Co-occurrence networks reveal more complexity than community composition in resistance and resilience of microbial communities

Plant response to drought stress involves fungi and bacteria that live on and in plants and in the rhizosphere, yet the stability of these myco- and micro-biomes remains poorly understood. We investigate the resistance and resilience of fungi and bacteria to drought in an agricultural system using both community composition and microbial associations. Here we show that tests of the fundamental hypotheses that fungi, as compared to bacteria, are (i) more resistant to drought stress but (ii) less resilient when rewetting relieves the stress, found robust support at the level of community composition. Results were more complex using all-correlations and co-occurrence networks. In general, drought disrupts microbial networks based on significant positive correlations among bacteria, among fungi, and between bacteria and fungi. Surprisingly, co-occurrence networks among functional guilds of rhizosphere fungi and leaf bacteria were strengthened by drought, and the same was seen for networks involving arbuscular mycorrhizal fungi in the rhizosphere. We also found support for the stress gradient hypothesis because drought increased the relative frequency of positive correlations.

Results: starting @ line 165 -"As noted above, the simple fact that fungi grow more slowly than bacteria...." I don't feel that this is a simple matter, bacteria "grow" as single-celled microorganisms through binary fission where, yes, doubling times can range in 10s of mins. Growth for fungi is something completely different; a (sometimes massive) mulicellular (generally) mass of hyphae (a mycelium) that grows by extension at the hyphal tip (unless we are talking about yeasts), where some taxa (e.g., Neurospora) can have relatively high growth rates (e.g., several mm per hour) at the hyphal tip. Therefore, a reductionist approach to growth rates is likely not warranted here.
Results: paragraph @ lines 165-179 -There are no results given here, this paragraphs has elements that may be more appropriate for the Methods section.

Reviewer #3 (Remarks to the Author):
In this manuscript, the authors report the effect of pre-flowering drought, post-flowering drought, and recovery after pre-flowering drought on fungal and bacterial communities and networks in/on roots, rhizosphere soil, bulk soil, and leaves of field-grown sorghum. They hypothesise, based on previous work, that fungal communities and network are more resistant but less resilient than those of bacteria. They test these hypotheses using previously published data for new analyses. They find that their hypothesis that fungal communities are more resistant and less resilient than bacterial communities is supported. Using all correlations between bacteria and fungi in the four compartments, they find that the frequency of positive correlations increased in pre-flowering drought, but using only significant positive correlations (ie co-occurrence networks), they find that pre-flowering drought disrupts networks in roots, rhizosphere and soil but increases their connectivity on leaves. Re-watering resulted in networks resembling control networks again, except for the network in soil (but note that I inferred those results myself from Fig. 3 as I found the description of the results hard to follow). They conclude that understanding microbial network response to stress might inform manipulating microbial communities for increased plant tolerance to stress in agricultural settings.
I enjoyed reading this mostly clearly written manuscript that addresses interesting hypotheses. However, I found the amount of results presented quite overwhelming and not always easy to follow/ interpret. The hypotheses stated are quite abstract and informed entirely by previous work on soil fungal and bacterial communities and network responses to drought, and in that sense the paper reads as largely confirmatory and leans heavily on the results from a few recent papers. I also feel that there is really a severe lack of context on why we want to understand how the communities/ networks in these different plant compartments respond to drought. To me, it would be much more interesting to focus in on the differences between these compartments. What drives the assembly of fungal and bacterial communities on leaves, and how is this different from those in roots and in soil? What would be the implications for their functioning and for plant health of the changes in these communities in response to drought? I am missing all of this in the manuscript, other than quite vague and general statements. I would suggest to focus on this, and I would also suggest ditching the post-flowering drought treatment, as there is no recovery phase after this drought, which makes it difficult to compare these data to the pre-flowering drought.
Moreover, while the manuscript focusses on networks, never is the reliability of these correlations and whether they actually represent interactions between microbes discussed. Positive correlations between microbes can simply indicate niche sharing or responding to the same drivers. Moreover, it is not clear which OTUs were used for correlations (all? Or the ones that occurred over a certain number of experimental units? Or the most abundant ones?), and on how many observations these correlations are based. From the methods it seems that there were 6 replicates of each treatment -does this mean that correlations were based on only 6 data points? Then I would seriously question the robustness of the resulting networks.
