Parasites as niche modifiers for the microbiome: A field test with multiple parasites

Parasites can affect and be affected by the host’s microbiome, with consequences for host susceptibility, parasite transmission, and host and parasite fitness. Yet, there are two aspects of the relationship between parasite infection and the host microbiome that remain little understood: the nature of the relationship under field conditions, and how the relationship varies among parasite species. To overcome these limitations, we assayed the within-leaf fungal community in a grass population to investigate how diversity and composition of the fungal microbiome are associated with natural infection by fungal parasites with different feeding strategies. We hypothesized that parasites that more strongly modify niches available within a host will thereby alter the microbial taxa that can colonize the community and be associated with greater changes in microbiome diversity and composition. A parasite that creates necrotic tissue to extract resources (necrotrophs) may act as a particularly strong niche modifier whereas one that does not (biotrophs) may not. Barcoded amplicon sequencing of the fungal ITS region revealed that the microbiome of leaf segments that were symptomatic of necrotrophs had lower fungal diversity and distinct composition compared to segments that were asymptomatic or symptomatic of other parasites. There were no clear differences in fungal diversity or composition between leaf segments that were asymptomatic and segments that were symptomatic of other parasite feeding strategies. This supports the hypothesis that within-host niches link infection by parasites to the host’s microbiome. Together, these results highlight the importance of parasite traits in determining parasite impacts on the host’s microbiome.


Introduction 57
The fungi, bacteria, and viruses that comprise an organism's microbiome can interact 58 with parasites in ways that have consequences for host susceptibility, disease severity, and 59 transmission (Berg & Koskella, 2018). Yet, while investigations of the association between the 60 microbiome and a single parasite species are becoming increasingly common (Aivelo & 61 Norberg, 2018;Libertucci & Young, 2019), how such associations vary among parasite species 62 is still poorly understood. Here, we evaluated potential mechanisms through which parasites alter 63 the fungal microbiome using a trait-based analysis of multiple co-occurring fungal parasites in a 64 grass host. Specifically, we test the hypothesis that parasites with a trait that more strongly 65 modifies the environment within the host act as niche modifiers and more strongly impact 66 microbiome diversity and composition. 67 Microbes may interact with parasites by competing for resources, releasing antimicrobial 68 compounds, or eliciting a host immune response (Graham, 2008;Raaijmakers & Mazzola, 2012;69 Bashey, 2015). Such interactions can influence host health, making the host more susceptible to, 70 tolerant of, or resistant to parasites (A. E. Arnold et al., 2003;Busby, Peay, & Newcombe, 2016;71 Hayes et al., 2010). The microbiome is also dynamic, and the introduction of a parasite can lead 72 to a change in microbiome diversity and composition (Barman et al., 2008;Jani & Briggs, 2014). 73 The link between parasites and host microbial diversity varies, with some studies showing no 74 relationship (Li et al., 2012;Williams et al., 2017), some studies showing a negative relationship 75 (Jani & Briggs, 2014;Leung et al., 2018;Mosca, Leclerc, & Hugot, 2016;Wu, Stanley, Rodgers, 76 Swick, & Moore, 2014), and still others showing a positive relationship between parasite 77 infection and within-host microbial diversity (Lee et al., 2014). One possible source of this 78 variation among studies is variation among the parasite species studied. Yet, there is a lack of 79 studies that examine variation among parasite species in their associations with host microbiota 80 (but see Aivelo & Norberg, 2018), perhaps because we have lacked a robust framework for 81 studying multiple parasites. 82 A trait-based approach might provide a framework for studying multiple parasite species 83 (Zanne et al., 2019). As well as being evolutionarily diverse (Weinstein & Kuris, 2016), parasites 84 vary in traits such as growth rate, generation time, and feeding strategy (Leggett, Cornwallis, 85 Buckling, & West, 2017). Different parasite feeding strategies can stimulate different immune 86 responses in a host and differently impact host performance (Glazebrook, 2005;Budischak, 87 O'Neal, Jolles, & Ezenwa, 2018;Halliday, Umbanhowar, & Mitchell, 2018). Parasites infecting 88 plants employ three typical feeding strategies. Parasites with a biotrophic feeding strategy keep 89 host cells alive to extract resources from them. Parasites with a necrotrophic feeding strategy kill 90 host cells to access their resources, creating necrotic tissue while they grow within their host. 91 Finally, hemibiotrophic parasites infect as biotrophs, then switch to become necrotrophic. Given 92 that parasites with different feeding strategies have different impacts on the host environment, 93 parasites with different feeding strategies may also have different associations with the host 94

