All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
After reviewing this revised version of your manuscript, I see that the main comments suggested by the reviewers have been included. Therefore, I am satisfied with the current version and consider it ready for publication.
[# PeerJ Staff Note - this decision was reviewed and approved by Nigel Andrew, a PeerJ Section Editor covering this Section #]
The authors have done a great job of responding to reviewer feedback, and the manuscript is much improved. The Introduction now reads much more clearly and cogently, and the justification for the methods now is quite clear. This work will make a great contribution for PeerJ.
No comment.
No comment.
After reviewing this revised version of your manuscript, I see that the main comments suggested by the reviewers have been included, while the suggestions not considered are justified in detail. However, there are still some details that need to be clarified before having a final version that can be published. The reviewer's comments focus mainly on the background of the introduction and the experimental design.
The authors have done a commendable job of improving the manuscript’s focus and adjusting the analyses in response to reviewer concerns. That being said, (1) there are still a number of passages throughout the manuscript that provide an overwhelming picture of past research or are otherwise unclear and (2) the analytical approaches could use more explicit justification to convince readers that this is the best approach for this task, even if the manuscript’s goals are purely exploratory.
To the first point above, while the Introduction now does a more thorough job of introducing the study, it still requires readers to juggle a great deal of information and contrasts (e.g., structural vs. functional, quantity vs. quality, affiliative vs. aggressive, strong bonds vs. weak bonds, uniplex vs. multiplex, etc.). As it stands, I think the sum of this information is rather overwhelming and (given that these contrasts are not always clear cut) sometimes ambiguous. The Introduction would be clearer if it focused solely on contrasts that (i) are directly relevant to the analysis at hand and/or (ii) appear throughout the manuscript (e.g., uniplex vs. multiplex, quality vs. quantity). For example, given that the manuscript does not explicitly delineate strong vs. weak bonds, devoting so much “real estate” to describing how previous studies have measured these bonds, how their effects on health and fitness are broadly equivocal, and how using these metrics inherently excludes potentially valuable information (L135-165) – although largely valid – is less relevant to the broader approach and framework used in the manuscript. Altogether, the Introduction could use some editing for improved brevity and focus.
To the second point, the justification for the analytical approach could be a bit more explicit. While the authors explain that they limited their comparisons to the 110 of the 630 possible combinations of the multiplex and uniplex predictors, it is unclear why those 520 other permutations were discarded from these comparisons. In other words, why were models using uniplex grooming metrics concurrently with multiplex contact sitting metrics not included within the larger final model set? While I understand that reducing the total model set is a good aim, I’m unsure of the quantitative rationale behind this.
Additionally, the authors use a suite of social network metrics to capture distinct facets of sociality (i.e., frequency, social capital, embeddedness, etc.) and measure their impacts on the immune markers of interest. This seems well-suited to the exploratory nature of the analysis, but I would have appreciated seeing these concepts introduced earlier in the Introduction and some brief descriptions of how these components might differentially impact health/fitness. I imagine there would be sufficient room for this if some of the extraneous information listed above was abbreviated.
Likewise, the Discussion does not explicitly address that their results indicate that embeddedness (and not frequency, social capital, etc.) is the strongest predictor of immune function across the model set. What implications does this have for the function of these social measures and the strategies that individuals might use to gain these social positions? Some more context and interpretation would be useful here, especially since the Introduction sets up the idea that different facets of sociality are likely functionally distinct.
No comment.
L60-63: This sentence is a bit awkwardly worded. Consider editing for clarity.
L71-92: See comment above. I’m not sure whether the structural vs. functional distinction adds much here, especially when the functional is not as easily measured in animals as it is in humans (unless you are focused on a behavior that has a specific, unambiguous functional consequence; e.g., grooming leading to reduced parasite counts, huddling leading to thermoregulatory benefits).
L100-103: I would veer away from focusing on aggression, since those behaviors were not addressed within the multiplex approach used here.
L106-108: This is where the distinction between structural vs. functional and quality vs. quantity gets a little blurred.
