In this study, authors focused on multiple anthropogenic stressors and their interactive effects on freshwater ecosystems. Aware of the complexity of the topic and of the need for more comprehensive scientific reviews, they used a machine learning framework to perform a very general and broad initial literature search, which would otherwise have been humanly infeasible. They ended up with a very large and seemingly robust corpus of >2000 studies reporting multiple stressors experiments in freshwaters. This corpus, made publicly available, is more than 100 times larger than in previous reviews, which is both impressive and promising for future studies. Authors then performed a rather interesting bibliometric analysis of the corpus (proposing for instance a “stressor taxonomy”), and identified some systematic flaws and research gaps in the corpus, which led them to recommend best practices and, most importantly, to propose future research directions.
The paper is very well written and pleasant to read. I only have a few comments that will help make it even better.

TITLE

I think this study is more about “a systematic corpus” rather than “ a systematic review of 2,396 multiple-stressor experiments"... but it's just a question of terminology.

INTRODUCTION

L.60-61 : the sentence could be rephrased to avoid suggesting that “Anthropogenic activities” are “home to a disproportionately high diversity of life”…

METHODS

In L.126 you mention an unbiased corpus but you later (L.135) acknowledge a bias towards experiments conducted in English-speaking countries. You should thus nuance the word “unbiased” in L. 126. Furthermore, I am not sure that this bias is towards English-speaking countries per se, but rather towards English-language science (sensu Amano et al. 2023; 10.1038/s41893-023-01087-8)… Scientific research performed in non-English-speaking countries often has to be published in English, mainly because the “the currency of science is publication in [English- language] academic journals.”(see 10.1038/s41562-023-01679-6 for a very recent (English) article on that matter).

L.141: You should mention that prior knowledge stemmed from Jackson et al. 2016, or at least refer to Supplementary Information (§3.3)

L.183: “broad” rather than “broadly”?

RESULTS

L.259: I like the way this review highlights “fashion trends” in research, with the case of nanoparticles and microplastics showcased in Fig. 3A and that of temperature (a proxy for “global warming”) showcased in Figure S7 as a quasi-systematic covariate in most experiments from the 2010s

L.272-273: Could you somehow highlight these disciplines in Fig. 4?

L.311: You should specify that there are “rare examples” of such higher-order interactions in the literature (because full-factorial experiments with more that two stressors are actually uncommon, cf. Fig 1D), but that they are probably much more common (if not the rule) in the real world, with very complex higher-order interactions among multiple stressors.

L.318-320: “If a stressor-response relationship is non-linear, however, even adding the same stressor twice would result in a non-additive combined effect despite that a stressor clearly cannot interact with itself”: I agree with the non-additive outcome of a non-linear stressor-response relationship (in the sense that, if a stressor-response relationship is non-linear, or, better said, that you do not know that such relationship is non-linear, then you cannot properly assess stressor interactions), but not exactly with the way the statement is formulated here. In my mind, a stressor can actually interact with itself, which just translates into a non-linear outcome. This is the very definition for instance of a quadratic (non-linear) effect: doubling the concentration of a chemical does not result in doubling of the biological response (Y~X) but in its quadrupling (Y~X^2): the chemical interacted with itself (AxB=A^2 if A=B) to shape the biological response. You should thus rephrase to clarify your point… or just stick to the next statement, which is clear.

FIGURES

Figure 1A: I think that the growth of multiple-stressor experiments in freshwater systems can be “higher” or “greater”, but not “faster” than that of academic publishing in general.

Figure 1A-inset: what do the provided percentages stand for? Are they compound annual growth rates? By the way, what were the annual growth rates over the whole time period for multiple stressor experiments and academic publishing in general?

Figure 2 (L.659-661): this sentence must be clarified. My understanding is that the numbers in brackets correspond to the number of occurrences (and not "frequencies") of stressors in the corpus, either by stressor class (all occurrences) or by stressor identity (displayed when the number of occurrences is > 5). For instance, the “disease class” in Figure 2 indicates 32 occurrences in the corpus, which is exactly the number I get when summing the corresponding stressor identities from Supporting Table 2…

Figure 3 (L.669): the six stressor classes (habitat alteration, temperature, UV light, nanoparticles, acidity and metals) are not illustrative of “physical, physical-chemical, chemical, chemical-biological, biological, and composite stressors”, as indicated in brackets.

SUPPLEMENTARY INFORMATION

Last sentence of § 3.3 to be checked for clarity

§3.4: “until ASReview suggested 50 irrelevant papers” : do you mean “in a row” ?