally dividing cells retained RGC hallmarks consistent with premature delamination (Asami et al., 2011). PAX6 mutations in patients lead to aniridia and complex malformations (OMIM 607108). Also, *LGN* (G-protein-signaling modulator 2, GPSM2) codes for a G-protein regulator that links the cell cortex and mitotic spindles (Du et al., 2001; Du and Macara, 2004). LGN protein was found concentrated on the apical side of the VZ and localized to the lateral cell cortex in dividing apical progenitors (Konno et al., 2008). In *LGN*mutants, mitotic orientations of progenitors were essentially randomized at E10.5 and E14.5, and Pax-6- and Tbr-2-positive cells were scattered into the SVZ and IZ. Non-surface apical progenitors were formed at the expense of attached RGCs, causing a decrease of approximately 30% in the thickness of the VZ. The average length of the progenitors' cell cycle and the production of neurons were unchanged in the mutant (Konno et al., 2008). Thus, changes in spindle orientation can cause detachment and misplacement of progenitors without apparently changing their identity. Although this model does not give clues about an overall resulting malformation, another similar model displaying ectopic Pax6-positive progenitors scattered in the VZ and IZ, is represented by the *HeCo* (*Heterotopic Cortex*) mouse (Croquelois et al., 2009). The gene mutated in this model is *Eml1*, coding for an MT-associated protein whose function in brain development is not known. *EML1* was found mutated in patients affected by a very severe form of periventricular and globular ribbon-like subcortical heterotopia, and *HeCo* mice show a similar phenotype to band heterotopia (Kielar et al., 2014). The spontaneously arisen *tish* rat model (Lee et al., 1997), and BXD29 mouse mutants (Croquelois et al., 2009; Rosen et al., 2013), with unknown mutations, also show a similar subcortical heterotopia. Certain other mutants mentioned above, e.g., conditional knockout of RhoA (Cappello et al., 2012) or overexpression of mInsc or Wnt3a, also have subcortical neurons, although they show either a severe displacement of the VZ and defects in the ventricular lining, or greatly increased numbers of basal progenitors, and hence causative mechanisms may not to be identical. In the case of the *HeCo* mouse, like LGN mutants, the VZ is largely intact although slightly reduced in thickness and only a proportion of Pax6-positive progenitors show a re-distribution into the SVZ and IZ (Kielar et al., 2014). Tbr2-positive basal progenitors are also found ectopically, although do not differ in overall number in the *HeCo* model. Human patients with mutations in EML1 also exhibit macrocephaly, in some cases associated with hydrocephaly. The analysis of the *HeCo* model showed defects in mitotic spindle orientation during the neurogenic period that might be the cause of the detachment of some apical progenitors from the VZ, which is likely to be the primary event which eventually leads to heterotopia formation in this case (Kielar et al., 2014). Perturbed RGC guides almost certainly contribute to the phenotype. It is hence clear from these and other data that subcortical heterotopia can arise via multiple mechanisms.

#### **Cytokinesis**

Another important step during progenitor division is cytokinesis, the final separation of the two daughter cells. RGCs divide using a polarized form of cytokinesis, which is not well understood. Cytokinetic furrowing starts on the basal side and ingresses toward the apical membrane, where the midbody is formed. Cytokinetic abscission is mediated by the midbody at the ventricle, only after the daughter nuclei have migrated away (Kosodo et al., 2008). The midbody is a structure that forms at the end of the furrow and it contains central spindle compacted MTs and other factors important for mediating abscission. Moreover, midbodies of progenitors have been shown, together with the primary cilium, to release extracellular membrane particles enriched in the stem cell marker Prominin-1, thus influencing the balance between proliferation and differentiation after division (Dubreuil et al., 2007).

Spindle functioning and cytokinesis are tightly linked processes, this explains why proteins which function at the spindle poles or at the spindle midzone are often found also at the midbody (e.g., dynein, Horgan et al., 2011). Although this has not been widely studied, another protein that seems to function at both spindle poles and the cytokinetic furrow is Aspm, involved in microcephaly. As mentioned above, Aspm has been shown to be involved in regulating the precise cleavage plane orientation required for symmetric, proliferative divisions (Fish et al., 2006). This protein was also found enriched at the midbody of neuronal progenitors and thus, it was hypothesized to coordinate spindle rotation with cell abscission. Another example is the *magoo* mouse mutant, carrying a recessive, perinatal lethal mutation in the *Kif20b* gene, with fully penetrant microcephaly (Janisch et al., 2013). The thickness of the cortex is reduced in these mutants, but the layered structure is preserved. The output of progeny by apical progenitors is greatly reduced, but their capacity to produce daughters with ordered layer fates is intact. Kif20b protein was detected in cytokinetic midbodies of progenitors at the ventricular surface, where it is thought to regulate midbody behavior by transporting different cargoes. In the *magoo* mutant, midbodies had an abnormal shape and appeared misaligned with respect to the ventricular surface, indicating defects in midbody formation, maturation or maintenance (Janisch et al., 2013). Thus, failures in abscission of progenitors are likely to have similar consequences as failures in mitotic spindle functioning, leading to microcephaly.

## **DISCUSSION**

We discussed in this review how different types of cortical malformation arise following perturbations in RGC structure and/or mechanisms (mouse mutant data schematically resumed in **Figure 2**). Due to the large number of players involved (**Figure 3**), the resulting phenotypes are sometimes difficult to classify into distinct categories and a certain degree of overlap remains. Indeed, mutations in a single gene can be causative of several distinct malformations, and conversely, single malformations can be linked to mutations falling in genes apparently involved in different pathways or mechanisms. Here we classified genes in areas where they have been shown to play a major role, making a link where possible, with resulting malformations. However, minimal modifications regarding one compartment are likely to have dramatic secondary effects. It is hence possible that some mutants may not yet have been explored enough in order to identify primary vs. secondary events.

#### **BASAL PROCESS**

We discussed how perturbations of the RGC basal process mainly lead to heterotopia phenotypes, polymicrogyria and type II lissencephaly. These perturbations can be caused by mutations that affect BM components and the attachment of glial endfeet to the pial surface, e.g., glycosyltransferases and laminins. The phenotypes can be related to a cause or a consequence of basal process detachment and disorganization, for example, neurons may not find an appropriate scaffold for migration and are prone to generate heterotopias. Also, similar phenotypes are revealed in the mouse due to mutations of proteins that have a more ubiquitous role in RGC structure suggesting that some ECM interactions in apical and basal regions are similar. Mutations in genes like *GPR56*, genes encoding integrin subunits or components of the integrin signaling system, genes like *RhoA*, and *Marcks*, not only disrupt the basal process but have a wider, and perhaps primary, effect influencing apical attachment and progenitor proliferation. Although we classified certain proteins as perturbing the basal process, since the malformations generated are consistent with this, it makes sense in the future to systematically examine both extremities, and at early timepoints. Similarly, the mechanisms leading to polymicrogyria are still under debate, but both the basal and apical attachments, and proliferation, are probably involved. Experiments aiming to film the detachment of RGC processes *ex vivo* may help to clarify the temporality and causality of apical and basal process defects.

#### **APICAL DOMAIN**

The apical domains of RGCs are even more complicated to dissect in terms of mechanisms leading to malformations. This domain is not only important for progenitor attachment to the ventricular surface, mechano-transduction, and maintenance of polarity, but it is also the place where primary cilium and junctional complexes between cells are located. Several proteins play a role in intercellular and ECM contacts in the VZ, and perturbations of their function mainly lead to ventricular lining breaks and PH. Sometimes these seem to be produced only by disorganization of RGCs, without altering the cell cycle and cell fate, such as for example after β1 integrin, laminin-α2, and FLNA mutations. More frequently, alteration of the junctional complexes involve proteins that are linked with signaling pathways and this not only produces disorganization but also affects cell cycle characteristics and the balance between proliferation and differentiation. Examples are mutations in Dchs1, Fat4, Mtll4 and Ephrin-B1. The primary cilium is a structure important for both mechano-transduction and signaling, and perturbations of this organelle lead to a special category of diseases called ciliopathies, which can include severe cortical disorganization, as mentioned in the case of Arl13b.

Disruption of proteins and complexes that form the apical polarity domain of RGCs have not been found in human cases of cortical malformations but studies in mice have clearly shown that they are important for the maintenance of ventricular lining integrity and the regulation of cell fate. Indeed, the αE- and β-catenin, Numb, Numbl, Par-complex components, their regulators such as ASPP2, and proteins such as Lgl1, are important to regulate the balance between progenitor expansion and neuron production (see **Table 2**). Moreover, the integrity of the VZ is often severely perturbed due to the close link these proteins have with adherens junctions and a potential role in the maintenance of their integrity. Cell biology studies could be further performed to dissect the exact role polarity proteins exert in the maintenance of adherens junctions between RGCs, this could help clarify why their mutations have consequences on cell fate. Indeed, a study in *Drosophila* neuroepithelium showed that Par-complex proteins act by regulating the endocytosis of molecules important for adherens junction stability (Harris and Tepass, 2008). This is an interesting area to pursue in the future.

#### **MITOSIS, SPINDLE FUNCTION AND CYTOKINESIS**

Due to the complexity of the mitotic machinery and the mitotic cycle, there are a number of steps potentially susceptible to perturbation and the exact mechanisms leading to cortical malformations are still difficult to dissect. Centrosomes are key structures, and mutations in associated proteins cause microcephaly in human as well as in mouse models. However, even if all the studied proteins have a role at the centrosome, the causes of the reduced production of neurons are not always the same. Indeed, in some cases, such as for example when proteins like MCPH1, STIL, and ASPM are mutated, a thinner cerebral cortex seems to be produced as the consequence of a premature exhaustion of the progenitor pool and premature neuronal differentiation. In these cases, premature differentiation may be the consequence of centrosomal perturbations that lead to spindle orientations that favor asymmetric divisions and cell detachment from the ventricular lining. In other cases, such as for example when CDK5RAP2, CENPJ, and PLK4 are mutated, microcephaly seems to be the product of an increased cell death affecting both the progenitor and neuronal populations, due to mitotic delay and/or aneuploidy. In these cases spindle mis-orientation may be a non-specific feature. The same malformation can hence arise due to potentially different mechanisms. Moreover, the different proteins can have an important role during different temporal windows and compensation mechanisms can also influence the different outcomes. A common feature of the two mechanisms is the elongation of M-phase, hence further studying cell cycle regulation may clarify why in some cases cell death is triggered and in other cases differentiation is favored.

We discussed proteins that regulate mitosis in RGCs without being specifically restricted to centrosome function. These proteins frequently show a localization that changes during the mitotic cycle. They can be involved in the regulation of the mitotic spindle at the poles influencing astral MTs and their interaction with the cell cortex, in the regulation of spindle MT assembly and dynamics, in chromosome congression at the metaphase plate, or in the generation of pulling and pushing forces necessary for chromosome segregation. What appears clear is that spindle orientation defects are always present when the mitotic process is affected, thus it appears as a consequence common to different underlying mechanisms, and probably not something specific to a certain malformation. The degree of the spindle orientation defect is, however, another variable that can influence the downstream cascade of events leading to a particular malformation. Cortical disruptions that derive from mutations in genes such as *NDE1*, *LIS1*, *PP4c*, *WDR62*, *CENPE*, *TCOF1*, *AGS3*, *VANGL2* and *HTT*, are classified as micro- or macrocephaly, and/or cortical disorganization and abnormal cortical architecture, lissencephaly and simplified gyral patterns. These are in general due to progenitor detachment, mis-regulation of cell death and changes in cell fate, sometimes associated with abnormal neuronal migration. In the case of models like mInsc and Wnt3a overexpression, the common feature is an increase in indirect neurogenesis through the generation of basal progenitors, leading to heterotopia and an increase in cortical thickness. The reason why sometimes there is only a defect in the number of neurons produced without alterations of cortical laminar architecture, and sometimes the defects in mitosis lead also to an incapacity of the neurons produced to fit in a layered structure, is still not clear. One of the causes of additional problems in lamination are likely to be the presence of anomalies of the RGC scaffold, or due to multiple roles of these proteins during corticogenesis. Moreover, we can imagine that different variables influence these phenotypes: different proteins have distinct roles and severities once perturbed, also temporal issues can play a role, together with compensation mechanisms that do or do not take place. Moreover, we must consider that different types of cell integrate in a highly complex general architecture, and their mutual interaction is a very influential variable. Indeed, basal progenitors and neurons, can communicate with RGCs to regulate their mode of division through signaling pathways, such as for example Notch signaling (Nelson et al., 2013) that also influences transcription factors and genetic regulation, having also a consequence on neuronal identity, essential for lamination. Also, regulatory inputs from different brain areas (e.g., the thalamus (Gerstmann et al., 2015)) during development may influence progenitor behavior at specific time-points. However, this area remains for the moment little explored.

We also presented genes such as *PAX6*, *LGN* and *EML1* that have been found, when mutated, to lead to progenitor detachment from the ventricular surface that nevertheless appears intact. Ectopic Pax6-positive progenitors hence constitute a source of cells that can be the cause of heterotopia formation. Interestingly, in these mutants, spindle orientation defects and progenitor detachment do not lead to premature differentiation as in the case of microcephaly genes. Indeed progenitors that leave their apical position retain their proliferative capacities and RGC identity. What we can hypothesize is that for reasons that still remain unclear, detachment of a proportion of RGCs from the apical surface occurs in a way that allows them to retain the molecular signatures necessary to not exit the cell cycle and remain in a similar proliferative state. These mechanisms deserve further studies to try to understand why, in some cases, loss of apical attachment is accompanied by a switch to a state more committed to differentiation, and in some others this does not happen. The explanation can be sought also in the steps that regulate cell cycle progression and in the signaling with other cell types. For example, detached progenitors could be insensitive to signals that tell a cell outside the VZ to become a neuron. Conversely, the environment could change and with it the signal. Also, in these models detached progenitors could be insensitive to cell death pathways activated in some microcephaly models, when centrosomes are absent or are too numerous. Indeed, in the CENPJ and PLK4 studies it has been shown that once the tumor suppressor p53 is inactivated, ectopic progenitors actively divide in the IZ, suggesting a role for the centrosome in the anchoring of progenitors, but not in their ability to divide once detached.

Cytokinesis is the last step of cell division, and its perturbation was also shown to produce defects similar to spindle malfunction. The midbody is a transient structure that forms during cytokinesis and can contain molecular signatures important for cell fate determination of daughter cells, which is probably why changes in its structure or function potentially produce cortical malformations. Cytokinesis and its molecular players in RGCs still remain poorly studied mechanisms that deserve further investigation. Molecules that take part in mitotic regulation are often later found located at the midbody, however their accumulation for degradation, or alternatively importance in this structure are still little understood.

#### **CONCLUSIONS**

In this review we hence correlated studies in mouse models and genes found mutated in cortical malformation patients. A large number of mutated proteins found to perturb cortical development in mouse models have not been identified in human. This could be sometimes due to the lethality of mutations in certain proteins in human, or simply to the fact that they remain to be identified by patient screening. In other cases, genes found mutated in patients have been studied in mouse models where they reproduce with different degrees the defects found in human. In rare cases mutations appeared spontaneously (e.g., *HeCo*) and were able to recapitulate, even if with certain differences, the phenotype found in patients carrying mutations in the same gene. In most cases, mouse models are generated by knocking-down genes found mutated in human diseases. This approach is being systematically generalized (e.g., international mouse knockout and phenotyping consortiums) as it furnishes an essential and very useful tool to be able to dissect the basic mechanisms leading to malformations studying the role of single proteins and integrating them in more complex networks. For this latter approach, mouse models furnish a good tool to perform genetic studies at a bigger scale through genome wide comparative studies in order to identify protein and/or regulatory networks important for normal and pathological brain development. These data will be further complemented in the future by knockin studies whereby individual mutations (e.g., missense) are studied in the rodent, to more closely match human pathological situations, and these studies will be extremely interesting to compare with previous knockout data.

Even if mouse models remain a very good tool to study human cortical malformations, such studies still remain a matter of debate as very often the consequences of mutations in the same gene are different when comparing the two organisms. With respect to this, it is important to consider essential differences between the mouse and human brain that can account for the different severities certain malformations have in the two organisms. The human cerebral cortex is much more complex and presents not only a much more extensive tangential and radial expansion but is also characterized by the presence of gyri and sulci. A larger number of progenitors is essential to produce this expansion and the interplay between the different cell types is different and more complex. Even if the classification of the different progenitor types in gyrencephalic species probably deserves further exploration, we know that the human cortex presents additional germinal layers, such as the inner and outer SVZ (i/oSVZ) that are located basal to the VZ (Fish et al., 2008; Borrell and Reillo, 2012). So far we know that in the oSVZ the so called outer or basal RG-like cells (oRGs or bRGs) are located and contribute to neuron production (Hansen et al., 2010). These cells are Pax6-positive but lack an apical attachment to the ventricular surface and have recently also been identified in small numbers in the mouse brain, however their role is still under debate (Shitamukai et al., 2011; Pilz et al., 2013). Such cells have rarely yet been studied in mouse models, but recently, the ventral telencephalon, the region of the murine brain with the largest SVZ, has been shown to be a useful model to study progenitor expansion. Indeed, novel and morphologically heterogeneous progenitor types have been identified in this area that can be traced to gyrencephalic cortices (Pilz et al., 2013). The process of gyrification also adds a degree of complexity to human cortical development and only recently some molecules have been identified potentially playing a role in this process (e.g., Trnp1, Stahl et al., 2013). Mouse models were shown to be useful in this respect, together with gyrencephalic species (e.g., the ferret). Thus, lissencephaly, as a characteristic of the mouse brain, can be viewed as a tool to study single proteins or networks that contribute to the formation of gyri and sulci in more evolved species. Recently, a three-dimensional culture system generating cerebral organoids has also been shown to represent a sensational tool to model human brain development and related cortical malformations (Lancaster et al., 2013). Similar studies will almost certainly help to fill the gap between mouse and human, and will contribute to the study of cortical malformations in the future.

#### **AUTHOR CONTRIBUTIONS**

Sara Bizzotto and Fiona Francis wrote the review.

#### **ACKNOWLEDGMENTS**

We thank Richard Belvindrah and Delfina Romero for helpful discussions concerning this manuscript. We also thank other members of the lab for further discussions on this subject. We are grateful for financial support from the Agence National de la Recherche (ANR13-BSV4-0008-01), as well as from INSERM, the CNRS and UPMC, the Fondation Bettencourt Schueller, the Région Ile-de-France, and the Fédération pour la recherche sur le cerveau. Sara Bizzotto is funded by the French MESR and is affiliated with the Complexité du Vivant Doctoral School. Authors are associated with the BioPsy Labex project and the Ecole des Neurosciences de Paris Ile-de-France network.

#### **REFERENCES**


evidence for an overlap with the cobblestone complex. *Brain* 133, 3194–3209. doi: 10.1093/brain/awq259


impaired expression of Filamin-A. *Hum. Mol. Genet.* 21, 1004–1017. doi: 10. 1093/hmg/ddr531


without muscular or ocular abnormalities. *Am. J. Hum. Genet.* 92, 468–474. doi: 10.1016/j.ajhg.2013.02.005


mutations in a glycosyltransferase, POMGnT1. *Dev. Cell* 1, 717–724. doi: 10. 1016/s1534-5807(01)00070-3


**Conflict of Interest Statement**: The Review Editor Veronique Marthiens declares that, despite being affiliated to the same institution as authors Sara Bizzotto and Fiona Francis, the review process was handled objectively and no conflict of interest exists. The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

*Received: 14 December 2014; accepted: 18 January 2015; published online: 12 February 2015*.

*Citation: Bizzotto S and Francis F (2015) Morphological and functional aspects of progenitors perturbed in cortical malformations. Front. Cell. Neurosci. 9:30. doi: 10.3389/fncel.2015.00030*

*This article was submitted to the journal Frontiers in Cellular Neuroscience*.

*Copyright © 2015 Bizzotto and Francis. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms*.

# MCPH1: a window into brain development and evolution

#### Jeremy N. Pulvers <sup>1</sup> , Nathalie Journiac2,3 , Yoko Arai <sup>4</sup> and Jeannette Nardelli 2,3 \*

<sup>1</sup> Sydney Medical Program, University of Sydney, Sydney, Australia, <sup>2</sup> U1141 Inserm, Paris, France, <sup>3</sup> Université Paris Diderot, Sorbonne Paris Cité, UMRS 1141, Paris, France, <sup>4</sup> Institut Jacques Monod, CNRS UMR 7592, Université Paris Diderot, Sorbonne Paris Cité, Paris, France

The development of the mammalian cerebral cortex involves a series of mechanisms: from patterning, progenitor cell proliferation and differentiation, to neuronal migration. Many factors influence the development of the cerebral cortex to its normal size and neuronal composition. Of these, the mechanisms that influence the proliferation and differentiation of neural progenitor cells are of particular interest, as they may have the greatest consequence on brain size, not only during development but also in evolution. In this context, causative genes of human autosomal recessive primary microcephaly, such as ASPM and MCPH1, are attractive candidates, as many of them show positive selection during primate evolution. MCPH1 causes microcephaly in mice and humans and is involved in a diverse array of molecular functions beyond brain development, including DNA repair and chromosome condensation. Positive selection of MCPH1 in the primate lineage has led to much insight and discussion of its role in brain size evolution. In this review, we will present an overview of MCPH1 from these multiple angles, and whilst its specific role in brain size regulation during development and evolution remain elusive, the pieces of the puzzle will be discussed with the aim of putting together the full picture of this fascinating gene.