In addition, while on close inspection the analyses seem robust and the results are mostly correctly interpreted, I found the figures quite hard to understand as the axes and legends are rather ambiguous. The clarity can be improved, and perhaps also the presentation, because as I said above the amount of data is overwhelming.
More detailed comments: L 164: yes, but also because of their hyphal growth form and thick cell walls, see Schimel et al. 2007 Ecology and Guhr et al. 2015 PNAS. L 175-184 and Figure 1: I found this section very hard to follow. Here, it says that resistance and resilience are calculated as 1-R2, but in the figure Bray-Curtis dissimilarities are reported (are similarities? This is not clear), and in the figure legend it says resistance and resilience. I am lost. It's also not immediately clear what is meant by inter-group and intra-group. L 205: can you be more specific? Which compartments? L 238-244: I found this section very hard to read, as pretty much every sentence mentions that vertices are dropped and rise, but in response to what and compared to what? I assume to drought, but this is never explicitly mentioned. L 252: The biotic interactions become even more complex than the control after rewatering. But is this resilience? Resilience means that the disturbed treatment is approaching or resembling the control. L 315-318: I don't understand this sentence L 325: not just in leaf in post-flowering drought, also in soil and root L 324: De Vries et al. 2018 Nat Comms also analysed combined bacterial-fungal networks -this is detailed in their supplementary material L 327-330: this sentence makes no sense to me. Hypotheses developed from one type of analysis? I would think that it is not about the analysis but about the concept. The analysis is just a means to test a hypothesis. L330-331: again, I have no idea what is meant here. Whole communities hide variation based on compartments? L332-334: I think it is rather stark to make inferences about applications in agriculture from these theoretical hypotheses Methods: I understand that these are previously published data but there's really more detail needed here. How large were the plots? What was the experimental layout? How were samples collected? What other analyses were done? Were there six replicates per treatment, and does this mean that correlations for network analyses were done only using 6 datapoints….?

Reviewer #4 (Remarks to the Author):
Cheng Gao and colleagues in their manuscript 'Resistance and Resilience in Microbes: Cooccurrence Networks Delve Deeper Than Community Composition' address two fundamental questions in the field of microbial community compositions: Resistance and resilience. To do so, they combine two very comprehensive previously published datasets analysing microbial communities on crop plants under extreme drought conditions and irrigation. The datasets are based on 16S and ITS amplicon sequencing and the analyses in the paper is primarily based on pairwise correlations of these datasets. Particularly the question if fungi are more resistant H1 but less resilient H2 than bacteria is certainly a key question in the field and addressed in depth in this manuscript. Besides direct analyses of correlation data, the authors use networks to get deeper insights into community structures. They identify a disruption of communities by drought and see an increase of positive correlations among bacteria, fungi and across kingdoms correlating bacteria and fungi. In combination with network analyses, this gives support for the stress gradient hypothesis. Based on their analyses, they can further underpin the importance of mycorrhiza fungi in stabilizing communities under drought.
In summary, the paper touches a very timely and relevant field and the authors show convincingly that their dataset can be used to infer their central hypothesis H1 and H2. Although I think this manuscript has great potential it would certainly benefit from more details and by addressing some of the following points: 1. As the authors state, key to the paper are pairwise correlations. The authors focus, however, only on Spearman's Rho or Spearman's rank-order correlation. This assumes a monotonic relationship. From the paper it is not clear if the authors have analyzed other correlations to show that this fits the best or have plotted the data to see if this really fits for all samples. Why not using Spearman's correlation, particularly for the networks this might be a better choice or a combination? 2. Further to the correlation analyses: How valid is it to correlate 16S and ITS data together to make conclusions about robustness and resilience? Both will result in completely different resolution. ITS is used to resolve on a species level, 16S will rarely branch that deep. Wouldn't it be better to compare 16S and 18S? Is it possible that bacteria are more resilient because of less resolution, meaning other bacteria move in following rewetting but they are seen as having the same 16S sequence while fungi move back in that show the same taxonomic distance but can be resolved? 