The manuscript entitled “Temperature change exerts sex-specific effects on behavioural variation” reports from a study on 129 Drosophila melanogaster that underwent repeated behavioural and metabolic assays at two different temperatures (25 and 28 degC). Univariate mixed models revealed that males were more active than females. Moreover, a marginally non-significant interaction suggested that temperature caused a greater increase in activity in males than females. Temperature also caused an increase in among-individual and residual variances in males, but not in females (in fact, there was a decrease in residual variance in females). As expected, metabolic rate increased with temperature, but in contrast to activity there were no sex differences in metabolic rate. Finally, a bivariate mixed model was used to test for an association between activity and metabolic rate, but the uncertainty around the correlation estimates were extremely large and overlapped with 0 (which might reflect a lack of power, given data-hungry multivariate models, or variance components that are estimated close to the boundary). Overall, this is a nicely written manuscript from a nicely conducted study, based on cutting-edge respirometry techniques and statistical analyses (variance partitioning), but there is room for improvement.

Global warming slant: One of the “selling point” included in this manuscript is the relevance to climate change, more specifically on the adaptive potential of populations in the face of global warming. To answer that question, one needs to know the strength and direction of selection on thermal reaction norms and their underlying genetic basis. None of these were quantified here, which represents the biggest limitation of the study (which arises to the extent that adaptive arguments are included/discussed). The current study is based on a relatively small sample size compared to typical selection/genetics studies on Drosophila. As such, it might be best described as a preliminary study that naturally paves the way to larger follow-up selection and quantitative genetics studies. Indeed, the adoption of a repeated measures design and application of a variance partitioning approach certainly represent a step in the right direction. There is enough value here to merit publication, without the need to oversell the relevance to adaptive potential to global warming.

Repeatability: despite the limitation above, there is certainly value in the work presented here. It is widely accepted that individual variation is important to study because it represents the raw material for selection. In studies focussed on individual variation, a very important metric is individual repeatability. However, there is no repeatability estimate reported in the manuscript. Of course, sex differences in thermal sensitivity are important, but what about individual variation relative to phenotypic variance? It would be informative to calculate and report the repeatability of activity and metabolic rate in males vs females at 25 vs 28 degC. It would also be crucial to report all of the estimates (fixed effect estimates and random variance components) from all models fitted (note: this is done for activity, but is lacking for meytabolic rate and the bivariate model).

A question that derives logically from the above is: do individuals differ in their sensitivity to temperature? A common way to answer this question is to adopt a reaction norm approach and model individual variation in intercepts and slopes. It seems like the authors have the data and analytical capabilities to provide these. But, we can already answer this question, at least for activity. Indeed, table 1 reports the among-individual correlation between 25 and 28 degC, which is 0.94 in males and 0.90 in females. Both estimates do not differ from 1, which implies activity measured at 25 and 28 degC represent the same trait (i.e., individuals do not differ in their thermal reaction norm slope). This aspect of the data/results needs development and discussion.

Hypotheses and Predictions: it could be useful to modify the last paragraph of the introduction. Perhaps the best way forward would be to present the study as a descriptive study (which is totally fine). Alternatively, it would be important to develop the hypotheses such that they present a clear causal (mechanistic) explanation for observed phenomena. As presented, the hypotheses appear more like expectations based on previous results and there are many logical gaps. For example, the prediction on line 111 reads like: “any effect in x may be more pronounced”. The lack of directionality, let alone refutability, suggest the predictions are not derived from a clear causal hypothesis.

There is a prediction that “male flies would demonstrate increased population-level (i.e. mean-level) thermal plasticity […] due to their smaller body size in comparison to females”. If this is the case, then sex differences in plasticity should “disappear” when conditioning on a “mass x temperature” interaction. To this effect, it would be interesting to look at the estimate for the “sex x temperature” interaction before vs after including a “mass x temperature” interaction. Pulling apart the sex vs mass effect on activity changes as function of temperature would be highly relevant to discussion on line 420, where it is concluded that “the reduced body size of male D. melanogaster, relative to females, does not result in males being more metabolically plastic on average”.

Regarding prediction coming from the second “hypothesis”: there are many logical steps required to go from temperature effects on mean MR to greater variance in activity. If increased metabolic rates at higher temperatures “provides more energy to fuel behavioural variation”, then one would intuitively predict higher activity at higher temperatures. However, the prediction is that “that rising temperatures would result in greater among- and within-individual variance in locomotor activity.” There are some logical gaps to fill to make sure readers clearly understand the causal mechanism linking temperature effects on MR to variation in activity.

Potential outliers: there are 3 are suspiciously low MR values in the uploaded dataset, two of them even negative (-0.14, -0.07, and 0.03 for individuals 9, 4, and 3). Should these be deleted? At least, a separate analysis without these could be provided.

It is remarkable that a double-hierarchical generalized linear mixed-effects model (DHGLMM) was fitted to the activity data. However, readers might wonder what was gained by fitting a DHGLMM over a regular mixed effect model. The only additional estimates that the DHGLMM provide over a mixed model are associated with the ID random effect fitted in the residual part of the model (i.e, the two bottom lines in Table 1). However, these estimates are not mentioned or discussed. What is gained by using a DHGLMM?