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan

#### Reviewed by:

Yuanyi Feng, Northwestern University, USA David A. Keays, Research Institute of Molecular Pathology, Austria

#### \*Correspondence:

Jeannette Nardelli, U1141 Inserm, Hôpital Robert Debré, 48 Blvd Sérurier, 75019 Paris, France Tel: +33 1 40 03 19 26, Fax: +33 1 40 03 19 95 jeannette.nardelli@inserm.fr

Received: 19 December 2014 Accepted: 28 February 2015 Published: 27 March 2015

#### Citation:

Pulvers JN, Journiac N, Arai Y and Nardelli J (2015) MCPH1: a window into brain development and evolution. Front. Cell. Neurosci. 9:92. doi: 10.3389/fncel.2015.00092 Keywords: MCPH1, microcephaly, brain development, brain evolution, mouse models, human

# Introduction: MCPH1 in Brain Development and Evolution

The study of mammalian neurogenesis and cortical development stands at a fascinating intersection between neuroscience, cell biology, developmental biology, genetics, and evolutionary biology (Molnár et al., 2014; Paridaen and Huttner, 2014; Sun and Hevner, 2014). The studies of the genes that cause autosomal recessive primary microcephaly (MCPH) are exemplary of this exciting synthesis of research fields (Woods et al., 2005; Kaindl et al., 2010; Gilmore and Walsh, 2013). One of the causative genes of this condition, MCPH1 (syn. BRIT1, Microcephalin), plays a role in brain development (Jackson et al., 1998, 2002), DNA damage repair (Xu et al., 2004; Lin et al., 2005; Peng et al., 2009), chromosome condensation (Neitzel et al., 2002; Trimborn et al., 2004; Yamashita et al., 2011), cancer (Chaplet et al., 2006; Rai et al., 2006; Richardson et al., 2011), germline function (Liang et al., 2010), and has also provided insights into brain evolution (Evans et al., 2004, 2005; Wang and Su, 2004; Ponting and Jackson, 2005). Many unanswered questions remain on this multifaceted gene, such as how the lack of MCPH1 leads to microcephaly, its molecular mechanisms in neurogenesis, and the key question of its role in the evolution of brain size.

The development of the cerebral cortex begins with formation and patterning of the neural tube (Lumsden and Krumlauf, 1996; Rubenstein et al., 1998; Copp et al., 2003), which is followed by the amplification of neuroepithelial cells, the primary neural progenitor cells, and their subsequent differentiation into downstream progenitors and neurons, or ''neurogenesis'' (Götz and Huttner, 2005; Paridaen and Huttner, 2014; Sun and Hevner, 2014). A constellation of processes follows to form a fully developed cerebral cortex, including neuronal migration (Sidman and Rakic, 1973; Nadarajah and Parnavelas, 2002; Marín and Rubenstein, 2003), axon guidance (Tessier-Lavigne and Goodman, 1996; Dickson, 2002) and synaptogenesis (Garner et al., 2002; Waites et al., 2005). In the context of brain development and evolution, the embryonic development of the mammalian cerebral cortex (neocortex) is the subject of prime interest, being the seat of higher brain functions, and has powerful implications for primate and human evolution (Rakic, 2009; Clowry et al., 2010).

Investigations into cortical malformations give profound insight into not only developmental and molecular mechanisms, but also provide a platform to investigate the evolution of brain size and function (Walsh, 1999; Mochida and Walsh, 2001; Sun and Hevner, 2014). Amongst these conditions, congenital microcephaly of genetic etiology is of particular interest, as they allow the dissection of fundamental molecular and developmental mechanisms. Interestingly, these mechanisms may be affected in congenital microcephaly linked to environmental intrauterine insults, such as viral infections (Cheeran et al., 2009), alcohol, or other extrinsic cues, exemplified by the finding that Mcph1, the mouse ortholog of human MCPH1, was shown to be down-regulated in a mouse model of microcephaly induced by early embryonic exposure to a VIP (vasoactive intestinal peptide) antagonist (Passemard et al., 2011). MCPH1 may be a common denominator in the pathway causing microcephaly, encompassing the spectrum of both environmental and genetic forms of microcephaly. Therefore, given its implication in diverse molecular and cellular mechanisms during brain development, investigating MCPH1 function is of particular interest. Here, an overview of the key issues relating to the function of MCPH1 in brain development and evolution will be reviewed.

# Autosomal Recessive Primary Microcephaly (MCPH)

Microcephaly is the clinical finding of a small brain, typically measured by head circumference (HC), compared to the population mean values of the age, sex, and ethnicity of the individual (Woods, 2004; Kaindl et al., 2010; Woods and Parker, 2013). HC, or more specifically occipito-frontal circumference (OFC) is commonly used as a surrogate measure of brain size (Woods et al., 2005); head size being a readily measurable approximation of brain size, and thus the terms microcephaly (small head) and microencephaly (small brain) are generally interchangeable (Gilmore and Walsh, 2013). An OFC of three standard deviations below the age- and sex-matched means (<--3 SD) is commonly accepted as a clinical definition of microcephaly (Woods and Parker, 2013). Furthermore, microcephaly is subdivided into primary and secondary microcephaly (Qazi and Reed, 1973). Primary microcephaly (microcephaly vera) is generally defined as being present at birth, with no obvious abnormalities other than gross brain size (Mochida and Walsh, 2001), and has a variety of genetic and non-genetic causes such as infections (Woods and Parker, 2013), prenatal radiation exposure (Plummer, 1952; Wood et al., 1967), and prenatal alcohol exposure (Ouellette et al., 1977; Spohr et al., 1993). Secondary microcephaly is generally defined as manifesting after birth (Woods and Parker, 2013). From the perspective of investigating brain development, the distinction between primary and secondary microcephaly is key, as primary microcephaly is likely to be related to neurogenesis, and secondary microcephaly may involve any of the downstream processes post-neurogenesis (Woods, 2004; Woods et al., 2005). Furthermore, microcephaly of genetic etiology, or autosomal recessive primary microcephaly, allows the detailed dissection of the relevant molecular mechanisms. The pathological defect in primary microcephaly is likely to fall temporally within the neurogenic interval and spatially within the neural progenitor cell compartment, making it ideal for the investigation of the regulation of brain size. However, in this context it is important to note that neuronal death due to alterations in diverse aspects of neuronal differentiation (including but not limited to neuronal migration and maturation) cannot be excluded, and may also account for the decrease in brain size seen in primary microcephaly. The main challenge in this regard is to determine how the alteration of these different processes at multiple stages of cortical development, ultimately impacts the final brain size.

Autosomal recessive primary microcephaly (MCPH) is a genetically heterogeneous condition (Mochida and Walsh, 2001), with 12 or more causative genes identified, all of which produce a clinically indistinguishable phenotype. The loci are numbered MCPH1 to MCPH12 (Kaindl, 2014), and the genes identified in this order are: MCPH1 (Jackson et al., 1998, 2002), WDR62 (Roberts et al., 1999; Nicholas et al., 2010), CDK5RAP2 (Moynihan et al., 2000; Bond et al., 2005), CASC5 (Jamieson et al., 1999; Genin et al., 2012), ASPM (Jamieson et al., 2000; Pattison et al., 2000; Bond et al., 2002), CENPJ (Leal et al., 2003; Bond et al., 2005), STIL (Kumar et al., 2009), CEP135 (Hussain et al., 2012), CEP152 (Guernsey et al., 2010), ZNF335 (Yang et al., 2012), PHC1 (Awad et al., 2013), and CDK6 (Hussain et al., 2013). Mutations in CEP63 have also been identified as causing primary microcephaly (Sir et al., 2011). ASPM (MCPH5) is reported as being the most common locus for MCPH, with WDR62 (MCPH2) being the second, followed by MCPH1 and the other loci being rarer causes of MCPH (Roberts et al., 2002; Darvish et al., 2010; Nicholas et al., 2010; Sajid Hussain et al., 2013).

It has long been noted that the MCPH genes are all related to the mitotic spindle or centrosome (Bond and Woods, 2006; Bettencourt-Dias et al., 2011), and most MCPH proteins are located at the centrosome or spindle. However the more recently identified genes do not appear to be related to these functions, such as PHC1 which plays a role in chromatin remodeling and DNA repair (Awad et al., 2013). It is interesting to note that MCPH1 traverses multiple functions related to: transcription activation (Yang et al., 2008), mitotic spindle and centrosome (Gruber et al., 2011), and DNA damage repair (Zhou et al., 2013), which raises a possible link between DNA repair and microcephaly. Although the hitherto known functions of the MCPH genes are linked to the cell cycle, the fact that the loss-of-function mutation of only one gene is sufficient to cause microcephaly excludes the possibility of functional redundancy. Nevertheless, such exclusion does not rule out genetic interactions between MCPH genes. Further investigations of the spatio-temporal expression patterns of these genes during cortical development and a detailed assessment of the neurodevelopmental defects occurring in MCPH patients and mouse model mutants will be necessary to answer the question of a crosstalk between MCPH genes, and the identity of a common centrosome- or spindlerelated mechanism regulating brain size (Bond and Woods, 2006).

# Genetics of MCPH1

MCPH1 was the first locus identified for MCPH, initially mapped to 8p22-pter (Jackson et al., 1998). The gene was later identified and the protein was named Microcephalin (Jackson et al., 2002). A feature unique to primary microcephaly caused by mutations in MCPH1 are defects in chromosome condensation, specifically premature chromosome condensation (PCC) in early G2-phase, and delayed decondensation post-mitosis (Trimborn et al., 2006). This condition was originally named PCC syndrome (Neitzel et al., 2002); however mutations in MCPH1 were later identified, and PCC and primary microcephaly caused by mutations in MCPH1 were found to be allelic disorders (Trimborn et al., 2004). PCC is detected in cytogenetic preparations, as a high proportion of prophase-like cells and poor quality metaphase G-banding (Trimborn et al., 2004). PCC is a pathognomonic feature of MCPH1, being absent in all other mutations of MCPH genes. Another autosomal recessive condition characterized by microcephaly with additional craniofacial features and mitotic and chromosomal defects (Tommerup et al., 1993), was later found to also be caused by mutations in MCPH1 (Farooq et al., 2010). MCPH1 was also identified independently in a screen for transcriptional repressors of hTERT (catalytic subunit of human telomerase) and named BRIT1, BRCT-repeat inhibitor of hTERT expression (Lin and Elledge, 2003), which was later shown to be the same gene as MCPH1 (Lin et al., 2005). Taken together, these molecular features of MCPH1 collectively reveal that the gene is implicated in diverse processes that are crucial not only for proper brain development, but also for the maintenance of genome integrity (Liang et al., 2010).

# MCPH1 Protein Structure and Function

In human and mouse (Mus musculus), the MCPH1 coding sequence contains 14 exons, distributed across 200 kb of genomic DNA. The protein contains three BRCT (**BR**CA1 **C**-**T**erminal) domains (**Figures 1**, **2**), which were first described in BRCA proteins and mediate protein-protein interactions (Koonin et al., 1996; Huyton et al., 2000). One domain is located at the N-terminal side of the MCPH1 protein (BRCT1), and a tandem of two domains spans the C-terminal sequence (BRCT2/3). These domains are predicted to be crucial for function (Jeffers et al., 2008) and may be differentially involved in the diverse roles of MCPH1 by mediating interactions with distinct partners. In line with this notion, repeats of BRCT domains, such as BRCT2/3, have been shown to preferentially interact with phosphorylated residues (Woods et al., 2012).

A number of studies have reported on MCPH1 interaction partners and differential functions of the N-terminal BRCT1 domain, and the more C-terminal BRCT2/3 domains (**Figure 1**). Two regions of human MCPH1: an N-terminal fragment containing BRCT1 (residues 1--195), and a central fragment (residues 381--435) interact with different subunits of condensin II to regulate chromosome condensation (Yamashita et al., 2011). The N-terminal sequence (residues 1--48) was shown to be necessary to recruit BAF170, a component of the chromatin remodeling complex SWI/SNF, to relax chromatin for DNA repair (Peng et al., 2009). This suggests a possible role of MCPH1 in chromatin conformation. BRCT2/3 domains have been shown to bind to E2F1 to form a complex able to transactivate BRCA1 and CHK1 (Yang et al., 2008). Interactions between MCPH1 BRCT2/3 domains and Cdc27, a subunit of the anaphasepromoting complex (Singh et al., 2012b), and a phosphorylated domain of H2A.X have also been reported (Singh et al., 2012a). On the other hand, a truncated MCPH1 protein lacking BRCT2/3, is able to complement the defective chromosome condensation in human MCPH1-deficient cells (Gavvovidis et al., 2012), indicating that these domains are dispensable for this function. Molecular interactions between MCPH1 and partners may likely be regulated by post-translational modifications. In line with this notion, Ser322 phosphorylation of MCPH1 by ATR is required for TopBP1 recruitment and sustained ATR signaling for DNA repair (Zhang et al., 2014). Further mechanistic dissection of the functions of BRCT1 and BRCT2/3 domains will allow a better definition of the different molecular and biochemical mechanisms impaired by MCPH1 mutations.

# Human MCPH1 Mutations

A large number of homozygous mutations in MCPH1 causing primary microcephaly have been identified (**Table 1**; **Figures 1**, **2**). These include large deletions of exons (Garshasbi et al., 2006; Darvish et al., 2010; Pfau et al., 2013), single basepair insertions and deletions (Trimborn et al., 2004; Darvish et al., 2010; Sajid Hussain et al., 2013), nonsense mutations (Jackson et al., 2002; Farooq et al., 2010; Hosseini et al., 2012), missense mutations (Trimborn et al., 2005; Darvish et al., 2010; Leung et al., 2011; Ghani-Kakhki et al., 2012), and one splice site mutation (Darvish et al., 2010). Interestingly, all missense mutations identified so far are located in exons 2 or 3 (**Table 1**; **Figure 1**), which is within the N-terminal BRCT domain (BRCT1; see previous section on MCPH1 protein). The missense mutations all lead to non-conservative changes in an amino acid residue: threonine to arginine (c.80C>G, p.Thr27Arg) which is polar uncharged to positively charged (Trimborn et al., 2005), histidine to glutamine (c.147C>G, p.His49Gln) which is positively charged to polar uncharged (Darvish et al., 2010), serine to leucine (c.215C>T, p.Ser72Leu) which is polar uncharged to non-polar (Darvish et al., 2010; Ghani-Kakhki et al., 2012), and tryptophan to arginine (c.223T>C, p.Trp75Arg) which is non-polar aromatic to positively charged (Ghani-Kakhki et al., 2012). These mutated residues are highly conserved during evolution, Thr27 being conserved in orthologs of MCPH1 in mammals and amphibians (Trimborn et al., 2005). Ser72 and Trp75 are conserved in all vertebrates and Drosophila, and these two residues are also conserved in BRCT domains of BRCA1 (Ghani-Kakhki et al., 2012). In one case a homozygous double mutation in consecutive codons (c.149T>G, c.151A>G; p.Val50Gly, p.Ile51Val) was found, which leads to conservative residue changes (both non-polar to non-polar); however still within the N-terminal BRCT domain. Molecular and biochemical studies of full-length MCPH1 proteins harboring these various amino acid changes may be a powerful tool in correlating MCPH1 genotype and molecular function with phenotype.

As previously noted for ASPM (Nicholas et al., 2009), correlations between mutation and phenotype, such as the severity of microcephaly, could not be established so far

and the positions of reported mutations in humans and mice. (A) Schematic of MCPH1 gene intron-exon structure (upper) and protein domain structures (lower), which are highly conserved between mouse and human. The MCPH1 coding sequence includes 14 exons shown as black rectangles numbered from 1--14. The three BRCT domains are shaded in green. Factors interacting with the BRCT domains are indicated below each domain. A domain including residues 381--435 in exon 8 and interacting with

condensin II is shaded gray, and the phosphorylation site Ser322 (S322P)

MCPH1 causing primary microcephaly are indicated on the gene schematic (see Table 1; Figure 2 for amino acid changes). The extent of the deletions is indicated as colored bars below the gene structure. (C) Schematic of the four reported Mcph1 mutations in mice (see also Table 2). For targeted deletions, the triangles overlaid on the intron flanks the exons that were targeted for deletion, with the allele name indicated by the same color below. For the gene trap mutation and knockout-first allele, the site of the insertion is indicated by a triangle above the intron, with the corresponding allele name in the same color.


proteins. Conserved amino acids are shown in bold characters. Blue upper lines indicate BRCT domains, and red underlines indicate nuclear localization signals (Gavvovidis et al., 2012). Amino acid residues mutated in microcephaly (either nonsense or missense; see Table 1; Figure 1) are in red.

(**Table 1**). One notable exception is the p.Thr27Arg missense mutation, which shows only a mild microcephaly (--2.4 SD at birth), and the fraction of prophase-like cells being only marginally higher than controls (Trimborn et al., 2005; Ghani-Kakhki et al., 2012). Meta-analysis of genotype and phenotype correlations are hampered however by the rarity of patients with MCPH1 mutations, and by the fact that the HC data is reported at birth in some cases (Trimborn et al., 2005) but often at older ages for others, and often only a range of HCs are shown for multiple individuals within a family (Darvish et al., 2010). Similarly, PCC is often reported qualitatively (Pfau et al., 2013), without a careful quantification of prophase-like cells as seen in other studies (Ghani-Kakhki et al., 2012). Molecular studies comparing the activities of the mutant proteins may aid in answering some of these questions (Leung et al., 2011), and will provide powerful clues in understanding the mechanisms of MCPH1.

# Mouse Models of Mcph1 Deficiency: Microcephaly Phenotype

A number of mouse models for Mcph1 loss-of-function have been reported (**Table 2**; **Figure 1**). These mutant mouse lines have been successful in recapitulating the features of primary microcephaly and PCC caused by mutations in MCPH1 in humans. The first study demonstrating a microcephaly phenotype in a Mcph1 null mutant mouse model (Gruber et al., 2011) was generated by targeted deletion of Mcph1 exon 4--5 (Mcph1tm1.1Zqw). A follow-up study of this mouse model was reported by the same group (Zhou et al., 2013). In homozygous Mcph1tm1.1Zqw mutant mice, the size of the newborn mouse brain was visibly smaller, and brain weight was also reduced, fulfilling the criteria of human primary microcephaly at a gross anatomical level. Histological examination of the newborn brains revealed an approximate 20% reduction in both the radial thickness and the lateral extent of the neocortex. The Mcph1 mutant brain at embryonic day (E) 13.5 was reported as reduced, and the cortical wall was thinner at E15.5 (Gruber et al., 2011), implying an early defect. Interestingly, in adult mutant mice, the anterior portion of the brain (defined in this case as the olfactory bulb, cerebrum, and thalamus) showed a more significant reduction compared to the remaining posterior portion. Moreover, histological examination of the cerebellum at post-natal day (P) 21 was reportedly normal suggesting that MCPH1 may not be required for the development of the cerebellar cortex (Zhou et al., 2013).

Importantly, MCPH1 mutant mice were also reported to have an approximate 20% reduction in body weight (at P0, P21, and P180; i.e., at birth, weaning, and adult) and the proportion of brain weight to body weight showed no significant difference compared to controls (Zhou et al., 2013). Short stature is a variable feature of human MCPH1 and is seen in some patients (Trimborn et al., 2004). However, a proportionate reduction in both brain and body weight, raises questions of the specificity of Mcph1 as a sole brain size regulator, at least in the mouse model. In another mouse model, a Mcph1 null mutant generated by a targeted deletion of exon 2 (Mcph1tm1.2Kali), growth retardation was also found, also showing an approximate 20% reduction in weight; however brain weights were not reported in this study (Liang et al., 2010). The answer to the question of a specific effect of Mcph1 on brain size, rather than an effect on overall body size, will require the analysis of mouse conditional mutants, with specific inactivation in the brain or in the neocortex. In contrast, reported mouse models of ASPM show a specific reduction in brain weight, with no or minimal reduction in body weight (Pulvers et al., 2010; Fujimori et al., 2014).

Two other mouse models of Mcph1 have been reported, one generated from gene trap ES cells (Skarnes et al., 2004), and another harboring a knockout-first allele (Skarnes et al.,



The genotype, predicted effect on protein ("?" if unknown or not described), the exon containing the mutation, standard deviation (SD) of the head circumference (HC), and the reference are shown. All mutations are homozygous.

#### TABLE 2 | Summary of mouse models of human MCPH1.


Genotypes of mouse Mcph1 mutations, the phenotypes of the mice, and the references are shown. Allele names obtained from Mouse Genome Informatics (MGI:2443308). MRI, magnetic resonance imaging; IR, ionizing radiation; Hom, homozygous mutant; wt, wildtype.

2011). The model containing a gene trap vector between exons 12--13 (Mcph1Gt(RRO608)Byg ), which leads to a protein lacking only the most C-terminal BRCT3 domain, showed normal body and brain weight, however did exhibit the PCC phenotype (Trimborn et al., 2010). The knockoutfirst allele mouse model with an insertion between exons 3--4 (Mcph1tm1a(EUCOMM)Wtsi) showed brain weight reduction (∼15%); however there was no evidence of growth retardation (Chen et al., 2013). Both of these Mcph1 models are hypomorphic mutants with detectable residual wild-type transcript, and it is thus difficult to interpret these phenotypes and compare them to the other Mcph1 null mutants and human MCPH1 patients.

# Mouse Models of Mcph1 Deficiency: Non-Brain Phenotypes

A number of other phenotypes have been reported in the Mcph1 mutant mice (**Table 2**). Infertility has been observed in the two null mutants (Liang et al., 2010; Gruber et al., 2011; Zhou et al., 2013) and one of the hypomorphic mutants (Chen et al., 2013). Phenotypes were observed in both spermatogenesis and in the ovary (Liang et al., 2010). Germline phenotypes have also been reported for mouse models of ASPM (Pulvers et al., 2010; Fujimori et al., 2014). This raises the important question of whether infertility, or testicular or ovarian atrophy, is seen in human primary microcephaly. So far no cases have been reported and the relationship between primary microcephaly and infertility in humans remains an interesting question, as it may provide clues to the mechanism of the gene, and also raises the possibility that the function of MCPH1 in the germline may have been the target of positive selection in the primate lineage (Woods et al., 2006; Dobson-Stone et al., 2007; Timpson et al., 2007).

In one of the Mcph1 hypomorphic mutants (Mcph1Gt(RRO608)Byg ), homozygous mice exhibited a shortened lifespan, specifically a significant difference in survival after 65 weeks of age (Trimborn et al., 2010). This observation of decreased overall survival in Mcph1 homozygous mutants was replicated in the Mcph1tm1.2Kali null mutant, which manifests an increased cancer susceptibility (Liang et al., 2014). This difference in survival was not reported in the other mutants (Gruber et al., 2011; Chen et al., 2013; Zhou et al., 2013); however as the difference becomes apparent when mice are older, a shortened lifespan may have been overlooked. This raises the question of whether patients with primary microcephaly exhibit a shortened lifespan, independent of other comorbidities that may be exacerbated by the associated mental retardation. Epidemiological studies may be difficult due to the rarity of MCPH. However, one post-mortem case report of a 77-year old with primary microcephaly exists (McCreary et al., 1996) and another study reports on MCPH patients ranging to their 70 s (Darvish et al., 2010) indicating that the condition is compatible with normal lifespan.

