PCI Zoology #215
 
Review, round 2 – comments to authors
 
Dear authors,
First of all, thanks for your thorough revision and patient clarifications. The paper reads very well now, and I have no further general comments (good job clarifying the intro and discussion!) - just a few minor ones.
 
L45: critical to fitness of such parasites -> critical to the fitness of such parasites
L133: remove parenthesis before Franceschi, move it to before 2008
L286: whatever -> regardless of
L327: to increase clarity, add “with” before “significantly”
L328: "increased activity according to time" could be changed to “increased activity with time”
L456: rephrase to: “Such hypotheses nevertheless explain neither why…, nor... “
L477: to avoid repetition of the word “behavior”, consider rephrasing to e.g. “the first component of the ‘protective’ behavioral manipulation”
L489, consider rephrasing to “while high and low refuge use are beneficial to the acanthella and cystacanth stages, respectively.”
 
Thanks again for your answers!

I have meticulously reviewed the revised version of the manuscript, the supplementary material, and the authors' response to the reviewers. I genuinely appreciate the effort with which the authors addressed and integrated our feedback. This is a very nice study. 

In my previous review, I had suggested rerunning the experiment to validate the results. I recognize the inherent challenges and constraints in doing so. My initial recommendation stemmed from the ambiguity and some over-interpretations present in the first version. In this revised manuscript the authors prudently focused on their more robust findings. I am now more confident about the replicability of the strong effects observed in this study.

That said, I'd like to point out some lingering inconsistencies, notably with the terms “not infected” and “uninfected.” For instance, I think that it can be observed in figure 1 and possibly on line 228. It would be good to go through the manuscript for that, since this detail created a fair amount of confusion for me in the earlier version.

To address further things that I pointed out before, and to go in the sense of the authors. The data, now in supplementary material, unambiguously indicates that both the exposed-non infected and non-exposed groups exhibit the same mortality rate. It is also reasonable to conclude that the behaviour of exposed-non infected individuals doesn't significantly differ from their non-exposed counterparts in a meaningful way. Therefore, it is justified to combined them in the analyses.

Lastly, I found the supplementary material to be a valuable complement to the manuscript. The French annotations in the code are, I believe, acceptable but it may be a decision of the journal.

In this study, Rigaud et al. are investigating the manipulation of a host (a shrimp) by a parasite (a helminth) throughout the parasite's development. This parasite has a complex life cycle and manipulates its intermediate host to increase its chances of being consumed by its final host, a fish. This behavior is well-documented and is employed by many parasites as a strategy. The current hypothesis is that the parasite initially protects its host from predators before exposing them to continue its life cycle. However, the timing of this shift in behavior is not known and is the subject of this study. Additionally, the study explores the effect of predator scent in the water, which personally, I found very intriguing and it brings an interesting perspective to the research. 

I enjoyed reading the introduction, but I am not up to date with the literature on the topic. I will entrust the author, the editor, and possibly other reviewers to better judge if something is missing.

Overall, I would conclude that it is a nice and interesting study. However, to ensure the validity of the conclusions, I would recommend repeating the experiment once more, ideally with an unexposed group for comparison. It is surprising that several outcomes changed once the infection could be clearly characterized (strong example being the survival). I hope my concern is evident in the comments I provided below. I wish the study had compared individuals exposed to the pathogen with individuals who were non-exposed (possibly discarding or analyzing individuals who resolved the infections in a separate category). I am confident that the effect would be clearer when compared to the unexposed group. The authors acknowledged that multiple handling could have influenced their conclusions. In a repeated experiment, they could use a new group for each day of experiments, ensuring that each individual is handled only once. Without such a repeat, I would exercise caution, as most confidence intervals largely overlap for many of the claims, and refrain from delving too deeply into the details, such as the effect observed in the early days.

Major comments:

* There is a need for improved clarity regarding the distinction between uninfected individuals who were exposed but did not contract the infection. While the authors have been cautious in their terminology by using "uninfected" and "infected," it is commonly assumed that uninfected individuals represent control subjects who were not exposed at all. To address this confusion, it is advisable to explicitly state this distinction in the text legends of the figures. For example, using labels such as "exposed, uninfected" versus "exposed, infected" (as done on line 238 for behavioural assays) would help clarify the terminology. Additionally, including a control group in the survival graphic would be beneficial to enhance clarity. This inclusion would help explain why "uninfected" individuals exhibit similar mortality rates to "infected" individuals early in the experiment.

