Edited by: Susan J. Sara, Collège du France, France
Reviewed by: Fabrício A. Pamplona, D'Or Institute for Research and Education, Brazil; Soyun Kim, University of California San Diego, USA
*Correspondence: Heather A. Bimonte-Nelson, Behavioral Neuroscience Division, Department of Psychology, Arizona Alzheimer's Consortium, Arizona State University, PO Box 871104, Tempe, AZ 85287, USA e-mail:
This article was submitted to the journal Frontiers in Behavioral Neuroscience.
†Equal contributions, shared lead authorship.
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.
We constructed an 11-arm, walk-through, human radial-arm maze (HRAM) as a translational instrument to compare existing methodology in the areas of rodent and human learning and memory research. The HRAM, utilized here, serves as an intermediary test between the classic rat radial-arm maze (RAM) and standard human neuropsychological and cognitive tests. We show that the HRAM is a useful instrument to examine working memory ability, explore the relationships between rodent and human memory and cognition models, and evaluate factors that contribute to human navigational ability. One-hundred-and-fifty-seven participants were tested on the HRAM, and scores were compared to performance on a standard cognitive battery focused on episodic memory, working memory capacity, and visuospatial ability. We found that errors on the HRAM increased as working memory demand became elevated, similar to the pattern typically seen in rodents, and that for this task, performance appears similar to Miller's classic description of a processing-inclusive human working memory capacity of 7 ± 2 items. Regression analysis revealed that measures of working memory capacity and visuospatial ability accounted for a large proportion of variance in HRAM scores, while measures of episodic memory and general intelligence did not serve as significant predictors of HRAM performance. We present the HRAM as a novel instrument for measuring navigational behavior in humans, as is traditionally done in basic science studies evaluating rodent learning and memory, thus providing a useful tool to help connect and translate between human and rodent models of cognitive functioning.
Spatial learning and memory, the ability to encode, store, and retrieve information about route navigation and object locations (Barnes et al.,
In the RAM task, rewards are typically not replaced once they have been located within each testing session, resulting in increasing task difficulty (i.e., the number of spatial locations the animal must avoid for successful performance) across trials, within each testing session. In the animal research literature, working memory is considered to be a form of short-term memory and is classically defined as information that is worked with, kept “online,” and updated. In the RAM, working memory demand is elevated with each trial; once a reward is located at the end of an arm, the animal must then remember to avoid that arm on future trials for optimal task performance. This complexity makes the RAM a valuable instrument for evaluating the ability to handle a systematic increase in working memory load. It is well documented in both rats and mice that RAM errors increase within each day as trials progress and working memory demand escalates; however, errors decrease across multiple testing sessions as animals learn the task (Olton and Samuelson,
Evidence supports the assertion that, in humans, spatial learning and memory involves multiple complex cognitive processes similar to those measured in rodents. For example, in order to form a cognitive spatial map, humans also acquire knowledge about environmental cues (Taylor and Tversky,
Rodent assessments of spatial memory are often also assessments of episodic memory, working memory capacity, as well as visuospatial ability. Rodent models have been critical to our understanding of spatial learning and memory, the brain regions and mechanisms that confer navigational skills, and potential therapies and pharmacological treatments to improve quality of life in populations suffering from cognitive impairments. Rodent RAM research, specifically, has produced a wealth of translational knowledge by allowing for pharmacological, genetic, and environmental manipulations that are not ethically or logistically possible in human populations. Data collected with the rodent RAM have led researchers toward numerous discoveries and new directions with the potential to enrich and optimize cognitive function in humans; use of this paradigm is essential to decipher the infinitely complex neural mechanisms associated with learning and memory, as well as the influence of aging, disease, environmental changes, and countless other factors. It is generally thought that rodent performance on the RAM depends on visuospatial ability, working memory capacity, and an intact episodic memory, but not general intelligence. These same cognitive domains are readily evaluated in humans; however, it remains unclear whether working definitions of these cognitive domains in rodent and human research are functionally equivalent. The extent to which rodent RAM research is directly translational to human learning and memory persists as a key scientific question.
One approach to this immensely complex and dynamic issue is to create an intermediate testing instrument by adapting experimental paradigms from animals to humans. Our aim in the present study was just that—to use a direct and literal translational approach to design a human task that measures the ability to remember and utilize information about spatial locations in a real world, walk-though environment, modeled after rodent RAMs. We assembled a complementary team of scientists with expertise in rodent maze learning, human perception and memory, navigational behavior, and diagnostic clinical neuropsychology. We constructed an 11-arm, walk-through human RAM (HRAM), aiming to make the task as similar as possible to the rodent RAM, and compared performance on the HRAM to performance on a battery of tests tapping cognitive domains that are hypothesized to underlie spatial learning and memory; namely, spatial reasoning ability, episodic memory, working memory, and general intelligence. The HRAM allowed us to translate and compare navigational error patterns, exactly as measured in rodent RAM studies, to performance on a battery of standard neuropsychological and cognitive tests in human participants.