3. Very much depends on the calculation of the networks. From the methods I can see igraph has been used and the implemented calculation of networks. To better understand the quality and robustness of the networks it certainly needs more information on the calculation. For example, how was sparsity addressed and how density of the networks. Based on the figures, density is a particular issue, as very dense networks are compared to extremely sparse networks. I would suggest to use at least one other method to calculate the networks correcting for abundance and sparsity or not correcting and comparing those to each other. In my opinion this is relevant to identify if modularity is robust, as this has been debated a lot. 4. As far as I understand from the data sets, the samples are not independent form each other but have a time factor: PRE-Drought, PRE-Rewatering, POST-Drought. To analyze stability it would be useful to track vertices over time and compare PRE and POST networks directly. Particularly positional stability of each vertex would be a good additional measure when comparing different network calculations. 5. A minor thing but relevant to understand what has been done: What are the Guilds and how have they been calculated? I guess this is based on Nguyen et al 2016 but I could not find any information. 6. Question concerning the experimental layout: The experiments have been set up in an area with extremely low precipitation. So any microbe in the soil would be adapted to cope with drought. In this case I would assume that regular irrigation is a perturbation to the community and not drought. Have samples been taken before the planting that could be compared? Is the drought state perhaps a communal 'recovery'? To address the reviewer's concern about the different resolution of 16S and ITS, we compared 21 bacterial 16S OTUs against both fungal communities recognized by ITS OTUs as well as fungal 22 communities recognized at the family level (roughly the taxonomic level determined by 18S 23 rDNA). The results of analyses using either fungal families or OTUs are consistent. Out of total 36 24 comparisons (15 root, 15 rhizosphere and 6 soil), different family and OTUs results were detected 25 in four instances. In two of these, significances detected by OTUs were not detected by family 26 (root, week 4 and 17) and, in the other two cases, significances detected by family were not 27 detected by OTUs (rhizosphere, weeks 7 and 8). In our revised manuscript we report only results 28 that are consistent in both analyses. Importantly, our key findings that fungi are (i) more resistant 29 than bacteria to drought stress but (ii) less resilient than bacteria when the stress is relieved by 30 rewetting are unaffected by this change because of the 23 significant comparisons supported by 31 both analyses from weeks 5 and 9-16 in root and weeks 4-6 and 11-17 in rhizosphere. Ecological resistance to drought stress is detected by comparing compositional dissimilarity of 39 between-group pairs (control-drought pairs) against within-group pairs (control-control pairs and 40 drought-drought pairs) at each of the droughted weeks (weeks 3 -8). Ecological resilience to 41 rewetting is detected by assessing, from before to after rewetting, the change in the difference 42 of compositional dissimilarity between within-group pairs and between-group pairs. Here, the 43 point just before rewetting was week 8 and the points after rewetting were weeks 9 -17. To 44 account for the different resolution of ITS and 16S, we compared bacterial 16S OTUs against both 45 fungal ITS OTUs as well as fungal families. In 32 of 36 cases, the results of fungal families and 46 OTUs are consistent. Different family and OTUs results were detected in two cases where 47 significances detected by OTUs were not detected by family (root, week 4 and 17), and in two 48 cases where significances detected by family were not detected by OTUs (rhizosphere, weeks 7 49 and 8). We report only results that are robust across these two conditions. Significantly higher 50 resistance to drought of fungi than bacteria was detected in root (week 5), rhizosphere (weeks 4 51 -6) and soil (weeks 4, 6 -8). Significantly higher resilience to rewetting of bacteria than fungi was 52 detected in root (weeks 9 -16) and rhizosphere (weeks 11 -17). Note that fungi exhibited 53 stronger resilience than bacteria at the first week of rewetting (week 9). The finding that fungal 54 community composition in soil is not shaped by drought prevented us from further detecting 55 resilience in this compartment. Note that fungal communities in early leaves are excluded from 56 analysis due to the high proportion of non-fungal sequencing reads. The detailed results at fungal 57 family levels can be found in Fig. S1. 