Role of immigrant males and muzzle contacts in the uptake of a novel food by wild vervet monkeys

  1. Pooja Dongre
  2. Gaëlle Lanté
  3. Mathieu Cantat
  4. Charlotte Canteloup
  5. Erica van de Waal

  • Curated by eLife

    eLife logo

    Evaluation Summary:

    Dongre et al. build on previous social learning research on wild vervet monkeys to investigate the role of a particular social behaviour, muzzle-muzzle contact, in aiding the acceptance of a novel food and provide interesting observations on the potential for male monkeys immigrating from one social group to another in spreading this novel behaviour. With a more robust and transparent analysis, this manuscript has the potential to provide significant insights into innovation and social learning in animals.

    This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.

Reviewed by

Read the full article See related articles

Listed in

Log in to save this article

Abstract

The entry into and uptake of information in social groups is critical for behavioral adaptation by long-lived species in rapidly changing environments. We exposed five groups of wild vervet monkeys to a novel food to investigate the innovation of processing and consuming it. We report that immigrant males innovated in two groups, and an infant innovated in one group. In two other groups, immigrant males imported the innovation from their previous groups. We compared uptake between groups with respect to the initial innovator to examine the extent to which dispersing males could introduce an innovation into groups. Uptake of the novel food was faster in groups where immigrant males ate first rather than the infants. Younger individuals were more likely overall, and faster, to subsequently acquire the novel food. We also investigated the role of muzzle contact behavior in information seeking around the novel food. Muzzle contacts decreased in frequency over repeated exposures to the novel food. Muzzle contacts were initiated the most by naïve individuals, high rankers, and juveniles; and were targeted most towards knowledgeable individuals and high rankers, and the least towards infants. We highlight the potential importance of dispersers in rapidly exploiting novel resources among populations.

Article activity feed

  1. Version published to 10.7554/elife.76486 on eLife
  2. Version published to 10.1101/2021.12.16.472640v3 on bioRxiv
  3. eLife

    Author Response

    Reviewer 1

    Strengths:

    This manuscript combines experimental, exploratory, and observational methods to investigate the big question in innovation literature--why do some animals innovate over others, and how information about innovations spread. By combining a variety of methods, the manuscript tackles this question in a number of ways, and finds support for previous work showing that animals can learn about foods via social olfactory inspection (i.e., muzzle to muzzle contact), and also presents data intended to investigate the role of dispersing animals in innovation and information spread.

    Using data from a previously-published experiment, the manuscript illustrates how investigators can numerous interesting questions while limiting the disturbances to wild animals. The manuscript's attempt at using exploratory analysis is also exciting, as exploratory analyses provide a useful tool for behavior research-indeed, Tinbergen insisted that behavior must first be described.

    Weaknesses:

    The manuscript's introduction is a bit unclear as to how the fact that dispersing males may be an important source of information ties to innovations in response to disruptions due to climate change, humans, or new predators, if at all. An introduction regarding the role of dispersed animals in introducing novel behaviors and social transmission would better prepare readers for the questions presented in the manuscript. As it stands now, the manuscript only provides one sentence discussing the theoretical relevance of investigating the role of dispersing animals in innovations.

    We have added some information about this to the introduction (lines 66 – 69 and 121-123) and maintain our discussion of it in the discussion.

    Additionally, while the manuscript attempts to use exploratory analysis, it does not provide enough theoretical background as to why certain questions were asked while the data were explored. While the discussion provides some background as to the role of dispersing males in innovation, the introduction provides little background, and thus does not properly frame the issue. It is unclear how dispersing males became of interest and why readers should be interested in them. As the manuscript reads now, it may be that dispersing males became interesting only as a result of the exploratory analysis-except that the predictions explicitly mentions dispersing males. Thus, manuscript at present makes it difficult to know if the questions surrounding immigrant males resulted from the exploratory analysis, or was a question the analyses were intended to answer from the beginning. If this question only came out after first reviewing the results, then this needs to be made clear in the introduction. I see no issue with reporting observations that were the result of investigations into earlier results, but it needs to be reported in a way that can be replicated in future research-I need to know the decision process that took place during the data exploration.

    We hope this is clearer from our new research aims (lines 125-173)

    The manuscript never clearly defines what counts as an immigrant male; presumably, in this species, all adult males in the group should be immigrants, as females are the philopatric sex. Sometimes, the manuscript uses "recently" to modify immigrant males, but doesn't define exactly what counts as recent, except to say that the males that innovated were in their respective groups for fewer than 3 months, but never explains why three months should be an important distinction in adult male tenure.

    We realise how we wrote about this previously was not clear and perhaps misleading. We noticed that the males that innovated had been in the group for less than three months. We do not know if this is necessary for them to innovate or not. We also added to the discussion a description of the male in AK19 who had been in the group for four months and did no innovate – as he had many other traits which we would expect to exclude him from criteria for innovation (e.g. very old, post-prime, and inactive – died within months of the experiment).

    Due to the above weaknesses, the provided predictions are a bit murky. It is not clear how variation between groups in accordance with who innovated, or initiated eating a novel food, or demographics is related to the central issue. The manuscript does contribute to the literature by looking at changing rates of muzzle contact over exposure to a novel food source, and provides a good extension of previous findings; that, if muzzle contacts help animals learn about new foods, then rates of muzzle contacts involving novel foods should decrease as animals become familiar with the food. However, this point isn't explicit in the manuscript.

    This is now addressed in the new aims paragraph (lines 125-173)

    Finally, it is also unclear as to why changing rates of muzzle contact AND whether certain individual level variables like knowledge, sex, age, and/or rank might influence muzzle contacts during opportunities to innovate.

    We are not sure exactly what the reviewer means here, but hope that the substantial revisions we have made now address their concern.

    As for the methods, the manuscript doesn't provide enough details as to why certain decisions were made. For example, no reason is given as to why only the first four sessions after an animal ate were considered, why the first three months of tenure (but not four, as seen on one group that didn't innovate) was considered to be a critical time for which immigrant males may innovate, why (including the theoretical reasons) the structure of models for one analysis was changed (dropping one variable, adding interactions), or even how the beginning and ending of a trial was decided, despite reporting that durations varied widely,-from 5 minutes to two hours.

