Unity makes strength: clustered extinctions have stronger, longer-lasting effects on metacommunities dynamics

The manuscript “Spatial distribution of local patch extinctions drives recovery dynamics in metacommunities” presents the effects of the extent and spatial clustering of an extinction event on the recovery of local and regional diversity, productivity, and on the recovery rate, in a metacommunity of competing species. It is based on a microcosm experiment and a mathematical model. Spatial clustering had a larger effect than the extent of extinction events.

The introduction is effective in setting the scene for this study, but I believe it could be made clearer, at the end of the intro, which phase of recovery dynamics are targeted here with respect to the diversity and productivity data (just after the disturbance? One generation later? Etc). That was not clear to me, and I also did not find this back in the methods. The methods are clear and well-described (apart from the time points associated with the diversity and productivity data). The results are well-structured (although I provide some detailed comments below). I find the discussion repeats the results a bit too much, which makes the discussion section relatively long and renders some of the interesting bits less visible.

As a general comment to the paper, which I generally liked and found to be based on an impressive amount of work and well thought-through designs, I have to say it took me a while to realize that the main variables (e.g. Fig1) were measured during a recovery trajectory. This also raises the question of how representative these measurements are for the ‘recovery dynamics’ as a whole. If possible (e.g. if the necessary data have been measured in the experiments) I think it would be cool to try to plot the rate of change (or some other summary stat) of (some of) these variables during the recovery period instead of the variables themselves. This might be even more effective in addressing the main questions.

Another general comment, which may seem a tad pedantic, but well: I would propose to replace ‘extinct patches’ with ‘disturbed’ or ‘perturbed’ patches. That is because (1) a patch cannot go extinct, only populations or species can; and (2) it prevents some odd constructions like ‘alpha diversity in an extinct patch’ (which, technically, should be zero).

Finally, I wonder if it would be possible to compare the measured dispersal rates and the competition parameters to see if the empirical system exhibited features akin to a competition-colonization tradeoff.

Specific comments

L13: ‘persist’ suggests long-term persistence; not sure if this is what is meant here.

L37: In a scenario where all biomass disappears, how can one trophic level benefit? If all are removed, it does not matter which trophic level a species belongs to, right? Please clarify.

L46-52: Although I like the intro, I feel the focus of this bit is less clear and evokes some elements that are not core elements, so my suggestion would be to trim it down. At this point, it was also not clear what was meant by ‘spatially structured’.

L69: I propose to use something like ‘extent’ instead of ‘rate’, as there’s no time component, right?

Eq4: I advise to use single-letter expressions for variables (e.g. m instead of n_links).

L221: I advise to use ‘interaction coefficients’ or ‘competition coefficients’ instead of ‘competitive abilities’.

L227: It would be nice to have a rationale for the scenario of no species interactions. I’m sure a few things can be learnt from that additional scenario, but spelling them out would be useful.

L258: “extinction rate alone only had a marginal effect on the outcome of the experiment”. I’d say that for beta diversity it had.

Fig.1: why isn’t there a ‘no extinctions’ scenario plotted for the experiment?

Fig.1: the variable ‘recovery time’ seems to be factorial (5 levels). Was this the case and if so, why, and how was it dealt with in the statistical analyses?

Fig.2: depending on what you want to convey as the main result, there might be more effective ways to plot these data. One way could be to make a biplot for (some of) the variables at 4 extinctions vs. 8 extinctions, to show they are all located near the 1:1 line, regardless of interaction scenario (coded with a symbol key for example) or the spatial arrangement of extinction (e.g. using a color key). A second panel could do exactly the same but would now compare the variables between the two spatial extinction arrangements. Just a spontaneous thought though.

More on Fig.2: It seems interesting that beta diversity is higher in perturbed networks than in nonperturbed networks when there are no species interactions. Is that then purely due to difference in species abundances? In general, it could be insightful to plot one of these beta diversity indices that only look at compositional similarity and not at similarity of abundance.

L265-266: I would be a bit more precise. What is meant by ‘reproduced these results’? What are ‘these results’? I suspect it is the relatively small effect of the number of extinctions compared to that of spatial arrangement but it would be good to repeat this here.