Another interesting phenotype of Mcph1 mutant mice reported in two studies (Liang et al., 2010; Chen et al., 2013) is a significantly reduced proportion of homozygous mice (10∼15% vs. the expected normal 25% Mendelian ratio) born from intercrossing of heterozygous Mcph1 mutant mice. This indicates the possibility of an embryonic lethal phenotype of varying penetrance caused by homozygous mutations in Mcph1.

The vastly complex cerebral cortex has its developmental origin in the germinal zone, or ventricular zone (VZ), of the dorsal telencephalon of the neural tube (Götz and Huttner, 2005). The VZ is composed of neuroepithelial cells, which line along the ventricles of the embryonic brain vesicles and spinal cord, and serve as the primary progenitor cells of all neural cells of the central nervous system. Neuroepithelial cells exhibit typical epithelial cell characteristics, such as apicalbasal polarity (Chenn et al., 1998), and its nuclei undergo a unique movement in the apical-basal axis correlated with the cell cycle, termed interkinetic nuclear migration (Sauer, 1935; Taverna and Huttner, 2010; Kosodo, 2012), which gives the tissue a pseudostratified histological appearance, and thus its classification as a pseudostratified epithelium. Neuroepithelial cells, or neural progenitor cells, initially proliferate to expand the progenitor pool, and later commence differentiative divisions into downstream progenitors and neurons (Paridaen and Huttner, 2014), the process of neurogenesis.

In cortical development, the lateral dimension of the laminar structure is largely dictated by the expansion of neural progenitor cells or proliferative units, and the radial dimension is a result of the extent of neurogenesis subsequent to progenitor proliferation within an ontogenetic column or radial-unit (Rakic, 1988, 1995, 2009). This framework allows the dissection of cortical phenotypes, where a reduction in the lateral dimension is likely a deficit in the initial progenitor pool and its subsequent proliferative expansion, and a reduction in the radial dimension may be caused by a reduced capacity for neurogenesis, shortening of the neurogenic interval, or neuronal loss. Conceptually, an analysis distinguishing the radial and lateral dimensions is useful in phenotypic dissection of mutant mice to further clarify their respective impact on the final neurogenic outcome.

Mcph1 inactivation has an impact on both lateral and radial dimensions (Gruber et al., 2011; Zhou et al., 2013). These characteristics have been related to a loss of progenitors and to a specific decrease in upper layer neuron output, possibly due to premature progenitor exhaustion. They appear to result from the deficiency of two distinct Mcph1 functions: the control of the centrosome cycle through the Chk1-Cdc25 pathway, and DNA damage repair. The relative contribution of each deficiency in this progenitor loss has not been addressed and will require further studies. Likewise, it will be interesting to address whether the aforementioned participation of Mcph1 in chromatin conformation during DNA repair could also be important for chromatin conformation with regards to progenitor fate determination during neurogenesis. In line with this notion, competitive interactions between BAF170 and BAF155 with Pax6 have been shown to play an important role in the choice between cell cycle maintenance vs. differentiation (Tuoc et al., 2013). Interestingly, mutations of several BAF genes, including BAF170 and BAF155, have been reported to be involved in brain disorders (Ronan et al., 2013), and it will be interesting to further explore how interactions between BAF and MCPH1 proteins impact brain size and function.

The full interpretation of the defects reported for Mcph1 mutant mice is hampered by the lack of detailed information on the pattern of Mcph1 expression during cortical development, such as the temporal dynamics of gene activation, the cell types expressing the gene and the sub-cellular localization of the Mcph1 protein. It remains thus difficult to establish a direct link between the described phenotypes and gene function. Moreover, gaining insight into the spatio-temporal mode of MCPH1 expression in mouse and human will also allow the delineation of how the function of the gene may have diverged between both species. In particular, it will be interesting to determine if MCPH1, along with other MCPH genes, are expressed in the OSVZ (outer subventricular zone) progenitors (Smart et al., 2002), and eventually how the MCPH genes may impact the proliferation rate and division mode of these progenitors, which are considered to be involved in the expansion of the surface area of the neocortex (Lui et al., 2011; Betizeau et al., 2013; Sun and Hevner, 2014).

# MCPH1 and Brain Evolution

Ever since the identification of MCPH genes, there has been intense interest in their possible roles in the evolution of brain size (Evans et al., 2004; Wang and Su, 2004; Ponting and Jackson, 2005; Woods et al., 2005). Microcepaly (primary microcephaly, or microcephaly vera) has long been the subject of evolutionary interest, and is commonly described in the literature as an atavistic (or ''throwback'') condition, a reversion to an ancestral form (Mochida and Walsh, 2001; Gilbert et al., 2005; Ponting and Jackson, 2005; Vallender et al., 2008). In fact, microcephaly was already proposed in the mid-19th centrury by Carl Vogt to be the reappearance of an ancestral primate (Komai et al., 1955; Richardson, 2011). Much of the intrigue and fascination of research into MCPH genes has been fueled by the remarkable phenotype of primary microcephay: a significant and specific reduction in brain size with the absence of other neurological and non-neurological abnormalities (Woods et al., 2005). However, as little is known about the actual disturbance in cortical structure and cytoarchitecture in MCPH, and since the neuroanatomy of ancestral primates and hominids may forever remain in the realm of speculation and extrapolation from the comparison of extant species (Holloway, 1968; Rilling and Insel, 1999; Falk et al., 2000), parallels between the pathological condition of microcephaly, and the evolutionary change in primate brain size must be made with caution. Moreover, MCPH cannot be described as ''atavistic'' in genetic terms, as all mutations identifyed so far, whether they result in loss-offunction or the truncated MCPH1 protein has a toxic effect, are mutations predicted to cause gene dysfunction, not a reversal to an ancestral sequence (Gilbert et al., 2005). However, the MCPH phenotype does provide insight into a cell biological mechanism of brain size regulation, which may indeed have been involved in primate brain evolution (Bond and Woods, 2006; Fish et al., 2008; Thornton and Woods, 2009; Sun and Hevner, 2014).

Intriguinly, Mcph1 loss of function in mice primarily affects the number of upper-layer neurons (Zhou et al., 2013), which are very important for intra-cortical connections and are involved in higher cognitive functions. The output of these neurons has increased during evolution correlating with increasing complexity of cognitive functions, as observed in humans and primates (DeFelipe et al., 2002; Molnar et al., 2006; Fame et al., 2011). The participation of MCPH genes in the development of upper-layer neurons (Lizarraga et al., 2010; Yang et al., 2012; Zhou et al., 2013) may thus represent a major clue for understanding the evolution of brain size and function.

Regarding studies on the evolution of MCPH genes, interest has generally centered on (i) the possibility of genetic variations in these genes being directly involved in brain size regulation during primate and human evolution; or (ii) through the analysis of the function of these genes a molecular mechanism or pathway regulating brain size may be identified.

# MCPH1: Positive Selection and Polymorphisms

With regards to variations in MCPH1, studies have focussed on either the analysis of positive (or Darwinian) selection of MCPH1 in extant primate species, or through the analysis of polymophisms in human populations. Positive selection in the context of protein evolution can be studied by the analysis of the ratio of non-synonymous (Ka) to synonymous (Ks) changes in DNA sequence (Yang and Bielawski, 2000; Hurst, 2002), and has commonly been used in investigating the link between brain development and evolution (Dorus et al., 2004; Gilbert et al., 2005). Briefly, since synonymous mutations in codons do not change amino acid sequence and therefore do not alter the biochemical properties of the protein, they are assumed to be selectively neutral and reflect the neutral mutation rate. Non-synonymous mutations on the other hand, alters the amino acid sequence which may in turn alter protein function, which more commonly would lead to gene dysfunction and an evolutionary disadvantage, and rarely may confer a gain-of-function or an evolutionary advantage (Woods et al., 2005). Therefore a non-synonymous/synonymous substitution ratio (Ka/K<sup>s</sup> or dN/dS) of >1 can be interpreted as evidence for positive selection. Utilizing these methods, MCPH1 was found to exhibit positive selection in the primate lineage, specifically from the common ancestor of great apes and humans (Evans et al., 2004; Wang and Su, 2004). Another study which investigated four microcephaly genes across 21 species of anthropoid primates identified positive selection correlating with neonatal and adult brain size for ASPM and CDK5RAP2, but interestingly not for MCPH1 and CENPJ (Montgomery et al., 2011). Positive selection of MCPH1 was also identified in cetaceans, which was however also not associated with variations in brain size (McGowen et al., 2011). Examination of positive selection of microcephaly genes across 33 eutherian mammal species revealed signs of positive selection of MCPH1 across nonprimate mammals, however MCPH1 did not correlate with neonatal brain size (Montgomery and Mundy, 2014). How MCPH1 may be involved in brain size regulation more broadly across vertebrates and whether variations show any correlations between gyrencephalic vs. non-gyrencephalic species remains an interesting question.

A large number of non-pathogenetic mutations or polymorphisms in MCPH1 are known, particularly in exon 8 and 13 (Scala et al., 2010). One of which, c.940G>C (p.Asp314His, non-conservative missense from negatively to positively charged, exon 8; rs930557), has received much attention. This polymorphism which is diagnostic for a haplotype with the derived C allele, designated as haplogroup D, has a relatively young coalescence age (i.e., time to single ancestral copy) of 37,000 years, however with a high population frequency worldwide, indicating strong positive selection among anatomically modern humans (Evans et al., 2005). Further analysis on the origin of haplogroup D indicated its divergence 1.1 million years ago from the human lineage and subsequent introgression of this derived allele into human populations 37,000 years ago, possibly due to interbreeding between humans and an archaic homo species, which was speculated as Neanderthals (Evans et al., 2006). Although Neanderthals were suggested as the possible archaic homo (Hawks et al., 2008), subsequent sequencing of MCPH1 of Neanderthals revealed the non-D (ancenstral) haplotype (Green et al., 2010; Lari et al., 2010).

A large number of studies have attempted at identifying associations between MCPH1 haplogroup D and a variety of brain-related phenotypes. In short, no associations have been identified with regards to HC, brain size by MRI, IQ or mental retardation (Woods et al., 2006; Dobson-Stone et al., 2007; Mekel-Bobrov et al., 2007; Rushton et al., 2007; Timpson et al., 2007; Bates et al., 2008; Maghirang-Rodriguez et al., 2009). One study identified an association with the population freqency of haplogroup D and linguistic tone (Dediu and Ladd, 2007), although a later study showed that the derived allele is not associated with lexical tone perception (Wong et al., 2012).

The derived allele of another MCPH1 polymorphism, c.2282T>C (p.Val761Ala, conservative missense from non-polar to non-polar, exon 13; rs1057090), was associated with an increase in cranial volume in Chinese males, but not females (Wang et al., 2008). Non-exonic common variants in MCPH1 have been associated with brain size and cortical surface area in females (Rimol et al., 2010), and another study has investigated MCPH genes and their association with sexual dimorphism in brain size in primates (Montgomery and Mundy, 2013). Mechanistically, how MCPH1 may contribute to sexuallydimorphic brain phenotypes remains unclear.

Key questions remain as to whether MCPH1 did play a role in the evolution of brain size in the primate lineage, and whether common variants of MCPH1 in human populations today are associated with any structural or functional brain phenotype. In light of the phenotypic data from mouse models (Liang et al., 2010; Trimborn et al., 2010; Gruber et al., 2011; Chen et al., 2013; Zhou et al., 2013) which implicate MCPH1 in a number of other non-nervous systems, notably the germline, positive selection in primates and humans may not be due to adaptive changes in brain size or function (Woods et al., 2006; Dobson-Stone et al., 2007; Timpson et al., 2007). In this context it is interesting to note that a large scale study of human and chimpanzee orthologs for evidence of positive selection revealed a large number of genes involved in tumor suppression, apoptosis, and spermatogenesis (Nielsen et al., 2005); functions where MCPH1 may play a major role. Interestingly, two MCPH1 polymorphisms have been associated with breast cancer risk (Jo et al., 2013), which further stresses the need of examining non-nervous system phenotypes in the context of MCPH1 evolution and also in primary microcephaly patients.

# MCPH1 Evolution: From a Cell Biological Perspective

Cortical development in rodents and primates share many features, however there are a number of important differences relevant for brain size evolution. A key difference and one that is of relevance to MCPH1, is the mechanisms which regulate the production of progenitors in the OSVZ, the germinal compartment which has enlarged strikingly in primates and humans and is considered as the seat of the evolutionary expansion of neocortical surface area (Smart et al., 2002; Lui et al., 2011; Sun and Hevner, 2014). The analysis and comparison of these differences in progenitor cell types and lineage relationships, germinal layer cytoarchitecture, and cell biological mechanisms will help in constructing a model of mammalian brain evolution from a developmental and cell biological perspective (Fish et al., 2008; Rakic, 2009; Fietz and Huttner, 2011; Lui et al., 2011; Sun and Hevner, 2014). Another clue recently emerged is the importance of DNA repair pathways, revealed by a preferential effect of mutations of genes implicated in such pathways, including MCPH1, on neural progenitors (Gilmore and Walsh, 2013). This suggests the existence of a specific cross-talk between DNA repair pathways and primary cell cycle functions in these progenitors, which might have become more critical during evolution. Integrating these molecular findings with genetics and evolutionary biology will be a powerful approach in investigating brain size evolution (Enard, 2014).

With regards to MCPH1, some studies have taken this approach in investigating its function in brain size evolution. One study identified an E2F1 binding motif in the MCPH1 promoter region, which is specific to primates and absent in mice and other vetebrates (Shi and Su, 2012). E2F1 is a transcription factor regulating genes involved in cell cycle and apoptosis (Ginsberg, 2002), and interestingly MCPH1 is involved in transcriptional regulation of several DNA repair, checkpoint and apoptosis genes, via interaction with E2F1 (Yang et al., 2008). Another study performed a cell line assay comparing human and rhesus macaque MCPH1 protein and its affects on down-stream gene expression, and found that human-specific amino acid changes in MCPH1 led to differences in expression of three downstream genes involved in cell cycle regulation and apoptosis (Shi et al., 2013). These studies go beyond correlating genetic changes with brain size, and attempt to experimentally test the hypothesis that MCPH1 is an important gene in brain size evolution. Futher approaches may be to generate humanized and primatized mice expressing human and other primate MCPH1 (Pulvers et al., 2010), or the use of cerebral organoids (Lancaster et al., 2013), an in vitro model of human cortical development, where microcephaly-causing mutations in humans and primate-specific variants in MCPH1 can be investigated in detail in a system amenable to experimentation (Enard, 2014). Gene expression profiling studies aimed at identifying pathways dependent on MCPH1 in mouse and human, as well as the characterization of molecular partners for both the mouse and human proteins, will provide major clues on the molecular mechanisms involving MCPH1 and its role in the evolution of brain size. Much may be learnt regarding the role of MCPH genes in brain size evolution, from constructing developmental and cell biological tools for analyzing evolutionary questions.

MCPH1 is well-positioned as a candidate gene for understanding the mechanisms of brain size evolution, as it is related not only to the other primary microcephaly genes with regards to its phenotype, but also harbors the same BRCT domains as BRCA1, which has also been shown to be important for brain development (Pulvers and Huttner, 2009; Pao et al., 2014) and shows positive selection in the primate linage (Huttley et al., 2000; Lou et al., 2014).

# Conclusions

Advances in medical genetics have greatly enhanced our understanding of the origins of many brain and nervous system development disorders; microcephaly in particular. Nevertheless, the molecular and cellular processes underlying such disorders remain poorly understood, and gaining insight into the pathological mechanisms has remained as a major challenge in developmental neurobiology. In this respect, microcephaly of genetic etiology represents a valuable context for the study of the mechanisms that control the final neurogenic output, and by extension to animal models, to assess how these mechanisms have been adjusted and modulated during

# References


evolution along with the remarkable expansion of brain size. So far, interests have focused mainly on aspects related to cell division and proliferation; however the pathological mechanisms associated with microcephaly may prove to be more complex and multifactorial. In line with this notion, MCPH1 appears to assume multifaceted functions, including but not limited to: brain development (Jackson et al., 1998, 2002), DNA damage repair (Xu et al., 2004; Lin et al., 2005; Peng et al., 2009), chromosome condensation (Neitzel et al., 2002; Trimborn et al., 2004; Yamashita et al., 2011), cancer (Chaplet et al., 2006; Rai et al., 2006; Richardson et al., 2011), and germline function (Liang et al., 2010), as reviewed here and elsewhere (Venkatesh and Suresh, 2014). Further comparative expression and functional studies in different species, including primates, will prove to be highly informative in further delineating the molecular and genetic networks controlled by MCPH1, and how they may have been tuned or co-opted to participate in the expansion of brain size. Such progress in the understanding of fundamental developmental mechanisms of the brain is expected to have a valuable impact not only in the understanding of clinical conditions such as microcephaly, but also to answer one of the most enduring questions in biology: the evolution of brain size.

# Acknowledgments

We are grateful to Pierre Gressens for critical reading of the manuscript. JN and NJ were supported by Inserm, University Paris 7 Denis Diderot, Grace de Monaco and Roger de Spoelberch Foundations, ARC (Association pour la Recherche sur le Cancer); and YA by ARC, FRM (Fondation pour la Recherche Médicale) and JSPS (Japan Society for the Promotion of Science).


MCPH1. Proc. Natl. Acad. Sci. U S A 109, 14381--14386. doi: 10.1073/pnas. 1212366109


a composite modulator of condensin II. J. Cell Biol. 194, 841--854. doi: 10. 1083/jcb.201106141


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Pulvers, Journiac, Arai and Nardelli. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Cytoskeletal proteins in cortical development and disease: actin associated proteins in periventricular heterotopia

#### Gewei Lian and Volney L. Sheen \*

Department of Neurology, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA, USA

The actin cytoskeleton regulates many important cellular processes in the brain, including cell division and proliferation, migration, and cytokinesis and differentiation. These developmental processes can be regulated through actin dependent vesicle and organelle movement, cell signaling, and the establishment and maintenance of cell junctions and cell shape. Many of these processes are mediated by extensive and intimate interactions of actin with cellular membranes and proteins. Disruption in the actin cytoskeleton in the brain gives rise to periventricular heterotopia (PH), a malformation of cortical development, characterized by abnormal neurons clustered deep in the brain along the lateral ventricles. This disorder can give rise to seizures, dyslexia and psychiatric disturbances. Anatomically, PH is characterized by a smaller brain (impaired proliferation), heterotopia (impaired initial migration) and disruption along the neuroependymal lining (impaired cell-cell adhesion). Genes causal for PH have also been implicated in actin-dependent processes. The current review provides mechanistic insight into actin cytoskeletal regulation of cortical development in the context of this malformation of cortical development.

Keywords: filamin, formin, RhoGTPases, actin cytoskeleton, proliferation, polarity, migration, periventricular heterotopia

# Introduction

The cerebral cortex originates from an expansion of neural tube, which consists of a singlecell-layered, pseudostratified neuroepithelial cells (neural stem cells; Götz and Huttner, 2005). Neuroepithelial cells extend their processes from the apical side (ventricular zone) to the basal lamina, and undergo interkinetic nuclear migration along the apical-basal axis throughout the cell cycle. The nuclei locate to the basal lamina during G1 phase, staying at the basal lamina during S phase, and transition back to the apical side during G2 phase. Mitosis of most neural stem cells occurs at the apical surface, where adhesion proteins like N-cadherin and cytoskeleton-associated proteins form dense and dynamic adhesion structures. As neuroepithelial cells divide, neural cells adopt different cell fate specification—one proliferating neural stem cells and another differentiating neuronal cells by symmetric and asymmetric divisions (Cremisi et al., 2003). At an early stage of mouse corticogenesis, the neuroepithelial cells mainly adopt symmetric cell divisions to expand the neuroepithelial plane, thereby promoting enlargement the ventricle surface. As development progresses, the neuroepithelial cells begin to be progressively biased toward

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan

#### Reviewed by:

Makoto Sato, Osaka University, Japan Zhiheng Xu, Institute of Genetics and Developmental Biology, China

#### \*Correspondence:

Volney L. Sheen, Department of Neurology, Beth Israel Deaconess Medical Center and Harvard Medical School, 3 Blackfan Circle, CLS-628H, Boston, MA 02115, USA vsheen@bidmc.harvard.edu

> Received: 01 December 2014 Accepted: 07 March 2015 Published: 01 April 2015

#### Citation:

Lian G and Sheen VL (2015) Cytoskeletal proteins in cortical development and disease: actin associated proteins in periventricular heterotopia. Front. Cell. Neurosci. 9:99. doi: 10.3389/fncel.2015.00099 asymmetric division to generate one self-renewed progenitor and another differentiating intermediate progenitors (basal progenitors) or post-mitotic neurons. The post-mitotic neurons attach and migrate along radial glial scaffolds out of the expanding ventricular zone (germinal zone) creating a new overlying layer of cells called the preplate. The laminar position of neurons is characteristic of their birthdate, such that younger neuroblasts migrate past their older counterparts to form the more superficial layers of the cortex. Finally, once neurons reach their cortical destinations, they differentiate and adopt the complex dendritic and axonal connections that are characteristic of fully mature cortical neurons (Ohnuma and Harris, 2003). At later stages, neurogenesis declines and astroglial differentiation begins.

While different actin cytoskeletal mechanisms are often independently implicated in regulation of neural proliferation, migration and differentiation (Duong et al., 1994; Schaar and McConnell, 2005; Munro, 2006; Witte and Bradke, 2008; Solecki et al., 2009; Moon and Wynshaw-Boris, 2013), a primary role for the actin cytoskeleton in facilitating crosstalk between various cell membrane and intracellular molecules would suggest that shared actin dependent pathways could mediate these developmental processes. For example, cell fate specification of neural stem cells along the neuroependymal lining depends upon the actin cytoskeleton to direct cytokinesis and localization of cell fate determining proteins. This actin dependent process dictates whether progenitors produce daughter progeny capable of self-renewal or postmitotic neurons (Knoblich, 2008). Under a constant rate of cell cycle, the greater the number of progenitors that undergo self-renewal, the larger the number of neural progenitors generated, giving rise to more postmitotic neurons and leading to bigger brain size. Conversely, increased differentiation will lead to a fewer neural progenitors over time, and ultimately fewer neurons and smaller brain size. In this respect, a shared actin cytoskeletal dependent pathway regulates both neural proliferation and differentiation, thereby maintaining a delicate tune of self-renewal vs. differentiation to direct formation of the cerebral cortex. Within this same framework, the same actin cytoskeletal pathways that regulate cell fate specification will also mediate neuronal migration, as intermediate progenitors and post-mitotic neurons will migrate toward the intermediate zone or cortical plate respectively, whereas progenitors remain restricted to the ventricular zone. In this review, we focus on a fundamental actin cytoskeletal pathway that singularly directs each of these developmental stages during corticogenesis.

