Abstract
In a retrospective analysis of data from three studies using a delayed response task in cynomolgus monkeys, we examined the subjects' search patterns and success rates. Twenty-seven monkeys of both sexes, divided into three age groups, were tasked with retrieving two food items hidden in an array of six identical opaque cups. Although the task was challenging for all subjects, generating a high level of guesswork, evidence of common behaviors when approaching the spatial memory test were found. The search patterns employed by the monkeys suggest the use of landmark cues, adaption in response to failure and chronological memory recall. These strategies appeared to be shared by most subjects, however, the overall success rate appeared to also depend on individual characteristics including age, gender and whether the subject had been born in caged captivity or not. By elucidating some of the underlying cognitive mechanisms, these findings may serve to refine interpretation of future studies using similar delayed response tasks in non-human primates.
Delayed response tasks are utilized for assessing working memory in non-human primates (1). These tests assess the subject's ability to retain information that varies in time and/or in content, which is the type of memory that is often affected in dementias (2). Consequently, delayed response tasks have been frequently used in studying memory decline and dementias in aging primates (1, 3). In addition, the complex behavior exhibited by non-human primates in these tests can be used to infer underlying cognitive processes, making them valuable tools for studies in cognitive neuroscience and comparative psychology (4-7).
The present report collates the results from three delayed response task studies in a retrospective analysis, focusing on aspects of the data that were beyond the scope of the original studies. To assess the working memory in cynomolgus monkeys (Macaca fascicularis) of differing ages, we utilized a type of delayed response task where the subjects were assessed to see whether they remembered the location of food items hidden in two of six identical opaque cups arranged in a straight line. Part of the data used herein have previously been published in a pair of studies exploring impaired delayed response task performance in aging cynomolgus monkeys and its association to biomarkers associated with Alzheimer's disease (8, 9). Additional data come from a study that relates delayed response task performance of old monkeys to clinical magnetic resonance imaging findings (10). The overall performance in the task referred to as the memory load test (MLT), which we will be focusing on in the present study, is summarized in Table I.
When analyzing the data of the aforementioned studies, it was noted that individuals within an age group would differ greatly in their performance, begging the question whether other characteristics might also influence test performance. Non-human primate studies often feature small experimental groups – it is fairly common that the subjects are identified by name – and the cognitive studies will often discuss individual personalities. With a fairly large sample size, we decided to use this opportunity to investigate potential sex differences and whether the type of breeding facility the subjects originated from would influence test performance. Moreover, food item baits in certain cup positions would be retrieved more often than others and incorrect answers conformed to patterns shared by all subjects. Our hope was that insights into these answering patterns might be further used to distinguish between errors of different types, and thus, might serve to refine the framework for interpreting results from similar tests. With monkeys' and apes' performance in (static) delayed response tasks being comparable (11), we aimed to compare and contrast the findings in cynomolgus monkeys to trends/strategies previously discussed in great apes. Of particular interest was whether the monkeys use landmark cues to locate baits, the underlying mechanisms leading to clustering of answers and whether the monkeys are able to approach the challenge of retrieving baits in a systematic fashion.
We hypothesized that the cynomolgus monkeys would exhibit common strategies – underlying the spatial response biases – in solving the delayed response task. Since delayed response task performance tends to decline with age (12), we further hypothesized that the strategies would differ between age groups. Lastly, we hypothesized that the sex and life-history of the subjects would influence the overall performance, in addition to the previously-found effects of age.
