Cancer and loneliness in Drosophila
An interaction between cancer progression and social environment in Drosophila
Drosophila flies may not be perceived as a quintessentially social animal, particularly when compared to their eusocial hymenopteran cousins. Although they have no parental care, division of labour or subfertile caste, fruit flies nevertheless exhibit an array of social phenotypes that are potentially comparable to those of their highly social relatives. In the wild, Drosophila adults cluster around food resources where courtship, mating activity and oviposition occur. Recent work has shown not only that social interactions in these clusters condition many aspects of the behaviour and physiology of the flies  but also, and perhaps more unexpectedly, that social isolation has a negative impact on their fitness .
Many studies in humans point to the role of social isolation as a source of stress that can induce and accelerate disease progression. The ultimate proof of the connection between social interaction and disease is however mired in confounding variables and alternative explanations so the subject, though crucial, remains controversial. With a series of elegant experiments using Drosophila flies that develop an inducible form of intestinal cancer, Dawson et al  show that cancer progresses more rapidly in flies maintained in isolation than in flies maintained with other cancerous flies. Further, cancerous flies kept with non-cancerous flies, fare just as badly as when kept alone. Their experiments suggest that this is due to the combined effect of healthy flies avoiding contact with cancerous flies (even though this is a non-contagious disease), and of cancerous flies having higher quality interactions with other cancerous flies than with healthy ones. Perceived isolation is therefore as pernicious as real isolation when it comes to cancer progression in these flies. Like all good research, this study opens up as many questions as it answers, in particular the why and wherefores of the flies’ extraordinary social behaviour in the face of disease.
 Camiletti AL and Thompson GJ. 2016. Drosophila as a genetically tractable model for social insect behavior. Frontiers in Ecology and Evolution, 4: 40. doi: 10.3389/fevo.2016.00040
 Ruan H and Wu C-F. 2008. Social interaction-mediated lifespan extension of Drosophila Cu/Zn superoxide dismutase mutants. Proceedings of the National Academy of Sciences, USA, 105: 7506-7510. doi: 10.1073/pnas.0711127105
 Dawson E, Bailly T, Dos Santos J, Moreno C, Devilliers M, Maroni B, Sueur C, Casali A, Ujvari B, Thomas F, Montagne J, Mery F. 2017. An interaction between cancer progression and social environment in Drosophila. BiorXiv, 143560, ver. 3 of 19th September 2017. doi: 10.1101/143560
Ana Rivero (2017) Cancer and loneliness in Drosophila. Peer Community in Evolutionary Biology, 100030. 10.24072/pci.evolbiol.100030
Revision round #201 Sep 2017
Decision round #2
Thank you for the replies to our comments and suggestions. There are still some issues that the reviewers feel are important to improve the quality of the paper and that were not fully addressed in your earlier replies.
If you choose to answer to these points, and to facilitate our work, I would ask you to please indicate in your replies where in the manuscript the changes have been made (provide line numbers)
Page 7 (Social environment choice)
(9) (second para.) I see how what you are saying in this paragraph fits well with what I see in Figure 3, however I got confused by the stats. For example, it looks like the target fly effect is only significant 7 days post induction, so I find it surprising that you have a significant target fly main effect but a not significant target fly*age interaction. As I read on, I’m not even sure whether what is being tested is the difference between control and cancerous targets, or between observed vs expected (random) choice. Where does this second chi-square value for age come from? A bit more info about how was this analysis done would help understand.
We now provide more information on the way the analyses were done. We apologize but found an error in the statistical analysis of the dual choice (age is in fact strongly significant, no change in the results of the other factors). We observed a general decrease in preference for the cancerous stimulus group, but we could detect such decrease only in cancerous target flies. This may explain the lack of significant interaction target fly*age.
RECOMMENDER REPLY #1 The analyses of the choice experiments still lack clarity. These issues have also been raised by REVIEWER 2, please see below.
(19) Was the “tube” effect taken into account in the model? i.e. is this a mixed model with treatment as a fixed factor and tube as a random one?
Because in the heterogeneous environment only one cancerous fly was present, we grouped fly guts randomly from different tubes (this is now mentioned in the manuscript). Thus tube was not a random factor and only treatment was included as a fixed factor.
RECOMMENDER’s REPLY #2
Does this mean that for the “alone” and the “heterogeneous” treatments the % of cancerous flies is given for an individual cancerous fly and that for the “homogeneous” treatment it is the mean for the 8 cancerous flies in the tube? But then what does it mean that you “grouped fly guts randomly from different tubes”? (and where in the manuscript is this point clarified? Please provide line numbers).
- I am confused with the statistical analysis in the choice experiments. The authors show the results of a logistic regression in the text, where they ask the questions whether age, cancerous state/stimulus, or the interaction between them affects choice. They are thus comparing different groups in their choice, but they do not directly test for one group whether they are more attracted to a certain cage over the other (just differences between groups). This is more or less tested in individual tests for deviation of random choice (0.5) with asterisks in figures 3-4, however, the method of this testing is not described but seems to be performed for each individual point separately and p-values should thus be adjusted for multiple testing. Was this done? I suggest the authors to have the principle analysis be done on whether cancerous flies are attracted to a certain social group, a secondary analysis would be whether there are differences between cancerous vs control and age of the fly.