Our primary goal was to determine whether the HRAM produces a similar pattern of errors to that seen in rodents both within and across testing sessions. We expected to see HRAM errors change as a function of WM load and testing session. Specifically, we predicted that HRAM performance would decline as working memory demand became elevated within each testing session, but that performance would improve across testing sessions, similar to the pattern of performance seen in rodents. An additional goal of this study was to explore the relationship between HRAM performance and performance on commonly used neuropsychological and cognitive tests. In order to better understand the relationship between some of the most commonly used rodent and human methodology, we aimed to determine how much variance in HRAM performance could be predicted by scores on standard tests of visuospatial ability, working memory, episodic memory, and general intellectual ability. Because RAM performance relies on working memory and knowledge of spatial locations, we hypothesized that participants' scores on tests (defined in Methods Section) of working memory capacity (OSpan, RSpan, RotSpan, SymSpan), and visuospatial ability (MRT, JLAP) would predict performance on the HRAM. We also tested whether performance on a measure of episodic memory (RAVLT) would predict HRAM performance. We also wanted to investigate whether a measure of general intelligence would predict performance on the HRAM. The final goal of this project was to determine whether tasks that measure different domains of cognition in humans would account for unique portions of variance in HRAM scores, that is, whether each class of tests (i.e., working memory capacity tasks, visuospatial ability tasks, episodic memory tasks) contributed distinctly to overall prediction of HRAM performance. We aimed to assess the extent to which the addition of neuropsychological tests to a standard battery of cognitive tests would improve prediction of human ability to navigate and learn in a real-world environment. We predicted that performance on each group of tasks would account for unique variance in our HRAM task. The overarching goal of this study was to help expand knowledge of both human and rodent cognition, to allow broader interpretations of existing data in both species, and to facilitate translational connections between animal laboratory, human laboratory, and human clinical research domains.
A total of 157 participants (54 men and 103 women) were recruited from several psychology courses at Arizona State University
We developed and constructed a novel HRAM to fit human proportions. A schematic and pictures of the HRAM are shown in Figure
Instructions were given to introduce the participants to the goals of the task and to prevent participants from simply sequentially proceeding down successive arms or every other arm. This was done to encourage participants to use utilize spatial strategies or to utilize more complex strategies than simple chaining to traverse the maze. Each participant was read the following instructions prior to maze testing:
“Money is under the mat at the end of each arm of this maze. Your goal is to find all of the money in the shortest amount of time. Once you find the money in an arm, it will not be replaced. Therefore, you should avoid going into any arm twice. Do not enter arms that are immediately next to each other or go in a pattern entering every other arm. Only travel into an arm immediately next to the one you previously entered if you absolutely must in order to obtain the remaining money, which means only do it when you are almost certain that you are on your last reward. If you do travel into an arm immediately next to the previous one you will be asked to stop and return to the center. Once you find money, please return it to the researcher located in the center of the maze. Then, wait until they tell you to go, and proceed to the rest of the arms to collect the remaining money. During the course of testing please do not ask the researcher how many rewards remain or any other questions regarding your performance, as they are not permitted to respond.”
Each participant started at the center of the maze; after receiving the instructions, the participant was told to begin collecting the rewards from each arm. For each trial, the researcher recorded the exact arm(s) the participant went down and recorded the time it took the participant to discover each reward. Upon locating a reward, the participant was instructed to return to the center of the maze, hand the reward to the researcher and then continue on to the next trial. This process was repeated until all 11 rewards were located, resulting in 11 total test trials (testing session A).
Following successful collection of all 11 rewards, participants were brought outside the maze and administered the WRAT-3, which served as a general measure of verbal intelligence. During this time (approximately 5 min), a second experimenter replaced all 11 rewards in the HRAM. After the WRAT-3, participants were tested on the HRAM a second time (testing session B), adhering to the same set of directions. Participants were scored based upon the number of total incorrect arm entries they made (HRAM Errors). Because all of the arms contain rewards at the beginning of each testing session, errors solely consist of repeat arm entries within each testing session and all errors are considered to be working memory errors. After completion of the HRAM and WRAT-3, participants were taken to a separate room and administered a general survey and the remaining cognitive tests.
