Edited by: Frédéric Lavigne, Université de Nice - Sophia Antipolis, France
Reviewed by: Maurizio Mattia, Istituto Superiore di Sanità, Italy; Hecke Schrobsdorff, Max-Planck-Institute for Dynamics and Self-Organization, Germany
*Correspondence: Elisa M. Tartaglia, Center for Neuroscience and Cognitive Systems @UniTn, Istituto Italiano di Tecnologia, Palazzo Fedrigotti, Corso Bettini 31, 38068 Rovereto (TN), Italy e-mail:
This article was submitted to Cognitive Science, a section of the journal Frontiers in Psychology.
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.
Human efficiency in processing incoming stimuli (in terms of speed and/or accuracy) is typically enhanced by previous exposure to the same, or closely related stimuli—a phenomenon referred to as priming. In spite of the large body of knowledge accumulated in behavioral studies about the conditions conducive to priming, and its relationship with other forms of memory, the underlying neuronal correlates of priming are still under debate. The idea has repeatedly been advanced that a major neuronal mechanism supporting behaviorally-expressed priming is repetition suppression, a widespread reduction of spiking activity upon stimulus repetition which has been routinely exposed by single-unit recordings in non-human primates performing delayed-response, as well as passive fixation tasks. This proposal is mainly motivated by the observation that, in human fMRI studies, priming is associated to a significant reduction of the BOLD signal (widely interpreted as a proxy of the level of spiking activity) upon stimulus repetition. Here, we critically re-examine a large part of the electrophysiological literature on repetition suppression in non-human primates and find that repetition suppression is systematically accompanied by stimulus-selective delay period activity, together with repetition enhancement, an increase of spiking activity upon stimulus repetition in small neuronal populations. We argue that repetition enhancement constitutes a more viable candidate for a putative neuronal substrate of priming, and propose a minimal framework that links together, mechanistically and functionally, repetition suppression, stimulus-selective delay activity and repetition enhancement.
In priming, stimuli are perceived/identified with greater accuracy and/or reduced reaction times if observers have been previously exposed to intact (i.e., repetition priming), noisy or related versions of them (i.e., semantic priming). Reaction times reductions can span from few milliseconds up to several dozen milliseconds, depending on different factors as, for example, the time between subsequent presentations (Vorberg et al.,
Identifying the neural mechanisms which mediate priming has been difficult, on one hand, because of the difficulties to study it via neurophysiological recordings. Forms of priming which involve semantic judgments are difficult to assess in non-human primates; visual priming evaluation in monkeys is usually impeded by task overtraining, which causes behavioral performance to be already at ceiling at the time of the recordings, leaving hardly any room for further improvement [with the remarkable exception of the study of McMahon and Olson (
Importantly, neurophysiological studies have shown that repetition suppression is, in many cases, accompanied by an–often smaller but significant- proportion of neurons whose response is enhanced upon stimulus repetition. Concurrently, a fraction of the recorded cells displays a selective activation in the delay period separating subsequent stimulus repetitions. In the context of neurophysiological recordings, repetition enhancement and persistent activity, involving a minority of the sampled neurons, have largely been neglected as plausible neural correlates of priming, while they are widely considered as implicated in short-term memory tasks (e.g., Desimone,
Here, we first review neurophysiological studies which have consistently exposed
A substantial amount of data on the neural correlates of repeated stimulus presentation has been collected via neurophysiological recordings
In a DMS task with simple geometric shapes, when comparing neuronal responses in inferotemporal cortex to the first (sample) and the repeated (match) stimulus presentation, Baylis and Rolls (
Interestingly, both the percentage of neurons showing repetition enhancement and the percentage of those exhibiting sample selective persistent delay activity increased in prefrontal cortex, reaching, respectively, 42% and 33% of the responsive cells; response enhancement upon stimulus repetition was not only more common in prefrontal than in inferotemporal cortex, it was also greater in magnitude [Miller et al.,
Variable proportions of (differentially responsive) cells showing repetition enhancement (from 5% to 60%), as well as persistent activity, have been reported in posterior parietal cortex in a spatial version of a DMS task, in which monkeys were required to judge whether the locations of two sequentially presented stimuli were the same or different (Steinmetz et al.,
Lui and Pasternack (
In a DMS task with distractors, Hayden and Gallant (