Furthermore, there is a concern regarding the certainty of categorizing individuals as uninfected during the early stages of infection, especially when the acanthellae are still small and not easily detectable. Although it is mentioned that Gammarus specimens were dissected upon death (line 184), it remains unclear why all treatment groups exhibit comparable mortality rates until the acanthellae become detectable through the cuticula. At that point, the uninfected individuals stop dying. Is it possible that many infected individuals died early, but their acanthellae went undetected? This uncertainty emphasizes the importance of including an unexposed control group in the figure. While it may be regrettable that only 20 individuals were included in the unexposed control group, it would still provide valuable insights, especially if there were no natural deaths throughout the course of the experiment. Please note that I agree that the non-exposed group should not be included in the survival analysis.

* It is a bit tricky to use a non-parametric test to justify the fact that exposed-uninfected individuals are not statistically different from unexposed gammarids in terms of their refuge use. If the result had been significant, it would have been more convincing. However, the opposite outcome could indicate a lack of statistical power. It is difficult to assess this because the data supporting these findings are not shown. I wonder why the parameter was not included in the model. By including variables for "infected" versus "uninfected" and "exposed" versus "non-exposed," the analysis would be more accurate and not necessarily more complicated. I may be mistaken in my assessment, but I feel that this could have very important consequences on the whole study and I am afraid that it is hiding some of the effect/biology.

Furthermore, I strongly recommend making Figure 2 and 4 wider and displaying the data points within the box plots. There are several arguments to support the inclusion of the data points. One simple reason is that a box plot with only four individuals does not provide much information (see Figure 2, day 83, infected). Additionally, this would be an opportunity to color-code the data points based on whether the individuals were exposed or not.

* Throughout the manuscript, it should be acknowledged by using more cautious wording that the major results concerning the difference between infected and uninfected individuals in scented versus control water are based on trends or tendencies. Despite the substantial overlap in the 95% confidence intervals, they have been disregarded.

* Lines 333- 336: “while almost no difference between infected and uninfected animals was found when tests were made in control water, as seen using effect sizes, G. pulex infected by acanthellae use more the refuges and are more inactive than uninfected ones in fish-scented water”. 

This is part of the discussion, and I do not see where the conclusion on refuge use comes from. It is important to note that there is no significant statistical interaction between infection status and water type (Figure 2), and Figures 3 and 5 do not provide evidence supporting this conclusion. Table 2 could support the idea that the strategy regarding the use of refuge over time depends on the infection status and water type since both the interactions "date x infection status" and "date x water type" are significant. However, the triple interaction is clearly not significant, which suggests that the strategy of using the refuge over time is not different between infected and uninfected individuals for different water types. If we disregard the statistics and focus on Figure 3 (including the 95% confidence intervals, which serve as graphical statistics), we can observe a tendency at day 55 regarding the use of refuge, with infected individuals showing increased refuge use. However, this is strangely the time point at which the parasite can be reliably detected. Was the difference hidden before because the detection was not reliable?

 * line 347: “The second information is that this protective manipulation was observed – in scented water – early during parasite ontogeny (as soon as 20 days post-infection).” 

Then, line 357: “However, this protective manipulation was not constant with time since we observed variation (in scented water) on the differences between infected and uninfected gammarids: the effect size oscillate between negligible values to medium values.”

The author acknowledges the inconsistency in their results but only attempts to explain it through biology. While I appreciate the amount of work done, it is important to recognize that the experiment has only been conducted once. Therefore, it should not be ignored that the reason for the variation may be that the effect is minimal or not present. Although there is a trend at 21 days, the confidence intervals largely overlap, and the effect disappears at 27 days. Why would day 21 be the correct day while 27 might be an artifact due to handling? 

Note, that I could agree with this “artifact” explanation, but it is one among several possibilities. Overall, most confidence intervals overlap significantly with zero (indicating no effect) and even more so with the other condition used for comparison (i.e., the water type). Therefore, I would recommend repeating the experiment to confirm the finding or to exercise caution in interpreting the results.