Materials and Methods
Working memory was assessed using a delayed response task similar to what has been previously used in apes (13-15), termed the [working] memory load test (MLT). In this test, six identical opaque cups (7 cm in diameter) were arranged on a 55×20 cm tray. The cups were placed upside-down on the tray, always going from left-to-right (directions are given from the experimenter's point of view throughout), and small pieces of fruit (sliced apple or guava) were placed under two of the cups. The experimenter made sure that the subject was oriented toward the test apparatus and was watching as the cups were baited. Either immediately after the baiting, or following a 30-second delay, the tray was placed within the subject's reach, allowing the monkey to designate a cup by touching it. If a baited cup was chosen, the subject received the fruit within; if an empty cup was designated, the experimenter showed the cup to be empty. The tray was left in front of the subject until two answers had been provided (or the monkey moved away from the front of the testing cage, signaling a loss of interest in the test), at which point the trial was terminated. The full testing procedure can be seen in films that have been supplied as online supplements to this article. The test was conducted in randomized modules consisting of all 30 possible trial combinations (15 possible bait combinations with or without a 30-second delay). Subjects would receive one module's worth of testing per day – no more. The baiting process is illustrated in Figure 1.
Originally experimentally-naïve to delayed response tasks, all subjects were trained in three types of tests (for information on the two memory tests not covered in this study we refer to Darusman et al. (8)). Training consisted of building up the test procedure by increments. In the first step, the subjects were trained to collect a bait from the testing tray, habituating to the testing equipment and experimenter. Once successful, the subject was trained to retrieve a bait by designating a cup, but where the bait was only partially hidden by the cup. Finally, the subject was trained to retrieve a fully hidden bait from a cup. At the time of testing, subjects voluntarily sat down and faced the experimenter for the baiting procedure. If a subject at any point during testing lost interest and would not observe the baiting process, testing was stopped, to be resumed another day. When the subjects reached a stable performance level in the tests and showed strong motivation for participating in the task, data collection commenced.
The MLT data were collected from 27 cynomolgus monkeys of both sexes, consisting of both island-bred, wild-caught animals and animals born in caged captivity (Table II). The cage-bred animals originated from the Primate Research Center (PRC) IPB breeding facility (Bogor, Indonesia), whereas the island-bred animals were collected from the PRC natural habitat breeding facility on Tinjil island. Inside the PRC breeding facility, the animals were housed in social groups (uni-male-multi-female in composition) with access to both indoor and outdoor areas. As an integrated standard operating procedure, all animals had to be quarantined for a month before being brought into the testing facilities to minimize the risk of their bringing in pathogens. The subjects were caged in pairs approximately one month prior to being quarantined. Since the quarantine itself took 30 days, the subjects were in essence habituated to their new environment – caging, cage-mates, diet, staff, and daily routines – for two months prior to training/testing. In the testing environment, the subjects were paired by age, adapting the age ratio of Moss et al. (16) from humans to macaques. The age groups were “Young” (4-9 years old), “Middle aged” (10-19 years old) and “Old” (above 20 years old). A finer age-subdivision, although desirable, was not possible. Whereas the ages of the cage-bred animals were known, the island-bred animals' (approximate) ages had to be determined through dental scaling (17, 18).
An implicit assumption underlying a subset of the analyses in this study is that the time delay employed in half of the trials influences the behavior of the subjects indirectly – and only indirectly – by affecting memory retention. The subjects are less likely to correctly recall the bait placement when a delay is employed; but it is the remembering, or not remembering, that influences the subjects' behavior. As the effect is of an indirect nature, we can choose to omit the time delay as an explanatory variable when analyzing answering patterns/behavior (however, not when analyzing performance); the time delay is only one of many unaccounted-for factors that may influence memory retention. Previously (8), inter-observer reliability had been estimated from a random subset (n=90) of videotaped tests scored by two independent observers and found to be high (Cohen's kappa: 0.97).
Most analyses were carried out at the level of the individual; however, for studying the clustering of answers in the MLT, all trials were considered independent observations, the hypothesis thus being that the clustering is, in all subjects, due to the same underlying mechanisms. To analyze which factors contributed to the overall retrieval accuracy of baits in the test, a general linear mixed effects model (GLMM) was applied. The retrieval accuracy (an ordinal variable) was modeled with a generalized logit link, subject identities were accounted for as random effects (with an unstructured covariance type), and age [classes], origin (cage-bred or island-bred), sex and directionality of answers were provided as explanatory variables. Finally, the interaction of these variables with the time delay was tested, as essentially there were two levels of difficulty – with and without a time delay – and it was conceivable that some individuals would be better at tackling the more difficult tests. To accommodate this exploratory model structure, Satterthwaite approximations were used for the degrees of freedom and robust covariances were used for the tests of fixed effects. For testing whether adjacent answers affected the rate of missing both baits, a similar GLMM was constructed. Here, a nested structure for the individuals was used, accounting for individuals, sex and age as a random effect. The double failure was considered a binary outcome and the effect of providing adjacent answers was put forth as an explanatory variable.