    Please see: above about the male with 4 month tenure; and top of document for description of our updated models.

    The discussion contains results that are never elsewhere presented in the manuscript- (2a) Individual variation in uptake of a novel food according to who ate first).

    It was just an error in the sub-title in the discussion – this is now amended. But all the other corresponding details were already there, in the list of research aims in the introduction and in the results as well.

    Finally, the largest issue with the manuscript is that its results are not as convincing as the conclusions made. An issue with all the analyses is that some grouping variables in some analyses but not others despite the fact that all of the analyses contain multiple groups (necessitating group as a grouping variable) and multiple observations of the same individuals (i.e., immigrant males tested in multiple groups, necessitating animal identity as a random effect), and not accounting for individual exposure to the experiment when considering whether animals ate the food in the allotted period (an important consideration given the massive differences in trial times), making these results difficult to interpret in their current forms. As for the results regarding muzzle contact, the analyses has a number of issues that make it difficult to determine if the claims are supported. These issues include not explaining why rank calculated a year before the experiments took place was valid or if rank was calculated among all group members or within age and sex classes, not explaining how rank was normalized, and not conducting any kind of formal model comparisons before deciding the best model.

    Mostly addressed at top of this document. Regarding rank calculations: rank was not calculated a year before the experiments, it was calculated using a year’s worth of data up to the beginning of the experiments – and ranks were calculated among all group members - we have made this clearer in the methods. We also explained our method of normalisation, and noted that it was an error to include non-normalised rank in one of the models – this has now been rectified

    As for the results regarding immigrant males and innovation, little is done to help the fact that these results are from very few observations and no direct analyses. It is possible that something that occurs relatively often but in small sample sizes, like dispersing animals, could have immense power in influencing foraging traditions, and observation is a necessary step in understanding behavior. However, the manuscript doesn't consider any alternative hypotheses as to why it found what it found. No other possible difference between the groups was considered (for example, the groups that rapidly innovated appear to be quite smaller than the groups that did), making the claim that immigrant males were what allowed groups to innovate unconvincing. This is particularly true given that some groups in this study population have experimental histories (though this goes unmentioned in the current manuscript), which likely influenced neophobia-especially given work by the same research group showing that these animals are more curious compared to their unhabituated counterparts.

    We have added more discussion of alternative hypotheses to the discussion (line numbers mentioned above).

    Regarding the comment about rapid innovation in smaller groups – we are not sure what the reviewer means here – all groups except BD were similar sized. The second largest group, NH, had one of the quickest innovations and a smaller group (KB) innovated only at the third exposure. Unless the reviewer instead refers to the spread of the innovation here? This is also not quite what we see in the data – BD is the largest group and one of the fastest to spread, and KB is the smallest group and the slowest to spread. Regarding groups experimental histories, all the five studied groups have already been used in field experiments. The group (LT) with the least experimental history was the one having the greatest proportion of individuals eating the novel food at the first and over the four exposures (see Fig. 2) while one of the groups with the most experimental history (NH) was one having a smaller proportion of individuals eating the food across the experiment. This is discussed in the discussion (lines 370-380).

    Reviewer 2

    I have separated my issues with the manuscript into three sub-headings (Conceptual Clarity, Observational Detail and Analysis) below.

    1. Conceptual clarity

    There are a number of areas where it would greatly benefit the manuscript if the authors were to revisit the text and be more specific in their intentions. At present, the research questions are not always well-defined, making it difficult to determine what the data is intended to communicate. I am confident all of these issues could be fixed with relatively minor changes to the manuscript.

    For example, Line 104: Question 1 is not really a question, the authors only state that they will "investigate innovation and extraction of eating the food", which could mean almost anything.

    We re-wrote the research questions paragraph and results with this advice in mind – hope it is clearer now. We keep the innovation part just descriptive and hope this is less problematic now.

    Question 2a (line 98) is also very vague in it's wording, and I'm left unclear as to what the authors were really interested in or why. This is not helped by Line 104 which refuses to make predictions about this research question because it is "exploratory". Empirical predictions are not simply placing a bet on what we think the results of the study will be, but rather laying out how the results could be for the benefit of the reader. For instance, if testing the effects of 10 different teaching methods on language acquisition-rate: Even if we have no a priori idea of which method will be most effective, we can nevertheless generate competing hypotheses and describe their corresponding predictions. This is a helpful way to justify and set expectations for the specific parameters that will be examined by the methods of the study. In fact, in the current paper, the authors in fact had some very clear a priori expectations going into this study that immigrant males would be vectors of behavioural transmission (clear that is from the rest of the introduction, and the parameters used in their analysis, which were not chosen at random).

    We have now updated the whole research aims (lines 125-173).

    The multiple references to 'long-lived' species in the abstract (line 16 and introduction (39, 56) is a bit confusing given the focus of this study. Although such categorisations are arbitrary by nature (a vervet is certainly long-lived compared to a dragonfly), I would not typically put vervet monkeys (or marmosets, line 62) in the same category as apes (references 8 and 9) or humans (line 62) in this regard.

    When we use “long-lived” in the introduction, we explain that we mean animals with slow generational turnover for whom genetic adaptation is relatively slow – too slow to adapt to very rapid environmental change. Within the distinctions the reviewer makes here, we feel that vervets and marmosets are much more similar to apes than to dragonflies etc. in this respect… and we think making the comparisons that we do are valid in this context (though we do agree that for other reasons we would not find it appropriate). We have modified the sentence in the introduction (line 4042) and hope this is clearer now. The study in reference 9 is about crop-raiding, which is something vervets can learn to do within one generation too. In addition, reference 8 is used as it was one of the earlier and long-standing definitions of innovation which we are using here – we are not comparing vervets to apes directly, but we do not think a different definition of innovation is required.

    This contributes a little towards the lack of overall conceptual focus for the manuscript: beginning in this fashion suggests the authors are building a "comparative evolutionary origins" story, hinting perhaps at the phylogenetic relevance of the work to understanding human behaviour, but the final paragraph of the study contextualises the findings only in terms of their relevance to feeding ecology and conservation efforts. I would recommend that the authors think carefully about their intended audience and tailor the text accordingly. This is not to say that readers interested in human evolution will not be interested in conservation efforts, but rather that each of these aspects should be represented in each stage of the manuscript (otherwise - conservationists may not read far into the Introduction, and cultural evolution fans will be left adrift in the Conclusion).