L118-134: This summation I think elides a couple conceptual issues. Some studies use multiple sociality metrics as predictors of some outcome, but it’s unclear which is the most meaningful given tight correlations between these measures. Other studies focus on one metric based on some pre hoc justification and therefore cannot assess the impact of the sociality metrics they did not use.
L128-130: If there are concerns about using betweenness given its correlation with degree, then what was the justification for using this metric within the present study? Were there high correlations among metrics within the dataset? Also, is there a citation referencing this strong correlation and its methodological consequences?
L141: What do you mean by main-effects model? I know I’ve seen this verbiage elsewhere, but I’m not sure whether it’s necessary in this context.
L135-165: As above, I’m not convinced that this is particularly relevant, as the present study does not delineate weak and strong bonds. Perhaps a brief note that multiplex relationships and uniplex relationships may – albeit using a wholly different quantitative procedure – may capture strongly and weakly bonded dyads?
L178-180: Is that true on the whole? Off the top of my head, I can’t think of any studies that have explicitly compared the influence of single affiliative behavioral indices vs. composite indices on any fitness proxy.
L216: It might be clearer to replace “reflect” with “engaged in.”
L230-232: It might make more sense if both 1) and 2) started with a verb. I.e., determine if the network…, link the relationship between, or something to that effect.
L309-315: As noted above, indicating what the functional significance of these metrics might be could be helpful for readers here.
L416-418: Are the uniplex networks simply too sparse for the modularity algorithms to detect clustering?
L451-452: This is oddly worded and makes it sound like the “all grooming” and “all contact sitting” networks are multiplex or uniplex
L471-473: Given that these behaviors are highly correlated, however, would it ever be likely that the functional significance of just one behavior is wholly driving the benefit?
L483-484: I would disagree that altering endocrine function is inherently a fitness benefit.
L504-507: I’m surprised that the top model set was (relatively) so unambiguous, given the fact that the predictors were so highly correlated.
Thank you very much for your manuscript titled “Differential effects of multiplex and uniplex affiliative relationships on biomarkers of inflammation” that you sent to PeerJ.
This study explores possible differences in the relationship between network centrality and markers of inflammation when considering unidimensional or multidimensional social networks. They also explored possible structural differences between unidimensional and multidimensional social networks that may help understand how network centrality influences markers of inflammation.
As you will see below, comments from two referees suggest a major revision before your paper can be published. Their comments should provide a clear idea for you to review, hopefully improving the clarity and rigor of the presentation of your work. I will be happy to accept your article pending further revisions, detailed by the referees, which largely focus on clarifying various aspects of your work.
Reviewer 1 In addition to pointing out different minor observations within particular lines of the text, it also considers that some discrepancies between the main objective of the study and the experimental design implemented should be clarified, including some points in the statistics used.
Reviewer 2 also points out some sentences that can be rewritten to improve the reading of the manuscript and also highlights greater clarity in the analytical procedures, their justification and the resulting interpretations.
Please note that we consider these revisions to be important and your revised manuscript will likely need to be revised again.
The authors aimed to explore whether there are differences in the relationship between network centrality and inflammation markers when unidimensional social networks (either grooming or huddling) or multidimensional (grooming and huddling) are considered. In addition, they also explored if there are structural differences between unidimensional and multidimensional social networks that may help understand how network centrality influences inflammation markers.
The article provides an interesting perspective on why it might be important to consider different types of social interactions when assessing the function of affiliative relationships. However, there are some major limitations in the set-up, but more importantly, in the methodology that refrain me from recommending the article for publication in its current state.
The set-up of the study is quite clear up to the point where the authors explained their aim and methodology (from Line 150 onward). The key aim of the paper, based on the title and Introduction, is to disentangle the differential effects of multidimensional affiliative relationships (considers two types of interactions) versus unidimensional affiliative relationships (considers a single type of interaction) on two markers of inflammation. To do so, the authors decided to construct uniplex (unidimensional) and multiplex (multidimensional) social networks based on two social interactions: grooming and contact sitting. Based on the aim, I would think that the authors would end up with 3 social networks per group: one for grooming interactions (uniplex), one for contact sitting (uniplex) and one for both grooming and contact sitting (multiplex). However, the authors constructed 6 social networks per group (uniplex grooming, uniplex contact sitting, multiplex grooming, multiplex contact sitting, all grooming and all huddling) without explaining clearly:
1) why are there two multiplex networks if the idea is to pool both types of interactions (grooming and contact sitting) together?