58 59 60 GENERAL CONCERN 2additional network analyses (Reviewers 2 and 4),  Soil   3  4  5  6  7  8  9  10  11  12  13  14  15  16  17   3  4  5  6  7  8  9  10  11  12  13  14  15  16  17   3  4  5  6  7  8  9  10  11  12  13  14  15  16 Soil   3  4  5  6  7  8  9  10  11  12  13  14  15  16  17   3  4  5  6  7  8  9  10  11  12  13  14  15  16  17   3  4  5  6  7  8  9  10  11  12  13  14  15  16  We also present the results using the Pearson and CoDa approaches as supplementary 70 information (Fig. S14, S15). 71 72 Our first conclusion, that drought in general disrupts microbial networks, was found in 11 of 13 73 Spearman networks, 10 of 13 CoDa networks, and 9 of 13 of Pearson networks. The two out of 74 13 cases where the Spearman result was not supported by other methods are: i) The BF network 75 in rhizosphere was judged to be disrupted by drought using the Spearman and CoDa methods 76 but was found to be enhanced by drought using the Pearson method; and ii) The FF network in 77 soil was judged to be disrupted by drought using the Spearman method but was judged to be 78 unchanged by the Pearson and CoDa methods. 79 80 Neither did these new analyses have any effect our second conclusion, that co-occurrence 81 networks among functional guilds of rhizosphere fungi and leaf bacteria were dramatically 82 strengthened by drought, because these same results are found in all the three methods.      the case with most field-based experimental designs, it is not possible to assess the effect of 157 habitat filtering and niche sharing. However, we can note that the role of dispersal limitation on 158 the co-occurrence network is weak. Based on our implementation of a taxon-taxon-space 159 association approach, the percentage of network links related to spatial distance was no more 160 than three percent (0 -2.94 %; Figure S13). This result echos the absence of a significant 161 relationship between spatial distance and dissimilarity of microbial community composition 162 reported in our previous study (Gao et al. 2020). Thus, dispersal limitation is not likely the driver 163 of microbial interaction and community composition in our small research site (~480m 2 ), which 164 has been cultivated for nearly six decades and was planted to one crop (sorghum) throughout 165 our study (Gao et al. 2020 Please refer to the full reports below for details. Without substantial revisions, we will be unlikely 172 to send the paper back to review. 173 174 Additionally, another reviewer who did not provide a full report raised a potential concern on the 175 public sorghum drought data that may have been included in the analysis, namely low quality 176 scores of some of the deposited sorghum data. This point should also be addressed. 177 178 GENERAL CONCERN 6low quality scores of some of the deposited sorghum data.

180
Response: We have found a high proportion of non-specific amplification in fungal data of early 181 leaf samples. We removed these data when making this revision of the manuscript. Because none 182 of the results concerning these data are key findings of our report, we no longer report that: i) 183 early leaf fungal community composition was not affected by pre-flowering drought; ii) early leaf 184 fungal correlations was not affected by drought; and iii) early leaf fungal network was not 185 changed by drought. 186 187 Added text: The proportion of fungal reads was low in early leaves (weeks 1-8) due to non-188 specific amplification (Gao et al. 2020), so we excluded these fungal data from our analyses. 189 190 If you feel that you are able to comprehensively address the reviewers' concerns, please provide community composition (although this concern was not among those listed by the editor, it was 198 raised by more than one reviewer). Importantly, our key findings are unaffected by this new analysis. As before, fungi being more 207 resistant to drought stress was supported at week 5 in root, weeks 4-6 in rhizosphere and weeks 208 4, 6-8 in soil, while fungi being less resilient than bacteria when drought stress is relieved by 209 rewetting was supported at weeks 9-16 in root and weeks 11-17 in rhizosphere. 210 211 Added text: We followed the approach of Shade et al. (Shade et al. 2012) to detect resistance 212 and resilience, which had been developed for univariate variables, e.g., richness. For multivariate 213 data, e.g., community composition, we modified it by calculating pairwise community 214 dissimilarity for two groups: within-group (control-control pair, drought-drought pair, or 215 rewetting-rewetting pairs), and between-group (control-drought pairs, or control-rewetting 216 pairs). Ecological resistance to drought stress is detected by comparing compositional 217 dissimilarity of between-group pairs (control-drought pairs) against within-group pairs (control-218 control pairs and drought-drought pairs) for each of the droughted weeks (weeks 3 -8). Ecological 219 resilience to rewetting is detected by assessing, from before to after rewetting, the change in the 220 difference of compositional dissimilarity between within-group pairs and between-group pairs. 