Between HRAM testing sessions, the WRAT-3 Reading subtest was administered, serving as a general measure of verbal intelligence. The WRAT-3 relies on the participants' ability to read aloud a list of increasingly less common irregularly spelled words, and is useful as an estimate of verbal intelligence (Lezak et al.,
The Rey Auditory Verbal Learning Task (RAVLT) was used to assess episodic memory ability. In this task, participants must listen to and verbally recall words from a 15-item word list (List A) in 5 consecutive recall trials (Trials A1-A5; Total Words Learned). List A is then followed by recall of a distractor list (List B) in a single trial (Trial B1), and an immediate recall of List A (Trial A6; Retroactive Interference), which is often used as a measure of retroactive interference and short-term memory. After 20 min, delayed memory/long term memory recall is assessed in a single recall trial (Trial A7; Delayed Recall). Participants were scored on the number of words recalled correctly on each trial. Scores for Trials A1–A5 (Total Words Learned) were the total number of correctly recalled words across all five trials. Finally, participants complete a recognition trial discriminating words from List A from foils. We did not standardize scores by age or sex, but rather acknowledged age and sex as potential demographic variables that may influence scores on multiple tasks. Given that the rodent RAM has been reliably shown to be sensitive to both age and sex, this best facilitated our goal to examine the relationship between variance in our cognitive test scores and variance in our HRAM scores.
Two paper-and pencil measures of visuospatial ability were used in this study. The first was a version of the Vandenberg and Kuse Mental Rotation Task (MRT) (Vandenberg and Kuse,
Working memory was assessed by a set of four computerized complex-span working memory tasks. These tests require participants to maintain mental memoranda (either verbal or spatial) in the face of completing a distracting task. These tests included verbal (Operation Span; OSpan and Reading Span; Rspan) and spatial (Symmetry Span; SymSpan and Rotation Span; RotSpan) working memory tasks (see Unsworth et al.,
The HRAM (testing session A) was the first task participants completed as a measure of spatial working memory. The WRAT-3 was administered between the two HRAM testing sessions as a measure of verbal intelligence. The WRAT-3 was followed by a second HRAM testing session (testing session B), to determine whether participants improved performance across testing sessions. After completion of the second session of HRAM testing, participants completed a survey regarding health and demographic factors. Participants were then administered the RAVLT Trials A1-A6, MRT, JLAP, RAVLT Trial A7, and computer tasks. The testing battery was given in the same order for all participants. Upon completion of all tasks, participants were debriefed. The total time from beginning to completion was approximately 2 h per participant.
HRAM data were analyzed using repeated measures ANOVA, with HRAM Errors on Trials 1–11 and Sessions A and B as the repeated measures. Relations between performance on the MRT, JLAP-15, RAVLT, WRAT-3, RSpan, OSpan, SymSpan, and RotSpan with HRAM performance were examined with correlations and multiple regression analysis. In order to determine the extent to which each task predicts performance on the HRAM, a real-world, immersive task requiring spatial navigation, learning and memory, individual regressions were run with each task serving as the predictor and total errors made on both sessions of the HRAM combined (HRAM Total Errors) as the dependent (predicted) variable.
