The manuscript entitled "Light availability and light demand of plants shape the arbuscular fungal communities in their roots" by Neuenkamp et al. describes an experiment aimed at evaluating the importance of light restriction as a determinant of AMF communities found in roots. This was done by both evaluating the importance of the light environment (by distinguishing between naturally shaded and exposed sites), and by selection of sets of plant species that are more or less light tolerant. Through a factorial sowing experiment they could then ask whether plants that naturally perform better in a shaded environment, have more AMF taxa than those that do not and vice versa for more light-exposed environments. This is an interesting question, which can "shed light" on our fundamental understanding of AMF community assembly and how plant-AMF interactions relate with this.

The experiment is well designed, with an acceptable level of replication of plant species and in multiple blocks. Also in general, the research question is well conceived and elaborated on in the MS. However, the presentation of the results, and the support it provides to the conclusions, is problematic. The main conclusions reported in the abstract (L55) are that "AM fungal community composition differed markedly in response to light conditions and host plant shade tolerance", and that "Root AM fungal diversity peaked in plant species growing in light conditions that were close to their ecological optima". Both of these results are presented in Appendix S6 – the ordination shown there doesn't show a marked difference between either light conditions or plant-shade tolerance types (conclusion 1), nor do the boxplots show a difference in richness between the two plant types or the different habitat type (conclusion 2). The reason for the mismatch between these data (which are more reliable – but see comments below) and the data used to produce plots and Venn's in the MS (and the PERMANOVA stats in Appendix S11), is that in the latter case data are further aggregated into these functional groups and total taxa observed counted. This is problematic, because 1) the dataset is unlikely to be perfectly balanced meaning some groups (plant type/site/block combinations) may have higher total richness just because of read numbers or sample numbers alone, and 2) replications is reduced to N=1 making it impossible to statistically underpin results.

The way the MS is currently written and the hypotheses are set up (which all clearly allude to individual based plant-fungal interactions, not some population level or even functional-group level processes), there is just no justifcation of testing these individual-based questions with group-level aggregated data. And besides, if it were justified, accounting for variation in sample # read #, etc. that are bound to be there, should be done much more rigorously and described better to make a convincing case that although N=1 (or four? if the four replicate blocks weren't forgotten beyond M&M;?) the patterns shown are real.

Concludingly, the conclusions may very well be true and the topic is worth pursuing, however the MS is currently unconvincing in supporting them and it is doubtful that when analyzed properly the data will do it (judging from Supplementary figure 6). My sympathy goes out to the authors because I know how frustrating this is, but there's no way around this conclusion I fear...

Specific comments
L84: here a general description of expectations concerning the effect of shading on plant-mycorrhizal fungal interactions are given. In this sentence, it is described that plants may to a lesser extent allocate C to high-quality partners, because allocation is shifted to other organs. This doesn't seem very logical; given that C becomes more limited shouldn't that actually lead to higher "pickiness"?
L150: I see how regionally collected seeds are sown into a restricted area within plots. However I also understand that these same species occur locally. How were the seeded individuals distinguished from the locally occurring ones?
L183: roots from bait plants were subjected to DNA isolation, where 1-3 individuals were mixed. Why is that? To obtain theh 70 mg of roots needed? This seems tricky because the interpretation is such that it assumes AMF richness colonizing a plant individual (more on that later). At the least, a convincing case should be made that just this fact alone hasn't caused some of the richness changes (that a difference in sampling intensity on plant individual level is not responsible for the apparent changes in richness).
L199: please clarify what you mean by singletons; VTs represented by a single read globally?
L213-217: this is the most tricky aspect of the study; that all sequences obtained from a given species/light-level combination are aggregated; from this point onwards, within-individual richness and between-individual/plot heterogeneity get confused and detached from the research-question and hypotheses (which engage with within-individual processes such as symbiont selection). Moreover, imbalances of plant individuals sampled will contribute to richness estimates.
L224: sequences = reads
L232: this is confusing; I think the following is done: (unrarefied) samples are aggregated to species level -> calculating species level AMF range (L217). Then, individual samples are rarefied -> aggregated at species/light/year level (L233). If this is the case, the question arises why the data for Appendix S4 (missing though?) were not rarefied on beforehand? If something else is done, please clarify....
L235: here it is suggested that differences in # of plant individuals sampled should be somewhat reflected in read count, and that essentially accounting for read count should negate this problem. I am not convinced by this. It would be much better standardize (rarefy) reads of individual samples (and calculate alpha diversity), without the need to additionally account for # of individuals (because why is gamma diversity of interest here anyway?). Not familiar with the iNEXT R package, does this workflow account for differences in # of samples per species?
L251: here I get confused again: so now samples are first aggregated by plant species (but surely not pooling treatments I suppose????), and only then rarefied? And then rarefied to median reads # per plant species - I don't understand this either, because not all comparisons are at the within-species level, for instance using shade-tolerance as a factor means you are comparing communities between groups of species. Therefore, all samples should be rarefied to the same level.
L290-91: this sentence is weird (with one species name and otherwise VT's) – please rephrase
L328: here the use of "conversely" leads to an ambiguity: you mean dissimilar in non-preferred habitat, or also similar but different?
L424-26: here it is alluded to that heterogeneity may cause changes in cumulative species richness (regardless of within-root AMF richness). This is another good reason not to conflate within-root and across root richness.
L455: Werner & Kiers 2015, not 2005.
Box 1: here I miss the mycocentric perspective: it is not unthinkable that some fungi opt-out from engaging with plants that are unable to allocate resources (e.g. shaded light-adapted plants), instead of it being a matter of "plant choice".
Box 2: concerning fungal culturability; this whole topic seems orthogonal to the whole question and results of the paper. The same goes for the results related to this. It should be better integrated if possible or otherwise omitted.
In Appendix S9B it can be seen that shade-avoiding plants in shaded sites are represented by fewer reads than shade-tolerant plants (at least if the plot labels are right; the legend says it's both light-environments pooled). This is perhaps expected, because their performance is lower in those sites, but it does pose a strong challenge to just pooling this data and estimate richness (which is underestimated for the shade-avoiding plants because it has fewer sequences).