Results
Analyzing the main factors influencing the retrieval accuracy in the trials (Table III), an effect of age, previously discussed by Darusman et al. (8, 9), was found as well as the previously established (Table I) effect of adding a delay to a trial. In addition, an effect of sex was found, where female subjects performed better than their male counterparts, and cage-bred subjects performed better in the trials with no delay than did island-bred monkeys. The latter effect, particularly prominent in the old subjects (Figure 2), probably affected all trials, but was more difficult to distinguish due to a floor-effect, with some subjects approaching chance levels, in the trials where a 30-second delay was utilized.
Finally, an effect of the direction of search – whether the two answers were provided going from left to right, or the opposite – was found to influence the overall retrieval accuracy. During the tests, the monkeys were, regardless of age, better at retrieving the last baited cup – the rightmost of the two baits (Figure 3). A subject who intent on retrieving both baits would, however, probably benefit from recreating the baiting process from memory, chronologically, moving from left to right, picking the baits in the order they were placed in the cups. The present data suggest such a strategy may indeed exist among the tested monkeys. Left-to-right chosen search-pairs were unevenly distributed across the age groups (χ2=7.3, p<0.05), occurring more frequently among the young monkeys than among the middle aged ones; this was, in fact, the predominant strategy in young monkeys (54% left-to-right chosen pairs in contrast to the 47% of old and middle-aged monkeys). Even though many of the cases have occurred by chance – a monkey choosing truly at random is after all as likely to go right-to-left as the opposite – there is good reason to assume that some of the data were due to a more deliberate, systematic, memory recall process: The left-to-right search approach yielded significantly better results in retrieving baits (Table III), the difference being an increased retrieval rate of both baits in a trial by approximately 40% (Table IV). Worth noting is that no interaction between directionality and age was found (data from a separate GLMM are not shown). That is: although predominant as a strategy among the young monkeys, choosing baits in a left-to-right order appears to have improved the retrieval accuracy of monkeys, regardless of age.
Whereas the overall test performance was contingent on the subject's age, sex, background/life history and ability to approach the challenge in a systematic fashion, there were distinct patterns to the answers shared by all subjects, suggesting common strategies were employed. All monkeys showed a clear bias toward the middle cups in the trials, regardless of bait positions. Since the baits were perfectly equally distributed across the cups in each module, correct answers, interspersed with perfectly random guesses should align along the dashed line in Figure 4. As this is clearly not the case, it would seem the monkeys gravitated towards the middle cups when choosing. The central tendency in cup preference was not positively correlated with retrieval accuracy however (Figure 5) – in more than a third (average: 35%) of the trials, the subjects chose cup number three with their first guess, but less than half of these choices were found to be correct (average: 42%). This can be contrasted with cup number six, which was initially chosen less than once in every ten trials (average: 8.9%), but these choices turned out to be correct more than four times in five (average: 82%).