microbiome. 95
Because feeding strategy can define how a parasite impacts the host environment, a trait-96 based approach grounded in parasite feeding strategy might help explain why some parasites 97 have larger impacts on the within-host microbial community than other parasites. When an 98 organism modifies its environment, it can change the number and types of niches available, and 99 in turn, impact the species that can reside and colonize within the ecological community via the 100 process of niche modification (Lewontin, 1983;Fukami, 2015). Niche modification is closely 101 related to the concepts of niche construction (particularly with respect to evolutionary 102 implications, Odling-Smee, Laland, & Feldman, 2003) and ecosystem engineering (Jones, 103 Lawton, & Shachak, 1994), and has been documented in numerous communities of free-living 104 organisms (Naiman et al., 2009;Fukami & Nakajima, 2011). Parasites may also act as niche 105 modifiers within host individuals by impacting the host environment in such a way that the host 106 is more or less suitable to new colonizers. Specifically, a parasite that produces necrotic tissue 107 can create new niches within the host, allowing colonization by new taxa, and thus may lead to a 108 particularly large change in microbiome richness and composition. A parasite that keeps host 109 cells alive while extracting resources may have a more subtle impact on niches within the host, 110 and thus may not change microbiome diversity or composition. Thus, the stability of the 111 microbiome in the face of a parasite infection, which can impact disease severity and host health 112 (Coyte, Schluter, & Foster, 2015), may be determined by parasite feeding strategy. 113 Few studies have investigated the associations between host-associated microbiota and 114 parasites under field conditions (but see Jani & Briggs, 2014). This lack of studies may result 115 from a limited number of suitable model systems for exploring these questions in the field. The 116 long-lived nature of some hosts, limited ability to detect diseases observationally in live hosts in 117 the field, difficulty of excising infected tissue from animals, and ethical concerns also limit the 118 utility of many study systems for field research (Borer et al., 2011). As such, most investigations 119 of parasite-microbiome relationships have been conducted under controlled settings (Leung et 120 al., 2018). In a rare study in which the relationship between parasite infection and the host 121 microbiome was studied in both the lab and the field, the direction and magnitude of the 122 relationship differed between laboratory and field conditions (Leung et al., 2018). This finding 123 underscores the importance of investigating parasite-microbiome associations in field settings. 124 Foliar fungal parasites are a valuable model system to investigate microbiome-parasite 125 interactions in field settings. These parasites are often readily identifiable by external, 126 macroscopic symptoms and morphology, which facilitates observational studies in the field. 127 While bacteria can be numerically more abundant than fungi within plant hosts (Lundberg et al., 128 2012), fungi are functionally important symbionts of plants, with clear impacts on plant health 129 (Christian, Whitaker, & Clay, 2015). Many studies have examined the relationship between the 130 fungal microbiome and a single species of plant parasite (A. E. Arnold et al., 2003;Busby, 131 Ridout, & Newcombe, 2016). In contrast, no plant microbiome studies to our knowledge have 132 simultaneously considered multiple parasites in a common environment, either in the lab or in 133 the field. 134 To fill this gap, we conducted a molecular survey of the within-leaf fungal community of 135 the grass, tall fescue, in a field population infected by three fungal parasites. These three 136 parasites differed in a key ecological trait: feeding strategy (biotrophic, hemibiotrophic, and 137 necrotrophic). We hypothesized that necrotrophic parasites act as niche modifiers for the 138 microbiome. More specifically, we hypothesized that necrotrophic parasites, by producing 139 necrotic tissue, create new niches within the host, and thus have larger impacts on microbiome 140 richness and composition than biotrophic parasites, which keep host cells alive while extracting 141 resources. We further hypothesized that when hemibiotrophic parasites produce necrotic tissue, 142 they, like necrotrophs, will act as niche modifiers, but to a lesser degree than necrotrophs 143 because they only employ the necrotrophic feeding strategy in later stages of infection. This 144 hypothesis predicts that the composition of the fungal community in leaves exhibiting symptoms 145 of necrotrophic parasites will differ from the fungal communities in asymptomatic leaves and in 146 leaves exhibiting symptoms of biotrophic parasites. 147