Notably, evidence of both an early enhancement and a later suppression of the average activity of a population of cells in inferotemporal cortex have been found upon stimulus repetition (Woloszyn and Sheinberg,
Electrophysiological recordings on monkeys performing a DMS task allow to assess neural activity in response to repeated stimuli (i.e., when the sample stimulus serves as a
In semantic priming, reaction times in response to a test stimulus are reduced if the test is preceded by a conceptually or semantically related prime. In a paired associate task, the prime and the subsequent test consist in a pair of visual stimuli that the monkey has previously learned to associate. In each trial the monkey has to recognize whether the prime is followed by its pair-associate or by a different test stimulus. Although facilitatory effects on the behavioral performance are hard to observe on overtrained animals, priming-like effects on reaction times related to neurons' spike rates have been observed when the prime is followed by its pair-associate test (Erickson and Desimone,
Neurophysiological recordings in the prefrontal cortex of monkeys performing a paired associate task–with pictures of real world objects- have exposed, in analogy with the DMS task, both enhancement and suppression of neural activity in response to the test stimulus following its pair-associate prime–when compared to the response to the same test stimulus following a neutral prime. Interestingly, in contrast with the DMS task, the proportion of neurons showing enhanced activity to the paired associate test was found to be significantly greater than the proportion showing suppression. Concurrently, out of 181 responsive cells, 87 showed selective persistent activity during the delay separating the prime and the test (Rainer et al.,
In a delayed match to category task, in which monkeys have to recognize whether the test matches the category of the previously presented sample–i.e., the related prime-, Freedman et al. (
Repetition enhancement as well as persistent delayed activity, along with repetition suppression, have also been found during passive fixation, a task that does not involve a recognition component and, therefore, does not require active maintenance of previously seen stimuli.
Qi et al. (
Vogels et al. (
In remarkable resemblance with priming (Ellis et al.,
Table
ITC | oDMS | 81% | 19% | 6% | Baylis and Rolls, |
HF + PHG | oDMS | 29% | 71% | 20% | Riches et al., |
TE + RH | oDMS | 66% | 34% | 20% | Riches et al., |
ITC | oDMS | 77% | 23% | 25% | Miller et al., |
ITC | oDMS | 65% | 35% | many | Miller and Desimone, |
ITC | oDMS | 72% | 28% | 19% | Miller et al., |
ITC | oDMS | 89% | 11% | 7% | Vogels et al., |
PFC | oDMS | 40% | 60% | 33% | Miller et al., |
PPC | sDMS | 40% | 60% | few | Rawley and Constantinidis, |
PPC | sDMS | 92% | 8% | ND | Steinmetz et al., |
PPC | sDMS | 95% | 5% | ND | Constantinidis and Steinmetz, |
MT | DDD | 66% | 34% | 54% early delay | Lui and Pasternack, |
V4 | oDMS | 23% | 77% | 49% | Hayden and Gallant, |
PFC | DPA | 42% | 58% | 48% | Rainer et al., |
PFC | DMC | 51% | 49% | 18% | Freedman et al., |
ITC | DMC | 37% | 63% | 9% | Freedman et al., |
PFC | PF | 68% | 32% | NR | Qi et al., |
PFC | sDMS | 50% | 50% | NR | Qi et al., |
PFC | sDMS | 73% | 27% | ND | Lueschow et al., |
All above-mentioned studies involve repetition of highly familiar stimuli in the context of highly trained tasks, involving or not a voluntary memory component. It has been suggested that stimulus familiarity could be related to the amount of repetition enhancement–i.e., only the repetition of familiar stimuli can elicit increased neural activation (Rugg et al.,
Several studies have pointed out that the duration of both repetition suppression and enhancement effects might depend on the stimulus familiarity (Miller et al.,
What are the mechanisms of repetition priming? What provokes the faster response to the repetition of a stimulus, compared to its first presentation? The basic tenet of all models that have been proposed is that network(s) underlying priming have a different response to the second stimulus than to the first, and that this difference in responses lead to the observed differences in behavior. A major constraint on models is the observation, in the majority of cells, of
The next question that needs to be answered is then what are the mechanisms leading to such changes? Conceptually, one could imagine at least three types of mechanisms for why the response to a repeated stimulus might be different to the response to the first: (i) a single neuron mechanism: the first stimulus could trigger an intrinsic ionic current, that would change the state of the neuron, therefore modifying its response to subsequent stimuli; (ii) a synaptic plasticity mechanism: the first stimulus could trigger changes in the efficacies of synapses, which would then lead to changes in the response to subsequent stimuli; (iii) a network mechanism: the first stimulus could switch the network to an activity state that differs from the state in which the network was before the stimulus, which would then modify the response of the network to subsequent stimuli. In the following, we will focus on mechanisms (ii) and (iii). The first scenario has, to our knowledge, not been explored by modeling studies.