2) how were the networks constructed for both multiplex networks and what is their difference? (I see some vague explanations in Lines 155-159 but this does not explain what an edge represents in the multiplex grooming network and the multiplex contact sitting network)
3) what are the “all grooming” and “all contact sitting” networks and why were they incorporated in the study? (how they were constructed and the rationale for inclusion were not explained at all in the Introduction)
Methodology:
1) There are some important limitations in the sampling effort that the authors should try to account for and acknowledge throughout the manuscript in the behavioural data and the inflammation markers. First, the authors sampled each group for a 6-weeks period and collected the data on social interactions when most studies often aggregate data for a full year when building social networks for non-human primates in semi-wild/wild conditions. Even more, the authors used scan sampling, which might biased the estimation of how well-connected individuals are towards those that are bolder or more easily observable. The authors could have overcome these limitations by accounting for sampling effort when constructing the social networks. That is, edges between dyads should represent the number of either grooming or contact sitting interactions observed divided by the total number of scans for both members of the dyad (Silk et al. 2013). Second, the authors sampled each animal only once for cytokine levels which might reflect chronic inflammation, as the authors argued, but could also reflect acute immune activation (for instance in response to injury, parasite infection or some other acute stressor). To conclusively state that individuals are in poorer health because of chronic inflammation the authors would need repeated samples showing that the inflammatory profile is consistent through time. I think is necessary for the authors to acknowledge this limitation when discussing their results.
2) Another important part of the paper is the comparison of structural properties, such as kin biases, grooming up to the hierarchy, rank disparity among others between uniplex and multiplex social networks to have an idea of the function of uni versus multidimensional interactions. To do so, and acknowledging their sample size limitation, the authors used paired t-tests to assess statistically these differences. Missing from this part is an explanation of how each of these structural measures were computed (only very brief explanations were provided between lines 286-295), values for the metrics for each group (at least median/average and SD) and the statistic parameters rather than a Table only summarising these results. Again, I believe this is another point the authors should acknowledge as a limitation when discussing their results. There might be group differences, that were not accounted for, that could be driving the differences between uniplex and multiplex networks.
3) There are also issues with how the authors analysed the network data. From what is stated in Lines 312-313 and 315, it seems that the authors excluded from the statistical analyses those individuals who were not observed engaging in the social interaction of interest (grooming or contact sitting). This is a very odd procedure because those individuals can indeed provide very valuable data on how social isolation (the other end of social integration) might relate to inflammation markers. The authors need to explain clearly, why those individuals were excluded and back up their approach with references (if any).
4) The most problematic step from my perspective is the model selection, which I struggled to understand even with the inclusion of the diagram listing the steps in Figure S3. From what I gathered, the authors picked many (if not all) network metrics and used model selection to pick the relevant variables that explained inflammation markers (as dependent variables). At this point, I would have appreciated an explanation for why the authors deemed it relevant to test for all those network metrics rather than focusing on a few for which there is evidence of an effect on health/inflammation. There are several studies (e.g. Ellis et al. 2019, Schulke et al. 2022) trying to disentangle the different functions that each network metric might represent, so I do not understand why the authors decided to dismiss all that information when one of the key parts of their paper is to explore the functional role of affiliative relationships.
5) Also related to the model selection part. The authors first selected as null a model that included only a random effect for group ID. Then, if I understood correctly, they ran univariate models for each of the network metrics (from each social network) to come up with a new null model (which they termed the “Overall Best model from steps 3 and 4” Supplemental Table 3 and Table 4). If the aim of the paper from the beginning was to compare uniplex versus multiplex network metrics why steps 1, 2 and 3 were necessary in the model selection process? Concerning the same, what conclusion about similarities and differences between uniplex and multiplex relationships can the authors really draw if they are contrasting different network metrics, with potentially different functions? (e.g., comparing strength from uniplex network with clustering from multiplex networks)
No comment
Minor comments:
Lines 29-33: I would rephrase this whole sentence. The aim is not particularly clear and it would benefit from a more concise explanation. Maybe splitting the aim into two sentences and also providing a bit more information on the other aspects to be explored (kin biases, modularity, hierarchies?), so it does not come as a surprise when reporting the results later on in the abstract.