# Actin Cytoskeletal-Associated Proteins

The majority of actin cytoskeletal-associated proteins participate in the formation of actin filaments by regulating the dynamic processes of actin polymerization and de-polymerization. Actin polymerization occurs under the effect of actin-nucleating proteins, including Arp2/3 and formin proteins. The Arp2/3 complex nucleates G-actin in the lamellipodia to form short, branched actin filaments, whereas formins nucleate actin polymerization into long, unbranched filaments, which are important for both stress fiber formation and contractile ring assembly in mitosis (Bovellan et al., 2014; Fried et al., 2014; Khaitlina, 2014; Pan et al., 2014; Zhang et al., 2015). In addition, the actin polymerization state is also regulated by many small GTP-binding proteins like Rho GTPases, which cycle between active and inactive states through GTP to GDP exchange (Ciobanasu et al., 2013; Chen and Friml, 2014; Murali and Rajalingam, 2014). Actin polymerization coincides with a homeostatic de-polymerization of actin filaments through actinsevering molecules like cofilin and gelsolin, which bind and dissociate G-actin-GDP from actin filaments (Nag et al., 2013; Hild et al., 2014). Finally, the stability and contractile state of the actin cytoskeleton are regulated by many actin-crosslinking proteins, like filamins, actinin and myosin (Otey and Carpen, 2004; Ma and Adelstein, 2014; Modarres and Mofradt, 2014). Cooperative interaction between these various cytoskeletalassociated proteins can subserve multiple fundamental cell functions.

As F-actin-binding proteins, filamins are highly conserved and expressed in all eukaryotic cells. Filamins can crosslink actin filaments into orthogonal networks in the cellular cortical cytoplasm and participate in the anchoring of membrane proteins to the actin cytoskeleton (Gorlin et al., 1990). They also cross-link parallel stress fibers, thereby forming an aligned array in fibroblasts. Structurally, filamins are composed of a tandem calponin-homology (CH) actin-binding domain (ABD) at the N-terminus, 24 immunoglobulin-like repeat domains and a C-terminal dimerization domain. In vertebrates, filamins consist of three isoforms (FlnA, FlnB and FlnC) with molecular weights of ∼280 kDa (Nakamura et al., 2011; Razinia et al., 2012). The FLNA gene is located on human chromosome X q28 and mouse chromosome X q37.89, encoding an approximate 2640-amino acid protein. FlnA is ubiquitously expressed in almost all the tissues, especially in developing brain. The FLNB gene resides on human chromosome 3 and mouse chromosome 14 and its encoded protein is predominantly expressed in bone, whereas FLNC is found in muscle tissue. Although FlnB and FlnC are expressed in the brain, they have not been clearly associated with neurological disorders (Sheen et al., 2002; Krakow et al., 2004; Okumura et al., 2013). FlnA, as a scaffolding protein, can interact with more than 45 proteins, including cell adhesion proteins (i.e., integrin) and cell cycle regulators. Recently, FlnA was found to be essential for formation of the E-cadherin-catenin adhesion complex (Feng et al., 2006; Ferland et al., 2009). FlnA also plays important roles in embryonic development by linking these important adhesion proteins and membrane receptors to the actin cytoskeleton. FlnA loss causes various tissue defects during embryonic development, including periventricular heterotopia (PH), skeletal malformation, disorders in vascular and cardiac development and intestinal defect (Fox et al., 1998; Sheen et al., 2001; Robertson et al., 2003). Missense mutations in FLNA gene, which are thought to lead to gain of function, are also associated with various human diseases, such as otopalatodigital syndrome, Melnick-Needles syndrome, thrombocytopenia, and intestinal pseudo-obstruction (Robertson et al., 2003). These roles for filamins in embryonic development are likely to be closely related to their functional mechanisms on diverse cellular processes in different types of cells, such as cell receptor signaling, endocytosis and exocytosis of membrane proteins through vesicle trafficking, and signaling transduction in the nucleus.

Formin(s) are members of a family of actin-nucleating, cytoskeletal-associated proteins, which stimulate actin nucleation at the barbed ends of actin to form linear filaments (Basu and Chang, 2007). The mammalian genome encodes more than 10 distinct formin proteins, such as mDia1-3, Daam1/2, and formin1/2. All the formin family of proteins includes two common domains, the formin homology 1 and 2 domains (FH1 and FH2). The FH1 domain contains a proline-rich sequence motif that can bind profilin and interact with certain proteins containing SRC Homology 3 (SH3) and WW domains. The WW domains are so named because of the presence of two conserved tryptophans (W) which are spaced 20–22 amino acids apart within the sequence (Bork and Sudol, 1994). Profilinassociated actin can be concentrated at the positive ends of actin filaments, and act as the major source of actin used for filament polymerization. The FH2 domain contains the actin-nucleating domain that can be associated with the fast-growing barbed end of actin filament and increase the actin polymerization rate by binding profilin/actin via its adjoining FH1 domains. The FH2 domain forms a ring-shaped dimer, which binds to the positive end of nascent actin filament and recruits two actin monomers for nucleation (Otomo et al., 2005). Flanking the FH1 and FH2 domains, the N-terminal GTPase-binding domain (GBD) and C-terminal autoregulatory domain (DAD) are also present in some subsets of formin proteins, and are important for formin activation (Gould et al., 2011). The binding of Rho GTPases to GBD is thought to open the DAD association with N-terminal autoinhibitory domain (DID), thereby changing the FH2 conformation and facilitating actin binding and polymerization. Except for the conserved FH2 domain, different subsets of formins display a distinct difference in other domain structures, which may endow formins with the capacity responding to varied sorts of cellular signals (Schönichen and Geyer, 2010). mDia 1-3 (Diaph1-3) genes are located in different genomic chromosomes, encode three 1100–1200 aa formin proteins carrying N-terminal GBD and DID domains, and C-terminal DAD domain. Loss of mDia proteins has been shown to disrupt apical adherens junctions, impact neuroepithelial polarity, and cause periventricular dysplasia in mouse and microcephaly in humans (Thumkeo et al., 2011). The formin 1 (Fmn1) gene is located in mouse chromosome 2 and encodes a 1466-aa protein. It is predominantly expressed in the brain, kidney and developing limb buds, implying a potential role in development of various systems. The formin 2 (Fmn2) gene is located on mouse chromosome 1 and encodes a 1578 aa protein. It is expressed in the central nervous system and developing mesenchyme, suggesting Fmn1 and Fmn2 may share some similar or common functions. Unlike the mDia proteins, it is not known whether Fmn1/2 contains GBD and DID domains. Collectively, expression of the formin genes in nervous system suggests that they may play diverse roles in neural cell adhesion, migration, proliferation and differentiation.

RhoGTPases are small GTP-binding proteins, the most commonly studied being RhoA, Rac and Cdc42 (Cook et al., 2014). They regulate many cell behaviors like polarity, adhesion, cell division and migration primarily by mediating actin cytoskeletal dynamics. Their regulation of actin polymerization is dependent on both Arp2/3- and formin1/2-nucleating processes. As master regulators of the cytoskeleton, RhoGTPases function between active and inactive states by exchanging GDP with GTP. The switch between activity states is modulated by three classes of regulatory proteins, referred to as guanine nucleotide exchange factors (RhoGEFs), GTPase activating proteins (RhoGAPs), and guanine nucleotide dissociation inhibitors (RhoGDIs). Although RhoA, Rac and Cdc42 all regulate actin remodeling and polymerization, they have distinctly different effects on actin reorganization through their respective downstream effectors. RhoA modulates actin polymerization, stress fiber assembly and focal adhesion formation through formins, Rock and myosin. Rac regulates actin filament polymerization through Arp2/3 and WAVE/WASP to produce lamellipodia and membrane ruffles at the leading edge. Cdc42 stimulates filament assembly and filopodia formation via WASP, Arp2/3 and formin. Collectively, RhoGTPases are expressed in neural cells during development of cerebral cortex, regulate downstream actin effectors (formins and Arps), and play pivotal roles in various neural cell developmental functions.

Periventricular Heterotopia is a malformation of cortical development, characterized by nodules of neurons ectopically located along the lateral ventricles. This disorder is thought to reflect impairments along several states of cortical development, including loss in neuroepithelial integrity, disrupted neural proliferation and cell fate specification, and impaired initial neural migration (Sheen, 2012). Mutations in the FLNA gene cause PH (Fox et al., 1998; Sheen et al., 2005). Filamins bind multiple cytoplasmic and cell surface receptors and molecules. Our work shows that filamins also bind formins and RhoGTPases, both of which have been implicated in some kinds of heterotopia formation (Thumkeo et al., 2011; Cappello, 2013). This trimeric complex provides a basic mechanism for modulation of broad actin cytoskeletal dependent processes, essential for cortical development and disease. Given the broad topic of cytoskeletal proteins in brain development, the current review focuses on filamin-associated proteins in corticogenesis, followed by their potential roles in causing PH.

# The Actin Cytoskeleton and Neuroepithelial Integrity

The neuroepithelium forms tight cell -cell or -matrix adhesion junctions at the apical surface of the cortical ventricle. The actin cytoskeleton is also assembled into dense actin cables along the apical surface and anchored onto these adhesion sites through cytoskeletal-associated proteins, such as filamin, formin and catenins. The apical lining is enriched for cytoskeletal proteins and actin filaments play determinant roles on the stability of adhesion junctions, as well as the polarity and integrity of the neuroepithelium. FlnA has been found to be essential for formation of the E-cadherin-catenin adhesion complex and its loss causes aberrant adherens junctions in multiple tissues (Feng et al., 2006; Ferland et al., 2009). Furthermore, FlnA-deficient neural progenitors exhibit poor adhesion to extracellular matrix proteins such as laminin. MDia1 and mDia3 formin proteins are also both expressed in the developing brain, and mDia3 is especially concentrated at the apical surface of the neuroepithelium. Loss of mDia1 and mDia3 impairs neuroepithelial cell polarity with attenuated apical actin belts and impaired apical adherens junctions (Thumkeo et al., 2011). Similar to the above findings, β-catenin is a cytoskeleton-associated protein, linking cadherin(s) to actin filament. Mice with conditional loss of β-catenin show several abnormalities in the neuroepithelium, including loss of adherens junctions, and impairment of radial migration of neurons toward the superficial layers (Machon et al., 2003). Finally, RhoGTPases such as RhoA and Cdc42 are highly expressed in neuroepithelium and essential for assembly and stability of apical actin cables. Conditional loss of RhoA and Cdc42 in central nervous system impairs apical localization of cadherin, apical accumulation of actin filament and cell-cell junctions. Collectively, these observations raise the possibility of a fundamental filamin-RhoGTPase-formin pathway in maintaining the location and function of adhesions molecules such as cadherins, which are required to ensure neuroependymal integrity.

# The Actin Cytoskeleton and Neural Progenitor Proliferation

Cytoskeletal-associated proteins may not regulate only the stability of cell-cell or cell-extracellular matrix adherens junctions, but also mediate cell proliferation by affecting cell cycle progression (Olson et al., 1995; Cappello et al., 2006; Woodhead et al., 2006; Katayama et al., 2011; Lian et al., 2012; Ercan-Sencicek et al., 2015). For example, actin regulates M (mitosis) phase progression, as the filaments are essential for cleavage furrow formation and completion of cytokinesis (Heng and Koh, 2010). Disruption of actin filaments by inhibitory agents such as latrunculin leads to cytokinesis failure due to a defect in the F-actin cable ring at the cleavage furrow (Lee and Song, 2007). FlnA shows strong expression in the cleavage furrow during mitosis. Additionally, functional loss of actin inhibits centrosome separation early in mitosis and leads to a delay in chromosome segregation late in mitosis (Rosenblatt et al., 2004; Cao et al., 2010). Finally, recent reports have suggested that the interaction of the cortical actin network with astral microtubules is crucial in establishing correct spindle orientation and in proper chromosome segregation in mammalian cells (Théry et al., 2005).

Prior to mitosis, G2-M phase entry requires remodeling of the actin cytoskeleton to change cell shape from an extended to a rounded morphology with cell retraction (Maddox and Burridge, 2003; Cao et al., 2010; Heng and Koh, 2010).This change in cell morphology initially requires activation of RhoA, which triggers a signaling cascade through formin and Rock to re-organize the actin cytoskeleton. Additionally, the cyclin dependent kinase 1 (Cdk1) promotes G2-M transition, and activated Cdk1 phosphorylates Rho GTPase activating protein (p190RhoGAP), down regulating p190RhoGAP hydrolysis of Rho-GTP and thereby promoting RhoA function (Maddox and Burridge, 2003). Our recent study shows that FlnA also regulates Cdk1 activity through Cdk1 phosphorylation and cyclin B degradation (Lian et al., 2012). Lastly, FlnA regulates RhoA activity through mediating p190RhoGAP accumulation in lipid rafts (Mammoto et al., 2007). Therefore, a potential signaling pathway underlying rearrangement of the actin filaments in G2-M phase involves a cascade beginning with FlnA, Cdk1, and p190RhoGAP, which then collectively mediate RhoA and formin/Rock function. This signaling pathway may also be crucial for cell cycle progression through M phase. Loss of FlnA impairs degradation of cyclin B1-related proteins, thereby delaying the onset and progression through mitosis (Lian et al., 2012). Furthermore, loss of FlnA increases the inhibitory phosphorylation of Cdk1 via its interaction with the kinase Wee1. In developing cerebral cortex, FlnA loss causes a decrease in proliferation rate of neural progenitors and a decline in neural progenitor pool size (Lian et al., 2012). This prolongation in cell cycle would relate to the impairment in actin filament rearrangement. Our understanding of the formin role in cell division is more limited, but insight can be gained from our understanding of formin function in other organ systems. Apart from a known interaction with filamins, Fmn1-deficient mice exhibit a reduction in digit number as well as the absence of a fibula due a defect in chondrocyte proliferation (Zhou et al., 2009). Fmn1 loss is linked to up-regulation of BMP and Msx1 but down-regulation of Fgf4 signals within the apical ectodermal ridge, which may influence mitosis. Finally, other RhoGTPase-related proteins also influence M phase progression. Both constitutively active and dominant negative Cdc42 inhibit cytokinesis (Drechsel et al., 1997), and loss of its downstream effector mDia3 causes chromosome misalignment during metaphase (Yasuda et al., 2004). Expression of dominant negative Rac1 retards adventricular nuclear migration, and promotes cytokinesis failures (Michaelson et al., 2008; Minobe et al., 2009).

The actin cytoskeleton mediates G1 phase progression after completion of mitosis. Disruption of actin polymerization by the cytochalasin D causes G1 phase arrest (Bohmer et al., 1996; Lian et al., 2012). The cytoskeletal-dependent effects on G1 progression is mediated through cyclin expression and cyclindependent kinase (Cdk) activation. More specifically, the actin cytoskeleton is required for anchorage-dependent expression of cyclin D1, activation of Cdk4/6, phosphorylation of the retinoblastoma protein and transition of G1 phase in nontransformed primary cells. In addition to this primary pathway, cytoskeletal-associated proteins regulate G1 phase progression through other secondary mechanisms. Inhibition of Cdc42, Rac1 and RhoA block G1 phase transition and serum-induced DNA synthesis (Olson et al., 1995; Leone et al., 2010). Active RhoA increases the expression of Skp2 protein, which promotes ubiquitinylation-dependent degradation of the Cdk inhibitor p27kip1 (Mammoto et al., 2004). Conversely, RhoA inactivation results in higher levels of p27kip1, thereby arresting cell cycle in G1 phase. RhoA inactivation or F-actin disruption are also shown to slow down the degradation of another Cdk inhibitor p21Waf/Cip1 (Coleman et al., 2006). With respect to filamins, our prior study suggests that FlnA also regulates neural progenitor proliferation in G1 phase (Lian et al., 2012) and directs cadherincatenin complex formation. β-catenin is known to mediate G1 phase progression and neural proliferation, and tethers cadherin to the actin cytoskeleton (Woodhead et al., 2006). Conditional deletion of β-catenin in developing mouse brain results in a dramatic defect in neural progenitor proliferation and severe brain malformation. In contrast, overexpression of a stabilized β-catenin causes a significant increase in neural progenitor number and massive expansion of the cerebral cortex (Chenn and Walsh, 2002). Similar to filamins, formins may play coordinative effects on cell proliferation through G1 phase. Our ongoing studies suggest that loss of filamin and formin can affect β-catenin translocation and cyclin D expression. Compared to loss of FlnB or Fmn1 alone, loss of both Fmn1 and FlnB in mice leads to a more severe reduction in body size, weight and growth plate length (Hu et al., 2014). These findings would suggest that these actin associated proteins can mediate cell proliferation in multiple organ systems.

From the discussion above, the filamin-formin-RhoGTPase pathway can potentially be implicated in several phases of the cell cycle through interactions with various cell cycle associated proteins. A prevailing role for these proteins in regulation of actin dependent vesicle trafficking would provide a common mechanism for the transport and degradation cell cycle and cell fate proteins, which oversee neural proliferation.

# The Actin Cytoskeleton and Cell Polarity and Fate Specification

The neuroepithelium comprises a distinctive epithelial structure with apical-basal polarity. The actin cytoskeleton is selectively concentrated at the apical side along the ventricle of the developing cerebral cortex, forming a dense and dynamic filament belt to support tight adhesive junctions, cilium stability, and cell polarity, and to maintain a membrane barrier. Acting as regulators of cell shape and binding-partners for polarity proteins, the actin filament and its associated proteins are of key importance for regulating cell polarity (Ohno, 2001; Sawin, 2002; Etienne-Manneville, 2004; Witte and Bradke, 2008; Wang et al., 2009; Gonzalez-Billault et al., 2012). As an example, Rho-GTPases like RhoA, Cdc42 and Rac1 regulate actin cytoskeleton remodeling, focal adhesion formation and cell polarity (Etienne-Manneville, 2004; Iden and Collard, 2008; Gonzalez-Billault et al., 2012). Activated Cdc42 forms a stable hetero-tetrameric complex with polarizing proteins Par3, Par6, and atypical protein kinase C (PKCζ) and recruits these molecules to the leading edge to guide the reorientation of the microtubule and centrosome (Ohno, 2001; Etienne-Manneville, 2004). Cdc42-deficient neural progenitors exhibit multiple apical polarity-related defects including disorientation of cell division, aberrant location of the Par complex and adherens junctions, and severe impairments in the extension of nestin-positive radial fibers (Chen et al., 2006; Peng et al., 2013). Further, Cdc42 loss also causes PH and holoprosencephaly. As effectors of Rho-GTPases, formins and non-muscle myosin II have been shown to be indispensable for cell polarity (Habas et al., 2001; Ma et al., 2007). Depletion of the formin homologous protein Daam1 prevents Wnt/Fz activation of Rho and planar cell polarity during Xenopus gastrulation. Ablation of non-muscle myosin II-B in mice results in loss of neuroepithelial adhesion and severe hydrocephalus. Upstream of Rho GTPases, the association of FlnA with Wnt co-receptor Ror2 is required for Wnt5a-induced JNK activation, appropriate orientation of the microtubule organizing center and cell polarity (Nomachi et al., 2008). Further, loss of FlnA leads to a transition from bipolar neuron to multipolar neuron, suggesting a FlnA effect on neuronal polarity (Nagano et al., 2004). Given their physical interaction, filamins and RhoGTPases might regulate formin dependent polarized actin nucleation. Polarized actin provides a mechanism for establishment of neuroepithelial polarity.

Cytoskeletal proteins regulate cell proliferation not only by affecting cell cycle progression, but also through their control over cell fate specification (Chenn and Walsh, 2002; Taverna et al., 2014). During neuroepithelial cell fate specification, neural stem cells must undergo a decision process to undergo self-renewal or differentiate into intermediate progenitors or neurons. This process is dependent upon the asymmetric inheritance of cell fate determining proteins, which is regulated by polarized actin, actin dependent trafficking and degradation of cell fate proteins. In this respect cytoskeletal proteins could indirectly mediate cell fate. Cell fate determinants like Par3, aPKC, and numb as well as stem cell niche molecules like integrin and cadherin all directly or indirectly interact with the actin cytoskeleton, such that cell fate is influenced by cytoskeletal dynamics (Guo et al., 1996; Barros et al., 2003; Cappello et al., 2006; Woodhead et al., 2006). Furthermore, conditional deletion of Cdc42 at different stages of neurogenesis in mouse telencephalon results in an immediate increase in basal mitoses and altered differentiation of neural progenitors. In mesenchymal cells, loss of p190RhoGAP, which inactivates RhoA activity through GTP hydrolysis, causes a concomitant up-regulation of RhoA activity and increase in myogenic differentiation (but decrease in adipogenesis) (Sordella et al., 2003). In conditional RhoA-deleted embryos, RhoA-deficient neural progenitor cells exhibit accelerated proliferation and reduction in cell-cycle exit, indicating a change in cell fate specification (Katayama et al., 2011). Prior studies have shown that FlnA can bind to RhoA and integrin. Some of our initial observations suggest that loss of FlnA impairs cell cycle exit, in part through disruption of spindle orientation of neural progenitors during mitosis. Moreover, FlnA interactions with formins which nucleate actin in a polarized fashion would provide a basis for asymmetric delivery and localization of cell fate determining proteins.