    We agree that the line running through the whole paper needed to be clearer and have tried to improve this.

    1. Observational detail

    There are a number of areas of the manuscript which I found to be lacking in sufficient detail to accurately determine what occurred in these experimental sessions, making the data difficult to interpret overall. All of this additional information ought to be readily available from the methods used (the experiments were observed by 3-5 researchers with video cameras (line 341)) and is all of direct relevance to the research questions set out by the authors.

    We added more details about the experiment in the method section.

    While I appreciate that it will take quite a bit of work to extract this information, I am certain that it would greatly improve the robustness and explanatory power of this study to do so.

    The data on who was first to innovate/demonstrate successful extraction of the food in each group (Question 1) and subsequent uptake (Question 2), as well as the actual mechanism by which that uptake occurred (the authors strongly imply social learning in their Discussion, but this is never directly examined) is difficult to interpret based on the information presented. Some key gaps in the story were:

    We did not intend to claim that muzzle contact was the specific mechanism by which individuals learned to extract and eat peanuts – we rather use this experiment to evaluate the function of muzzle contact in the presence of a novel food.

    We did not record observation networks in all groups during experiments and cannot obtain accurate ones from all our videos – we hope it is clearer in our text now. Our group’s previous study (Canteloup et al., 2021) already shows social transmission of the opening techniques using data of two of our groups (NH and KB).

    • Which/how many individuals encountered the food and in what order? I.e., were migrants/innovators simply the first to notice the food?

    No, and we have now added some info about other individuals approaching the box and inspecting the peanuts before innovation took place

    • Did any individuals try and fail to extract the food before an "innovator" successfully demonstrated?
    • How many tried and failed to extract the nuts before and after observing effective demonstrators?

    We have added the number of individuals that inspected the peanuts (visually and with contact)

    • Were individuals who observed others interact with the food more likely to approach and/or extract it themselves?
    • Did group-members use the same methods of extraction as their 'innovators'?

    Yes – this is the topic of Canteloup et al. 2021 – and these data are not presented again here. That study was on two of the groups presented here (KB and NH), and with up to 10 exposures in each of those groups and present a fine-grained analysis of peanuts opening techniques used by monkeys. We hope this is clearer now in the text where we refer to this paper.

    • How many tried and succeeded without having directly observed another individual do so (i.e. 'reinvention' as per Tennie et al.)?

    For this, and the above points: We did not record an observation network for the groups added in this study and are not able to answer this – it is not the focus of this study. For this reason, we do not make claims in this line in the present study, and are cautious with our social learning related language. Whilst we examine the role of muzzle contact in acquiring information about a novel food, we do not expect this behaviour to be a necessary prerequisite in being able to extract and eat this food – indeed many individuals who learned to eat did not perform muzzle contacts. This aspect of the study is about using this novel food situation to explore whether muzzle contact serves information acquisition – which our evidence suggests it does.

    Moreover, the processing of this food is not complex and is similar to natural foods in their environment, and we do expect individuals to be capable of reinventing it easily (and this point with Tennie’s hypothesis is actually discussed in Canteloup et al. 2021 paper) – but the point here is that their natural tendency is to be neophobic to unknown food, and therefore they do not readily eat it until they see a conspecific doing so, after which they do. And we also used this opportunity, though in a very small sample size, to investigate which individuals would overcome that neophobia and be the first to eat successfully.

    The connective tissue between the research questions set out by the authors is clearly social learning. In short: the thesis is that Migrants/Innovators bring a novel behaviour to the group, then there is 'uptake' (social learning), which may be influenced by demographic factors and muzzle-contact (biases + mechanisms). Given this focus (e.g. lines 224-264 of the Discussion), I would expect at least some of the details above to be addressed in order to provide robust support for these claims.

    See above – the reason we talk about ‘uptake’ rather than social learning is that we really see this as a case of social disinhibition of neophobia, rather than more detailed social learning such as copying or imitation, as it would be in a tool-use setting, for example (though in Canteloup et al. 2021 paper, evidence is found that the specific methods to open peanuts are socially transmitted).

    Question 2a (Lines 136-146): This data is hard to interpret without knowing how much of the group was present and visible during these exposures.

    Please see response to reviewer 1 on this.

    For example: 9% update in NH group does not sound impressive, but if only 10% of the total group were present while the rest were elsewhere, then this is 90% of all present individuals. Meanwhile if 100% of BD group were present and only experienced 31% uptake, then this is quite a striking difference between groups.

    Experiments were done at sunrise at monkeys’ sleeping site in AK, LT, NH and KB where most of the group was present in the area; we added more precision on this point in the Method section (lines 615-619).

    Of course, there is also an issue of how many individuals can physically engage with the novel food even if they want to - the presence of dominant individuals, steepness of hierarchy within that group, etc, will significantly influence this (and is all of interest with regards to the authors' research questions).

    We discuss this with respect to the result showing that higher rank individuals were more likely to extract and eat the food at the first exposure and over all four exposures

    Muzzle-contact behaviour: The authors use their data to implicate muzzle-contact in social learning, but this seems a leap from the data presented (some more on this in the Analysis section).

    We hope our distinction between information acquisition and information use is clearer now.

    For example:

    • What is the role of kinship in these events?

    We did not analyse kinship here, but we see a lot of targeting towards adult males, and we do not have reliable kinship data for them. We also checked (see response to reviewer 3) the muzzle contacts initiated by knowledgeable adult females, and they are mostly towards adult males, not towards related juveniles (see new figure 4D and lines 497-500).

    • Did they occur when the juvenile had free access to the food (i.e. not likely to be chased off by a feeding adult)?

    We recorded muzzle contacts visible within 2m of the box, so individuals were not necessarily eating at the box at the time of engaging in muzzle contacts. However, the majority of muzzle contacts that we could record took place directly at the edge of the box – at the location where the food is accessed – so an individual would not likely be if they were not able to have access to the food. It is possible they could be there and not eating, but they would not have been chased off, otherwise they would not be able to engage in muzzle contacts there. But it is not entirely clear what the reviewer’s point is here.