Lines 37-28: “…which were more strongly linked with social status” This is not clear, what does it mean? That individuals more connected in the uniplex network were often high-status animals? If so, I would say that more explicitly.
Lines 38 to 41: I’m wondering if these results, instead, suggest then that huddling rather than the "multiplex relationships" promotes health? Given that grooming by itself does not?
Lines 47-49: I find this sentence difficult to follow, the structure does not seem correct.
Line 53: “tenor” I'm not familiar with this term in this context. Could the authors maybe use another word or provide a brief definition?
Lines 79-85: I am not convinced by this argument. The articles cited by the authors have explored specific network metrics and drawn conclusions on the functions based on that. Moreover, the study of Ellis et al. 2019 specifically tackles what the authors argue. That is, comparing the effect of different social networks on lifespan.
Line 88: I'm not entirely clear why there is a special emphasis on betweenness, among all the possible network metrics that have been studied. Is it related to the specific aims of the study? If not, I would keep this part more general or explain why betweenness is/has been particularly important when exploring the links between social relationships and health/fitness.
Additionally, this paragraph would benefit from brief definitions of the network metrics mentioned as not all readers might be familiar with social network analyses (I see that the authors have network metric definitions in Table 2, so might be useful to reference it earlier in the MS).
Line 93: Is the “quality of affiliative social relationships” equivalent to the functional aspect mentioned earlier in the MS? If so, I would try to make the link with the general message (dichotomy between functional and structural aspects) more clearly.
Line 99: acute hormonal response in which direction?
Line 102: The function of strong bonds was not mentioned before introducing the function of weak bonds.
Lines 108-112: The argument in this sentence is key for understanding the whole manuscript. However, the authors do not make it clear in which regard affiliative relationships based on grooming versus those based on grooming and proximity (for this specific example) might differ. Is it in their function? If so, how?
Lines 115-116: “…to disentangle the impact of structural and functional social relationships” could the authors expand on this part and explain how comparing multiplex to uniplex networks allows them to do so?
Lines 123-125: The first and second points are not clear. What do the authors mean by "the nature" of those relationships?
For the second point, if I got it right, the authors meant that the potential function can be understood from the structure, which seems a bit contradictory to the argument that has been laid out in the introduction.
Lines 132-134: Going back to my point in the Abstract: did this study (Balasubramanian et al. 2016) show that huddling alone did not explain the relationship with parasite infection?
If the authors want to make the argument that considering multiple types of social interactions might provide different information than exploring single interactions, they first need to rule out that each social interaction considered does not explain the pattern observed when pooled together.
Line 136: I would be more specific and mention the behaviours explored (grooming and contact sitting/huddling)
Line 147-148: Is contact sitting the same as huddling? Why “similar” instead of “equivalent”?
Lines 148-149: could the authors explain better what do they mean by social buffering? Social buffering of stress? buffering of infections?
Unlike their explanation for grooming where its transactional function is quite clear, for huddling the "social buffering" needs a bit more explanation (how a huddling social partner can buffer infection/stress)
Lines 189-192: This is repeating the statement in the introduction
Line 196: I'm not sure I understand what the authors mean with a matrilineal social group.
Lines 215-216: specify which behaviour had priority (grooming or contact sitting)
Lines 218-219: The study cited does not demonstrate that the sampling effort was sufficient.
To demonstrate this, simulations would be required. Maybe provide some more information on this study, like what was the mean sampling effort per female per group? I would think the sampling effort was probably not that high to capture all the dyadic interactions in the group but this can be acknowledged and discussed later on.
Lines 222-224: Maybe the explanation about why dominance rank was relevant to explore should come earlier than the data collection protocol for aggressive behaviours.