# The Actin Cytoskeleton and Neural Migration

Cell migration is a highly dynamic process involving cell adhesion, extension, protrusion of filopodia and lamellipodia at leading edge and formation of contractile structure at rear edge. All the events require the dynamic remodeling of actin cytoskeleton (Rottner and Stradal, 2011). A variety of literatures have reported that actin cytoskeleton and its associated proteins are essential for cell migration (Fox and Walsh, 1999; Ridley et al., 2003; Raftopoulou and Hall, 2004; Govek et al., 2005; Broussard et al., 2008). For instance, disruption of actin filaments with drug cytochalasin C in migrating cerebellar granule cells can completely block cell migration (Rivas and Hatten, 1995). Further, melanocytes lacking FLNA show defects in filopodia formation and abnormal surface blebbing, implying a necessary role for FLNA in promoting the assembly of the cortical actin network. Several types of FLNA-deficient cells like macrophages, melanocytes and Dictyostelium amoebae cells display profound defects in motility and chemotaxis (Cunningham et al., 1992; Cox et al., 1996). In contrast, re-expression of FLNA in the cells can rescue each of these phenotypes, further establishing the essential effects of FLNA on migration. FLNA also is concentrated at rear edge of migrating leukocytes, implying FLNA may execute some important functions on rear edge retraction via FLNA-crosslink-driven force (Ohta et al., 2006). Notably, recent studies using transgenetic FlnA-deficient mouse model result in some contradictory conclusions to FLNA effects on neural cell migration (Fox et al., 1998; Feng et al., 2006; Hart et al., 2006). FlnA loss does not affect in vivo migration of neural crest cells into neural-crest-derived tissues like endocardial cushion, and the membrane ruffling, locomotion and migration of in vitro cultured fibroblast cells also appear normal. However, BrdU pulse labeling in embryonic day 15 null FlnA brains shows that the migration of BrdU<sup>+</sup> cells into the cortical plate is slower than that of age matched wild type cells. Most BrdU<sup>+</sup> cells from null FlnA cerebral cortices migrate into intermediate zone, but not into the cortical plate, as seen in their wild type littermates by 3 days post-labeling (Zhang et al., 2013). Further, cultured neural progenitor cells from E13 FlnA-null cerebral cortex also show poor spreading on laminin-coated coverslips, implying impairment in cell adhesion on extracellular matrix (Zhang et al., 2013). Finally, FlnA regulates the stability and turnover of adhesion and migratory associated proteins such as paxillin (Zhang et al., 2012, 2013). These migratory defects may in part be related to FlnA interactions with Filip, which regulates the degradation of the actin binding protein and thereby influences neural migration and cell specification (Nagano et al., 2002, 2004). These findings would suggest that filamins may regulate neural motility but do not completely abolish the capacity of progenitors to migrate to their intended site. Thus, these cytoskeletal associated proteins may not form the primary mechanism required to allow cells to migrate, but rather mediate processes (i.e., trafficking of particular receptors/molecules) that influence the rate at which neural progenitors move.

RhoGTPases such as RhoA, Rac1 and Cdc42 regulate remodeling of actin cytoskeleton, thereby serving as key regulators for cell morphology and migration (Raftopoulou and Hall, 2004). Their effects on neural migration have been extensively studied. Basically, Cdc42 and Rac1 stimulate formation of filopodia and lamellipodia (Yang et al., 2012), to direct neurite outgrowth and promote neural cell migration (Chen et al., 2007), whereas RhoA promotes retraction of rear edge and nuclear translocation of neural cell but impairs neurite extension. Other small GTP-binding proteins such as Rnd2 and Rnd3 have also been implicated in neural migration, although their relationship with filamins and formins is not known (Heng et al., 2008; Azzarelli et al., 2014). As downstream effectors of Rho GTPases, formins play potential important roles in polymerization and remodeling of actin filament. For instance, mDia deficiency impairs tangential migration of cortical and olfactory inhibitory interneurons (Shinohara et al., 2012). mDia-deficient neuroblasts exhibit reduced separation of the centrosome from the nucleus, retard nuclear translocation and concomitantly impair F-actin movement and condensation at the cellular rear. In non-neural cells, mDia1 has been shown to regulate formation of stable actin filaments and turnover of focal adhesions. mDia1 deletion impairs focal contacts, and decreases lamellipodial thickness in migrating cells (Yamana et al., 2006). Other formins like Fmn1 and FLR also are shown to function in cellular migration (Zhou et al., 2009; Hu et al., 2014). Fmn1 protein is more similar in sequence to dishevelledassociated activator of morphogenesis-1 (Daam1). Mutations in mouse Fmn1 gene result in limb deformities and incompletely penetrant renal aplasia, and display altered cell protrusion at the leading edge, defective cell spreading, and less focal adhesions. Our recent findings show that formin(s) can physically interact with filamin(s) and they co-express in developing bone and brain tissues (Hu et al., 2014). Loss of both formin and filamin results in serve defects in cell proliferation and migration in developing mouse thoracic wall and brain, suggesting that filamin and formin play cooperative roles in cell proliferation and migration.

Cell migration is a complex and cooperative process with protrusion at cell leading edge and with concomitant retraction at rear edge. The molecular mechanisms underlying neural migration may include some signaling pathways from G proteincoupled receptors, tyrosine-kinase receptors, PI3K, MAPK and Rho GTPases and cytoskeletal reorganization (Witte and Bradke, 2008; Jurberg et al., 2014). Here, it needs to be underscored that these processes of signal transductions require directional and robust vesicle trafficking that cytoskeletal proteins regulate. For instance, directional vesicle trafficking from cell rear edge to leading edge may play crucial roles on cell migration. Thus in a manner similar to establishment of cell polarity by directing the localization of fate determining proteins, filamins, formins and RhoGTPases can coordinate similar processes during neural migration.

The functional roles of actin cytoskeletal genes in adhesion junctions and cell migration are also mutually interdependent. In the developing cerebral cortex, radial glial cells extend their end feet onto the apical lining of ventricle, where they form dense cell to cell/matrix connections via cell adhesion proteins like cadherins and integrins at the tips of the end feet. Newborn neuron or pre-neuron migrate out of ventricle along the radial glial fibers. Thus, loss of cytoskeletal-associated proteins causes destabilization of the interdependent adhesion complex (adhesion proteins and actin filament) and disconnection of radial glial at the apical lining. Impairment in the connection of radial glial end feet would interrupt migration along the radial glial tracks, thereby contributing to the neuronal migration defect. In addition, actin cytoskeletal and adhesion proteins are also enriched within the growth cone of migrating neuron. Loss of these proteins and/or disruption of their trafficking likely influence neuronal migration itself.

Finally, neural migration is dependent on the differentiation state of the precursor. For example, Rac1 is necessary for neural progenitor transition from G1 to S phase, at least in part by regulating cyclin D levels and retinoblastoma protein phosphorylation. Loss of Rac1 in progenitors impairs the migration of ventral GABAergic neurons into the cortical plate. Ablation of Rac1 from postmitotic progenitors does not result in similar defects (Vidaki et al., 2012).

# Cytoskeletal-Associated Vesicle Trafficking: A Common Thread in Brain Development

Vesicle trafficking maintains the apical-basal polarity of neuroepithelium through directional vesicle transport and membrane protein sorting (transcytosis), thus underlying cell polarity, migration and asymmetric division (Rodriguez-Boulan et al., 2005). Vesicle trafficking includes endocytosis, exocytosis and endosomal recycling and sorting (Symons and Rusk, 2003). Endocytic trafficking is characterized by the internalization of plasma membrane and extracellular molecules via several distinct pathways: micropinocytosis, phagocytosis, and clathrin and caveolae-mediated endocytosis. This process is dependent on actin cytoskeletal-associated proteins like Rho GTPases (Ridley, 2006). For example, dominant-negative Cdc42 or Rac1 can block macropinocytosis, while constitutively active Cdc42 and Rac1 can restore the pinocytosis in immature dendritic cells (Garrett et al., 2000). Actin dynamics is also crucial for phagocytosis. Both Cdc42 and Rac participate in FcγR-mediated phagocytosis with Cdc42 directing pseudopod extension, and Rac functioning in pseudopod fusion and phagosome closure (Massol et al., 1998). During clathrin-mediated endocytosis, overexpression of constitutively active Rac1 and RhoA inhibits clathrin-mediated internalization of transferrin and EGF receptor in Hela cells (Lamaze et al., 1996), whereas RhoA can stimulate this process in polarized MDCK cells, suggesting that Rho GTPase function is dependent on cell polarity state (Leung et al., 1999). Caveolae are comprised of cholesterol-enriched internalized plasma membrane and are closely associated with actin filament. Integrins with caveolin have been shown to mediate RhoA activity in endothelial cells. Both RhoA and caveolin must co-localize and interact to mediate RhoA dependent actin remodeling (Nuno et al., 2009; Yang et al., 2011).

Formin and filamin proteins have been increasingly implicated in vesicle trafficking. MDia, together with myosin II, controls the initiation of E-cadherin endocytosis in the epithelium by regulating the lateral clustering of E-cadherin (Levayer et al., 2011). In oocytes, long-range transport of vesicles is regulated by Fmn2 through assembling an extensive actin network from the vesicles' surfaces to plasma membrane. The vesicles move directionally along these actin cables to reach the cell surface (Schuh, 2011). Recent work also reveals that filamin binds to the caveolae marker, caveolin, and is required for endocytic trafficking of caveolin (Muriel et al., 2011). The endocytic trafficking of caveolae towards a recycling endosome is impaired in FLNA-deficient HeLa and M2-melanoma cells.

Besides the effect on endocytic trafficking, actin cytoskeletalassociated proteins may also be involved in exocytosis. Actin-associated proteins (filamin, myosins and Cdc42) are present in the Golgi complex, and actin filament may function as tracks for the myosin-driven movement of vesicles. Expression of Cdc42 mutants slows the exit of the basolateral marker N-cadherin from the trans-Golgi network while also stimulating the exit of the apical marker neurotrophin receptor p75 (Musch et al., 2001). RhoGTPases associate with the Golgi apparatus in an ARF-dependent manner. The ADP ribosylation factor guanine exchange factor 2 (ARFGEF2 encodes for BIG2) is a guanine nucleotide-exchange factor for ARF1/3, which play an important role in vesicular trafficking from Golgi complex to plasma membrane. Inhibition of BIG2 disrupts membranous localization of adherens junction protein E-cadherin and betacatenin by preventing their transport from the Golgi apparatus to the cell surface, leading to PH formation (Sheen et al., 2004; Zhang et al., 2012, 2013). Given that filamin can physically bind and interact with ARFGEF2 (Zhang et al., 2012), filamin and its associated proteins might orchestrate the vesicle exocytosis from Golgi to plasma membrane.

# Genetics of Periventricular Heterotopia

Periventricular heterotopia is one of the most common malformation of cortical development and causes seizures, dyslexia, and psychiatric disturbances (Sheen, 2012). PH is characterized by bilateral ectopic neuronal nodules found along the lateral ventricles (Lu and Sheen, 2005). The nodules are caused by impaired migration from the ventricular zone and disruption in the integrity of the neuroependyma (Sheen, 2012). Mutations in the causative genes also cause microcephaly (meaning small brain) (Sheen et al., 2003, 2004).

The most common form of PH is inherited in an X-linked dominant fashion from mutations in the FLNA gene (Fox et al., 1998; Sheen et al., 2001, 2005). A second form of autosomal recessive PH with microcephaly (ARPHM) has been associated with mutations in the ARFGEF2 gene. ARFGEF2 encodes brefeldin-A inhibited guanine exchange factor-2 (BIG2) (Sheen et al., 2004). BIG2 is a protein kinase A anchoring protein (AKAP) which regulates Golgi-vesicle trafficking through its Sec7 domain. Recent work has identified cadherin receptor ligands in causing PH due to mutations in FAT4 and DCHS1 (Cappello et al., 2013).

Several mouse genes have been shown to cause PH formation and can be functionally linked. αSNAP is a SNARE-related protein, involved in vesicle fusion. Prior reports have shown that αSNAP mediates VE-cadherin localization through a β1-integrin-associated process (Andreeva et al., 2005). FlnA binds β1-integrin (Calderwood et al., 2001). Mekk4 binds and regulates FlnA (Chi et al., 2005), and therefore could indirectly regulate caveolin mediated endocytosis. The RhoGTPases bind FlnA and direct various aspects of intracellular actin dynamics, which are required for endosomal vesicle transport (Cappello et al., 2006; Katayama et al., 2011). Deficiency of the formin associated mDia disrupts integrity of neuroepithelium and causes periventricular dysplasia (Thumkeo et al., 2011). Similarly, Spred1 is a multidomain scaffolding protein that contains an ENA/VASP domain which modulates actin stress fiber remodeling, like filamins. Spred1 is also associated with specific endosomal vesicles (Phoenix and Temple, 2010). Finally, SCF-ckit affects several downstream pathways including RAS/ERK and JAK/STAT pathways, both of which have been associated with Mekk4 and Spred1 activity (Rönnstrand, 2004; Soumiya et al., 2009).

Similar radiographic or anatomical findings of PH and potentially linked functions suggest that genes causal for PH might be involved in a common molecular pathway important in neural progenitor development. Prior studies have demonstrated a shared interaction between FlnA and Big2 in activating Arf to form vesicles at the cell membrane, and thereby regulate turnover/stability of cell adhesion molecules (Zhang et al., 2012, 2013).

# Anatomical Phenotypes Associated with PH

Periventricular heterotopia refers to heterotopic neurons along the lateral ventricles which are caused by impaired motility/migration and disruption of the neuroependymal lining. For example, loss of FlnA results in fewer post mitotic cells reaching the cortical plate compared to wild type (WT) following BrdU incorporation. Cells remain in the intermediate and ventricular zones, consistent with a cell autonomous migratory defect (Zhang et al., 2013). Disruption in the neuroependymal lining has also been reported with FlnA inhibition (Adams et al., 2012). Loss of neuroependymal integrity is the primary cause of heterotopia formation.

While human and mouse genes associated with PH show brain heterotopia, they all also regulate neural proliferation. Human ARFGEF2 mutations cause microcephaly. While microcephaly is not seen in females with FLNA mutations (likely due to mosaicism), males die at birth and have been reported to have thinner cortices, and loss of cortical convolutions consistent with underlying microcephaly (Guerrini et al., 2004). Human mutations in cadherin-associated DCHS1 and FAT4 alter progenitor proliferation (Cappello et al., 2013). FlnA null mice also have microcephaly (Lian et al., 2012). Disruption of Cdc42 and RhoA cause PH in mice and lead to changes in brain size and/or progenitor proliferation (Cappello et al., 2006; Katayama et al., 2011). Dysregulation of PH associated Napa, Stem Cell Factor 1 (SCF) and Spred1 genes also alter progenitor proliferation in mice (Chae et al., 2004; Ferland et al., 2009; Phoenix and Temple, 2010). Mekk4 loss causes PH and a smaller brain with increased cell death (Chi et al., 2005). Formins have been implicated in proliferation of mouse neuroepithelial cells (Thumkeo et al., 2011). Lastly, a nonsense

mutation of the formin related DIAPH1 (human mDia1) in humans has been found to cause microcephaly (Ercan-Sencicek et al., 2015).

neural proliferation.

There are several reasons to believe that the heterotopia formation, disruption in neuroependyma integrity, and impairments in neural proliferation are integrally linked. First, from an anatomical basis, FlnA loss leads to a smaller brain through prolongation in cell progression through mitosis (M) phase (Lian et al., 2012). In cortical development, M phase occurs at the neuroepithelial lining suggesting a shared mechanism with PH formation (defects in adhesion/migration also occur at the neurependyma). Second, from a molecular standpoint, we have shown that Big2/FlnA regulate turnover of adhesion molecules (catenins and cadherins) at the neuroependyma (Zhang et al., 2013). These molecules have been show to regulate brain size and their associated molecules cause PH (Chenn and Walsh, 2002; Cappello et al., 2013). Lastly, a delay in differentiation would also lead to a delay in neural migration into the cortical plate.

#### Cellular and Molecular Mechanisms in PH

Periventricular heterotopia was initially thought to derive from a simple failure in neuronal migration given FLNA's regulation of the actin cytoskeleton (Fox et al., 1998). However, several observations indicate that PH formation may be more complex, derived from diverse processes associated with cytoskeletal dynamics. First, BrdU labeling shows that loss of FlnA in developing mouse brain results in slower migration of neurons, but no PH formation, implying that the impairment in neuronal migration may not be the primary reason for PH formation (Zhang et al., 2013). Second, conditional deletion of RhoA in developing brain causes severe PH formation, but loss of RhoA does not impair neuronal migration (Katayama et al., 2011; Cappello, 2013). Third, as more genes causative for PH are identified, their shared function would argue that adherens junctions and actin cytoskeletal dynamics along the ventricular lining may be the primary pivotal factor for PH formation (Brault et al., 2001; Machon et al., 2003; Chae et al., 2004; Sheen et al., 2004; Kadowaki et al., 2007; Ma et al., 2007; Thumkeo et al., 2011; Peng et al., 2013). Consistent with this view are the findings that the actin filament network around the heterotopia in the PH brains is disrupted (Ferland et al., 2009). Concomitantly, the expression of neuroepithelial polarity and adherens junction proteins along the apical lining decreases or disappears.

Recent studies from this laboratory have begun to reconcile how two seemingly dissimilar proteins Big2 and FlnA can give rise to PH (Zhang et al., 2012, 2013). Either acute or chronic loss of either Big2 or FlnA leads to impairments in neural migration during development of the cortex. Migratory neural cells show defects in filopodia extension and attachment onto extracellular matrix coated surfaces. Both proteins physically bind and interact within neural progenitors, and loss of protein expression of either FlnA or Big2 leads to compensatory upregulation of the other. As with many proteins that FlnA binds, Big2 localization is dependent on phosphorylation of the actin binding protein which redirects Big2 from the Golgi to the cell membrane. Relocalization to the membrane allows Big2 to activate Arf1. Arfs have been shown to reside at the cell surface with ARF1 and ARF3 mediating endocytosis (Dong et al., 2010; Kondo et al., 2012). The reciprocal regulation likely reflects a negative feedback, where loss of Big2 promotes phospho-FlnA expression to enhance redistribution of Big2 to the membrane. Conversely, loss of FlnA enhances Big2 expression to allow for Big2 delivery to the membrane. These studies begin to suggest an integral relationship between FlnA dependent actin dynamics and Big2 dependent regulation of vesicle formation and trafficking.

# Periventricular Heterotopia as a Disorder of Vesicle Trafficking

Changes in Arf-dependent endocytosis have the potential to disrupt several cell developmental processes and provide a hypothetical model for PH formation (see the cartoon in **Figure 1**; Sheen, 2012). First, the primary anatomical defect leading to the PH phenotype (loss in neuroependymal integrity, impaired migration, and reduced proliferation) occurs within neural progenitors along the neuroepithelial lining. Genes implicated in PH (FlnA, Big2) regulate the stability, turnover and degradation of cell adhesion molecules (β-catenin, N-cadherin) and cell-ECM receptors (integrins, paxillin adaptor proteins). FlnA phosphorylation targets Big2 to the membrane, allowing for Arf-dependent activation and vesicle formation. Endocytosis occurs through a caveolin dependent mechanism leading to internalization of catenin/cadherin and integrins. Specificity for these molecules extends from their binding of filamins (Calderwood et al., 2001). Several associated PH genes (RhoA, αSNAP and Spred1) would be expected to participate in this process given their implied association with trafficking and/or endocytosis but their specific roles are not known. The downstream mechanisms that regulate endosomal and lysosomal/proteosomal processing in contributing to PH are also not known, although this review would point to a RhoA and Fmn dependent process.

In short, disruption of Big2 and FlnA (and presumably Fmn and RhoGTPases) would alter the stabilization, turnover and degradation of the cadherin, catenin and integrin proteins through impaired vesicle trafficking. Loss of cell adhesion (via altered cadherin) would disrupt the neuroependymal lining. Impaired cadherin stability would also alter catenin localization and function, thereby leading to impaired proliferation. Alteration of cell fate would be closely linked to proliferation and also effect neural migration. For example, slower progression through the cell cycle (as seen with loss of FlnA) causes delayed differentiation of neural progenitors at a given age (as seen with enhanced symmetric cell divisions from FlnA loss). This delayed differentiation would also lead to slower neural migration from the ventricular zone into the cortical plate. In this respect, many of the phenotypes seen with PH can be explained through changes in actin cytoskeletal regulation via FlnA-RhoA-Fmn2 (and Big2) within neural cells along the ventricular lining.

# Conclusion

Actin cytoskeletal-associated proteins play a variety of diverse and key roles in cerebral cortex development. A basic mechanism for filamins, formins and RhoGTPases may extend from their regulation of dynamic vesicle trafficking, in directed neural progenitor development. Directional transport of actin cytoskeletal-associated vesicles toward apicolateral or basolateral membranes along the neuroepithelium may be crucial for establishing polarity and maintaining adherens junctions. Loss of this function appears to be the primary pathology underlying PH formation. Cytoskeletal-regulated vesicle trafficking may transmit extracellular signals into cytoplasm and nucleus, as well as mediate degradation of cell cycle associated proteins, thereby regulating cell cycle progression and proliferation. PH has been associated with microcephaly. Finally, directional vesicle trafficking, as well as regulation of turnover of proteins toward the leading edge of migratory neural cells, would be required for proper neuronal migration, and account for the impaired migration seen in this disorder.

# Acknowledgments

This work was supported in part by the National Institutes of Health (1 R01 NS092062 01 to VLS).

# References


filamin): a molecular leaf spring. J. Cell Biol. 111, 1089–1105. doi: 10.1083/jcb. 111.3.1089


subventricular and intermediate zones during radial migration. J. Neurosci. 24, 9648–9657. doi: 10.1523/jneurosci.2363-04.2004


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Lian and Sheen. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Impact of prenatal environmental stress on cortical development

#### Seiji Ishii <sup>1</sup> and Kazue Hashimoto-Torii 1, 2, 3 \*

*<sup>1</sup> Center for Neuroscience Research, Children's National Medical Center, Children's Research Institute, Washington, DC, USA, <sup>2</sup> Department of Pediatrics, Pharmacology and Physiology, School of Medicine and Health Sciences, The George Washington University, Washington, DC, USA, <sup>3</sup> Department of Neurobiology, School of Medicine, Kavli Institute for Neuroscience, Yale University, New Haven, CT, USA*

Prenatal exposure of the developing brain to various types of environmental stress increases susceptibility to neuropsychiatric disorders such as autism, attention deficit hyperactivity disorder and schizophrenia. Given that even subtle perturbations by prenatal environmental stress in the cerebral cortex impair the cognitive and memory functions, this review focuses on underlying molecular mechanisms of pathological cortical development. We especially highlight recent works that utilized animal exposure models, human specimens or/and induced Pluripotent Stem (iPS) cells to demonstrate: (1) molecular mechanisms shared by various types of environmental stressors, (2) the mechanisms by which the affected extracortical tissues indirectly impact the cortical development and function, and (3) interaction between prenatal environmental stress and the genetic predisposition of neuropsychiatric disorders. Finally, we discuss current challenges for achieving a comprehensive understanding of the role of environmentally disturbed molecular expressions in cortical maldevelopment, knowledge of which may eventually facilitate discovery of interventions for prenatal environment-linked neuropsychiatric disorders.