Study System 149
Leaf segments were collected in a grass-dominated field within the Blackwood Division 150 of the Duke Forest Teaching and Research Laboratory in Orange County, North Carolina (35˚ 151 58'N, 79˚ 5'W). This field was chosen based on proximity to the lab (to expedite sample 152 processing), and abundance of tall fescue (Lolium arundinaceum) and its foliar fungal parasites. 153 This study focused on disease symptoms previously identified in another field of the Duke Forest 154 Teaching and Research Laboratory (Halliday et al. 2017) as representing parasites with three 155 different parasite feeding strategies (Table 1). 156

Leaf Segment Collection 157
Leaf segments were collected at a total of 36 points, spaced every 20 meters along 6 158 transects; the transects were 100m long, parallel, and spaced 20 meters apart ( Figure S1). Leaf 159 segments were collected over the course of two days in late September 2016 (September 26 and 160 September 30), which is when parasites tend to peak in their abundance in this system (Halliday, 161 Umbanhowar, & Mitchell, 2017). 162 At each point along each transect, we collected four whole leaves-one leaf with 163 symptoms of a necrotrophic parasite (and no other parasites), one leaf with symptoms of a 164 hemibiotrophic parasite (and no other symptoms), one leaf with symptoms of a biotrophic 165 parasite (and no other parasites), and one asymptomatic leaf. While coinfection is common in 166 this system (Halliday et al., 2017), we avoided collecting coinfected leaves. Only one leaf of 167 each symptom was collected at each point, and therefore leaves with the same symptom were 168 always collected at least 8 meters apart. This minimum distance was selected to minimize the 169 effect of spatial autocorrelation on the structure and composition of the microbiome (A. E. 170 Arnold, Henk, Eells, Lutzoni, & Vilgalys, 2007;Higgins, Arnold, Miadlikowska, Sarvate, & 171 Lutzoni, 2007). At each sampling point, for each symptom, leaves were haphazardly collected by 172 looking away, placing a finger on a plant, and then selecting the nearest tall fescue tiller. To 173 standardize the relative age of sampled leaves, we always sampled the oldest fully expanded 174 non-senescing leaf on the tiller. For each leaf, we estimated the percent of leaf area infected with 175 each parasite (infection severity) by visually comparing leaves to reference images of leaves of 176 known infection severity (James, 1971, e.g., Mitchell, Tilman, & Groth, 2002Halliday, 177 Heckman, Wilfahrt, & Mitchell, 2019). 178 At each of the 36 sampling points, we collected seven leaf segments: symptomatic and 179 asymptomatic segments from leaves infected with each of the three parasites, as well as an 180 asymptomatic segment collected from a leaf with no signs of disease. Thus, we collected a total 181 of 252 leaf segments. All segments were of approximately equal size. The two segments 182 collected from each symptomatic leaf were spaced at least 10 centimeters apart within the leaf. 183 We stored each leaf segment in an individual plastic bag that was then placed on ice. 184 Within four hours, all segments were processed back in the laboratory. Leaf segments 185 were washed under running DI water for 30 seconds to remove fungi that were on the surface of 186 the leaf but not attached to the leaf. Following surface washing, leaf segments were stored in a -187 80C freezer. 188