Minor comments:
At many places it would be preferable to use “locomotor activity” instead of “behaviour”, as the former is more precise and appropriate given the scope of the study.

Variance usually increases with the mean. Therefore, the expectation that within- and among-individual variances increase with temperature might be simply due to an increase in the mean. Perhaps it would be useful to re-analyse the data after log-transforming the response variables (activity and MR), to get an insight into how the mean-variance relationship might drive the results in males?

Lines 71-77: some revision might be needed here to avoid naïve readers think that the argumentation provided rests on group selection arguments. After a reference on ant colony fitness, we read that “within-individual variation may evolve as an adaptive strategy for dealing with heightened predation risk”, which could be taken as a suggestion that it is adaptive for a group (population, species) to display greater within-individual variance. However, studies on “predictability” are focussed on among-individual variation in within-individual variance, which is a slight subtle but important distinction to make here.

Line 90: "behavioir"

Line 110: “furthemrore”.

Were repeated measures on a given individual done in the same or different metabolic chambers? If the same chamber was used, then individual ID and chamber are confounded.

Line 222 and Table 1: unclear what insight is gained by the adoption of the double-hierarchical generalized linear mixed-effects model. The variance estimates (i.e., 4.05 and 1.53) are extremely small compared to other variance components that range between 79,000 up to 285,181.

Table 1: Presentation of 6 digit figures using a space to denote the 1000’s demarcation is confusing. (it took time figuring out why there were two separate numbers reported).

Table 1: the among-individual correlations between 25 and 28 degC are strong and close to 1. This aspect of the results needs consideration and discussion. First, one could see activity at 25 and 28 degC as the same trait. It also implies there are no individual differences in thermal sensitivity.

It would be useful to have a table similar to Table 1, but for SMR and AMR. How was the structure of fixed effects determined for each model?

Repeatability estimates are entirely lacking from the manuscript. Please calculate and report R for all traits and temperature/sex categories.

Lines 253-274: When activity and MR are analysed separately, the ID random effect is modelled separately according to sex and temperature. In the bivariate model, the correlations are estimated separately according to sex, but not to temperature. It is unclear why. Perhaps it would be useful to state that activity at 25 and 28 degC represented the same trait (based on the among-individual correlation of 0.94 and 0.90 in Table 1) and therefore temperature-specific correlations were not different.

Table 1: for among-individual variance, please use “V[i]” or “V[ind]” instead of V[a] (should be restricted to additive genetic variance).

Line 285: the mass effect is marginally non-significant and it is written that “there was no substantial effect of body mass”. By contrast, the also marginally non-significant “sex x temperature” interaction was interpreted as “males displayed a moderately greater increase in their activity in response to increased temperature when compared to females”. Why the suggested interpretations differ for two relatively similar effect sizes?

Line 319: "increased" isn’t appropriate in this context. This is the amounts of SD in variance at 28C. Increased would only be relevant if we were comparing SD at 28 to that at 25C, with an associated CI.

Line 339: the text says “sex × temperature interaction in the residual model; Table 1”, but this interaction cannot be found in table 1. The use of the term “interaction” is misleading.

Line 404: unclear what is meant by “More specially”. Specifically would make more sense.

Line 434: here the term “metabolic rate” is employed, but it may be important to specifically refer to SMR instead (unless the intended meaning is different, in which case some clarification is needed because increases in activity causes increases in metabolic rate, not the other way around).

The authors have followed most of my suggestions. Perhaps one last point would be to make sure the distinction 100% clear between A) residual (within-individual) variance versus B) variance in the within-individual variance. Reading throughout the text, it seems like the term “within-individual variance” is used to refer to mostly to residual variance, but some of the arguments about the adaptive nature of within-individual variation (i.e., in relation to predation) are potentially confusing A and B. In the current study, residual variance differed between males and females, which may arise from sex differences in measurement error, plasticity in response to one or more unaccounted factors, or “true” unpredictability. Perhaps it would be useful to remind readers about these 3 possible causes. It would be surprising that measurement error differs between the sexes in the current study, given measurements were taken with the same equipment. However, differences in (unaccounted) plasticity cannot be ruled out, and this should be mentioned as an alternative to the “unpredictability as a anti-predator strategy” explanation. In other words, the sex-specific effects of temperature on within-individual variation in activity rates found in the current study could be the result of different plasticity – it could be that male and female behaviour is sensitive to different sources of micro-environment variation and that the relative importance change as function of temperature (thus causing changes in residual variance).

Looking at bottom two lines of Table 1, we can see that males were significantly more variable in their residual variance (4.05 vs 1.53). In other words, there is greater among-individual differences in within-individual variance in males than females. Even if this was not the focus of the study, it remains odd that this significant result is not even mentioned. Would it be possible to integrate this result in the discussion about residual variance?

Line 287: the part “We allowed correlations to vary among intercepts (i.e. individuals)” sound like you estimate separate correlations for each individual, which is impossible. Perhaps a proper and simpler way to say this is that you estimated sex-specific among-individual correlations between SMR and activity.