Synaptic plasticity refers to a broad range of phenomena that describe dynamics of synaptic efficacy over time scales of hundreds of seconds (short-term plasticity) to much longer time scales (long-term plasticity).
Gotts (
The model is able to reproduce the so-called
Models implementing long-term synaptic plasticity mechanisms assume that long-term synaptic changes (i.e., long-term potentiation and depression) induced by a single presentation of a stimulus are strong enough to induce observable changes in the response of the network to a subsequent presentation of the stimulus. Intuitively, Hebbian-like synaptic changes typically tend to favor the stability of the network state that was driven by the stimulus; this then leads to a faster response to a second presentation of the stimulus (Becker et al.,
In a recurrent network model originally designed to explain the improvement of performance observed in a delayed-match-to-multiple-sample task with novel vs. familiar stimuli, a single stimulus repetition, via a
Moldakarimov et al. (
In a spiking neurons model version of their previous work, Moldakarimov et al. (
A different class of models relies on attractor dynamics. Hebbian, long-term synaptic modifications induced by external stimuli can lead to the creation of selective attractor states, in which a subpopulation of neurons maintain an elevated, persistent activity following the presentation of a particular stimulus (Amit,
This idea was investigated by Brunel and Lavigne (
The model is able to reproduce the dynamics of priming effects and, in particular, the dependency of reaction time reductions on the duration of the delay period separating prime and test presentation -or stimulus onset asynchrony (SOA)- observed in the context of semantic priming. Moreover, the model is consistent with neurophysiological recordings of monkeys performing a paired-associate task, in which a prime is followed by its paired-associated test–previously learned- or by a neutral test stimulus. The model relies on persistent delay activity, which leads to enhancement of the neural response upon associated test presentation, but leaves unaddressed the issue of neural activity suppression widely observed in the context of priming effects.