DNA extraction and Sequencing 189
Surface-washed leaf segments were ground under liquid nitrogen with a mortar and pestle 190 and transferred to 96-well plates for DNA extraction. DNA extraction was performed with the 191 DNEasy PowerSoil kit according to the manufacturer's protocol (Qiagen). 192 We assayed the fungal communities of tall fescue leaves by sequencing the internal transcribed 193 spacer (ITS) region. The ITS is a region of the nuclear ribosomal RNA cistron that is often used 194 as a DNA barcode for fungi, as it has less intraspecific variation than interspecific variation 195 (Schoch et al., 2012). We amplified the first part of the internal transcribed spacer (ITS1) with a 196 version of the primer set ITS1F and ITS2 modified for parallel sequencing on the Illumina 197 MiSeq platform (White, 1990;Smith & Peay, 2014). Each 25 uL PCR reaction had 2.5 uL of 10x 198 PCR Buffer, 3.5uL of MgCl2, 1uL of ITS1-F, 1uL of ITS2, 0.5uL of dNTPs, 0.63uL of Taq 199 polymerase, 13.12uL of water, and 3uL of DNA. The reactions were performed with the 200 following cycle conditions: initial denaturation 95C for 1 minute followed by 40 cycles of 94°C 201 for 30 seconds, 52°C for 30 seconds, 68°C for 30 seconds and a final elongation at 68°C for 7 202 minutes. We visualized PCR products using gel electrophoresis, cleaned samples with AMPure 203 beads (Lundberg, Yourstone, Mieczkowski, Jones, & Dangl, 2013), and concentration-204 normalized (using Qubit Fluorometric Quantitation, Life Technologies, Germany). The cleaned 205 amplicons were pooled in one run on an Illumina MiSeq instrument (Illumina, San Diego, CA, 206 USA) at the UNC High Throughput Sequencing Facility using a paired-end 2 x 250 bp kit. A 207 spike of 10% PhiX was added to the library to increase sample heterogeneity. All raw sequence 208 data is deposited in the National Center for Biotechnology Information Sequence Read Archive 209 (accessions XXXXXX-XXXXXX) and data is available at 210 (https://doi.org/10.5061/dryad.5x69p8d0v). 211

Fungal Community Analysis 212
All statistical analyses were performed with leaf segment as the unit of observation, and 213 in the R environment, version 3.6.0 (R Core Development Team 2012). Fungal sequences from 214 the pooled samples were assigned to individual leaf segments (i.e. demultiplexed) using Illumina 215 bcl2fastq pipeline (v.2.20.0), and sequencing adapters were removed from the fungal ITS 216 sequences using cutadapt (v.1.15) (Martin, 2011). Illumina-sequenced amplicon data of microbial 217 communities is often clustered into operational taxonomic units (OTUs) based on a fixed 218 dissimilarity threshold. This clustering reduces the rate at which sequencing errors are 219 misinterpreted as biological variation. The DADA2 package in R models and corrects Illumina-220 sequenced amplicon errors and infers exact amplicon sequence variants (herein referred to as 221 taxa), meaning these taxa are biological variants and not sequencing noise (Callahan et al., 222 2016). This method can resolve biological differences at a high resolution, and the output can be 223 directly compared between studies without the need to reprocess the pooled data. We therefore 224 employed DADA2 in this study. Quality control of sequencing reads for each leaf segment 225 consisted of truncating reads at the first quality score of 2 (a quality score of 2 indicates a portion 226 of the sequence that contains mostly low-quality reads of Q15 or less), and removing any read 227 with ambiguous base calls or greater than two expected errors. Reads shorter than 50 bases after 228 quality trimming were removed. 229

