The authors did a very good job in the revision, I really like the new Table 2.

Just some minor things:

1. You may want to add Nyman et al. 2019 (Early wasp plucks the flower: disparate extant diversity of sawfly superfamilies (Hymenoptera: ‘Symphyta’) may reflect asynchronous switching to angiosperm hosts.) to #6 in Table 2 – to add to the butterflies also a sawfly/hymenoptera study.

2. Figure 2 is cool - a possible problem with c) scenario H1 and H2 is that H1 may still show adaptive radiations in morphology even if shifts in diversification are not detected – following your definition of adaptive radiation in the glossary (although it looks like all recent clades will show a diversification rate shift as compared to the background rate). And similarly, H2 would possibly show the fastest rate of diversification in the range 3 clade (opposite of what is suggested), because of the long branch suggesting extinction but also fast speciation (high turnover), whereas range 1 and 2 show gradual accumulation of diversity – and thus not necessarily adaptive radiation. Last, for the radiation to be adaptive, one would have to show that within the range there is partitioning of resources (e.g. through the reconstruction of traits related to this partitioning – comparable to the beak-size in Darwin finches), something that is currently missing.

Jousselin et al. give an interesting (and probably quite comprehensive) overview of hypotheses and studies testing phytophagous insect diversification in relation to their host plants. This is an interesting topic, and the review is generally clear and well written (could be a bit more concise in places, and needs in parts clarification, check for a couple of minor spelling mistakes, some I indicated below). I especially like the Perspectives part.

I have a couple of suggestions to improve the ms, indicated below. More generally, it may be useful to make a table indicating support for the different hypotheses (e.g. for different study groups/taxonomic groups [would be useful to use taxonomic symbols/silhouettes], and indicate the reference, and the method used, for example) (this would support the text with the sum-up of examples, which is quite long at the moment). The same (a table, or maybe better a figure) could perhaps be done for the Perspectives ideas.

Other comments:

L48 Not clear whether hybrids will thus be selected against, or that this intermediate phenotype has the potential to become a new species through reproductive isolation. Or both. In case of both, perhaps some more information about the commonness of ‘selection against hybrids’ vs. ‘hybrids become new species’ can be presented.

L76 This overview is interesting, but not particularly novel. To really move forward in this field, I guess it would be great to incorporate the available fossil evidence, to calibrate the radiations / estimate diversification rates, but perhaps also to obtain a better idea of insect-hostplant associations in the past, if such things fossilize (or co-occurrence of species through insect/plant fossil assemblages perhaps).

L78 an extra ‘and’

L88 Why only speciation rates? I would say diversification rates, which would include both speciation and extinction. I imagine certain host-plant relationships are less vulnerable to extinction (e.g. more generalized as opposed to specialized relationships?)

L174 less fewer

L182 The musical chairs hypothesis needs a bit more detail. For example, the description does not really match the figure (1c), and it’s only indicated how it differs from the second hypothesis, not how it differs from the first, and more importantly, what makes it ‘unique’.

L203 What about the methods that test for host switching and co-diversification – are those not required to test some of these hypotheses as well?

L212 insect insects

L220 But a lack of support for coevolution has also been often found, or?

L230 and more generally – some predictions remain quite vague and could do with a bit more detail. For example, concepts/terms such as ‘high lability’ (in speciose lineages breadth) high compared to what? ‘Few transitions in host breadth’ – few compared to what? What is, and how does one measure ‘lability in host association’? ‘following the capture of a new host-plant lineage’ does that not directly indicate that there is host tracking? Or how can there be shifts to new host-lineages without tracking? ‘Conservatism of host-plants’ – compared to what? Conservatism in what? How does conservatism shift on a phylogeny – is it (or linked to) ‘intrinsic’ traits? ‘recurrent transitions in host breadth’ – what is recurrent, how often, how commonly?

In short – always make the comparison, i.e. larger (e.g. geographical range) than what. Make clear what ‘lability’ and ‘conservatism’ are – these are vague and difficult concepts.

L234 host-plant adapted traits – such as?

L237 But I guess these studies have not found support for all predictions from the hypothesis, or? Maybe write something about the methods they used and how reliable their conclusions are.

L265 How can host history be mapped on a tree when there are multiple associations per species? Or how are host associations classified?

L282 What if the host shift is to the nearest living relative of the current host plant? Or would that be called host tracking instead of host switching? Perhaps make clear.

L294-295 rephrase sentence

L300 few species of what? Plants? Angiosperms?

L326 I wonder whether the availability of (suitable, e.g. chemically matching) hosts also plays a role in the host-breadth evolution of insects. This therefore depends on the diversification of particular host plant lineages. Perhaps this could be mentioned here, or elsewhere.

L333 Here, and more generally: what is specialized and what is generalized? Should that always be in comparison to something else (e.g. other species or previous/ancestral state?)