221 Here, the point just before rewetting was week 8 and the points after rewetting were weeks 9 -222 17. A t-test was used to assess the statistical significance of the differences in resistance or 223 resilience between bacterial and fungal communities at each time point for each compartment. 224 To account for the different resolution of ITS and 16S, we compared bacterial 16S OTUs against 225 both fungal ITS, species-level OTUs as well the fungal family level (Fig. S1). The results of analyses 226 using either fungal families or OTUs are consistent. Out of total 36 comparisons (15 root, 15 227 rhizosphere and 6 soil), different family and OTUs results were detected in four instances. In two 228 of these, significances detected by OTUs were not detected by family (root, week 4 and 17) and, 229 in the other two cases, significances detected by family were not detected by OTUs (rhizosphere, 230 weeks 7 and 8). (Fig. 1). We report only results that are consistent at both the species and family 231 levels (Fig. 1). 232 In line with our first hypothesis, H1, we found that the resistance to drought stress for 233 fungal mycobiomes was consistently stronger than that for bacterial microbiomes for weeks 5 in 234 root, weeks 4 -6 in rhizosphere, and weeks 4 and 6 -8 in rhizosphere (Fig. 1, S1). In support of 235 our second hypothesis, H2, when the stress of pre-flowering drought was relieved by rewetting, 236 we found that the resilience for the bacterial communities was consistently higher than that for 237 the fungi in weeks 9 -16 in root, and weeks 11 -17 in rhizosphere (Fig. 1, S1). 238 Surprisingly, we found that resilience was stronger for fungal than bacterial communities 239 in the first week (week 9) of rewetting in rhizosphere (Fig. 1, S1). This high resilience of fungi may 240 be associated with the quick growth of sorghum roots when rewetted. The rhizosphere zone 241 around these newly formed roots may be quickly colonized by soil fungi, a community that was 242 weakly affected by drought. This result suggests that re-assembly of rhizosphere microbial 243 community is more complex than previously expected. 244 The finding that fungal community composition in soil is not shaped by drought prevented 245 us from further detecting resilience (Fig. 1). Note fungal community in early leaves was excluded 246 from analysis due to the high proportion of non-fungal reads in sequencing (Gao et al. 2020). 247 248

Fig. 1. Resistance and resilience of bacterial and fungal community composition. 250
Ecological resistance to drought stress is detected by comparing compositional dissimilarity of 251 between-group pairs (control-drought pairs) against within-group pairs (control-control pairs and 252 drought-drought pairs) at each of the droughted weeks (weeks 3 -8). Ecological resilience to 253 rewetting is detected by assessing, from before to after rewetting, the change in the difference 254 of compositional dissimilarity between within-group pairs and between-group pairs. Here, the 255 point just before rewetting was week 8 and the points after rewetting were weeks 9 -17. To 256 account for the different resolution of ITS and 16S, we compared bacterial 16S OTUs against both 257 fungal ITS OTUs as well as fungal families. In 32 of 36 cases, the results of fungal families and 258 OTUs are consistent. Different family and OTUs results were detected in two cases where 259 significances detected by OTUs were not detected by family (root, week 4 and 17), and in two 260 cases where significances detected by family were not detected by OTUs (rhizosphere, weeks 7 261 and 8). We report only results that are robust across these two conditions. Significantly higher 262 resistance to drought of fungi than bacteria was detected in root (week 5), rhizosphere (weeks 4 263 -6) and soil (weeks 4, 6 -8). Significantly higher resilience to rewetting of bacteria than fungi was 264 detected in root (weeks 9 -16) and rhizosphere (weeks 11 -17). Note that fungi exhibited 265 stronger resilience than bacteria at the first week of rewetting (week 9). The finding that fungal 266 community composition in soil is not shaped by drought prevented us from further detecting 267 resilience in this compartment. Note that fungal communities in early leaves are excluded from 268 analysis due to the high proportion of non-fungal sequencing reads. The detailed results at fungal 269 family levels can be found in Fig. S1.  This study investigated the resilience and resistance of Sorghum-associated bacterial and fungal 278 communities against drought. The strength of the study for me is that it targeted both bacteria 279 and fungi and studies these communities in all relevant soil and plant compartments (soil, 280 rhizosphere and leaf). The authors rightly point out that most studies have focused on bacteria 281 alone, and often on a single microbial compartment. 282 I found the study interesting and believe it will interest others in the field, as there is considerable 283 interest in understanding resilience and resistance in microbial communities, and this study's 284 comprehensive experimental design makes it a likely important article for those in the field. 