Additionally, hierarchical regression analysis was utilized to determine whether tasks measuring different domains of learning and memory offered unique predictive value to a regression equation predicting HRAM scores. Tasks that emerged as significant predictors of HRAM performance were entered into a regression equation in sequence, starting with the tasks that accounted for the largest proportion of variance in HRAM scores. Tasks were entered in clusters according to which cognitive domain they measure. The dependent variable for all equations was total errors made on both sessions of the HRAM combined (HRAM Total Errors). Four measures of working memory capacity, OSpan, RSpan, RotSpan, and SymSpan accounted for the largest proportion of variance in HRAM Total errors, and were entered as a first block of predictors to yield Equation (1):
Two measures of visuospatial ability, MRT and JLAP were added to yield Equation (2):
Gain in prediction Equations (1) and (2) assessed prediction from visuospatial ability measures over and above working memory capacity
Means, standard deviations, and ranges for scores on our measured variables are presented in Table
There was a significant main effect of Trial [
WRAT-3 scores did not correlate with HRAM Total Errors (Table
As shown in Table
As shown in Table
As shown in Table
The baseline regression equation Equation (1), including working memory capacity predictor variables, accounted for a significant proportion of variance in HRAM Total Errors [Table
The addition of two visuospatial tasks, MRT and JLAP, as predictor variables [MRT: β = −0.28; 95% CI (−0.55, −0.02);
The current study employed a human-sized, walk-through version of the RAM that was modeled after the rodent version used commonly in learning and memory research. The RAM has been used for decades to study spatial memory in the rodent. Notable landmark work includes that of Tolman in the 1940s utilizing the structurally-similar sunburst maze (Tolman,
One major goal of this study was to explore the translational relationship between human performance on the HRAM and commonly used neuropsychological tests that tap spatial ability, episodic memory, working memory, and intelligence. Evaluating these relationships allowed us to determine which tests commonly used in clinical settings and cognitive psychology account for variance in performance on the HRAM, a commonly used rodent task adapted to humans. Of the battery of tests we administered in this study, the JLAP emerged as the strongest predictor of HRAM performance, accounting for 10% of HRAM Total Errors. The verbal working memory capacity tasks, the Ospan and Rspan, surfaced as the next strongest predictors, predicting 8 and 9% of the total variance in HRAM error scores, respectively. The MRT predicted 5% of variance on HRAM Total Errors, and the predictive value of the spatial working memory tasks (the RotSpan and SymSpan) was similar to the predictive value of the MRT, each predicting 4% of the total variance in HRAM error scores (Table
When evaluating the nature of these tasks, plausible explanations for the observed relationships emerge. The predictive ability of the MRT and JLAP-15 may be attributable to the proposed use of a mental visuospatial sketchpad (Baddeley,
Performance on the HRAM did not correlate with the estimate of general verbal intelligence used in this study (reading subtest of the WRAT-3) or with new learning and long term delay measures of episodic memory, but did correlate with specific measures of visuospatial ability and working memory capacity, suggesting that the HRAM requires utilization of specific cognitive abilities of working memory and visuospatial skills rather than reliance on episodic memory or general verbal intelligence, as measured by the WRAT-3. Thus, our findings indicate that tasks measuring working memory (e.g., maintaining performance within the context of increased load or distracting stimuli) and visuospatial skills are correlated with performance on the RAM, a task used widely in rodent literature that we have fully adapted to human proportions.
Hierarchical regression analysis indicated that the proportion of HRAM error variance accounted for by each group of predictor variables (working memory capacity and visuospatial ability) was unique to that group of variables. Scores on the MRT and JLAP accounted for 9% of variance in HRAM scores, in addition to the 9% of variance accounted for by the four working memory capacity variables (Table
In conclusion, our collaborative research group created a three-dimensional, fully immersive, walk-through version of the RAM designed specifically for human use, in order to create an intermediary translational instrument. The results indicate that human performance on the HRAM is notably similar to rodent performance on the RAM, in that there is an exponential increase in errors as trials progress and task difficulty increases, but with the human error pattern revealing a larger processing capacity compared to rodents. The total number of errors per trial in humans remains low until the trial number exceeds a total similar to the classically defined human working memory capacity of 7 ± 2 items (Miller,
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.
This research was funded by grants awarded to HAB-N from: the National Institute on Aging (AG028084), a Diversity Supplement to National Institute on Aging grant AG028084, the state of Arizona, ADHS, the APA Diversity Program in Neuroscience, the NIH Initiative for Maximizing Student Development (IMSD) program (R25GM099650), the More Graduate Education at Mountain States Alliance (NSF), the Western Alliance to Expand Student Opportunities Louis Stokes Alliance for Minority Participation Bridge to the Doctorate (WAESO-LSAMP-BD) National Science Foundation Cooperative Agreement HRD-1025879, and Barrett Honors College at Arizona State University. LCB is supported in part by the State of Arizona and National Institutes on Aging (P30 AG19610) and the State of Arizona. The authors would also like to thank the entire Bimonte-Nelson lab, McBeath PEARL lab, Brewer lab, and Mark A. Wiener for their assistance in researching and organizing the project. We extend a special heartfelt appreciation to Bimonte-Nelson Laboratory undergraduate researcher Keley Rose Schaefer, who persevered and contributed wholeheartedly to this project while undergoing difficult cancer treatments. We are proud of the battle she fought, we miss her, and we dedicate this paper to her.
1Three participants were omitted from analyses due to missing data on one or more predictors. Four additional participants were omitted from analyses due to unusual scores on the HRAM tasks; HRAM Total Error scores for these participants were outside of two standard deviations of the mean. Thirteen participants were excluded from analyses of WRAT-3 Reading subtest performance because English was their second language, and two others were excluded due to abnormally low scores on the WRAT-3. All missing data were handled using listwise deletion.
2Change statistics are based on the unadjusted R 2 values