Keywords: cortical development, prenatal environmental stress, alcohol, autism, schizophrenia, maternal immune activation, gene-environment interaction, iPS cells

# Introduction

The development of the cerebral cortex consists of very intricate multifaceted steps including proliferation/differentiation of neural progenitor cells, neuronal migration and maturation (Whitford et al., 2002; Kriegstein and Noctor, 2004; Kriegstein et al., 2006; Ayala et al., 2007; Barnes and Polleux, 2009; Rakic, 2009; Rakic et al., 2009; Evsyukova et al., 2013; Lewis et al., 2013), and it can be impaired by exposure to environmental stress (Ben-Ari, 2008; Deverman and Patterson, 2009; Thompson et al., 2009). Even subtle disturbances in the development of the cerebral cortex impair cognitive and memory functions (Berger-Sweeney and Hohmann, 1997; Arnsten, 2009). Accordingly, ever increasing attention is being paid to understanding the underlying non-genomic alterations thought to govern impairment.

Alcohol is known as one of the most prevalent prenatal environmental stress, and prenatal alcohol exposure-linked impairments are categorized under the term "Fetal Alcohol Spectrum

#### Edited by:

*Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan*

#### Reviewed by:

*Richard S. Nowakowski, Florida State University, USA Mladen-Roko Rasin, Robert Wood Johnson Medical School, USA Eric C. Olson, State University of New York Upstate Medical University, USA*

#### \*Correspondence:

*Kazue Hashimoto-Torii, Center for Neuroscience Research, Children's Research Institute, Children's National Medical Center, and Department of Pediatrics, Pharmacology and Physiology, The George Washington University School of Medicine and Health Sciences, 111 Michigan Avenue N.W., M7633, Washington, DC, 20010-2970, USA KHTorii@childrensnational.org*

> Received: *01 February 2015* Accepted: *13 May 2015* Published: *27 May 2015*

#### Citation:

*Ishii S and Hashimoto-Torii K (2015) Impact of prenatal environmental stress on cortical development. Front. Cell. Neurosci. 9:207. doi: 10.3389/fncel.2015.00207*

**Abbreviations:** iPS cells, induced Pluripotent Stem cells; FASD, fetal alcohol spectrum disorder; HSP(s), Heat Shock Protein(s); HSF1, Heat Shock Factor 1; MIA, maternal immune activation; IL-6, interleukin-6; DISC1, disrupted-inschizophrenia-1.

Disorder (FASD)." FASD patients show higher rates of comorbidity with various types of neuropsychiatric problems, such as attention deficit hyperactivity disorder (ADHD) and epilepsy (Mattson and Riley, 1998). Histological analysis using postmortem tissues from FASD patients documented various anomalies in the brain, including heterotopias, microcephaly, hydrocephaly, and agenesis of the corpus callosum (Clarren and Smith, 1978; Roebuck et al., 1998; Muralidharan et al., 2013). Many of these morphological phenotypes, as well as behavioral phenotypes of human patients, have been reproduced by non-human primate, rodent and other vertebrate models of fetal alcohol exposure, and therefore, these animal models have been used for understanding etiology of FASD and other health problems linked to prenatal alcohol exposure (Miller and Nowakowski, 1991; Kelly et al., 2009; Wilson and Cudd, 2011; Patten et al., 2014). Furthermore, these animal studies found that fetal alcohol exposure particularly affects the development of the cerebral cortex, in multiple cellular events including proliferation, differentiation, apoptosis, migration, synaptogenesis and dendritogenesis, depending on the regimens and timing of exposure (Lindsley et al., 2006; Thompson et al., 2009; Miranda, 2012).

Similarly, clinical and epidemiological studies identified a variety of environmental stressors, exposure to which increases the risk of neuropsychiatric diseases (Schmitt et al., 2014). Importantly, rodent and non-human primate models of prenatal exposure to those environmental factors, including hypoxia (Golan et al., 2009; Howell and Pillai, 2014), drugs such as cocaine (Gressens et al., 1992; Cabrera-Vera et al., 2000; Stanwood et al., 2001; Lidow and Song, 2001a,b; Crandall et al., 2004; Thompson et al., 2009), and heavy metals such as methylmercury (Kakita et al., 2001; Hashimoto-Torii et al., 2014), have shown that these factors cause similar structural anomalies in the cortex as well as similar abnormal behaviors (Thompson et al., 2009). These findings imply that different environmental challenges provide common impacts on cortical development, thereby resulting in similar endophenotypes.

Here, we review recent publications that found molecular mechanisms underlying pathological cortical development elicited by exposure to prenatal environmental stress and discuss how various types of prenatal environmental stress similarly affect cortical development and increase the risk of neuropsychiatric disorders.

# Early Response Genes That Protect or Disturb Cortical Development under the Conditions of Exposure to Environmental Stress

Based on recent findings using prokaryotes, genes that respond (either by increase or decrease of expression) to environmental stress can be classified mainly into two groups (Mitchell et al., 2009; Levine et al., 2013; Young et al., 2013). The first group consists of genes that exhibit altered expression immediately upon exposure to multiple types of environmental insult. The second group consists of genes that exhibit altered expression profiles only upon exposure to specific types of environmental stress and are generally altered gradually post exposure. Thus, orchestrated changes in the activities of these two types of genes are likely to occur in developing cortices. The following section focuses on the first group of genes that immediately respond to environmental stress and may lead to common endophenotypes (Gluckman and Hanson, 2004), discussing how these genes change the molecular landscape of cortical development and contribute to the pathogenesis elicited by prenatal environmental stress.

#### Stress Responsive Signaling

The cellular stress activates multiple signaling pathways that are well-positioned to help restore homeostasis upon sudden environmental changes, or, in the long run, enforce a new gene expression program so cells can tolerate the new environment. These signaling pathways and genes include molecular chaperone encoding genes, genes involved in the unfolded protein response, Mitogen-Activated Protein Kinase (MAPK) and Growth Arrest and DNA Damage 45 (GADD45) signaling pathways (Yang et al., 2009). The Heat Shock Protein (HSP) pathway is a major molecular chaperone signaling pathway, the activation of which has been identified as one of immediate molecular responses to various types of environmental stress, including alcohol, heat, heavy metals and viral infection (Nollen and Morimoto, 2002; Hashimoto-Torii et al., 2011, 2014).

Our recent study using knockout mice of Heat shock factor 1 (Hsf1), a canonical transcription factor that controls transcription of Hsp genes revealed that activation of this signaling is required to reduce the risk of cortical malformation, such as heterotopias and small size of the cortex, upon prenatal exposure to various types of environmental stress, thereby reducing susceptibility to epilepsy (Hashimoto-Torii et al., 2014). Histological analysis immediately after prenatal stress exposure revealed that the increase of these cortical malformations in Hsf1 knockout mice is due to the increase of cell death and suspension of cell cycling, suggesting Hsf1's roles in cellular protection against environmental stress. Interestingly, the canonical downstream targets of Hsf1, Hsps mediate proapoptotic effects of Hsf1 but not the effects on cell cycling (**Figure 1**). El Fatimy et al. (2014) showed that, many cortical genes that are critically involved in the control of cell cycling/proliferation and the neuronal migration are under the control of Hsf1 and the family gene Hsf2. Thus, the activation of HSF1 immediately alters expressions of various types of genes to protect the embryonic cortex from environmental stress.

Another example of a stress responsive transcriptional factor that protects the fetal brain from prenatal environmental stress is Nuclear Factor Erythroid 2-Related Factor 2 (Nfe2l2/Nrf2). The transcriptional activity is increased in response to such as alcohol (Narasimhan et al., 2011), kainate induced excitotoxic damage (Rojo et al., 2008a) and hydrogen peroxide induced oxidative stress (Rojo et al., 2008b). The target genes include multiple genes that encode antioxidant proteins (Dong et al., 2008; Muramatsu et al., 2013). Prenatal exposure to methamphetamine (speed) plus Nrf2 loss of function lead to reduced motor activity, smaller body weight etc. in the offspring (Ramkissoon and Wells, 2013).

Interestingly, the gender dependent differences were observed in the severity of the phenotypes.

These lines of evidence suggest that multiple cellular mechanisms provoked by the stress response genes act to ensure fetal cortical tolerance to environmental stress, and thus decrease the prevalence and severity of ensuing neuropsychiatric diseases (Hashimoto-Torii et al., 2014).

#### MicroRNAs

Post-transcriptional controls have been demonstrated to be critically involved in the control of normal cortical development (Grabowski, 2011; DeBoer et al., 2013; Yano et al., 2015). MicroRNAs (miRNAs) are non-coding RNAs that are involved in post-transcriptional regulation of the expression of a wide variety of genes (Ambros, 2004). Because of their nature as short RNAs for post-transcriptional regulation of genes, they are likely to change the molecular landscape of the cell immediately and temporally in response to environmental challenges (Leung and Sharp, 2010).

In a comprehensive miRNA profiling study using a neurosphere model of alcohol exposure, Miranda and his colleagues found a reduction in expressions of miR-21, miR-335, miR-9, and miR-153 24 h after exposure (Sathyan et al., 2007).

MiR-9 knockout mouse displays smaller brain size (Shibata et al., 2011). The analysis of those embryonic brains suggested that impaired proliferation and differentiation of neural progenitor cells in stage dependent manner may lead to the smaller brain. Consistent with this in vivo observation, miR-9 knockdown inhibited the proliferation and promoted the migration of the neural progenitor cells in vitro (Delaloy et al., 2010). The control of these biological events by miR-9 may be mediated by controlling expression levels of the downstream targets such as Forkhead box G1 (Foxg1/Bf1) (Shibata et al., 2008, 2011), embryonic lethal, abnormal vision, Drosophila like 2 (Elavl2/HuB) (Sathyan et al., 2007), Fibroblast growth factor receptor 1 (Fgfr1) (Pappalardo-Carter et al., 2013), Forkhead box P2 (Foxp2) (Pappalardo-Carter et al., 2013), Stathmin 1 (Stmn1) (Delaloy et al., 2010), Nuclear receptor subfamily 2, group E, member 1 (Nr2e1/Tlx) (Zhao et al., 2009; Shibata et al., 2011), Inhibitor of DNA binding 4 (Id4) (Shibata et al., 2008), Paired box 6 (Pax6) (Shibata et al., 2011), Meis homeobox 2 (Meis2) (Shibata et al., 2011), GS homeobox 2 (Gsh2) (Shibata et al., 2011), Islet1 (Isl1) (Shibata et al., 2011), RE1-silencing transcription factor (Rest) (Packer et al., 2008), and Actin-like 6A (Actl6a/BAF53a) (Yoo et al., 2009). Thus, reduced expression of miR-9 by alcohol exposure is also likely to inhibit those events by the similar mechanism. The miR-153 and miR-21 also similarly control the cellular proliferation (Zhong et al., 2012; Wu et al., 2013).

Reduction of miR-9 expression and the target gene expressions in the zebrafish whole-embryo (Tal et al., 2012) and the embryonic forebrain (Pappalardo-Carter et al., 2013) exposed to alcohol also supports this hypothesis. However, in the conditions of exposure to different contexts of maternal stress induced by such as restraint of the body and forced swimming, expression of miR-9 was increased in the brain of offspring (Zucchi et al., 2013). Similarly, the expression of miR-21 has also been reported to be increased in the different ambience, such as in the mouse brain exposed to ionizing radiation (Shi et al., 2012a), in the endothelial cells under the exposure to shear stress (Weber et al., 2010), and in the embryonic fibroblasts exposed to arsenite (Ling et al., 2012). The expression of miR-153 is also upregulated by hydrogen peroxidase induced oxidative stress (Narasimhan et al., 2014) and nicotine exposure (Tsai et al., 2014). These lines of evidence indicate that the microRNAs are susceptible to the environmental changes and that the overall changes of various types of microRNAs may determine the phenotypes specific to types/regimens of the environmental stress exposure. The fact that miR-335 knockdown reverses the effects of miR-21 knockdown in the cell proliferation and death also supports this possibility (Sathyan et al., 2007).

# Maternal, Placental, and Extracortical Tissues Exhibit Indirect Effects as a Result of Environmental Stress

Beside direct molecular changes within embryonic cortical cells, evidences exist that indirect impacts of environmental stress from maternal, placental, and other extracortical tissues exert a critical influence on cortical development (Velasquez et al., 2013).

Maternal infection is well defined by epidemiological studies as a risk factor for neurodevelopmental disorders such as autism and schizophrenia (Hagberg et al., 2012; Depino, 2013; Meldrum et al., 2013). Mouse offspring that have been exposed to maternal infection display abnormalities reminiscent of the behavioral, histological, and molecular characteristics of autism (Patterson, 2011), while fetal brain infection does not cause these abnormalities (Meldrum et al., 2013). Mouse offspring exposed to maternal immune activation (MIA), which is elicited by poly-riboinosinic-polyribocytidylic acid or lipopolysaccharide, also reproduce the behavioral and histological abnormalities of autism (Meyer et al., 2006; Smith et al., 2007; Hsiao et al., 2012; Carpentier et al., 2013), suggesting that activation of maternal immune system triggered by infection is critical for manifestation of deficits. These early findings have proven MIA model useful in the investigation of the molecular mechanisms at play in unraveling maternal effects on the pathophysiology of autism.

Smith et al. (2007) demonstrated that a proinflammatory cytokine interleukin-6 (IL-6) supplied from the maternal tissues might mediate the MIA effects on the fetal cortex. A single maternal injection of IL-6 in the middle of corticogenesis causes deficits in prepulse inhibition and lateral inhibition in the offspring (Smith et al., 2007), both of which are linked to autism and schizophrenia (Solomon et al., 1981; Wynn et al., 2004; Bertone et al., 2005; Perry et al., 2007). They also demonstrated that inhibition of IL-6 by application of the antibody or using the knockout dam, significantly ameliorated such as cognitive and exploratory deficits in mouse offspring exposed to MIA (Smith et al., 2007). The gene expression profiles were also reversed by inhibition of IL-6 in the cortices of the MIA offspring. These results provided evidence that IL-6 may owe the indirect effects of MIA on fetal cortical development.

Indirect effects of MIA on cortical development may also involve the effects from gastrointestinal tissues of offspring. Autism is often associated with gastrointestinal barrier defects (Buie et al., 2010; Coury et al., 2012), and rodent MIA models reproduce these defects (Hsiao et al., 2013). Hsiao and colleagues made an interesting observation that probiotic treatment of gastrointestinal barrier defects improved behavioral abnormalities such as anxiety-like behavior, decreased prepulse inhibition, and deficits in ultrasonic vocal communication in the MIA offspring. Their study also suggested the possibility that gastrointestinal barrier deficit-induced increase of serum metabolites such as 4-ethylphenylsulfate, indolepyruvate, glycolate, imidazole propionate, and N-acetylserine, may contribute to behavior abnormality in the MIA offspring (Hsiao et al., 2013). Of these, the most dramatically affected metabolite, 4-ethylphenylsulfate, has been known as a uremic toxin, and the administration of this metabolite induces anxiety-like behavior in the mouse (Hsiao et al., 2013). As a recent study suggested the link between the uremic toxin and the depression in the chronic kidney disease (Hsu et al., 2013), the 4-ethylphenylsulfate in serum may be the common factor that affects the brain function in various pathophysiological conditions.

Serotonin derived from placenta may also indirectly affect embryonic brain development. Recent studies demonstrated that the placenta is the major source of serotonin at early embryonic stage, while the dorsal raphe nuclei in the hindbrain take over from late embryonic stage to adulthood (Bonnin et al., 2011). Abnormal serotonin levels in the brain have been linked to autism (Chugani et al., 1999; Whitaker-Azmitia, 2001; Gaspar et al., 2003), and the role of serotonin in the normal development of thalamocortical projections also has been reported (Bonnin et al., 2007). In addition, it has been demonstrated that prenatal intake of selective serotonin reuptake inhibitors increases the risk of cognitive impairment in mouse progeny (Smit-Rigter et al., 2012; Kinast et al., 2013). Importantly, serotonin level is lower in the cortices of the offspring exposed to environmental stress such as maternal infection (Fatemi et al., 2008; Wang et al., 2009) and cocaine (Cabrera-Vera et al., 2000). Therefore, environmental stressors may indirectly affect the cortical development as a result of disruption in the synthesis/release of serotonin in/from the placenta (Velasquez et al., 2013).

## Interaction between a Susceptible Genotype and Environmental Risk Factors

Genome wide association studies have shown a polygenic component contributes to the risk of schizophrenia and autism (Purcell et al., 2014). Similarly, many epidemiological studies as well as the aforementioned results from studies of animal exposure models have shown these disorders also include a "polyepigenetic" component that is influenced by various types of environmental stress (Weinberger, 1987; Caspi and Moffitt, 2006; Ben-Ari, 2008; van Os et al., 2010; Bregant et al., 2013). However, just how the polyepigenetic component increases the risk of disease manifestation by interacting with polygenic component is largely unknown.

One relatively new approach to help answering this question is the use of induced Pluripotent Stem (iPS) cells taken from subjects diagnosed with polygenic diseases such as schizophrenia or autism. iPS cells are not only becoming useful tools to obtain functional human cortical neurons (Mariani et al., 2012; Shi et al., 2012b; Espuny-Camacho et al., 2013; Lancaster et al., 2013) for understanding the pathogenesis of disease, but are also being utilized for drug screening (Han et al., 2011). To examine potential interactions between genetic predisposition and the environmental risk factors, we recently used iPS cells derived from schizophrenia patients, and exposed the differentiated neural progenitor cells to environmental stress including alcohol, methylmercury and hydrogen peroxide. Single cell RNA detection revealed augmented cell-to-cell variable activation of HSF1-HSP signaling in the schizophrenia patients' neural progenitor cells, individual cell lines of which carry different genetic risks for schizophrenia (**Figure 2**). This finding suggests that variable responses of HSF1-HSP signaling among a population of neural progenitor cells exposed to environmental

stress is predetermined by genetic predisposition and may increase the risk of the onset of schizophrenia as well as other neuropsychiatric diseases (Hashimoto-Torii et al., 2014; Brennand et al., 2015).

Using Disrupted-in-schizophrenia-1 gene (Disc1) mutant mice combined with MIA, in vivo evidence for the interaction of gene and prenatal environment in the pathogenesis of schizophrenia and depression was also provided. The Disc1 is one of the risk genes for psychiatric disorders such as schizophrenia and mood disorders (St Clair et al., 1990; Millar et al., 2000). The transgenic mice expressing the dominant negative form of Disc1 that was found in the patient (Millar et al., 2000), displayed hyperactivity and impaired social interaction (Pletnikov et al., 2008). When this transgenic mouse was subjected to MIA, neurobehavioral phenotypes such as anxiety, depression-like behavior, and a decrease in social interaction and an increase in aggressiveness were unraveled (Abazyan et al., 2010). Two other Disc1 mutant mouse lines with point mutations at Q31L and L100P, which show schizophrenia and depression related phenotypes, respectively (Clapcote et al., 2007), were also subjected to MIA. MIA exposure augmented the impairment in prepulse inhibition, lateral inhibition, spatial object recognition, and social motivation of those Disc1 mutant mice (Lipina et al., 2013). Importantly, the production of IL-6 was concomitantly increased by the combination of Disc1 mutations and the MIA in the fetal mouse brains (Lipina et al., 2013). Thus, these mouse models that combine Disc1 mutation and MIA will become powerful models for understanding the molecular mechanisms underlying interactions between the gene and prenatal environmental factors that increase the risk of the psychiatric diseases.

# Outlook

As outlined in this review, research on polyepigenetic mechanisms associated with many types of environmental stress that disturb cortical development and on potential prophylactic or preventative interventions of these disturbances are just beginning to emerge. To further facilitate this type of

# References


research, patient-derived iPS cells will become one of several powerful tools. Although there are a number of limitations in their use, easy application of environmental stress and the potential for high throughput analysis substantiate their usefulness. Challenges include: (1) limited availability of iPS cell lines that are fully characterized; (2) lack of validated differentiation protocols for specific types of neurons; and (3) lack of validated in vivo approaches (e.g., efficient transplantation methods to animal models, etc.) that allow observation of the iPS cells during cortical development.

A type of the environmental stress can lead to various phenotypes in the cerebral cortex, however, this variability cannot be explained exclusively by different regimens of exposure. Recent studies have revealed potential factors that may affect the resultant phenotypes, including gender (Mooney and Varlinskaya, 2011; Ramkissoon and Wells, 2013) and probabilistic molecular responses of individual cells to the environmental stress (Hashimoto-Torii et al., 2014) (**Figure 2**) etc. Thus, the next important questions will be: (1) if such molecular differences of individual cells elicited by environmental stress are sustained for long periods of time and ultimately result in altered cortical function, and (2) which molecules mediate the gender specific effects of prenatal environmental stress.

Another recent interesting observation that needs to be addressed at the molecular level is the transgenerational effects of prenatal exposure to environmental stress, as reported in the cases of alcohol (Govorko et al., 2012). This observation opens up a whole new field of research that might eventually lead to an understanding of why FASD and other environment-linked disorders show familial and geographical linkages.

# Acknowledgments

We thank Miki Masuda, Drs. Alexander I. Son, and Masaaki Torii for their comments on the manuscript. We also thank NIH/NIAAA R00AA1838705, CTSI-CN, NARSAD/Scott-Gentle Foundation and ABMRF for their respective support.


targeting multiple transcription factors. J. Neurosci. 31, 3407–3422. doi: 10.1523/JNEUROSCI.5085-10.2011


**Conflict of Interest Statement:** The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Ishii and Hashimoto-Torii. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

# Lost highway(s): barriers to postnatal cortical neurogenesis and implications for brain repair

Aslam Abbasi Akhtar 1,2 and Joshua J. Breunig1,2,3 \*

<sup>1</sup> Board of Governors Regenerative Medicine Institute, Cedars-Sinai Medical Center, Los Angeles, CA, USA, <sup>2</sup> Department of Biomedical Sciences, Cedars-Sinai Medical Center, Los Angeles, CA, USA, <sup>3</sup> Samuel Oschin Comprehensive Cancer Institute, Cedars-Sinai Medical Center, Los Angeles, CA, USA

The genesis of the cerebral cortex is a highly complex and tightly-orchestrated process of cell division, migration, maturation, and integration. Developmental missteps often have catastrophic consequences on cortical function. Further, the cerebral cortex, in which neurogenesis takes place almost exclusively prenatally, has a very poor capacity for replacement of neurons lost to injury or disease. A multitude of factors underlie this deficit, including the depletion of radial glia, the gliogenic switch which mitigates continued neurogenesis, diminished neuronal migratory streams, and inflammatory processes associated with disease. Despite this, there are glimmers of hope that new approaches may allow for more significant cortical repair. Herein, we review corticogenesis from the context of regeneration and detail the strategies to promote neurogenesis, including interneuron transplants and glial reprogramming. Such strategies circumvent the "lost highways" which are critical for cortical development but are absent in the adult. These new approaches may provide for the possibility of meaningful clinical regeneration of elements of cortical circuitry lost to trauma and disease.