285 286 The authors analysed and discussed positive microbial interactions and related this to resilience 287 and resistance. As many of the studies that linked positive interactions with resilience and 288 resistance are based on macro-ecological studies, it would be relevant to provide some context 289 for microbial studies which also looked more specifically at positive associations in networks. 290 There are a few studies that looked at the ratio of positive interactions in microbial networks in 291 relation to ecological status, especially ecological succession, and the authors didn't mention 292 these studies. I suggest including some of these studies in their discussion: e. Response: we agree with the reviewer that correlation does not equate with interaction but can 331 be ascribed to habitat-filtering or dispersal limitation. However, as pointed out by the reviewer, 332 it is not possible to remove the effect of habitat filtering in our case. Regarding dispersal 333 limitation, we used a taxon-taxon-space approach to find that only a small amount of network 334 links (0 -2.94 %) is related to spatial distance (Fig. S13). These results are consistent with our 335 study in a homogenous, ploughed, one crop farmland. 336 337 Added text: Please see General Concern 5, above. 338 339 In lines 176-179 the authors state that resistance is 1-R2 (when comparing control and droughted 340 communities) and resilience is 1-R (when comparing control and re-wetted communities). 341 However, in lines 180-186 (and in the figures and tables), the authors detail and discuss R2 values, 342 rather than 1-R2. Perhaps this can be simplified? Since R2 is related to the level of change 343 between treatments, perhaps the 1-R2 definition is not needed? 344 In Figure 1 is the significance indicated in every compartment for what comparison exactly? Could 345 the authors detail this in the legend? As the authors use the R2 as a measure of resilience and 346 resistance, to claim for instance that "that the fungal mycobiome is more resistant than the 347 bacterial microbiome to both pre-and post-flowering drought", it would be important to show 348 that these differences in R2 between bacteria and fungi are significant. 349 350 Response: We appreciate the reviewer's suggestion and recognize the confusion caused by the 351 usage of 1-R 2 . Now we directly calculate resistance and resilience following the methods of Shade 352 et al 2012 and have removed the text about 1-R 2 throughout our ms. As a result of this change, 353 we now use a T test to assess significance in the differences in resistance and resilience between 354 fungal and bacterial communities. 355 356 Added text: Please see General concern 7, above 357 358 consider, "For example, it has been proposed that positive microbial interactions should increase 490 in frequency under stress scenarios, such as drought, a response explained by the stress gradient 491 hypothesis (SGH). Further, stress studies of microbes on Arabidopsis leaves, roots, and the 492 surrounding soils suggest that within-taxonomic group microbial interactions tend to be positive, 493 while those between-taxonomic groups are negative. Ecological modeling also indicates...." 494 Further, microbial interactions, which biological/ecological in nature, should not be confused 495 with correlation, which is simply a statistical method. For example, positive correlations related 496 to shifting microbial abundances might be interpreted as mutualist interactions (or facilitation), 497 while negative correlations might be interpreted as antagonistic interactions (or competition). 498 The paper tends to confuse these concepts a bit (see comments immediately above and below), 499 and the authors should bear in mind that they are attempting to view/interpret microbial 500 interactions through the lens of statistical correlation (e.g., correlations metrics are appropriate 501 for the results, but the interpretation (i.e., in discussion) should focus on the interactions. 502 503 Response: We thank the reviewer for the suggested revision of text as well as the interpretation 504 of the results regarding correlation and interaction. We accepted all these suggestion in 505 preparing the revised ms. 506 507 Added text: Please see General Concern 5, above. 508 509 Introduction: starting @ line 112 -"Using these studies to frame hypotheses at the all-correlation and state that fungi are "(i) more resistant but (ii) less resilient than bacteria" (we assume this 526 refers there respective status under the stress or drought), while the H1 and H2 mentioned here 527 focus on interactions. 528 529 Response: Our hypotheses and the ways in which we evaluate them are a bit more complex than 530 presented by reviewer #2. 531 532 We test our hypotheses that "fungi are (i) more resistant but (ii) less resilient than bacteria" at 533 three levels: a) using community composition, b) using all-correlations (we follow de Vries 2018), 534 and c) using just correlations limited to those that are significant and positive as determined from 535 a co-occurring network. In the part referred by reviewer #2, we focused on the test of these two 536 hypotheses at the all-correlation level. 