Finally, we outline a model of repetition priming that reconciles the three neural correlates consistently observed upon stimulus repetition—repetition suppression, repetition enhancement and persistent delay activity (Tartaglia et al.,
In our model, implemented via a standard attractor neural network of excitatory and inhibitory spiking neurons, both repetition suppression and repetition enhancement arise from the dynamical interplay of the broad selectivity of visual responses–consistently with several observations that cortical neurons typically show visual responses to a large fraction of stimuli, even after such stimuli have become fairly familiar (Woloszyn and Sheinberg,
Following the first stimulus presentation (sample/prime), the population of neurons which responds the most to it—a small fraction of the whole network- keeps firing persistently, serving as the memory trace of the stimulus. Such persistent activity, in turns causes, by increasing the overall network inhibition, a suppression of activity in the remaining populations, which constitutes the majority of neurons. Upon stimulus repetition (match/target), the changes in the patterns of neural activity brought about by the persistent activity, induce an enhanced response in the most selective neurons and a suppressed response in the less selective ones. Hence, as in the original sharpening model, the responses of the weakest neurons decrease when a stimulus is repeated. However, the concurrent response enhancement of a small fraction of neurons now carries the critical information about the stimulus and can yield, via a suitable read-out mechanism, the more accurate/faster behavioral performance observed in priming (Figure
Our model essentially relies on persistent delay activity whereby neurons belonging to the memory representation–i.e., those that fire persistently during the stimulus retention period- respond more promptly upon stimulus repetition–as in Brunel and Lavigne (
In our model, repetition enhancement targets preferentially the cells which exhibit the strongest response to the sample, consistently with some of the data reviewed above (Woloszyn and Sheinberg,
Such an explanation would be also consistent with the fact that no repetition enhancement has been observed in those studies which reported proportional scaling (Li et al.,
One of the essential constituents of our model is persistent delayed activity, widely considered to be a major neuronal correlate of temporary memory storage. Several studies reviewed above have shown that persistent activity, as well as repetition suppression and repetition enhancement, are concurrently observed in several areas, not only during short-term memory tasks, but also during repeated passive exposure, which does not require any active maintenance of previously seen stimuli. These observations support the intriguing assumption that, upon stimulus presentation, some form of memory trace is automatically activated, regardless of task demands. Such memory trace, encoded in the sustained patterns of neural activity, can then support different mnemonic processes such as recognition in DMS tasks, associative recall in paired-associate tasks, or simply faster stimulus processing in priming, by modifying the tuning properties of the underlying neural representations. In these terms, our mechanistic interpretation is consistent with the so called “abstractionist” theories according to which priming stems from the reactivation of (voluntary or involuntary) preexisting memory representations (Henson,
Suppression of neural activity upon repetition of a prime has been thought to be a neural correlate of priming. Even though suppression often involves the majority of the recorded neurons, a fraction of the sampled population responds more vigorously to the prime repetition–i.e., repetition enhancement effect. Interestingly, in almost all the instances in which repetition suppression and repetition enhancement are observed, also persistent delay activity is observed. These patterns of neural activity seem to co-occur in several brain areas and in the context of tasks that implicate or not an explicit mnemonic component.
Here, we propose a new perspective according to which repetition enhancement and persistent activity, although involving a small fraction of neurons–which is nonetheless quantitatively consistent with the fraction of selective neurons typically observed in neurophysiological studies- have a pivotal role in the neural machinery underlying priming effects. In this framework, the modulatory effects of neural activity upon stimulus repetition are conditional to persistent delay activity, which, disregarding whether it represents active or passive temporary storage of the prime—is instrumental to priming effects.
In our account, the presentation of a prime entails changes in the
Our scenario gives rise to intriguing experimental predictions. Any manipulation (e.g., pharmacological) weakening or destroying persistent activity should have a significant impact on priming, severely diminishing its amplitude or completely hindering it. Support to this idea comes from the observation that both priming and working memory are impaired by pharmacological interventions which enhance the GABA-ergic neurotransmitter system, e.g., via benzodiazepines. It has been shown that benzodiazepines hinder repetition suppression (Thiel et al.,
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.
We acknowledge the financial support of the SI-CODE project of the Future and Emerging Technologies (FET) programme within the Seventh Framework Programme for Research of the European Commission, under FET-Open grant number: FP7-284553.
1FMRI stimulus repetition paradigms have consistently reported reduced latency of the BOLD response peak as well as a decrease in its magnitude (on much longer time scales than those of reaction time reductions typical of priming.). It has been suggested that changes in the BOLD response upon stimulus repetition could be interpreted as a net shorter duration of neural activity, which would be the signature of the faster stimulus processing underlying priming effects (Henson,
Here, we focus on repetition suppression as measured in single cell recordings. We invite the reader to refer to Grill-Spector et al. (
2Evidence of repetition enhancement have been found also via ERP and fMRI (Schacter et al.,
3Note, however, that increasing the network heterogeneity level e.g., by increasing the sparseness of the synaptic connections, might yield response enhancement also on cells which neither preferentially respond to the sample nor exhibit persistent delay activity, but, by chance, receive a stronger recurrent input upon stimulus repetition.