Keywords: reprogramming, transdifferentiation, regeneration, neurogenesis, brain repair, ventricular zone, gliogenesis

# Introduction

Comprising 10–26 billion neurons and a similar number of glia (Azevedo et al., 2009; Herculano-Houzel, 2009), and possessing roughly a trillion synapses per cubic centimeter (Drachman, 2005), the mammalian cerebral cortex is one of the most complex structures known to man. Through tightly orchestrated interactions between these neurons and glia, along with communication to and from other brain regions, the cortex mediates higher-order cognition, coordinates motor function, visual perception, somatosensory perception, memory, and a host of other processes (Shipp, 2007). Because of this complexity, developmental disorders often lead to lifelong impairment or disability (Francis et al., 2006). Further, trauma or disease resulting in neurodegeneration in the cortex almost inevitably leads to impairment of psychosocial, cognitive, motor, visual or somatosensory function, depending on the age of the individual and site(s) of neuronal loss (Van Hoesen et al., 1991; Johansson, 2000; Rosema et al., 2015).

To compound this, the primate cortex has almost no proclivity for de novo neurogenesis after birth (Rakic, 1985, 2002a,c; Kornack and Rakic, 2001; Spalding et al., 2005; Bhardwaj et al., 2006). The evidence for and against postnatal neurogenesis in the cortex of primates and other mammals has been extensively reviewed elsewhere and less recent findings will not be discussed here

#### Edited by:

Takeshi Kawauchi, Keio University School of Medicine/PRESTO, JST, Japan

#### Reviewed by:

Emanuel DiCicco-Bloom, University of Medicine and Dentistry of New Jersey/Robert Wood Johnson Medical School, USA Kenneth Yu-Chung Kwan, University of Michigan, USA Carlos Cardoso, Institut de Neurobiologie de la Méditerranée/Institut National de la Santé et de la Recherche Médicale, France

#### \*Correspondence:

Joshua J. Breunig, Board of Governors Regenerative Medicine Institute, Cedars-Sinai Medical Center, AHSP A8109, Los Angeles, CA 90048, USA joshua.breunig@cshs.org

> Received: 30 January 2015 Accepted: 21 May 2015 Published: 16 June 2015

#### Citation:

Akhtar AA and Breunig JJ (2015) Lost highway(s): barriers to postnatal cortical neurogenesis and implications for brain repair. Front. Cell. Neurosci. 9:216. doi: 10.3389/fncel.2015.00216 (Rakic, 2002b; Breunig et al., 2007, 2011; Feliciano and Bordey, 2013). Nevertheless, despite the contradictory data from model organisms, carbon-14 dating of human cortical neurons under normal and post stroke conditions indicates that these cells are born prenatally (Huttner et al., 2014), though adult neurogenesis was detected with similar methods in the human hippocampus where it had been previously observed (Eriksson et al., 1998; Spalding et al., 2013). Therefore, cortical neurogenesis is unlikely to significantly contribute to plasticity after cortical injury.

Taken together, directed regeneration of cortical circuitry, however difficult the prospect may be, is one of the few approaches available to potentially ameliorate functional deficits due to cortical neurodegeneration. Though, it is possible that functional recovery after neurodegeneration may turn out to be a matter of generating the appropriate cell numbers and types; developmental cortical disorders such as those seen in Reeler mutants—where neurons are generated normally but migrate inappropriately, resulting in mental retardation—suggests that a more meticulous reconstruction of circuitry, one that mimics the pattern of normal development may be necessary (Caviness and Rakic, 1978). Here we review cortical development in this context and describe emerging approaches to generate cortical neurons for cell therapy or in situ replacement.

# Cortical Development

The majority of cortical neurons and glia arise from radial glia stem cells which reside along the ventricle (**Figure 1A**) and provide highway-like migratory substrates from the ventricle to the pial surface (Rakic, 1972; Noctor et al., 2001; Breunig et al., 2011). During neurogenesis, radial glia divide asymmetrically

(A) Schematic representation of neocortical development from neurogenesis through postnatal stages. Approximate location denoted by green box in coronal brain. Proliferative radial glia (pRG) asymmetrically divide to generate intermediate progenitors (INPs) and short neural precursors (SNPs) which proliferate and contribute to the migratory cortical neuron (mCN) population. This population migrates along radial glia and into the cortical plate (CP), where they exit the RG fiber and begin the maturation process. As neurogenesis procedes, progressively more superficial layers are generated (i.e., Layer VI, then Layer V, etc.) Depicted here is the generation of Layer V subcortical projection neurons (SCPN) and the later generation of an upper layer immature callosal neuron (imCN). (Other layers are not depicted for the sake of space and clarity). In parallel, migrating interneurons (mINs) from the ventral telencephalon invade and can proliferate locally prior to maturation into functional interneurons (INs). After neurogenesis ceases, gliogenesis commences en force with the conversion of some RG into immature and subsequently mature astrocytes (iACs and ACs, respectively). Also, oligodendrocyte progenitor cells (OPCs) are born and differentiate into

myelinating oligodendrocytes (MOs) in the white matter (WM). Other RG transition into subventricular zone neural stem cells (NSCs), and ependymal cells (ECs). NSCs can give rise to INPs, which generate migrating olfactory bulb-destined neurons (mOBNs). (B) In the postnatal and adult brain, significant barriers to regeneration are present. Radial glia are exhausted and become a "lost highway" to any neuronal migration. Similarly, cortical neurons are no longer generated and thus virtually no neurons can be found migrating into the cortex. Neuronal plasticity becomes significantly attenuated, preventing the type of plasticity observed prior to developmental critical periods. Interneuron progenitors and mINs disappear. Parenchymal glia do not cross lineage boundaries and become reactive after injury and degeneration. Moreover, the axon lengths become many fold longer in the adult due to the growth of the organism. For example, a SCPN may reach almost a meter in length while the initial axon started at a few millimeters before progressive lengthening. Finally, in humans there is minimal subventricular zone (SVZ) neurogenesis in the adult, inhibiting strategies which might utilize such cells. Abbreviations: PS, pial surface; IZ, intermediate zone; MGE, medial ganglionic eminence; LGE, lateral ganglionic eminence; Ctx, cortex.

to generate developing neurons which migrate along these scaffolds and exit off into their respective cortical layers (**Figure 1A**; Noctor et al., 2001; Rakic, 2009). The cortex is distinctly organized into six horizontal layers which are generated embryonically in an inside-out fashion (Rakic, 2009). As development progresses, generation of lower layers tapers off into genesis of progressively more superficial layers due to an underlying transcriptional program (Kwan et al., 2012). As neurogenesis completes, the ''gliogenic switch'' shifts the fate of radial glial stem cells towards the genesis of astrocytes and oligodendrocyte progenitor cells (**Figure 1A**; Rowitch and Kriegstein, 2010). Hence, the majority of cortical neurogenesis occurs prior to birth, while gliogenesis is a perinatal and postnatal phenomenon (Bhardwaj et al., 2006; Yeung et al., 2014).

The majority of neurons in the cortex are glutamatergic while the remaining (roughly 20%) of neurons are GABAergic interneurons (Chu and Anderson, 2015). These interneurons are generated in the subcortical areas of the ventral telencephalon in the rostral forebrain, specifically in the medial ganglionic eminence (MGE) and caudal ganglionic eminence (Anderson et al., 2002; Nery et al., 2002; Wonders and Anderson, 2006; Southwell et al., 2014). Immature interneurons migrate to the developing and neonatal cortex, attain their mature fate (**Figure 1A**), and integrate their circuitry wherein they function to modulate neuronal activity through inhibitory neurotransmission (Nery et al., 2002; Wonders and Anderson, 2006; Hansen et al., 2013). In humans, evidence exists for the local generation of interneurons in the cortical ventricular and subventricular zone (VZ and SVZ), respectively during embryogenesis, but this remains contentious (Letinic et al., 2002; Hansen et al., 2013; Radonji´c et al., 2014). Moreover, there is evidence that interneuron progenitors can proliferate extensively during their migration, including in the cortex and, in addition, the tail end of this migrating progenitor population can be observed postnatally (Costa et al., 2007; Inta et al., 2008; Breunig et al., 2012; Levy et al., 2014).

Due to the intricate spatial and temporal cues that orchestrate cortical development, the task of regenerating the cortex in adulthood possesses many challenges (**Figure 1B**). Physically speaking, projection neurons in the cortex reach their respective targets early in development (e.g., the spinal cord for Layer V neurons and the thalamus for Layer VI neurons). After axonal targeting is achieved in embryogenesis, these processes progressively lengthen with age from a distance of millimeters and centimeters to almost a meter for an adult corticospinal projection neuron. Furthermore, the cortical radial glia population is exhausted in early postnatal life leaving no scaffold for migration to respective cortical layers (Tramontin et al., 2003). The absence of a path to cortical layers from the ventricular zone represents a significant ''lost highway'' that may impede regeneration (**Figure 1B**). Further, neurons are postmitotic cells that do not divide once they have reached their final position in the cortex. And though precursors do exist around the lateral ventricles in the human, akin to development, they largely lose their neurogenic ability in vivo (**Figure 1B**; Sanai et al., 2011; discussed in detail below).

# Postnatal Forebrain Cell Genesis

The neurogenic niches of the adult brain have received a great deal of attention over the past fifteen years due to their ability to contribute to brain plasticity and serve as a model and perhaps substrate for regeneration. The two neurogenic niches in the adult mammalian brain are the subventricular zone (SVZ) of the lateral ventricles and the subgranular zone (SGZ) of the hippocampus (Song et al., 2002; Alvarez-Buylla and Lim, 2004; Ming and Song, 2011). The SVZ neural stem cells are derived from embryonic and perinatal radial glia (Merkle et al., 2004). As these radial glia transition into neural stem cells of the astrocyte lineage, they lose the long basal process that they possess (and therefore the migratory substrate for neurons to migrate into the parenchyma—the eponymous ''lost highway''). These Gfap<sup>+</sup> neural precursors also serve to populate the cortex and striatum with astrocytes (Ge et al., 2012) and oligodendrocytes (Menn et al., 2006), as well as seed the olfactory bulb (OB) with neurons (Alvarez-Buylla and Lim, 2004; Ming and Song, 2011). Specifically, SVZ precursors generate transit amplifying cells which proliferate and generate neuroblasts (Doetsch et al., 1999). These neuroblasts migrate in a clustered, tangential fashion to the OB through the rostral migratory stream (RMS), a structure of ensheathing glial fibers and migrating neurons between the lateral ventricle and OB core (Lois et al., 1996). Upon reaching the OB core, the immature neurons migrate radially towards the OB periphery, establish synapses, and mature into OB interneurons (Carleton et al., 2003; Ming and Song, 2011). Importantly, OB neurogenesis does not appear to be lifelong in humans and instead may be supplanted by striatal interneuron genesis, which does not appear to occur in other non-human primates and rodents (Ernst et al., 2014). Unlike neural precursors, the glial subtypes generated by SVZ radial glia extensively divide perinatally throughout the forebrain (Ge et al., 2012; Yeung et al., 2014). Astrocyte turnover attenuates perinatally in rodents, but polydendrocytes of the oligodendrocyte lineage are believed to be the most proliferative cell type in the adult brain (Burns et al., 2009; Geha et al., 2010).

In the postnatal hippocampal SGZ, radial glia-like neural stem cells give rise to intermediate progenitors (INPs) which in turn generate neuroblasts (Seri et al., 2004). These neuroblasts migrate to the inner granule cell layer and differentiate into granule cells which project to the CA3 region of the hippocampus (Ming and Song, 2011). Several studies in mice have associated changes in hippocampal neurogenesis to learning and memory (Deng et al., 2009, 2010; Mu and Gage, 2011) though a new genetically modified rat allowing for ablation of adult neurogenesis failed to find a similar correlation (Groves et al., 2013). Importantly, evidence of human hippocampal neurogenesis is observed late into life (Eriksson et al., 1998; Spalding et al., 2013). Taken together, the adult brain does have active sites of neurogenesis, but postnatal cortical neurogenesis has seemingly been selected against by evolution. Nevertheless, these neurogenic zones have provided insight into the mechanisms of postnatal neurogenesis and have also served as experimental testing platforms for various genetic experiments which aim at using stem cells for postnatal cortical neurogenesis. In particular, the postnatal cortical SVZ is of increased interest due to its proximity to the cortex. Furthermore, in addition to techniques such as viral-mediated transgenesis, which can manipulate progenitors in the SVZ and SGZ (Braun et al., 2013; Zuccotti et al., 2014); novel techniques such as electroporation allow rapid transgenic manipulation of SVZ progenitors pre- and postnatally, without the need of virus generation (Breunig et al., 2007).

# Learning from Development

In light of the radial glia-to-cortical neuron development paradigm which exists embryonically, neuroscientists have strived to achieve adult cortical neurogenesis from several angles, namely: (1) the redevelopment of radial glia to serve as substrates for differentiation to cortical neurons or as scaffolds for nascent neurons generated through other means; (2) the transplantation of neural stem and progenitor cells to the cortex; and (3) the analysis of developmental genetic factors which direct stem and progenitor cells to become cortical neurons, and as such, the effect of their misexpression in astrocytes and neurons. Below we detail examples of the work neuroscientists have achieved which show glimpses of postnatal neurogenesis in a region of the brain that was initially thought to be cemented during early development. Due to the numerous reviews present in this field, we will focus on recent findings in respect to cortical regeneration and direct the reader towards other reviews for an in-depth analysis of postnatal neurogenesis (Breunig et al., 2007; Urbán and Guillemot, 2014).

# Promoting Radial Glia Reemergence and Alternate Migratory Substrates

Radial glia are necessary for the development of the cortex as they divide asymmetrically to generate immature neurons and are scaffolds for migrating neurons as they travel to their respective layers (Rakic, 2003; Breunig et al., 2011). Efforts have been made to generate radial glia in adulthood to serve as: (1) a source of progenitors for replacing cortical neurons; and (2) scaffolds for migration of neurons into the cortex. In this regard, expression of the tyrosine kinase receptor, ErbB2, in cortical quiescent mature astrocytes for 3 weeks enabled a subset of these astrocytes to assume a radial glia-like phenotype (Ghashghaei et al., 2007). In addition to possessing elongated processes, these ''induced radial glia'' had increased expression of the Notch 1 ligand which has been shown to be essential for the maintenance of radial glia identity. Other transcription factors required for radial glial maintenance, Sox2, Pax6, and Hes5 (the downstream targets of Notch1) were also upregulated in the induced cells. Interestingly, only astrocytes surrounding the ventricles, but not those in the cortical parenchyma, were able to re-assume their radial glia identity, suggesting that not all astrocytes in the cortex are the same. Induced radial glia were able to give rise to new neurons in vitro and in vivo, and also supported the migration of transplanted embryonicday 16 cortical neurons. Similar work was done by Gregg and Weiss (2003) where epidermal growth factor (EGF) expression in cultures of forebrain neural stem cells resulted in an adoption of radial glia morphology, expression of Nestin and RC2, and supported the migration of immature neurons. Infusion of EGF into the adult forebrain lateral ventricle also led to the generation of radial glia-like cells within the adult forebrain ependyma (Gregg and Weiss, 2003). In addition, the endocytic adaptor proteins Numb and Numbl have been identified as regulators of radial glia adhesion and polarity (Rasin et al., 2007). Though originally identified in Drosophila neural progenitors for its role in promoting neuronal cell fate by inhibiting notch signaling, when Numb and Numbl were inactivated in mouse radial glia, adheren junctions and polarity were disrupted through disrupted cadherin trafficking, resulting in progenitor cell dispersion and disordered cortical lamination (Rasin et al., 2007). In contrast, when Numb or cadherin was overexpressed, radial glia were maintained postnatally beyond the standard neurogenic period in the cortex (Rasin et al., 2007).

Investigation into the migrational modalities used by OB neurons may also provide insight regarding different migratory substrates that may potentially be utilized for induced cortical neurogenesis. As previously mentioned, OB neurons are one of the few neuronal populations continuously generated in the adult rodent. These nascent neurons migrate radially from the OB core to the OB periphery. However, the OB does not contain radial glia-like projections or an RMS-like glial sheath to serve as scaffolds for these migrating neurons. Upon investigating the mode of travel of these neurons, it was found that these nascent interneurons use the OB vasculature as a scaffold for migration (Bovetti et al., 2007; Tanaka et al., 2011). This new modality, termed vasophilic migration is facilitated through an interaction between the extracellular matrix and perivascular astrocyte end feet (Bovetti et al., 2007). Therefore, it is possible that vasculature in the cortex may serve as a modality for migration of nascent neuronal populations to their respective cortical layers. Indeed, there is evidence that upon focal microlesions, SVZ cells can generate neurons in the cortex (Magavi et al., 2000; Brill et al., 2009) but more recent work failed to find new neurons after employing two different genetic lesion paradigms (Diaz et al., 2013).

# Employing Developmental Lessons to Facilitate Transplantation

The creation of induced pluripotent stem cells (iPSCs), which led to Shinya Yamanaka receiving the 2012 Nobel Prize, energized the field of regenerative medicine in several ways. First, the finding that pluripotency can be induced in postmitotic cell types opens the door to myriad possibilities in terms of reprogramming and directed differentiation. Secondly, iPSCs now provide a virtually unlimited source of patient-matched cells for transplantation. Over the last two decades, numerous groups have generated various cortical neuronal subtypes from embryonic and induced pluripotent stem cells in vitro (Gaspard et al., 2008; Hansen et al., 2011; Shi et al., 2012; Espuny-Camacho et al., 2013; Michelsen et al., 2015). In parallel with these findings, investigators have also demonstrated that a three-dimensional culture of pluripotent stem cells can ''selforganize'' into a complex tissue with striking similarities to the cerebral cortex, suggesting that neural stem cells may possess the inherent cues needed for cortical formation (Lancaster et al., 2013). This raises the prospect of transplanted stem cells to replace damaged or diseased neurons in the adult cortex. This possibility would exist if the adult cortex maintained the intrinsic spatial cues of development in order to direct the differentiation of temporally relevant stem cells. In this regard, (Ideguchi et al., 2010) derived mouse embryonic stem (ES) cells from the E4.5 blastocyst and differentiated them for 7 days before transplanting into various regions of the cortex. The cells exhibited region specific projections, with cells transplanted to the motor or visual cortex projecting to their respective targets. In a similar experiment, (Gaspard et al., 2008) transplanted cells were differentiated for 12–17 days (instead of 7) and did not present the region specific projections as all cells projected to visual cortex targets, despite being transplanted to the frontal cortex. This suggests that a variation of 5–10 days of in vitro differentiation plays a role in the success of graft integration and specificity, and highlights the need for temporal specificity of the transplanted cells. Furthermore, recent work further illustrates the need for matching the areal identity of transplanted neurons (Michelsen et al., 2015). In this study, visual cortex neurons were differentiated from mouse ES cells in vitro for 14 days and transplanted into the adult visual cortex after a focal neurotoxic lesion. The transplanted cells integrated into their respective pathways and electrophysiological recordings revealed the cells were responsive to visual stimuli. Similar to previous studies, significant integration and engraftment were not observed when ES cell-derived visual cortex neurons were transplanted into the motor cortex, or when motor cortex neurons were transplanted into the visual cortex, indicating the need for areal specificity. Importantly, recent work suggests that in vitro differentiation of ES-derived neural progenitors towards cortical fates fails to properly recapitulate in vivo development in many aspects, including aberrant progenitor specification and stalled differentiation (Sadegh and Macklis, 2014). It should be noted that this report (Sadegh and Macklis, 2014) used mouse ES cells cultured in a monolayer and allowed to mature in a manner previously used for cortical differentiation. Taken together, these findings suggest that neuronal replacement using pluripotent stem cells combined with specific differentiation protocols holds promise. However, they also suggest that it will be important to spatiotemporally match donor cells for transplantation and this will require a requisite understanding of the mechanisms of areal identity in order to specify the appropriate populations from pluripotent cells. In this regard, much work remains to be done to understand how to appropriately derive specific cortical cells and tissues from pluripotent cell types despite an increasing understanding of how this process occurs during development (Rakic et al., 2009; Kwan et al., 2012; Greig et al., 2013). Clinically, the long-term safety issues of iPSC-based cell therapy will need more thorough evaluation in light of recent findings of oncogenic transformation of iPSC-derived NSCs when transplanted into the spinal cord (Nori et al., 2015). However, the iPSCs from this study were generated using lentivirus (Nori et al., 2015) and it remains to be seen if similar problems will emerge from non-integrating iPSC generation methods where continued transgene mis-expression is unlikely.

In the interim, the use of iPSCs for generating neurons for disease modeling and drug screening will likely be of great importance.

#### Interneuron Transplantation

As previously mentioned, unlike other neuronal subtypes in the cortex which are generated locally from radial glia, interneurons are generated from the ganglionic eminences of the ventral telencephalon and migrate to the cortex (among other regions of the brain) where they mature and integrate into their respective networks (Noctor et al., 2004; Southwell et al., 2014). While projection neurons of the cortex target various intra- and extra-cortical regions, cortical interneurons project onto other neurons in the cortex and inhibit neurotransmission within this region. This balance of inhibitory/excitatory signals is critical for the proper function of the cortex. Conditions such as epilepsy and schizophrenia are thought to be related at least in part to dysfunctional interneurons, giving them the title of ''interneuronopathies'' (Kato and Dobyns, 2005; Southwell et al., 2014).