    • Did they primarily occur when adults had a mouthful of food? (i.e. could it simply be attempted pilfering/begging)

    This is not typical of this species. Very few specific individuals remove food from others’ mouths, and they do it with their hands, usually beginning with grooming their face and cheekpouches, before prising their mouth open and removing food from the victim’s cheekpouches

    • What proportion of PRESENT (not total) individuals were naïve and knowledgeable in each group for each trial (if 90% present were knowledgeable, then it is not surprising that they would be targeted more often)?

    We agree somewhat with this statement, but given the multiple ways we show the effect of knowledge – both at the individual level and the group level (effect of exposure number i.e. overall group familiarity) – we feel we present enough evidence to establish the link between knowledge of the food and muzzle contacts. We find that the model showing the interaction between exposure number and number of monkeys eating on the overall rate of muzzle contacts actually addresses this issue, because we see that when many monkeys are eating during later exposures, when many were indeed knowledgeable, the rate of muzzle contacts is massively decreased. Moreover, if 90% of the individuals present are knowledgeable, then only 10% of the individuals present are naïve, and we show both that knowledgeable individuals are targeted, but also that naïve individuals are initiators.

    • Did these events ever lead to food-sharing (In other words, how likely are they to simply be begging events)?

    We do not observe food-sharing in vervets.

    • Did muzzle-contact quantifiably LEAD to successful extraction of the food? If the authors wish to implicate muzzle-contact in social learning, it is not sufficient to show that naïve individuals were more likely to make muzzle-contact, they must also show that naïve individuals who made more muzzlecontact were more likely to learn the target behaviour.

    We disagree here, because there is a distinction between information acquisition and information use - obtaining olfactory information about a novel resource that conspecifics are eating is not the same as learning a complex tool use behaviour for which detailed observation of a model is required. We are not claiming that that muzzle contact is THE mechanism by which the monkeys learn how to eat the food – but we do believe that the clear separation between naïve individuals initiating and knowledgeable individuals being target, and the decrease of the rate of this behaviour as groups’ familiarity with the food increases – is good evidence that this behaviour functions to acquire information about a novel food.

    1. Analysis

    There are a number of issues with the current analysis which I strongly recommend be addressed before publication. Some of these are likely to simply require additional details inserted to the manuscript, whereas others would require more substantial changes. I begin with two general points (A & B), before addressing specific sections of the manuscript.

    A) My primary issue with each of the analyses in this manuscript is that the authors have fit complex statistical models for each of their analyses with no steps to ascertain whether these models are a good fit for the data. With a relatively small dataset and a very large number of fixed effects and interactions, there is a considerable risk of overfitting. This is likely to be especially problematic when predictor variables are likely to be intercorrelated (age, sex and rank in the case of this analysis).

    We have now checked for overfitting in our models.

    The most straightforward way to resolve this issue is to take a model-comparison approach. Fitting either a) a full suite of models (including a 'null' model) with each possible permutation of fixed effects and interactions (since the authors argue their analysis is exploratory) or b) a smaller set of models which the authors find plausible based on their a priori understanding of the study system. These models could then be compared using information criterion to determine which structure provides the best out-of-sample predictive fit for the data, and the outputs of this model interpreted. Alternatively, a model-averaging approach can be taken, where the effects of each individual predictor are averaged and weighted across all models in the set. Both of these approaches can be performed easily using the r package 'MuMIn'. There are also a number of tutorials that can be found online for understanding and carrying out these approaches.

    Please see our answer at the beginning of the document, detailing how we have updated our models.

    B) It does not seem that interobserver reliability testing was carried out on any of the data used in these analyses. This is a major oversight which should be addressed before publication (or indeed any re-analysis of the data).

    We have added this now and mention it above already.

    Line 444: Much more detail is needed here. What, precisely, was the outcome measure? Was collinearity of predictors assessed? (I would expect Age + Rank to be correlated, as well as Sex + Rank).

    This is now addressed (please see details above) – we use VIFs to assess multicollinearity of predictors in our models and find they are all satisfactory (see R code).

    Line 452. A few comments on this muzzle-contact analysis:

    The comments below are a little confusing as some seem to refer to the muzzle-contact rate model (previously line number 452), and some seem to refer to the initiator/receiver model. We have tried to figure out which comments refer to which, and answer accordingly.

    "We investigated muzzle contact behaviour in groups where large proportions of the groups started to extract and eat peanuts over the first four exposures"

    What was the criteria for "a large proportion"?

    All groups are now included in this analysis.

    The text for this muzzle-contact analysis would indicate that this model was not fit with any random effects, which would be extremely concerning. However, having checked the R code which the authors provided, I see that Individual has been fit as a random effect. This should be mentioned in the manuscript. I would also strongly recommend fitting Group (it was an RE in the previous models, oddly) and potentially exposure number as well.

    The model about muzzle contact rate never contained individual as a random effect because individuals are not relevant in this model – it is the number of muzzle contacts occurring during each exposure. However, the reviewer might refer here to the model that we forgot to provide the script for. Nonetheless, we have substantially revised this model, it now (Model 3) includes all groups, and has group as a random effect.

    Following on from this, if the model was fit with individual as a random effect it becomes confusing that Figure 3 which represents this data seemingly does not control for repeated measures (it contains many more datapoints than the study's actual sample size of 164 individuals). This needs to be corrected for this figure to be meaningfully interpretable.

    Figure 3 is not related to the model described in (original) line 452.

    The numbers were referring to the number of muzzle contacts, and this was written in the figure caption. However, we no longer present these details on the new figure (see Fig 4).

    Finally, would it make sense to somehow incorporate the number of individuals present for this analysis? Much like any other social or communicative behaviour, I would predict the frequency of occurrence to depend on how many opportunities (i.e. social partners) there are to engage in it.

    We have included the number of monkeys eating in our muzzle contact rate model now (Model 3) as upon further thought, we found that this was the issue leading us to want to exclude exposures, and only include the groups where many monkeys were eating. We have resolved this now by including all groups and not dropping exposures, and rather we include an interaction between number of monkeys eating and exposure number. We feel this addresses our hypothesis here much more satisfactorily. We hope these updates also address the reviewers concerns adequately.