L349 Why would these be the result of range expansion? What is that based on?

L378 disruptive selection to what - host plants? Make explicit, because disruptive selection could have still played a role during speciation.

L438 I wonder whether it may not be just phytophagy vs. non-phytophagy driving the high diversification rates, but phytophagy in interaction with other (extrinsic, or intrinsic) traits (such as host plant diversification / availability). This would require more sophisticated trait and interaction-dependent diversification models. Or, alternatively, a similar trait-based approach to Onstein et al. 2017 (Nature Ecology & Evolution) could be taken, in which the authors infer speciation rates based on important interaction-relevant traits (in that case of frugivory). Those traits would tell you something about the interaction (e.g. chemical traits, phytophagy and or host-breadth) as well as the consequence for speciation based on host-plant characteristics (e.g. host-plant diversity or diversification, or their variation in chemical traits).

L498 changed changes

L509 geography, or other factors. Explain why geography is likely in this case.

L591 Is such a phylogeny available?

L735 Indicate which definition of adaptive radiation you follow – in some cases the radiation does not have to be rapid, just morphologically. Rapid refers to the necessity to also test for a shift in diversification rate at that point in time, which was not mentioned in the main text.

L747 What is a ‘wide niche’? And what a ‘narrow niche’?

This paper is an important contribution: it goes beyond the hand waving arguments for diversification to get at which mechanisms may credibly lead to observed patterns.

Major suggestions

A persistent problem throughout the paper is that diversification and speciation are treated as synonymous, but they are not. For example, one idea for escape and radiate is not that the plant speciation rate would go up (for what it's worth, I agree with the authors that this probably wouldn't happen) but extinction rate of plants could go down (less herbivore pressure). Both would increase diversification; only one relates to speciation. I would go through and any time there is diversification, change it to speciation if that is meant; if that is not meant, think about how varying extinction could lead to a diversification pattern.

I would love another table: scenario on the rows, papers on the columns, and a symbol indicating which hypotheses were examined and which ones were supported. Much of the text is a verbal description of this, but doing a table like this would really help clarify the results: in situations were both hypothesis A and C were evaluated, C won 75% of the time, but comparing B and C, C won 17% of the time. One could lump the columns by focal taxon (butterflies, beetles, etc.). It would also help highlight which hypotheses have not been examined much, and which ones have been only looked at on their own without rigorous comparison to others.

Lines 289-313: an important caveat is that methods looking at host shifts and speciation shifts will miss some: a shift, then a shift back, for example. I suspect that missed speciation events are far more common than missed host shifts -- if anything, multiple missed speciation events and still few host shifts is even more support for the idea that most speciation doesn't involve a host shift, supporting the conclusion. I'd still mention this caveat. It also comes into the discussion on 321-325 -- one problem with these approaches is they assume all speciation events and host switches are on the tree, which is not true (it makes me wary of such methods).

One caveat about diversification models that could be more strongly emphasized is that under any of these models where phytophagy affects diversification rate, we don't expect that is the ONLY factor (this could be relevant to bring up around line 451). Gaining or losing flight likely has a substantial effect on diversification rate, for example, and that this happens sometimes does not mean phytophagy does or does not also matter.

The interaction network approach is compelling -- I am glad it is in this paper.

Table 1 is a good contribution.

The paper would be stronger if there were a section explaining more clearly how to compare the models in Table 1 in one study (this related to line 707: the paper argues for a standardization of these predictions, but show better how this can be used in practice). Say I have a fairly well sampled insect phylogeny where I have host families for all of them. What parameters, model selection, would help me select which macroevolutionary scenario best explains my data? Some of the predictions in table 1 are vague ("high lability", "few transitions", "conservatism") -- how do these work operationally?

Minor suggestions

Line 162: Figure misspelled

Line 553: "hwas" typo.

Line 557: Shouldn't be parentheses around author names

I would say BiSSE and related methods are referred to more as SSE or *SSE rather than -SSE methods.

Line 613: This is a good point, but it's even worse than that: there could be many changes and it's still not robust to say a state led to a diversification rate increase. See Beaulieu & Donoghue (2013) https://doi.org/10.1111/evo.12180 -- a single change in one clade that correlates with diversification increase could make it seem like that change always increases rate, but it's not true.

615: I'm not sure sister group comparisons are "very" sensitive to sampling biases in the way SSE models are. Sensitive, yes -- but when one is comparing a clade of 1000 vs one with 50, you'd have to do some pretty extreme sampling to get the wrong sign.

Closing on transcriptomes was odd -- it's not bad, but I could see counter reasoning (selection in the present doesn't necessarily correlate with selection when switching hosts; this does not get at the speciation genes). It does not mean cut this, only that I might move this somewhere less prominent -- it's an approach, not the ultimate approach.

Line 717: should be sister lineages.

Left align the contents of Table 1: justified looks terrible