537 538 Based on the stress-gradient hypothesis (stress increases frequency of positive microbial 539 interactions), the hypothesis that fungi will be more resistant than bacteria can be extended from 540 the community composition level to the all-correlation level. The expectation is that drought will 541 increase the proportion of positive correlation more strongly for B-B correlations than F-F 542 correlations. It is also possible to extend the Resilience hypothesis (Bacteria > Fungi) to the all-543 correlation level, i.e., rewetting will decrease the proportion of positive correlations more 544 strongly for B-B correlations than F-F correlations. 545 546 The original framework for evaluating resistance (Fungi > Bacteria) and resilience (Bacteria > 547 Fungi) was limited to interactions within fungi or within bacteria and did not have expectation on 548 the interaction between bacteria and fungi (B-F). We added these inter-domain interactions 549 based on the results of research on Arabidopsis (within-taxonomic group microbial interactions 550 tend to be positive, while those between-taxonomic groups are negative) and ecological 551 modeling (negative interactions promote stability). In adding B-F interactions to resistance, we 552 hypothesized that drought would increase the proportion of positive correlation more strongly 553 for within-taxonomic group microbial interactions (B-B and F-F) than between-taxonomic groups 554 (B-F). In adding B-F interactions to resilience, we hypothesized that rewetting would decrease 555 the proportion of positive correlations more strongly for within-taxonomic group microbial 556 interactions (B-B and F-F) than between-taxonomic groups (B-F). 557 558 Putting all these items together, our resistance hypothesis is that "H1, under drought we expect 559 an increase in the proportion of positive correlation most strongly for B-B, followed by F-F, and 560 lastly by B-F correlation"; and our resilience hypothesis is that "under rewetting, we expect a 561 decrease in the proportion of positive correlation most strongly for B-B, followed by F-F, and 562 lastly by B-F correlation" 563 564 We do not expect that "drought [would] enhance facilitation within taxonomic groups (i.e., 565 positive correlations for B-B and F-F) and enhance competition between taxonomic groups (i.e., 566 negative correlation for B-F using these studies to frame hypotheses focusing on all-correlations, for our resistance 589 hypothesis, H1, under drought we expect an increase in the proportion of positive correlation 590 most strongly for B-B, followed by F-F, and lastly by B-F correlation; and for our resilience 591 hypothesis, H2, under rewetting, we expect a decrease in the proportion of positive correlation 592 most strongly for B-B, followed by F-F, and lastly by B-F correlation. 593 594 Introduction: paragraph @ line 118-136 -I find this paragraph to be confusing and repetitive with 595 respect to the hypotheses (and see above) overall, the discussion of "nonintuitive outcomes" is 596 a bit obtuse and appears to be splitting hairs (to justify results/methods?). Also, "Simplifying 597 matters by focusing on just the significant, positive correlations" -if a correlation is not significant 598 then it should not be considered as a result at all; further, the paragraph above and H1/H2 stress 599 the importance of validating negative correlations. This paragraph appears to be justification for 600 the methods used in the co-occurence network analysis part of the study, but the case could be 601 more clearly and directly made (i.e., this is a common method for such analyses). 602 603 Response: We agree with the reviewer and simplified this paragraph. We follow the approach of 604 de Vries by including both significant and non-significant correlations in all-correlation analysis, 605 and only significant, positive correlations in the co-occurrence network. 606 607 Revised However, in response to one of my other comments, it appeared that the networks in this study not only include datapoints from the 6 true field replicates, but also lump together the various time points during the progressing drought (6 time points over 6 weeks) and during the recovery period (8 timepoints over 8 weeks). This approach is not mentioned explicitly and not justified, and it seems rather inappropriate to me. It is clear that during those periods, microbial communities go through large changes (as can be seen in Fig. 1, although no information is presented on shifts in community composition here) and not only am I wondering what networks of these combined time points actually represent, as far as I am aware, no other studies constructing networks have lumped time points, which means that they can't be compared to these. This also bring me back to my most important issue, which is that it is hardly explored what these networks/ interactions actually mean ecologically.