Stars on figure 3 and 4 represent significant deviation from random choice calculated for each line and age. We disagree with the necessity of doing the adjustment for multiple testing as we are not comparing the treatments among them in this analysis (compared to figure 1 and 2).
We believe that this representation allows the reader to see at which age and state there is a significant effect and can conclude that there is for example aversion of the cancerous flies by the control ones when tumours are well developed. We clarified each analysis done and detail the statistics.
RECOMMENDER REPLY #3
3.1. We agree that this is a useful way of looking at the data, but the question asked by reviewer to was how was the significant deviation from random choice tested. Please state what statistical test was used.
3.2. Contrary to what the authors state, the adjustment for multiple testing is required irrespective of whether you’re comparing treatments with each other or, as is the case here, comparing observed vs expected (random) 6 different times (one for each day x fly state combination). You could easily do a (back of the envelope) Bonferroni correction by dividing your critical alpha (0.05) by the number of tests (0.05/6 = 0.008). Thus, you would then only consider tests with p<0.008 (or to round it up p<0.01) to be significant.
- P7, paragr3: “This was especially pronounced when flies were young i.e. at the very beginning of the tumor development”. However, the interaction between age and target fly is not significant so this is not a significant effect.
We made this clear that here we were not comparing cancerous vs control
RECOMMENDER REPLY #4
Not that clear, actually. The analyses in Lines 178-179 seem to be comparing control vs cancerous (this is the “target fly” effect, right?). As stated by the reviewer, since the interaction age x target fly is not significant, stating that “this was especially pronounced when flies were young” is, statistically speaking, incorrect (even if there seems to be a trend in that direction).
- P7, paragr3: “However, at later … P<10-3).” I believe this p-value must be based on the individual datapoint analysis that is not described. This becomes confusing because you first report a non-significant interaction (see point above), but here you don´t talk about a difference between groups, but a difference from random choice. Please rewrite the results so these distinctions become clearer.
We hope this is now clarified
RECOMMENDER REPLY #5
Not really. Please clarify what analyses have been done (i.e. what is it exactly that you’re testing?) in lines 180-181 and lines 183-184 and how they differ from the analysis reported in lines 178-179. I’m guessing 178-179 is cancerous vs target, while the other analyses test for a departure from randomness for cancerous (180 181) and control (183-184) flies. If so, this is far from clear from the text. Please see also #6 regarding clarifying the significance of the intercept.
- P7, paragr4: “Cancerous flies showed … P=0.44”, same point as above, you report the nonsignificant result but mention a significant effect. Report statistical analysis.
We apologize, the significance of attraction towards the social stimulus could only be understood by mentioning the p value of the intercept. It has been added.
RECOMMENDER REPLY #6
6.1 Thanks for the clarification. As an intercept can be interpreted in many different ways depending on the type of model, the fact that in a logistic regression a significant intercept indicates a departure from randomness (which may or may not be explained by the explanatory variables) needs to be clearly stated here (Line 454-455 is lost in the depths of the m&m and does not suffice). If possible, please provide a reference.
6.2 I take it the individual departure from each of the treatment combinations (indicated by the stars in Figure 4) was done separately? If so please state how (as per #3.1 and #3.2).
6.3 Why are there no stats comparing Cancerous vs Control target flies?
- In the concluding paragraph of the discussion there is a referral to the contribution of this study to the evolutionary ecology of cancer, it would be great if the authors could expand a few sentences on this. What are the evolutionary benefits? Could such behavior be adaptive for cancer or is it an unintended consequence of a non-specific infection avoiding behavior? The tumor cells in this study do not impact fitness, could this bias any conclusions drawn from this study?
We modified the discussion accordingly to this comment
RECOMMENDER REPLY #7
I have failed to see any substantial change to the discussion. Could you please be more precise as to how and where (line numbers) these changes have been made?
- P8, paragr4 “Even if not… with being sick”: if this is a general response, shouldn´t cancerous flies also avoid other cancerous flies (which they don´t)? If it is a general response, they may want to avoid flies with contagious infection despite themselves having cancerous cells. Please discuss.
Even if we have not observed clear ‘avoidance’ of the cancerous flies by other cancerous ones we still see a decrease in preference. We believe that there could be a balance between avoidance of the potentially contagious individual and attraction of individual of the same type. This balance may vary with cancer progression.
RECOMMENDER REPLY #8
Please clarify whether this clarification has been included in the manuscript and if so where.
Revision round #115 Jul 2017
Decision round #1
This is a very interesting article that addresses an important issue on the correlation between cancer and social environment, using a relevant experimental system. However, both referees have raised issues with the statistical analyses: they lack clarity and at times they seem to be at odds with what it's written in the text. Would be happy to recommend provided the statistical analyses are clarified and all the individual points raised by the referees (statistical, or otherwise) addressed.