Statistical analyses: Diversity 230
To compare the diversity of fungal communities among asymptomatic and symptomatic 231 leaves of tall fescue, we quantified Hill's series of diversity. Hill's series of diversity (Hill, 1973) 232 is comprised of three orders (q) of diversity that summarize information about the number and 233 relative abundances of taxa. In Hill's series, the values of q (0, 1, 2) indicate the relative weight 234 applied to relative abundance (Bent & Forney, 2008). We estimated fungal richness (Hills' N0, 235 q=0), exponentiated Shannon entropy (Hill's N1,q=1), and inverse Simpson diversity (Hill's N2,236 q=2) (Jost, 2006). Because Shannon entropy and Simpson diversity are less sensitive to the 237 detection of rare taxa than species richness, they each place more weight on abundant taxa. To test whether fungal diversity is associated with symptom type, we used linear mixed 243 models to explain Hill's N0, N1 and N2. In order to meet assumptions of normality and 244 homoscedasticity of the residuals, we log-transformed diversity. We included symptom type (7 245 levels: the asymptomatic and symptomatic segments from leaves with each of the three focal 246 symptoms, plus asymptomatic segments from asymptomatic leaves) as fixed effects. High-247 throughput sequencing of pooled samples can result in samples that differ in sequencing depth; 248 we accounted for observational bias stemming from this difference by incorporating sequencing 249 depth into the models as another fixed effect, following Bálint et al., 2015. Leaf ID collection 250 point were included as random effects, with leaf ID nested within collection point. Linear mixed 251 effects models were assessed in nlme, and we used emmeans (Lenth, 2018, version 1.3.2) to 252 evaluate the estimated marginal means of diversity indices for each explanatory variable level, 253 adjusted for multiple comparisons (Tukey HSD). After accounting for the variation explained by 254 random effects and sequencing depth, we compared the partial residuals of Hill's numbers 255 among the treatments with Tukey's HSD. 256 To quantify any changes in within-host microbial diversity as disease progressed, we 257 used disease severity (percent leaf area exhibiting symptoms) as a measure of disease 258 progression. Specifically, for each of the three parasite symptom types, we fit three linear mixed 259 models using R package nlme (version 3.1-142, Pinheiro, Bates, DebRoy, Sarkar, & Team, 260 2013) with Hill's N1, Hill's N2, or Hill's N3 as the dependent variable and disease severity as an 261 independent variable. Hill's N1, Hill's N2, and Hill's N3 were log-transformed to meet 262 assumptions of homoscedasticity and normality. Each model also included the square root of 263 sequencing read numbers obtained for a leaf segment as an independent variable, and collection 264 point as a random effect. Thus, each model had the following form: Hill ~ sqrt(readNumbers) + 265 Symptom + 1|Collection Point. A clear relationship between disease severity and fungal 266 diversity, in the same direction as the overall association of fungal diversity and symptom type, 267 would suggest that the parasite progressively impacts the fungal microbiome as the disease 268 severity increases. No relationship between disease severity and fungal diversity, or a 269 relationship in the opposite direction to the overall association of fungal diversity and symptom 270 type, would suggest that any association between the parasite and microbiome is not due to a 271 progressive impact of the parasite on the microbiome. 272

Statistical Analyses: Community Composition 273
To test the hypothesis that parasites that modify niches within their host by creating 274 necrotic tissue alter fungal community composition, we tested whether fungal community 275 composition was correlated with symptom type. Bray-Curtis distances were calculated among 276 leaf segments separately and visualized using non-metric multidimensional scaling (NMDS) 277 implemented in the phyloseq package (version 1.24, Mcmurdie & Holmes, 2013). We performed 278 a permutational MANOVA using the adonis function in the vegan package. The predictors were 279 collection point and symptom type. The adonis function is sensitive to the order in which 280 variables are added, so models were run multiple times, varying the order of predictors, to verify 281 important predictors and we report predictors that were significant regardless of order (following 282 Vannette, Leopold, & Fukami, 2016). 283 We also investigated whether the homogeneity of the composition of the fungal leaf 284 microbiome varied with symptom type. For each leaf segment, we quantified the distances from 285 each measured Bray-Curtis distance to the centroid of Bray-Curtis distance for that leaf 286 segment's symptom type. We then compared the dispersion of the measurements within each 287 symptom type across categories using the betadisper function in the vegan package. 288

Statistical Analyses: Interpretation 289
To improve statistical inference, we report our results using the language of the 290 "statistical clarity concept," instead of emphasizing statistically significant results (Dushoff, 291 Kain, & Bolker, 2019). This approach puts forward that the results of null hypothesis 292 significance testing are most usefully interpreted as a guide to whether a certain effect can be 293 seen clearly in a particular context. 294