Reviewer #4 (Remarks to the Author): Response to reviews of our revised manuscript. Reviewer #1 (Remarks to the Author): The authors have carried out a considerable revision of their manuscript, and in general, have 11 addressed most of my concerns. I have some remaining concerns, which I detail below. 12 13 I find interpreting Figure 1 and S1 difficult. I particularly struggled with the shaded vs unshaded 14 data. Could the authors help the reader somehow, perhaps by indicating in the text discussing 15 the figure whether they are referring to the shaded or unshaded parts of the graph? 16 17 Response: We appreciate this comment and have revised the figure and figure legend to clarify 18 matters. 19 20 Lines 791 and 795 in change-tracked manuscript: Revised legend of Figure 1 and also of 21 Figure S1: Ecological resistance to drought stress is … at each of the droughted weeks (weeks 3  22 -8, the grey shaded area). Ecological resilience to rewetting is … after rewetting were weeks 9 23 -17 (the gold shaded area). 24 25 Regarding the general concern 3 about correcting p-values with FDR. This seems an appropriate 26 response, however, without statistics regarding how many nodes or edges were removed it is 27 hard to assess the impact of FDR in their networks. 28 29 Response: We assessed the impact of applying a FDR to network structure and provided the 30 results in the supplementary Table S3. Out of the 64 networks examined, 16 were affected by 31 FDR correction, and the proportion of edge removal ranged from 19.49% to 94.76% and the 32 proportion of vertices removal ranged from 10.84% to 90.40%. Information added in line 573 of 33 change-tracked manuscript 34 Fig. S14 Minimal spurious association was detected in using the approach of Coenen et al 2020 from 6 and 8 independent random walks over 6 temporal series. The analysis was repeated 10 times and results from one run are shown here. (A) Six time-series of six independent random walks mimicking the drought period. (B) For the 15 correlations among six time series of six independent random walks, at most 0-1 significant spurious associations were detected (none were found in this example). (C) Six time-series of eight independent random walks mimicking the rewetting period. (D) For the 15 correlations among six time series of eight independent random walks, at most 1-3 significant spurious associations were detected (The one in this run is marked with an asterisk in the example). First, as with Dai et al 2022, we used the MENAP to comparison the empirical network against random networks, finding that all networks are non-random (Table S4).
We propose adding text to our manuscript and a supplemental table S4, as shown below.
Added text in lines 568-570: In addition to FDR, we used Random Matrix Theory (RMT) to assess the robustness of correlations as implemented in the Molecular Ecological Network Analyses Pipeline (MENAP) 70 . We found that all empirical networks were non-random (Table S4). Random networks were generated at the Molecular Ecological Network Analyses Pipeline (MENAP) by randomly rewiring all the links while keeping the numbers of nodes and links of the empirical network.
Next, we compared the association of networks based on Spearman correlations as filtered by either the FDR or RMT approaches. As shown in the following figure of average degree, the results of these two different methods are consistent. The results of the two methods continue to support our first conclusion, that drought in general disrupts microbial networks. This result was found in 11 of 13 FDR networks, and 10 of 13 RMT networks. There was only one inconsistent case, concerning roots during drought, where the FF network showed disruption using the FDR approach but was unchanged using the RMT approach. We propose adding text to our manuscript a supplemental figure S15, as shown below.
Added text in lines 571-573: We then compared the association networks based on Spearman correlations as filtered by either the FDR or RMT approaches, finding that results of these two different methods are consistent in terms of drought response (Fig. S15-S16).

Fig. S15
Consistent responses to drought of average degree of association networks based on Spearman correlations as filtered by either the false discovery rate (FDR) or random matrix theory (RMT) approach. Note that in only one case, roots, is there disagreement where the FF network showed disruption using the FDR approach but was unchanged using the RMT approach.
Finally, neither did application of the new, RMT analyses affect our second conclusion, that cooccurrence networks among functional guilds of rhizosphere fungi and leaf bacteria were dramatically strengthened by drought, because these same strengthening is found with both approaches. We propose adding a supplemental figure S16, as shown below.

Fig. S16
Spearman Rho co-occurrence networks of rhizosphere fungi and leaf bacteria were dramatically strengthened by drought, whether measured by FDR-or RMT-based approach.
To reiterate, both FDR and RMT approaches support the key findings that: (i) In general, drought disrupts microbial networks based on significant positive correlations among bacteria, among fungi and between bacteria and fungi. (ii) In contrast, co-occurrence networks among functional guilds of rhizosphere fungi and leaf bacteria were dramatically strengthened by drought.