L282-283: It could be interesting to compare the distributions of interaction strengths across model scenarios (possibly scaled by intrinsic growth), as it might explain the results for recovery rate.

Fig.3: This caption mentions “at the last two measurement points”: what are these and why was this info not mentioned in Fig.1 (and 2)?

The section starting L357 starts by mentioning the consistency of the results across response variables, but then focuses on the recovery rate only. It might be useful to, even if briefly, mention how exactly effects on the recovery rate explains the effects on the other variables.

L366-367: If the data are available to do so, it could be nice to substantiate this further by testing how the number of neighboring non-extinct patches affects recovery rate.

L400-401: I would find it more logical to state “extinct patches had a higher biomass”.

Summary

Saade et al. (bioRxiv 2020.12.03.409524) evaluated how the rate and spatial distribution of extinctions affected metacommunity diversity. Using a simplified ciliate metacommunity of three species in a factorial manipulation of extinction rate (0, 4, and 8 patches out of 16) and spatial distribution (dispersed vs. clustered) paired with numerical simulations, Saade et al. demonstrate that recolonization dynamics and community recovered depended more on spatial distribution or the interaction between spatial distribution and extinction rate than extinction rate alone. By combining the experiment with numerical simulations, Saade et al. were further able to identify that interspecific interactions and, specifically, a competition-colonization tradeoff as a possible mechanism. It is important to note that these numerical simulations were based on empirical measurements of the study organisms and more directly link the numerical to experimental (and vice versa) in the study. The authors were also very clear when discussing their results and the broader implications, presenting any uncertainty and tempering any broader claims with appropriate context; in doing so, I think it makes the link between this study to the broader body of work stronger because it accepts limitations while also presenting the strengths. This study contributes very strong evidence of how spatial configuration needs to be considered when looking at extinction or extirpation dynamics. This study also provides an important “proof-of-concept” that spatial distribution of extinctions should be considered to understand community recovery. In an applied sense, this study suggests that exclusion of spatial distribution from analysis or management of extinctions could be faulty. In summary, I found this paper very compelling, clearly written and ‘mentally replicable’ (i.e., while reading the methods and looking at the figures, I could replicate the study), and with no major criticism or comments. I only have a few minor suggestions to improve clarity and presentation of the results and some comments for the discussion.

Minor Comments

1. Line 187: What R package was used for the mixed-effects models? It would be good to know whether it was lme4 or nlme, for the sake of reproducibility.
2. Line 195: On which parameters was relative importance calculated and how was it determined? It is likely a simple thing I just missed, but this component spilled over to Table 1 and Table 2 where it was not clear exactly what information was being provided.
3. Line 258: Extinction rate is said to have a “marginal effect” on community recovery, and this seems accurate based on Table S2 but not Table 1. Looking at Table 1, I would consider the effect of extinction intermediate but Table S2 I can see how it is marginal. The in-text presentation could therefore appear misleading when calling the effect marginal yet readers only immediately seeing Table 1.
4. Table 1 and Table 2 captions: As written, I do not know exactly what the relative importance is derived from. Following my second comment (Line 195), a more detailed clarification in the methods would make a brief, if any, clarification in the table caption helpful.
5. Line 344: It is reported that “β -diversity was fairly low because the patches ended up being homogeneous” but the results are not shown. This happens again on Line 397. Would it be possible to add a figure to the supplement?
6. Line 517: Effects of metacommunity synchrony on extinction dynamics are discussed, but asynchrony and ‘portfolio effects’ could also be discussed. While the experiment only used synchronized extinction treatments, it would be good to at least mention asynchrony and how this could have affected results and conclusions.
   • More broadly, is it possible to use the metacommunity model to parameterize temporal asynchrony in the extinction treatments and numerical responses? Spatial asynchrony would also be interesting, along with the interaction of spatial and temporal asynchrony.
   • I completely understand that this is not really the point of the paper and going too far into this topic would dilute the message of the present study, but it would be good to at least note potential effects of asynchrony and potentially parameterize temporally and/or spatial asynchronous models.
7. Figure S5: It looks like the default graphing parameters re-organized the time points so that the x-axis does not go in chronological order. In ggplot (and wrappers like ggpubr), this could be done with the scalexdiscrete() function.