    Line 460: "For BD and LT we excluded exposures 4 and 3, respectively, due to circumstances resulting in very small proportions of these groups present at these exposures"

    What was the criterion for a satisfactory proportion? Why was this chosen

    See above – this is now addressed.

    Line 461: "We ran the same model including these outlier exposures and present these results in the supplementary material (SM3)."

    The results of this supplemental analysis should be briefly stated. Do they support the original analysis or not?

    We no longer present this like this. We revised the model examining muzzle contact rate substantially and actually included the number of individuals eating in the model rather than excluding groups where this number was low. The results of the new model show good support our hypothesis.

    Line 465: "Due to very low numbers of infants ever being targets of muzzle contacts, we merged the infant and juvenile age categories for this analysis."

    This strikes me as a rather large mistake. The research question being asked by the authors here is "How does age influence muzzle-contact behaviour?"

    Then, when one age group (infants) is very unlikely to be a target of muzzle-contact, the authors have erased this finding by merging them with another age category (juveniles). This really does not make sense, and seriously confounds any interpretation of either age category.

    Yes we agree with this issue, and no longer do that. Rather we remove the infant data from this model, which is now Model 6, because of the large amounts of error they introduced into the model due to the small sample size. We show the process in the R code, and we describe our reasons in the text (lines 713-719). Since we are now only comparing within age- and sex-categories (see below) we do not find this decision introduces any bias.

    Lines 466-474: Why was rank removed for the second and third models? Why is Group no longer a random effect (as in the previous analysis)? The authors need to justify such steps to give the reader confidence in their approach.

    This is now addressed and discussed in descriptions of our new models.

    Furthermore - because of the way this model is designed, I do not think it can actually be used to infer that these groups are preferentially targeted, merely that adult female and adult males are LESS likely to target others than to be targeted themselves, which is a very different assertion.

    Because the specific outcome measure was not described here, this only became apparent to me after inspecting Figure 3, where outcome measure is described as "Probability of (an individual) being a target rather than initiator" - so, it can tell us that adults are more often targeted rather than initiating, but does not tell us if they are targeted more frequently than juveniles (who may get targeted very often, but initiate so often that this ratio is offset).

    We thank the reviewer for noticing this as we had indeed chosen an inappropriate model for what we were intending to measure – this has been addressed now with two additional models (Models 4 and 5; see details at the top of document). We nonetheless found the aspects of this model to still be highly interesting, so have re-framed it to focus on them.

    Lines 467-473: "Our first simple model included individuals' knowledge of the novel food at the time of each muzzle contact (knowledgeable = previously succeeded to extract and eat peanuts; naïve = never previously succeeded to extract and eat peanuts) and age, sex and rank as fixed effects. Individual was included as a random effect. The second model was the same, but we removed rank and added interactions between: knowledge and age; and knowledge and sex. The third model was the same as the second, but we also added a three-way interaction between knowledge, age and sex."

    This is a good example of some of the issues I describe above. What is the justification for each of these model-structures? The addition and subtraction of variables and interactions seems arbitrary to the reader.

    For Model 6, we no longer include rank at all, because we had not hypothetical reason to (see lines 723-725). We now begin with the three-way interaction, and only remove this, because it is not significant, and the model had problems converging as well, due to its complexity. We show this in the R script. We retain only the two separate interactions, and we do not include group as a random effect in this model due to the complexity AND because we do not think there is a theoretical requirement for it to be included here (this is explained in lines 730-735- in the manuscript. We report the results of the 3-way interaction in the supplementary material – SM3 Table S2).

    Reviewer 3

    In this study, the authors introduce a novel food that requires handling time to five vervet monkey groups, some of which had previous experience with the food. Through the natural dispersal of males in the population, they show that dispersing individuals transmit behavioral innovations between groups and are often also innovators. They also examine muzzle contact initiations and targets within the groups as a way to determine who is seeking social information on the new food source and who is the target of information seeking. The authors show that knowledgeable adults are more often the target of muzzle contacts compared to young individuals and those that are not knowledgeable.

    This is a very interesting study that provides some novel insights. The methods employed will be useful to others that are considering an experimental approach to their field research. The data set is good and analyzed appropriately and the conclusions are justified. However, there are several areas where the paper could be improved for readers in terms of its clarity.

    1. It wasn't until the Discussion that it became clear to me that the actual physiological and personality traits of dispersers were being linked with innovation. From the Title, Abstract, and Introduction, it seemed as though the focus was on dispersing males bringing their experience with a novel food to a new group to pass it on. I think it needs to be made clear much earlier in the manuscript that the authors are investigating not only the transmission of behavioural adaptation but also how the traits of dispersers might may make them more likely to innovate.

    We have now addressed this above.

    1. Early in the paper on line 28, the authors state that continued initiation of muzzle contacts by adult females could have been an effort to seek social information. This is true but another interpretation is that females were imparting or giving social information. It seems important here and elsewhere (lines 322-323) to consider and report the target of these initiations. If these were directed at more knowledgeable individuals, it supports the idea that this was social information seeking. If muzzle contacts were directed to younger or unknowledgeable individuals, it would imply a form of teaching, which is possible but perhaps unlikely, so I think the authors need to be totally clear here.

    We thank the reviewer for pointing this out We looked into our data and now present figure 4D, showing that almost all knowledgeable adult females’ muzzle contacts were targeted towards knowledgeable adult males and talk about it in the discussion (lines 499-500).

    1. The argument made on lines 344-350 needs more fleshing out to be convincing or it should be deleted. The link between number of dispersers, social organization, and large geographic range seems a little muddled. There are many dispersing individuals in species that are not typically in large multi-male, multi-female social organizations. Indeed, in many species both sexes disperse. Think of pair living birds where both sexes disperse and geographic range can be enormous. There are also no data or references presented here to show that species in multi-male, multi-female social organizations do have larger geographic ranges than those that are not in these social organizations. It seems to me that, even if this is the case, niche is more important than social organization, for instance not being dependent on forests to constrain much of your range.