In addition to the central tendency, the monkeys tended to provide clustered answers: More than four out of five (82%) of all provided answer-pairs were adjacent cups, which can be contrasted with only one third (33%) of all bait positions being adjacent. Whereas this adjacent-answers bias naturally made the monkeys better at obtaining both baits when they were in adjacent positions, they failed at retrieving any of the baits just as often as they would with non-adjacent baits (F1,1.98=0.22, p=0.64). This indicates that the monkeys were no better at recalling adjacent positions than they were at recalling non-adjacent positions; the difference simply lies in the answering strategy. The central tendency in guessing (Figure 4) will naturally increase the odds for adjacent answer-pairs. But if we assume that the guesses were independent of one another, the resulting frequency of adjacent answer-pairs accounted for is still only 45%. Clearly the answers were not independent of one another and the high level of answer adjacency cannot be attributed solely to the central tendency. This strategy may instead in part be a simple search pattern where the monkey picks the neighboring cup following a mistaken answer. Looking at only the non-adjacent bait positions, when the monkeys provided a correct answer on the first attempt, the incidence of non-adjacent answers approximately doubled (Table V; Relative risk 95% confidence interval (CI): 1.84-2.94; χ2=54, p<0.001). Since correct answers in part abolished adjacent answers (for non-adjacent baits), it is fair to assume that they were, at least in part, due to a natural search pattern where the monkeys adjust their initial guesses. In the case of a correct first answer, however, the subjects would still choose adjacent cups in 71% of the (non-adjacent bait) trials, which is considerably higher than the expected 45% (accounting for central tendency).
A complete data set, comprising the raw data from all 1,980 trials, has been made available under a free-to-use license (refer to the section on Additional Online Materials).
Discussion
While all subjects performed at higher-than-chance levels, it is clear that the task was challenging for cynomolgus monkeys of any age. Even the best performing subjects could not reliably retrieve both baits under the best of conditions, and with an added 30-second delay, the old monkeys' performance barely exceeded chance levels. It is clear from the data that not only the age of the subjects predicted their test performance; sex and life history seemed to have a significant influence as well.
A study highlighting sex-differences in rhesus monkeys solving spatial tasks has found, much like the present study, that females would out-perform their male counterparts (19). This was hypothesized to be due to a superior ability of the females in relating baited positions to (local) landmark cues. Although this study is the only one we are aware of investigating the phenomenon in primates, the theory is supported by studies in rodents and humans, where it has been shown that males tend to rely highly on distal cues for spatial orientation, whereas females tend to be more flexible, but primarily use local cues (20). In the present study, the high retrieval accuracy of baits placed at the ends of the array (cups 1 and 6 – refer to Figure 5) mirrors previous findings (15) in great apes. As with apes, we suggest that this pattern emerges due to the subjects' using landmark cues to locate the hidden baits – i.e. the end cups are more easily remembered/retrieved by their not being flanked on both sides by other cups or by relating their positions to the edges of the tray. When examining the retrieval accuracy in the present study by sex (Figure 6), we find support for the hypothesis proposed by Herman and Wallen (19), also in cynomolgus monkeys. Whereas all monkeys excelled at retrieving baits from the last-baited cup of the array (cup 6), the females were also proficient at retrieving baits placed at the other extreme end of the array (cup 1); their male counterparts were not. This, in turn, can be seen as a symptom of the male subjects being worse at utilizing local landmark cues, which may help explain the overall performance difference between sexes.
Why cage-bred monkeys would perform better overall than did the island-bred monkeys can be debated. An explanation could be that cage-bred monkeys were more accustomed to working with an experimenter, having a long-standing history of interacting with personnel during their daily feeding and cage cleanings. They would also be more adjusted to the tools and treats used, to the environment, to being subjected to arbitrary rules etc., which may have made them ultimately perform better. The effect of enculturation has been discussed in length with greater apes and, although the concept is a source of great controversy and the term itself may not be a good fit for monkeys (21), there is some evidence for monkeys being better-equipped for tackling human-designed games if they have lived in close proximity to people (22, 23). In addition, the island-bred subjects can essentially be considered wild-caught, and may thus have found the testing experience – in particular, interacting with an experimenter – to be much more stressful than did the cage-bred subjects. They may furthermore have been subjected to prolonged stress, if we assume that they adapted poorly to captivity. Acute stress has been shown to negatively impact cognitive test results in monkeys (24-26) and chronic stress is well-known to affect the hippocampus (27), which plays a crucial role in solving delayed response tasks.