As cortical interneurons exhibit a developmental paradigm which involves long distance migration from extra-cortical regions followed by maturation and integration in the cortex, the use of interneuron progenitors in transplant therapy for interneuron replacement may be more feasible compared to transplant efforts aimed at replacing cortical projection neurons. For a detailed review of interneuron transplants, we direct the reader to Alvarez Dolado and Broccoli (2011) and Southwell et al. (2014). Indeed, interneurons do not require radial glia-like scaffolds to migrate from the ganglionic eminences to the cortex, and the neuronal subtypes they project on are not as distant as the respective targets of projection neurons. To this, several groups have achieved promising results transplanting interneurons generated from the embryonic MGE, ES cells, and iPSCs (Tanaka et al., 2011; Maroof et al., 2013; Nicholas et al., 2013; Southwell et al., 2014). Transplanted interneurons migrate from the graft site throughout the cortex and successfully integrate into the young and adult rodent cortex (Maroof et al., 2013; Nicholas et al., 2013). Unlike the transplantation of other cell types which has resulted in decreased transplant survival or a decrease in the native population, which may be detrimental, transplanted interneurons increase the overall pool of cortical interneurons (Southwell et al., 2010). Notably, the cortex has a limit of how many interneurons it can support, which is reported to be roughly 10% more than the native population (Southwell et al., 2010, 2014). Transplanted interneurons display spontaneous and induced synaptic currents and further, transplantation has been shown to increase inhibitory signals in glutamatergic neurons (Southwell et al., 2010; Bráz et al., 2012). Additionally, several groups have reported functional benefits from cortical interneuron transplants in mouse models of epilepsy and schizophrenia (Baraban et al., 2009; Tanaka et al., 2011). Specifically, (Baraban et al., 2009) transplanted precursor cells from the embryonic E13.5 MGE into the postnatal day 2 mouse brain and reported their maturation into GABAergic interneurons when analyzed 30 days after transplantation. The transplanted cells integrated and dispersed throughout the cortex as indicated by immunohistochemistry, electron microscopy, electrophysiology, and increased GABA-mediated synaptic inhibition on pyramidal neurons. Bilateral grafts of the embryonic MGE cells into experimental epileptic mice reduced the duration and frequency of spontaneous electrographic seizures. (Tanaka et al., 2011) transplanted MGE cells of the same embryonic age (E13.5) into the medial prefrontal cortex and observed similar functional integration as well prevention of phencyclidine-induced cognitive defects.

While diseases that independently affect cortical projection neurons may not directly benefit from the increased inhibition provided by transplanted interneurons, these migratory neurons may be used as a vehicle for delivery of trophic factors. Traditionally, astrocytes have been used in this regard as they have been viewed as a resilient supportive cell type for neurons. Among the characteristics that have made astrocytes more appealing are their ability to divide after transplantation, migrate to sites of injury, ensheathe neurons, and their ease of generation in vitro (Svendsen et al., 1997; Behrstock et al., 2006, 2008; Suzuki et al., 2007). However, recent reports of successful interneuron generation from iPSCs, engraftment after transplantation, and transgenic manipulation may suggest interneurons as tools for factor delivery to areas of the cortex (Southwell et al., 2014).

# Genetic Engineering and Reprogramming

Over the last decade, ''direct reprogramming'' or ''transdifferentiation'' has allowed for terminally differentiated cells to directly assume fate of another differentiated cell type without having to go through a state of pluripotency. Direct reprogramming of neurons from a host of different terminally differentiated cells such as fibroblasts, pericytes, hepatocytes, and other neural cells has been described (Heinrich et al., 2010, 2014; Vierbuchen et al., 2010; Marro et al., 2011; Karow et al., 2012). Much of the discovery and success of these direct reprogramming experiments have come to fruition due to our increasing knowledge of the molecular control of neurogenesis and the advances in transcriptomics over the last decade. For example, neurogenesis follows a general pattern (**Figure 2A**) whereby a neuronal progenitor is generated by Notch-mediated asymmetric cell division of a neural stem cell (Ables et al., 2011). Among the resulting pair of cells after mitosis, the NotchHigh cell will remain a stem cell and the NotchLow cell will subsequently upregulate proneural genes such as Neurog1 and Neurog2 (or Ascl1 in other regions) and become an intermediate progenitor (**Figure 2A**; Bertrand et al., 2002; Ables et al., 2011). This progenitor will continue to terminally differentiatiate under the regulation of basic helix-loop-helix transcription factors such as NeuroD1 (**Figure 2A**) until later maturation (Guillemot,

FIGURE 2 | Molecular control of neurogenesis and its use in reprogramming. (A) Radial glia utilize Notch signaling to self-renew during mitosis. (\*-during asymmetric division the cell inheriting the basal process exhibits high Notch activity while the other daughter displays diminished Notch activity). Neuronal daughter cells upregulate proneural genes such as Ngn1 and Ngn2 in the cortex. († -Ascl1 is the predominant proneural gene in the ventral telencephalon among other regions). Basic helix-loop-helix (bhlh)

transcription factors such as NeuroD1 regulate the terminal differentiation of migrating neurons into mature projection neurons. Fezf2 cooperates with all of these factors in the specification of subcortical projection neurons (SCPNs), leading to the eventual expression of a diverse array of transcription factors involved in the postmitotic identity of this neuronal subtype. (B) Transgenic misexpression strategies for reprogramming of disparate cell types to neurons.

2007). Finally, any number of transcription factors involved in neuronal subtype specification will be expressed to control neuronal identity (Kwan et al., 2012). Transcriptomics has led to the discovery of several ''master regulators'' which specify particular neuronal subtypes (Kwan et al., 2012; Greig et al., 2013) such as Fezf2 (discussed in detail below). As mature cortical projection neurons cannot be transplanted, the ability to directly differentiate or reprogram other cell types into precise neuronal subtypes with the end goal of replacing neurons lost due to damage or neurodegenerative disease would be invaluable. Below we discuss the burgeoning field devoted to directed differentiation of neurons.

## Glia to Neuron Reprogramming

The reasoning behind using cortical glia as substrates for reprogramming to neurons stems from several characteristics of glial cells. Glial cells are plentiful in the cortex (Azevedo et al., 2009) and are actively generated throughout life (Colodner et al., 2005; Geha et al., 2010; Yeung et al., 2014). Therefore, a lack of substrate cells is not an issue. Furthermore, in response to injury and disease, glial cells become reactive and increasingly proliferative, allowing for an increased number of cells to reprogram at the injury site.

One of the first studies that investigated the ability of single transcription factors to induce directed differentiation of cortical astroglia to specific neuronal subtypes was done by isolating cortical glia from postnatal day 5–7 mice followed by retrovirusinduced expression of Neurog2 or Dlx2 (Heinrich et al., 2010). As mentioned above, Neurog2 is a proneural gene that regulates the early differentiation of glutamatergic neurons (Hevner et al., 2006; Heng et al., 2008; Zhang et al., 2013). In contrast, the Dlx2 homeobox protein is normally expressed in progenitor cells derived from the ventral telencephalon and has been shown to promote the generation of GABAergic interneurons (Petryniak et al., 2007; de Chevigny et al., 2012). Expression of Neurog2 in astroglial cultures led to the generation of glutamatergic-like neurons which expressed Tbr1 and Tbr2, the T-box transcription factors expressed by glutamatergic neurons in the forebrain. The Neurog2-induced neurons also generated synapses, expressed vGluT1 around their soma and MAP2<sup>+</sup> processes, and acquired projection neuron-like morphology. In contrast, the Dlx2 induced neurons expressed markers of GABAergic interneuron lineage GAD67 and vGAT, and generated functional synapses. Interestingly, the Dlx2-mediated reprogramming was not as efficient as Neurog2-induced reprogramming.

The majority of glia to neuron conversion studies in the brain have been reported in regions other than the cortex. Several groups have reported successful conversion of glia to neurons in the striatum by the expression (or knockout) of various factors (Niu et al., 2013; Torper et al., 2013; Magnusson et al., 2014). The expression of Brn2a, Ascl1, and Mytl1 (also known as BAM factors) was previously shown to convert fibroblasts and hepatocytes to neurons in vitro (Vierbuchen et al., 2010; Marro et al., 2011). Expression of the BAM-factors mediated by a Cre-inducible lentiviral injection into the striatum of GFAP-Cre heterozygous mice converted striatal glia to neurons. These induced neurons expressed NeuN and had neuronal morphology; however the functionality of the induced neurons was not assessed (Torper et al., 2013).

Notch signaling is involved in various stages of cortical neurogenesis and generally functions to inhibit neuronal differentiation (Ables et al., 2011). A recent study has shown that an experimental model of middle cerebral artery occlusioninduced stroke results in transient neurogenesis in striatal astrocytes (Magnusson et al., 2014). Notch1 signaling was reduced in striatal astrocytes after stroke while ectopic activation in astrocytes inhibited stroke-induced neuroblast production. Furthermore, blocking notch signaling in the absence of stroke promoted astrocytes in the striatum and medial cortex to enter a neurogenic program and express markers of immature and mature neurons, such as Ascl1, Dcx, and NeuN (Magnusson et al., 2014). Taken together, this suggests that Notch signaling actively suppresses the neurogenic potential of parenchymal astrocytes in the striatum (**Figure 2B**). Lastly, several studies have shown that single factors can convert glia to neurons after injury (Guo et al., 2014; Heinrich et al., 2014). These are discussed below.

# Cortical Neurogenesis after Inflammation and Injury

Over the last decade, plentiful information has been discovered about the relationship between inflammation and neurodegenerative disease. Previous dogma understood neuroinflammation to always have a negative role in disease progression. However, recent studies have shown that certain aspects of the immune system are beneficial to the host in neurodegenerative conditions such as amyotrophic lateral sclerosis (ALS), Alzheimer's disease and multiple sclerosis (McCombe and Henderson, 2011; Weitz and Town, 2012; Breunig et al., 2013; Guillot-Sestier and Town, 2013; Hussain et al., 2014). In respect to cortical neurogenesis, neuroinflammation is a double-edged sword. Numerous groups have reported that inflammation promotes cortical neurogenesis and conversion of various cell types to neurons in the cortex (Guo et al., 2014; Heinrich et al., 2014; Magnusson et al., 2014). As mentioned above, several groups reported that laser-induced injury and apoptosis in deep layers of the cortex can stimulate precursors in situ to increase division and differentiation into glutamatergic neurons (Magavi et al., 2000; Brill et al., 2009). However, a follow up study which used a neuronal promoter-driven caspase as well as a neuronal promoter-driven diphtheria toxin method of induced apoptosis reported increased proliferation of microglia, but did not promote generation of glutamatergic neurons (Diaz et al., 2013). In respect to transgene-induced conversion, the retrovirusmediated misexpression of Sox2 in the adult mouse cortex following stab wound injury induced the conversion of NG2 cells to interneuron-like cells (**Figure 2B**), whereas misexpression under normal (non-stab wound) conditions did not result in conversion (Heinrich et al., 2014). The resulting induced neurons exhibited voltage-and time-dependent conductance and received synaptic connections from endogenous GABAergic neurons. This study used NG2 glia cells as a substrate, a cell type that is abundant and proliferating in the adult brain (Dimou and Gotz, 2014). Other studies have shown that NG2 cells are recruited to the site of injury and may be equally or more permissive to conversion than astrocytes, suggesting that future reprogramming efforts should not look over this cell type (Buffo et al., 2005, 2008; Hughes et al., 2013). Interestingly, (Buffo et al., 2005) reported that stab wound injury increased expression of Olig2 by immunohistochemistry and mRNA expression, and 26% of these Olig2<sup>+</sup> cells were NG2 cells. Furthermore, increased Olig2 expression after stab wound was accompanied by decreased Pax6 expression. This Pax6-Olig2 relationship is also seen in the developing spinal cord and subependymal zone (Mizuguchi et al., 2001; Hack et al., 2005). Inhibition of Olig2 by retrovirus encoding a dominant-negative form of Olig2 two days after stab wound resulted in increased Pax6 expression and increased neurogenesis as evidenced by the generation of Dcx<sup>+</sup> neuroblasts. The retroviral-mediated overexpression of Pax6 two days after stab wound injury also resulted in similar increases in Pax6 expression and neurogenesis.

It was also reported that retroviral expression of NeuroD1 in the cortex of adult stab wound-injured mice resulted in conversion of: (1) astrocytes to glutamatergic neurons; and (2) NG2 cells to GABAergic and glutamatergic neurons (**Figure 2B**; Guo et al., 2014). Electrophysiology performed on slice cultures revealed that the NeuroD1-converted neurons exhibited spontaneous and evoked synaptic responses. Similar results were seen in the cortex of Alzheimer's disease mice, but not in control (un-injured or non-diseased) mice (Guo et al., 2014). Similar work was done in vitro where adult cortical astrocytes where isolated after stab wound injury and converted to glutamatergic neurons by Neurog2 or GABAergic interneurons by Dlx2 as previously discussed (Heinrich et al., 2010). Taken together, these reports suggest the importance of the neuroinflammatory response for conversion to neuronal subtypes in the adult mouse cortex under certain conditions.

Of increased interest in the Heinrich et al. (2014) and Guo et al. (2014) reports, is the idea that transcription factors can exhibit opposing roles depending on the context. Developmentally, Sox2 promotes self-renewal in neural stem cells and prevents their differentiation into neurons (Graham et al., 2003). However, when Sox2 is expressed in NG2 cells after injury, it promotes their reprogramming to interneuronlike cells as described above Heinrich et al. (2014). A similar study in the adult mouse striatum reported the conversion of astrocytes to other neuronal subtypes upon induction of Sox2 (Niu et al., 2013). The mechanism of these differential responses remains to be determined. It may reflect different expression levels of the transgene, the epigenetic state of the transduced cells, or extrinsic cues associated with the neuroinflammatory state of the substrate cells and their microenvironment. Going forward, an integrated analysis of genome-wide binding sites, interacting proteins (e.g., other transcription factors and cofactors), chromatin configuration, and other epigenetic marks in these varied spatiotemporal contexts will likely illuminate the disparate responses to transcription factor misexpression (Amador-Arjona et al., 2015).

Interestingly, the majority of in vivo glia-to-neuron conversion studies have been carried out in an inflamed neocortex. Importantly, the activation of the immune system in response to injury or disease usually results in perpetuated and chronic neuroinflammation, which eventually leads to neurotoxicity. The glial scar is an example of this as the resulting activated astrocytes after injury are initially beneficial but eventually form a chronic scar which serves as a barrier to neuronal regeneration (Sofroniew, 2009; Cregg et al., 2014). In respect to the examples discussed above regarding expression of Sox2 or NeuroD1 in NG2 cells (Guo et al., 2014; Heinrich et al., 2014), future studies will need to determine what exactly is different in the NG2 cells or their environment after injury that allows for their reprogramming. Is it only that they are cycling more rapidly? Is the inflammatory microenvironment transiently more permissive to reprogramming in the cortex? And if so, can these features be mimicked in the un-inflamed cortex for a more translatable approach to replacing cortical neurons? Indeed there is some indication that activation of innate immunity is necessary for reprogramming. Specifically, it was noted that TLR3 pathway activation induced notable epigenetic changes leading to chromatin alterations which enhanced the pluripotent stem cell conversion of fibroblasts by four reprogramming factors (Li et al., 2009, c-Myc; Lee et al., 2012). When TLR3 (or its adaptor protein TRIF) was knocked down, reprogramming efficiencies decreased. Interestingly, the authors noted that conventional retroviral methods of transgene expression activate the TLR3 pathway, and nonviral methods of transgene expression complemented with ectopic TLR3 activation increased reprogramming efficiencies. Specifically, a change in the methylation status of the Oct4 and Sox2 promoters was observed. These authors claim that there is an optimal window of immune activation necessary for reprogramming (Cooke et al., 2014). They term this the ''Goldilocks zone.'' A greater understanding of such a phenomenon in the context of neuronal reprogramming might yield more powerful control over the process. In this regard, it is also possible that the neuroinflammatory state present in the successful neuronal reprogramming experiments discussed above involve an activated TLR3 pathway resulting in an open, ''pro-reprogramming'' chromatin configuration.

## Other Cell Types

Reprograming strategies in cell types other than neurons and glia may also show promise for cortical neurogenesis. (Brill et al., 2009) isolated human pericytes from the adult cerebral cortex and converted them to neurons by the misexpression of Ascl1 and Sox2. Pericytes are involved in regulation of blood flow in the brain and the establishment and maintenance of the blood brain barrier (Armulik et al., 2011). Interestingly, they have also been reported to be multipotent mesenchymal stem cell-like and can give rise to cartilage, muscle, and bone lineages (Armulik et al., 2011). Since these cells are dispersed throughout the cortex and are more abundant than neural stem cells, they may be an attractive substrate for in vivo reprogramming to cortical neurons.

#### Neuron to Neuron Reprogramming

While the above studies provide important insight into the phenomenon of neuronal reprogramming, they all employ transcription factors broadly expressed in diverse types of neurons (**Table 1**). And while it is clear that neurons are generated, detailed characterization awaits. Given these facts, the precise nature of the resulting reprogrammed populations are hard to predict and may be extremely heterogeneous. For example, this approach might yield the types of populations seen after differentiation of pluripotent cell types, namely many neurons exhibiting ''stalled'' phenotypes and aberrant expression of phenotypic markers (Sadegh and Macklis, 2014). Going forward, it may be necessary to employ additional specific factors to direct the differentiation of precise subtypes. One such example of a protein which may be utilized is the transcription factor Fezf2. During embryogenesis, Fezf2 is necessary for the specification of corticospinal motor neurons and Fezf2−/<sup>−</sup> mice lack corticospinal motor neurons (Chen et al., 2005a,b; Molyneaux et al., 2005). Embryonically, forced expression of Fezf2 is sufficient to reprogram progenitors destined to generate upper layer neurons or striatal neurons to corticospinal motor neurons (Chen et al., 2005b; Molyneaux et al., 2005; Rouaux and Arlotta, 2010). Elegant work by Rouaux and Arlotta (2013) has shown that early postnatal overexpression of Fezf2 in postmitotic Layer 2/3 callosal projection neurons which typically project interhemispherically via the corpus callosum, results in reprogramming to Layer 5-like corticofugal neurons (**Figure 2B**). The newly generated neurons acquired the molecular properties of Layer 5 neurons expressing markers ER81 and CRYM, and downregulating CUX1. Axonal projections in the newly reprogrammed cells were also redirected from interhemispherical targets to subcortical targets. (De la Rossa et al., 2013) has shown that early postnatal overexpression of Fezf2 in Layer 4 spiny neurons changes their identity to Layer 5-like corticofugal neurons. Taken together, these two reports indicate that postmitotic neurons, which are already in their respective cortical layer and have defined their projections, still maintain a sense of plasticity to reprogramming catalyzed by single transcription factor overexpression in early postnatal life. However, both groups reported decreased reprogramming efficiencies at postnatal day 21, suggesting that the plasticity of these postmitotic neurons declines with age and will likely be absent in the adult brain. Interestingly, when Fezf2 is expressed in SVZ progenitors destined to become OB GABAergic interneurons, it directs fate towards a glutamatergic phenotype (Zuccotti et al., 2014). The resulting Fezf2-respecified OB neurons have features akin to pyramidal cells including larger cell bodies, elaborative dendritic trees, and pyramidal neuronlike electrophysiological outputs. However, the reprogramming reported does not appear to be as complete as is seen in the studies performed embryonically or perinatally. Interestingly, a recent study has also reported high Fezf2 expression in a subset of callosal neurons in the adult cortex, suggesting that Fezf2 expression may not be restricted solely to corticospinal motor neuron fate (Tantirigama et al., 2014). Taken together, the findings of the reprogramming studies discussed above suggest a reduction in reprogramming ability by Fezf2 with age. Given the complex combinatorial transcriptional co-regulation of neuronal specification during development (Kwan et al., 2012; Greig et al., 2013), this is perhaps not surprising.

Furthermore, as suggested by various examples discussed above, a single transcription factor may have several respective roles and its expression may result in different outcomes depending on the time and context of its expression. For example, Sox2 promotes self-renewal in neural stem cells and is one of the Yamanaka factors for induced pluripotency (Graham et al., 2003). However, when Sox2 is expressed in NG2 cells after stab-wound injury or in adult astrocytes in the striatum, these cells can be reprogrammed to interneuron-like cells (Heinrich et al., 2014) or neuroblasts respectively (Niu et al., 2013). These results suggest the need to carefully sculpt strategies for directed differentiation by incorporating developmental logic, knowledge of the donor cell properties, knowledge of the host tissue, and forethought in avoiding situations which might be deleterious clinically (e.g., tumor growth due to transgene misexpression; Nori et al., 2015). Employing such developmental metrics combined with precise tools for genetic manipulation will likely be increasingly necessary in bolstering reprogramming efficiencies to generate precise neuronal subtypes (Victor et al., 2014; Akhtar et al., 2015). Lastly, the use of accurate


disease-specific models systems and transplant hosts—including aged genetically modified rodents, non-human primates, and potentially human organoids—will likely be of the utmost importance to assess the translational capacity of potential novel therapies.

# Conclusions

Grafting and transplantation of neuronal tissue has proceeded for decades with little or no progress towards clinical therapies for cortical trauma and degeneration. However, a critical mass of findings in multiple subfields promises to improve this situation. The development of iPSC technology has reformatted our conception of cell differentiation. Moreover, this technology provides virtually unlimited ''starting material'' for the generation of cells for transplant. Our knowledge of cortical development has greatly informed our ability to rationally direct differentiation of neurons from pluripotent stem cells. Specifically, employing developmental paradigms to direct neuron subtype differentiation holds the promise of enabling mix and match generation of neurons lost to injury and disease. Nevertheless, as we have detailed, meaningful

# References


clinical replacement of cortical circuitry will be among the most difficult problems for modern science to solve. Generating the appropriate cell types will only partially solve this challenge. The next step is to bridge the gap between the precise specificity of circuits created during development with our ability to transplant or reprogram cell types in a manner that allows for the recapitulation of these circuits. With a trillion synapses per centimeter of lost tissue, this will be no small feat. However, given the lack of alternatives, we have no choice but to ''beat on, boats against the current,'' in attempting to rise to the challenge of creating efficacious clinical treatments for those who would otherwise suffer lifelong disability.

# Acknowledgments

We apologize to colleagues whose work we could not include due to space constraints. We thank P. Rakic and N. Sestan for many insightful discussions on these subjects and assistance with figure elements. We thank M. Dutra-Clarke for critical review. JJB was supported by funding from the Samuel Oschin Comprehensive Cancer Institute Cancer Research Forum Award and the Board of Governors Regenerative Medicine Institute of Cedars-Sinai.

factor to parkinsonian rodents and aged primates. Gene Ther. 13, 379–388. doi: 10.3410/f.1030023.349469


zone of the adult brain. J. Neurosci. 26, 7907–7918. doi: 10.1523/jneurosci.1299- 06.2006


phencyclidine-induced cognitive deficits. J. Neurosci. 31, 14116–14125. doi: 10. 1523/jneurosci.2786-11.2011


**Conflict of Interest Statement**: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2015 Akhtar and Breunig. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution and reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.