    We have removed this section

  4. eLife

    Evaluation Summary:

    Dongre et al. build on previous social learning research on wild vervet monkeys to investigate the role of a particular social behaviour, muzzle-muzzle contact, in aiding the acceptance of a novel food and provide interesting observations on the potential for male monkeys immigrating from one social group to another in spreading this novel behaviour. With a more robust and transparent analysis, this manuscript has the potential to provide significant insights into innovation and social learning in animals.

    This preprint has been reviewed by eLife. We include the public reviews from the reviewers here; the authors also receive private feedback with suggested changes to the manuscript. Reviewer #3 agreed to share their name with the authors.

  5. eLife

    Reviewer #1 (Public Review)

    Strengths:

    This manuscript combines experimental, exploratory, and observational methods to investigate the big question in innovation literature--why do some animals innovate over others, and how information about innovations spread. By combining a variety of methods, the manuscript tackles this question in a number of ways, and finds support for previous work showing that animals can learn about foods via social olfactory inspection (i.e., muzzle to muzzle contact), and also presents data intended to investigate the role of dispersing animals in innovation and information spread.

    Using data from a previously-published experiment, the manuscript illustrates how investigators can numerous interesting questions while limiting the disturbances to wild animals. The manuscript's attempt at using exploratory analysis is also exciting, as exploratory analyses provide a useful tool for behavior research-indeed, Tinbergen insisted that behavior must first be described.

    Weaknesses:

    The manuscript's introduction is a bit unclear as to how the fact that dispersing males may be an important source of information ties to innovations in response to disruptions due to climate change, humans, or new predators, if at all. An introduction regarding the role of dispersed animals in introducing novel behaviors and social transmission would better prepare readers for the questions presented in the manuscript. As it stands now, the manuscript only provides one sentence discussing the theoretical relevance of investigating the role of dispersing animals in innovations.

    Additionally, while the manuscript attempts to use exploratory analysis, it does not provide enough theoretical background as to why certain questions were asked while the data were explored. While the discussion provides some background as to the role of dispersing males in innovation, the introduction provides little background, and thus does not properly frame the issue. It is unclear how dispersing males became of interest and why readers should be interested in them. As the manuscript reads now, it may be that dispersing males became interesting only as a result of the exploratory analysis-except that the predictions explicitly mentions dispersing males. Thus, manuscript at present makes it difficult to know if the questions surrounding immigrant males resulted from the exploratory analysis, or was a question the analyses were intended to answer from the beginning. If this question only came out after first reviewing the results, then this needs to be made clear in the introduction. I see no issue with reporting observations that were the result of investigations into earlier results, but it needs to be reported in a way that can be replicated in future research-I need to know the decision process that took place during the data exploration.

    The manuscript never clearly defines what counts as an immigrant male; presumably, in this species, all adult males in the group should be immigrants, as females are the philopatric sex. Sometimes, the manuscript uses "recently" to modify immigrant males, but doesn't define exactly what counts as recent, except to say that the males that innovated were in their respective groups for fewer than 3 months, but never explains why three months should be an important distinction in adult male tenure.

    Due to the above weaknesses, the provided predictions are a bit murky. It is not clear how variation between groups in accordance with who innovated, or initiated eating a novel food, or demographics is related to the central issue. The manuscript does contribute to the literature by looking at changing rates of muzzle contact over exposure to a novel food source, and provides a good extension of previous findings; that, if muzzle contacts help animals learn about new foods, then rates of muzzle contacts involving novel foods should decrease as animals become familiar with the food. However, this point isn't explicit in the manuscript. Finally, it is also unclear as to why changing rates of muzzle contact AND whether certain individual-level variables like knowledge, sex, age, and/or rank might influence muzzle contacts during opportunities to innovate.

    As for the methods, the manuscript doesn't provide enough details as to why certain decisions were made. For example, no reason is given as to why only the first four sessions after an animal ate were considered, why the first three months of tenure (but not four, as seen on one group that didn't innovate) was considered to be a critical time for which immigrant males may innovate, why (including the theoretical reasons) the structure of models for one analysis was changed (dropping one variable, adding interactions), or even how the beginning and ending of a trial was decided, despite reporting that durations varied widely,-from 5 minutes to two hours.

    The discussion contains results that are never elsewhere presented in the manuscript- (2a) Individual variation in uptake of a novel food according to who ate first).

    Finally, the largest issue with the manuscript is that its results are not as convincing as the conclusions made. An issue with all the analyses is that some grouping variables in some analyses but not others despite the fact that all of the analyses contain multiple groups (necessitating group as a grouping variable) and multiple observations of the same individuals (i.e., immigrant males tested in multiple groups, necessitating animal identity as a random effect), and not accounting for individual exposure to the experiment when considering whether animals ate the food in the allotted period (an important consideration given the massive differences in trial times), making these results difficult to interpret in their current forms. As for the results regarding muzzle contact, the analyses has a number of issues that make it difficult to determine if the claims are supported. These issues include not explaining why rank calculated a year before the experiments took place was valid or if rank was calculated among all group members or within age and sex classes, not explaining how rank was normalized, and not conducting any kind of formal model comparisons before deciding the best model.

    As for the results regarding immigrant males and innovation, little is done to help the fact that these results are from very few observations and no direct analyses. It is possible that something that occurs relatively often but in small sample sizes, like dispersing animals, could have immense power in influencing foraging traditions, and observation is a necessary step in understanding behavior. However, the manuscript doesn't consider any alternative hypotheses as to why it found what it found. No other possible difference between the groups was considered (for example, the groups that rapidly innovated appear to be quite smaller than the groups that did), making the claim that immigrant males were what allowed groups to innovate unconvincing. This is particularly true given that some groups in this study population have experimental histories (though this goes unmentioned in the current manuscript), which likely influenced neophobia-especially given work by the same research group showing that these animals are more curious compared to their unhabituated counterparts.

  6. eLife

    Reviewer #2 (Public Review)

    I have separated my issues with the manuscript into three sub-headings (Conceptual Clarity, Observational Detail and Analysis) below.

    1. Conceptual clarity

    There are a number of areas where it would greatly benefit the manuscript if the authors were to revisit the text and be more specific in their intentions. At present, the research questions are not always well-defined, making it difficult to determine what the data is intended to communicate. I am confident all of these issues could be fixed with relatively minor changes to the manuscript.