Minor comments: 

* Generally speaking, I would recommend including a concise conclusion in every figure legend. Many readers tend to focus primarily on the figures rather than reading the paper thoroughly, so it is important to ensure that the key message you want to convey is understood from there.

* The interaction between water type and infection status shows borderline significance in the survival analysis, but the tendency reveals a clear reversal that should not be ignored. This could be attributed to the relatively low sample size for the control group (175 - 132 = 43 exposed uninfected and 20 non-exposed).

Moreover, the analysis itself is likely a contributing factor. It appears that the current Cox regression performed may not be entirely appropriate for this situation. Cox models assume that the hazard ratio (HR) remains constant over time, which is not the case here (a condition that can be tested using cox.zph in R). It seems that there are two distinct phases in the survival pattern, which may warrant a slightly different analysis approach (such as using stratification or a time-dependent covariate). While the Cox regression was conducted in JMP, it might be worthwhile to perform it in R as well. This could provide additional options and improve the repeatability of the analysis, but I don’t know enough JMP. 

Note that this analysis may be affected by my concerned that some dead infected were categorized as uninfected just because the parasite was not found during dissection.

* I would also plot the HR ratio as a second figure associated with the survival, this allow to give the 95 % CI, and the sample size under each treatment. Providing HR and 95% CI also prevents sentences like “was almost supported statistically” based on pvalue. 

* In the analysis section, I would provide the exact models used to analyse the data and the package when R was used (which is recommended for reproducibility).

* There are no details about the analysis of the impact of the parasite intensity on survival (beside that the analysis was done as either infected with 1 parasite versus infected with more than one, or as a continuous variable). I guess that the individuals from the treatment “Scented water” and “Control water” were included but how? Was treatment as a covariate? (I would have done that, and it might be worth testing if the treatment affects the outcome).

* Line 329: “This shift was nevertheless not observed for the activity rate, another behavior involved in predation avoidance. In addition, our study brings new information on this anti-predatory behavior.” Was not observed or not tested?

* Figure 2: Why do some outliers appear larger than others? As I see dots side by side, I guess it does not mean “more dots”.

If the objective is to track behavioral changes over time, it might be more effective to plot the individual's behavioral variations on a continuous x-axis and connect the dots from a same individual. The lines could terminate when individuals die, but this representation may better capture the data points and align more closely with the analysis (it currently suggests that data point are independent across time points). Such an approach would likely provide a stronger visual representation of the statement, 'The refuge use by gammarids varied considerably with time (Figure 2, Table 2), being moderate at the beginning of our survey and progressively intensifying until reaching high values around 30 to 40 days' (lines 240-243).

Additionally, it is mentioned that 'This phenomenon was found whatever the infection status and the water type, but was found to be more rapid under scented water than control water (Figure 2), with the interaction between water type and date being statistically significant' (lines 252-254). However, the table only presents a Wald statistic and lacks an estimate of the effect. Without additional context, it is difficult to discern the specific influence of water type on the rate of refuge usage. Can't the method used in Figure 3 be used to show this effect between water types?

Regarding the analysis, it is not explicitly specified in the methods whether time is treated as a continuous or ordinal variable. It would be helpful to provide clarification in this regard.

* The Cliff's effect size estimation is appropriate; however, a description of what it entails should be included in the method section. For instance, it should be explained that the Cliff's delta measures the frequency of values in one distribution being larger than the values in another distribution. Additionally, it is important to clarify whether a value of 0.2 is considered a small or medium effect size and provide the corresponding scale. Currently, the discussion provides the only indication of this information, which requires careful reading. Therefore, it would be beneficial to include details on how the Cliff's effect size was calculated.

* Infected individuals are using less refuge and are more inactive than non-infected individuals when at Day 69. Aren’t they simply dying from the infection on the ground? I would address that with in the manuscript because this would be the most expected for most infections although it is here (including in the abstract) presented as something surprising.

* Figure 3 & 5: I like it very much, but the fact that 95% CI are largely overlapping is often ignored. I am aware that part of the conclusion are based on the model (Table 3) but looking at the data, it is likely that the effect is not linear and I wonder how the model deals with that. 

* Figure 4: Most of the comments given for Figure 2 are relevant here too.