This manuscript presents new data linking social network metrics with inflammatory immune markers in a captive population of highly social primates. Individuals that were well-connected in ‘multiplex’ grooming networks (where only dyads that also engaged in physical contact retained their edges) showed lower levels of inflammation, while individuals that were well-connected in the ‘uniplex’ grooming network (where only edges that engaged in grooming alone were retained) showed higher levels of inflammation. The manuscript is generally well-written, although there are a few sentences where rewording for clarity would be helpful for readers.
MINOR COMMENTS
Lines 73-76: This sentence is a bit clunky - consider rewording.
Line 83: What does “social role” mean here?
Lines 87-91: Consider rewording to get to the point more quickly. Also, readers may not be immediately familiar with social network metrics (some of which are defined later on in Table 2).
Lines 126-128: This is vague. What does “the nature of those social relationships” entail?
Line 154: “represent” might read more cleanly as “engaged in”
Lines 204-206: This was unclear on first read – each group was sampled for one six-week period each, correct? Were groups resampled over successive years?
Line 358-360: This line is somewhat misleading. The results don’t differ between the two methods simply because directionality and reciprocity are not discernible from an undirected network.
Line 394-396: Switching “animals” for “dyads” might be less confusing to the reader.
Lines 406-409: This sentence reads to me as odd. Consider rewording.
By dividing networks into uniplex and multiplex subnetworks and linking these measures with inflammatory markers, the authors suggest a more nuanced link between sociality and health in primates (and by extension other social mammals). However, I have some reservations about the analytical procedures, their justification, and the resulting interpretations.
In short, I’m not entirely clear on the biological meaning of the social network filtering procedures, where multiplex networks were composed of ‘edges’ that both groomed and contact sat. It’s not simply that contact sitting in addition to grooming suddenly “switches on” the benefits of grooming. Rather, dyads that both groom and contact sit tend to comprise stronger, more stable relationships, which in turn may generate more health benefits than relationships that affiliate less frequently, less dependably, and using fewer types of affiliative behavior. In practice, this is simply an analytical approach, no different than using a cutoff to characterize strong vs. weak bonds or implementing other network regularization procedures. I would have appreciated more justification as to why this procedure is comparable or even preferable to other approaches (i.e., dichotomizing “strong” vs. “weak” bonds).
MINOR COMMENTS
Line 116: How do multiplex networks better capture the structural and the functional? This is not clear to me.
Lines 128-134: Is this necessarily a 1:1 match with the logic underlying multiplex networks though? I.e., this analysis detected an interaction effect at the individual-level, whereas multiplex networks are filtered depending on affiliative tendencies at the dyadic-level.
Line 162-163: I think it could be clearer that multiplex and uniplex ties – in a way – map onto predictions regarding the effects of ‘strong’ and ‘weak’ bonds.
Line 228-231: This procedure means that multiplex grooming and multiplex contact networks are structurally identical except for their edge weights – correct? If so, then some network-level measures of cohesion should be effectively identical.
Line 234: Were these adjusted for sampling effort? In theory, the number of scans for each individual could be the same for all members, but did the denominator ever differ (i.e., if individuals were out of sight during a scan sample)?
Line 311-313: What does this mean in practice? I.e., did that female groom all of the individuals she also ‘contact sat’ with? If so, then why not include her as an isolated individual in the uniplex network? I’m not sure I understand this rationale.
Line 327-328: How would the results differ if all terms that did not generate improved model fits were removed from Step 4? I know conclusions often vary depending on the model winnowing strategy used, so more justification would be reassuring to the reader.
Line 328-331: It wasn’t immediately clear to me that all of the models in Step 4 included two predictors each that were either (1) drawn from the same behavior or (2) drawn from the full unfiltered networks of both behaviors.
Lines 376-379: I’m not sure I understand the logic here. Why is contact sitting relevant here when it’s not in the best model set?
The authors provide all necessary data to replicate the findings and thorough supplementary tables documenting the model selection process. The conclusions are well stated and clearly supported by the data provided. The logic behind the authors' conclusions is sound and compelling. However, as noted above, I would appreciate more context and explanation regarding the analytical choices and their practical meaning.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.