    For example, Line 104: Question 1 is not really a question, the authors only state that they will "investigate innovation and extraction of eating the food", which could mean almost anything.

    Question 2a (line 98) is also very vague in it's wording, and I'm left unclear as to what the authors were really interested in or why. This is not helped by Line 104 which refuses to make predictions about this research question because it is "exploratory". Empirical predictions are not simply placing a bet on what we think the results of the study will be, but rather laying out how the results could be for the benefit of the reader. For instance, if testing the effects of 10 different teaching methods on language acquisition-rate: Even if we have no a priori idea of which method will be most effective, we can nevertheless generate competing hypotheses and describe their corresponding predictions. This is a helpful way to justify and set expectations for the specific parameters that will be examined by the methods of the study. In fact, in the current paper, the authors in fact had some very clear a priori expectations going into this study that immigrant males would be vectors of behavioural transmission (clear that is from the rest of the introduction, and the parameters used in their analysis, which were not chosen at random).

    The multiple references to 'long-lived' species in the abstract (line 16 and introduction (39, 56) is a bit confusing given the focus of this study. Although such categorisations are arbitrary by nature (a vervet is certainly long-lived compared to a dragonfly), I would not typically put vervet monkeys (or marmosets, line 62) in the same category as apes (references 8 and 9) or humans (line 62) in this regard. This contributes a little towards the lack of overall conceptual focus for the manuscript: beginning in this fashion suggests the authors are building a "comparative evolutionary origins" story, hinting perhaps at the phylogenetic relevance of the work to understanding human behaviour, but the final paragraph of the study contextualises the findings only in terms of their relevance to feeding ecology and conservation efforts. I would recommend that the authors think carefully about their intended audience and tailor the text accordingly. This is not to say that readers interested in human evolution will not be interested in conservation efforts, but rather that each of these aspects should be represented in each stage of the manuscript (otherwise - conservationists may not read far into the Introduction, and cultural evolution fans will be left adrift in the Conclusion).

    1. Observational detail

    There are a number of areas of the manuscript which I found to be lacking in sufficient detail to accurately determine what occurred in these experimental sessions, making the data difficult to interpret overall. All of this additional information ought to be readily available from the methods used (the experiments were observed by 3-5 researchers with video cameras (line 341)) and is all of direct relevance to the research questions set out by the authors.

    While I appreciate that it will take quite a bit of work to extract this information, I am certain that it would greatly improve the robustness and explanatory power of this study to do so.

    The data on who was first to innovate/demonstrate successful extraction of the food in each group (Question 1) and subsequent uptake (Question 2), as well as the actual mechanism by which that uptake occurred (the authors strongly imply social learning in their Discussion, but this is never directly examined) is difficult to interpret based on the information presented. Some key gaps in the story were:

    - Which/how many individuals encountered the food and in what order? I.e., were migrants/innovators simply the first to notice the food?
    - Did any individuals try and fail to extract the food before an "innovator" successfully demonstrated?
    - How many tried and failed to extract the nuts before and after observing effective demonstrators?
    - Were individuals who observed others interact with the food more likely to approach and/or extract it themselves?
    - Did group-members use the same methods of extraction as their 'innovators'?
    - How many tried and succeeded without having directly observed another individual do so (i.e. 'reinvention' as per Tennie et al.)?

    The connective tissue between the research questions set out by the authors is clearly social learning. In short: the thesis is that Migrants/Innovators bring a novel behaviour to the group, then there is 'uptake' (social learning), which may be influenced by demographic factors and muzzle-contact (biases + mechanisms). Given this focus (e.g. lines 224-264 of the Discussion), I would expect at least some of the details above to be addressed in order to provide robust support for these claims.

    Question 2a (Lines 136-146): This data is hard to interpret without knowing how much of the group was present and visible during these exposures.

    For example: 9% update in NH group does not sound impressive, but if only 10% of the total group were present while the rest were elsewhere, then this is 90% of all present individuals. Meanwhile if 100% of BD group were present and only experienced 31% uptake, then this is quite a striking difference between groups.

    Of course, there is also an issue of how many individuals can physically engage with the novel food even if they want to - the presence of dominant individuals, steepness of hierarchy within that group, etc, will significantly influence this (and is all of interest with regards to the authors' research questions).

    Muzzle-contact behaviour: The authors use their data to implicate muzzle-contact in social learning, but this seems a leap from the data presented (some more on this in the Analysis section).

    For example:
    - What is the role of kinship in these events?
    - Did they occur when the juvenile had free access to the food (i.e. not likely to be chased off by a feeding adult)?
    - Did they primarily occur when adults had a mouthful of food? (i.e. could it simply be attempted pilfering/begging)
    - What proportion of PRESENT (not total) individuals were naïve and knowledgeable in each group for each trial (if 90% present were knowledgeable, then it is not surprising that they would be targeted more often)?
    - Did these events ever lead to food-sharing (In other words, how likely are they to simply be begging events)?
    - Did muzzle-contact quantifiably LEAD to successful extraction of the food? If the authors wish to implicate muzzle-contact in social learning, it is not sufficient to show that naïve individuals were more likely to make muzzle-contact, they must also show that naïve individuals who made more muzzle-contact were more likely to learn the target behaviour.

    1. Analysis

    There are a number of issues with the current analysis which I strongly recommend be addressed before publication. Some of these are likely to simply require additional details inserted to the manuscript, whereas others would require more substantial changes. I begin with two general points (A & B), before addressing specific sections of the manuscript.

    A) My primary issue with each of the analyses in this manuscript is that the authors have fit complex statistical models for each of their analyses with no steps to ascertain whether these models are a good fit for the data. With a relatively small dataset and a very large number of fixed effects and interactions, there is a considerable risk of overfitting. This is likely to be especially problematic when predictor variables are likely to be intercorrelated (age, sex and rank in the case of this analysis).

    The most straightforward way to resolve this issue is to take a model-comparison approach. Fitting either a) a full suite of models (including a 'null' model) with each possible permutation of fixed effects and interactions (since the authors argue their analysis is exploratory) or b) a smaller set of models which the authors find plausible based on their a priori understanding of the study system. These models could then be compared using information criterion to determine which structure provides the best out-of-sample predictive fit for the data, and the outputs of this model interpreted. Alternatively, a model-averaging approach can be taken, where the effects of each individual predictor are averaged and weighted across all models in the set. Both of these approaches can be performed easily using the r package 'MuMIn'. There are also a number of tutorials that can be found online for understanding and carrying out these approaches.