Whereas the overall performance differed greatly between monkeys of different sex, of different ages and from different breeding facilities, the strategies and resulting answering patterns were remarkably ubiquitous. The central tendency in cup preference – although most striking – appears to be the result of guesswork. When not remembering the location of a bait, but prompted to make a choice, it seems that the subjects would choose a centrally placed cup. This may be simply due to convenience, as the central cups would be physically nearest to the subject when being centered across from the tray. Alternatively, if the subjects found it, in some trials, hard to distinguish individual cups, but rather treated the entire array as an object; it would be natural for them to aim for the center of the “object.”
It appears, however, that the guesswork exhibited by the monkeys was not without some rationale. The high incidence of adjacent answers mirrors findings from similar studies, primarily involving great apes. Call (13) noted that, when prompted to provide multiple answers – to retrieve two hidden baits from a linear array of possible caches – great apes and human children were more likely to provide adjacent answers than could be explained by chance. It was suggested that the pattern arose from the subjects' poor inhibition control – the positive trait of the baited caches “carrying over” to their neighbors. This was disputed by Beran and collaborators (14) who, with a larger array, could not find evidence to support the theory in chimpanzees. Instead, adjacent answers were explained as the results of incorrect memory recall. The merits of this reasoning were acknowledged by Barth and Call (28), who nevertheless pointed-out that the disproportionate number of adjacent answers probably had a multi-factorial explanation as poor memory recall alone could not explain the answering patterns.
Providing mostly adjacent answers did not, in an obvious way, improve the monkeys' overall retrieval accuracy in the trials. The pattern does, however, betray a conscious strategy: The fact that the frequency of adjacent answers was conditional on whether the first answer was a success suggests that the monkeys were able to improve on an incorrect answer by guessing at an adjacent cup. This helps explain the clustering of answers seen in these tests. However, even accounting for this adaptive search pattern and the central tendency in answers, the frequency of adjacent answers (for non-adjacent baits) was far higher than could be expected by chance. The remaining unaccounted-for adjacent answers cannot be explained using the memory hypothesis of Beran et al. (14) but may well be an expression of poor inhibition control.
The central tendency in answers combined with the disproportionately high number of adjacent answers suggests a high level of impulsivity. The monkeys tended to answer rashly, reaching for cups that were physically proximal to them once the tray was placed within their reach; a majority of first answers were provided within the first two seconds (unfortunately only a subset of the trials were filmed, thus preventing a formal analysis). The rashness is a well-known characteristic of the cynomolgus monkey: Amici et al. (4) previously reported that cynomolgus monkeys would, when challenged with a simple task of retrieving a food item from behind a translucent screen, unlike many closely-related primates, more often than not, attempt to go through the screen, rather than first finding an appropriate hole or door allowing them to reach the bait. They also found that cynomolgus monkeys were poor at postponing gratification. The answering strategy employed by the subjects in the present study may also, in part, be related to the social structure of the cynomolgus monkeys. The animal caretakers identified most of the studied subjects to be socially subordinate, based on their behavior. Living in a highly despotic, strict dominance hierarchy (29), alpha males/females have first pick when it comes to food items available to the colony. This, in turn, may force the lower ranked animals – comprising the majority of the colony – to adjust their strategy in retrieving the available food. Priority may be given to acting fast, possibly at the expense of accurate decision making. In future studies, it would be interesting to see if social rank, as well as sex, age and breeding facilities, have an influence on the overall success in similar cognitive tests. It is also worth noting that the monkeys were not overly penalized for guessing in the MLT paradigm, as they were more likely to obtain at least one bait per trial, than not (pure guesswork would yield an average of 0.66 baits per trial). It can be speculated whether this may have inflated the level of guesswork by making the subjects somewhat complacent.