    B) It does not seem that interobserver reliability testing was carried out on any of the data used in these analyses. This is a major oversight which should be addressed before publication (or indeed any re-analysis of the data).

    Line 444: Much more detail is needed here. What, precisely, was the outcome measure? Was collinearity of predictors assessed? (I would expect Age + Rank to be correlated, as well as Sex + Rank).

    Line 452. A few comments on this muzzle-contact analysis:

    "We investigated muzzle contact behaviour in groups where large proportions of the
    groups started to extract and eat peanuts over the first four exposures"

    What was the criteria for "a large proportion"?

    The text for this muzzle-contact analysis would indicate that this model was not fit with any random effects, which would be extremely concerning. However, having checked the R code which the authors provided, I see that Individual has been fit as a random effect. This should be mentioned in the manuscript. I would also strongly recommend fitting Group (it was an RE in the previous models, oddly) and potentially exposure number as well.

    Following on from this, if the model was fit with individual as a random effect it becomes confusing that Figure 3 which represents this data seemingly does not control for repeated measures (it contains many more datapoints than the study's actual sample size of 164 individuals). This needs to be corrected for this figure to be meaningfully interpretable.

    Finally, would it make sense to somehow incorporate the number of individuals present for this analysis? Much like any other social or communicative behaviour, I would predict the frequency of occurrence to depend on how many opportunities (i.e. social partners) there are to engage in it.

    Line 460: "For BD and LT we excluded exposures 4 and 3, respectively, due to circumstances resulting in very small proportions of these groups present at these exposures"

    What was the criterion for a satisfactory proportion? Why was this chosen?

    Line 461: "We ran the same model including these outlier exposures and present these results in the supplementary material (SM3)."

    The results of this supplemental analysis should be briefly stated. Do they support the original analysis or not?

    Line 465: "Due to very low numbers of infants ever being targets of muzzle contacts, we merged the infant and juvenile age categories for this analysis."

    This strikes me as a rather large mistake. The research question being asked by the authors here is "How does age influence muzzle-contact behaviour?"
    Then, when one age group (infants) is very unlikely to be a target of muzzle-contact, the authors have erased this finding by merging them with another age category (juveniles). This really does not make sense, and seriously confounds any interpretation of either age category.

    Lines 466-474: Why was rank removed for the second and third models? Why is Group no longer a random effect (as in the previous analysis)? The authors need to justify such steps to give the reader confidence in their approach.

    Furthermore - because of the way this model is designed, I do not think it can actually be used to infer that these groups are preferentially targeted, merely that adult female and adult males are LESS likely to target others than to be targeted themselves, which is a very different assertion.

    Because the specific outcome measure was not described here, this only became apparent to me after inspecting Figure 3, where outcome measure is described as "Probability of (an individual) being a target rather than initiator" - so, it can tell us that adults are more often targeted rather than initiating, but does not tell us if they are targeted more frequently than juveniles (who may get targeted very often, but initiate so often that this ratio is offset).

    Lines 467-473: "Our first simple model included individuals' knowledge of the novel food at the time of each muzzle contact (knowledgeable = previously succeeded to extract and eat peanuts; naïve = never previously succeeded to extract and eat peanuts) and age, sex and rank as fixed effects. Individual was included as a random effect. The second model was the same, but we removed rank and added interactions between: knowledge and age; and knowledge and sex. The third model was the same as the second, but we also added a three-way interaction between knowledge, age and sex."

    This is a good example of some of the issues I describe above. What is the justification for each of these model-structures? The addition and subtraction of variables and interactions seems arbitrary to the reader.

  7. eLife

    Reviewer #3 (Public Review)

    In this study, the authors introduce a novel food that requires handling time to five vervet monkey groups, some of which had previous experience with the food. Through the natural dispersal of males in the population, they show that dispersing individuals transmit behavioral innovations between groups and are often also innovators. They also examine muzzle contact initiations and targets within the groups as a way to determine who is seeking social information on the new food source and who is the target of information seeking. The authors show that knowledgeable adults are more often the target of muzzle contacts compared to young individuals and those that are not knowledgeable.

    This is a very interesting study that provides some novel insights. The methods employed will be useful to others that are considering an experimental approach to their field research. The data set is good and analyzed appropriately and the conclusions are justified. However, there are several areas where the paper could be improved for readers in terms of its clarity.

    1. It wasn't until the Discussion that it became clear to me that the actual physiological and personality traits of dispersers were being linked with innovation. From the Title, Abstract, and Introduction, it seemed as though the focus was on dispersing males bringing their experience with a novel food to a new group to pass it on. I think it needs to be made clear much earlier in the manuscript that the authors are investigating not only the transmission of behavioural adaptation but also how the traits of dispersers might may make them more likely to innovate.

    2. Early in the paper on line 28, the authors state that continued initiation of muzzle contacts by adult females could have been an effort to seek social information. This is true but another interpretation is that females were imparting or giving social information. It seems important here and elsewhere (lines 322-323) to consider and report the target of these initiations. If these were directed at more knowledgeable individuals, it supports the idea that this was social information seeking. If muzzle contacts were directed to younger or unknowledgeable individuals, it would imply a form of teaching, which is possible but perhaps unlikely, so I think the authors need to be totally clear here.

    3. The argument made on lines 344-350 needs more fleshing out to be convincing or it should be deleted. The link between number of dispersers, social organization, and large geographic range seems a little muddled. There are many dispersing individuals in species that are not typically in large multi-male, multi-female social organizations. Indeed, in many species both sexes disperse. Think of pair living birds where both sexes disperse and geographic range can be enormous. There are also no data or references presented here to show that species in multi-male, multi-female social organizations do have larger geographic ranges than those that are not in these social organizations. It seems to me that, even if this is the case, niche is more important than social organization, for instance not being dependent on forests to constrain much of your range.

  8. Version published to 10.1101/2021.12.16.472640v2 on bioRxiv
  9. Version published to 10.1101/2021.12.16.472640v1 on bioRxiv