With the array of cups being visible throughout the trial – as opposed to similar tests where the array is hidden in the interim between baiting and challenge – non-mnemonic solutions to the test can be utilized. The subject could place itself, spatially, in front of one of the baited cups, thus eliminating the need for correctly recalling its position (30). As previously concluded by Zola and collaborators (31) in a delayed non-matching to sample task, however, no evidence for a reliance on similar techniques could be found in the cynomolgus monkeys. The purported prominent use of landmark cues does not, in itself, rule out the possibility that the subjects also utilized egocentric or non-mnemonic strategies. But, even though the positioning of the subjects during the individual trials was not recorded, we would, nevertheless, expect a high reliance on these to produce a much more uniform retrieval accuracy, at least for the first chosen cups, than is evident from Figure 5 and 6. In addition, it was noted that subjects would frequently move laterally to the array in the testing cage during trials, thus making non-mnemonic strategies reliant on their own position hard to implement.
Data further suggest that the acts of remembering (/recalling) and retrieving should be separated for a fruitful discussion. The cups at the ends of the array were baited just as often as the central cups, but the monkeys seemed to direct their guesses toward the ends much less frequently. This is in stark contrast to these cups being associated with the highest retrieval accuracy. In order to retrieve a hidden bait, the subject needed to first recall its position and then correctly identify the remembered position in the array. Evidence suggests that it is the latter task that is easier with the cups at the extreme ends of the array. A possible interpretation of the data is thus that the monkeys do not find it easier remembering the outermost cups – they may even have more trouble remembering them – but when they do, the outermost cups are more easily (correctly) retrieved due to landmark cues.
Furthermore, since the directionality of search determines the retrieval accuracy, we theorize that the monkeys may be capable of recreating the baiting process from memory for improved results. Humans remember the order in which events occurred, using multiple cognitive mechanisms, while it has been suggested that adult rhesus monkeys, rely heavily on a mechanism that appears to encode order of occurrence relative to other events, rather than in absolute time (32). With aging, this mechanism, suggested to be controlled by the hippocampus (33), deteriorates. This agrees with our findings (Figure 3) where young and middle-aged subjects, but not old subjects, seem to gravitate toward retrieving the last hidden bait. Serial position effects, where the subjects adapt to take advantage of recent memories being stronger, have previously been noted in macaques (34-36).
In conclusion, the cynomolgus monkeys exhibited common strategies in solving the MLT. Impulsive guesswork was highly prevalent in our studies but, looking beyond this, we found evidence of the use of landmark cues, of adaptive answering and possibly evidence of systematic (/chronological) memory recall. Whereas we can expect these strategies to have influenced the overall MLT success rate, individual characteristics – age, sex and which breeding facility the subjects originated from – appeared to exert a clearer effect on the monkeys' overall performance. Clearly, subject characteristics and life history need to be considered when managing test variation and assay capability and when interpreting data from cognitive tests on cynomolgus monkeys.
Acknowledgements
This research was supported by the Directorate of Higher Education, Ministry of Education and Republic of Indonesia through a PhD fellowship for Dr. H.S. Darusman. The authors wish to thank Dr. Josep Call of the Max Planck Institute for valuable insights into the method and its interpretation.
Footnotes
-
This article is freely accessible online.
-
Ethical Standards
The experiments were conducted in August-October 2011, April-June 2012 and March-May 2013 in the AAALAC-accredited Primate Research Center IPB (Bogor, Indonesia). Subject housing conditions (before, during and after the experiment) and the test procedures were approved by the Primate Research Center IPB's Animal Care and Use Committee; license number IPB PRC-13-A002.
-
Conflicts of Interest
The Authors declare that they have no conflicts of interest.
-
Additional Online Materials
A complete data set comprising all trials has been made available online under a free-to-use license together with films demonstrating the MLT procedure. They can be accessed using the links listed below. Complete data set: http://dx.doi.org/10.6084/m9.figshare.1038346
Testing procedure without delay, example 1:
http://dx.doi.org/10.6084/m9.figshare.1038348
Testing procedure without delay, example 2:
http://dx.doi.org/10.6084/m9.figshare.1038349
Testing procedure with delay, part 1:
http://dx.doi.org/10.6084/m9.figshare.1038354
Testing procedure with delay, part 2:
- Received May 26, 2014.
- Revision received July 9, 2014.
- Accepted July 10, 2014.
- Copyright © 2014 International Institute of Anticancer Research (Dr. John G. Delinassios), All rights reserved