Results 295
From the 252 leaf segments, Illumina generated 6,650,600 ITS1 reads. Of these, 296 4,483,694 reads passed quality filtering. This represents a mean number of reads per leaf 297 segment of 17,792. Using DADA2, we identified 2961 unique amplicon sequencing variants 298 (taxa). This represents a mean number of taxa per leaf segment of 70.8 (median of 62). All taxa 299 placed within the kingdom fungi. 12.5% of taxa could not be placed lower than the kingdom 300 fungi. Of the taxa that could be placed lower than the kingdom fungi, 99.2% placed within 301 Ascomycota (1459) or Basidiomycota (1110). At the class level, most taxa within Ascomycota 302 were assigned to Dothideomycetes (808) or Sordariomycetes (307), and most taxa within 303 Basidiomycota were assigned to Agaricomycetes (691) or Tremellomycetes (186). The following 304 analyses consider these 2961 taxa delineated by DADA2. 305

Diversity 306
After accounting for variation in sequencing depth, symptom type strongly and clearly 307 predicted variation in all three numbers in Hill's series (ANOVA P < 0.0001). There were fewer 308 fungal taxa (Hill's N0) and there was lower diversity (Hill's N1 and N2) in leaf segments that 309 exhibited symptoms of necrotrophic parasites compared to leaf segments that were asymptomatic 310 or symptomatic of either hemibiotrophic or biotrophic parasites (Table S1, Figure 1, Tukey's 311 HSD, P < 0.01). Specifically, when comparing the mean of each of Hill's numbers between leaf 312 segments symptomatic of necrotrophic parasites and all other leaf segments,  55.0% lower, 314 there were no clear differences in fungal richness or diversity between asymptomatic leaf 315 segments and those symptomatic of hemibiotrophic or biotrophic parasites (p>0.05). Finally, 316 there were no clear differences in fungal richness or diversity between asymptomatic leaf 317 segments of any type, whether from leaves with any of the parasite symptoms or leaves that were 318 wholly asymptomatic. Thus, the leaf segments symptomatic of a necrotrophic parasite came 319 from leaves that were not detectably different from any other leaves, which suggests that the 320 fungal diversity of the leaves symptomatic of necrotrophic parasites was not lower than other 321 leaves prior to infection by necrotrophic parasites. 322 To investigate the lower richness and diversity of leaf segments symptomatic of 323 necrotrophic parasites, we considered how the diversity metrics as defined in Hill's series varied 324 with estimated disease severity (percent leaf area exhibiting symptoms of a given parasite 325 feeding strategy) within the segments symptomatic of necrotrophic parasites. We had predicted 326 that if a necrotrophic parasite decreases the diversity of the fungal microbiome as it grows within 327 its host, fungal diversity would decrease with necrotrophic parasite disease severity. Instead, 328 fungal diversity weakly increased with necrotrophic parasite disease severity (Figure 2, Table  329 S4; Hill's N0, P = 0.010; Hill's N1, P = 0.043; Hill's N2, P = 0.064). Because these three 330 correlations were not negative, these results suggest that fungal diversity, particularly richness, 331 does not decrease progressively as a necrotrophic parasite spreads through the leaf. 332

Community Composition 333
We analyzed variation in community composition using the Bray-Curtis distance metric. 334 The fungal community composition of leaf segments with symptoms of necrotrophic parasites 335 differed from leaf segments that were asymptomatic or symptomatic of other parasites (Figure 3, 336 Table S2, PerMANOVA, P = 0.001). In other words, necrotrophic symptoms were associated 337 with not only fewer fungal taxa, but also a different assemblage of fungal taxa compared to the 338 other parasites. We had predicted that if a necrotrophic parasite alters the assemblage of fungal 339 taxa as it grows within its host, fungal composition would change with estimated disease 340 severity. Within the leaf segments that exhibited necrotrophic symptoms, disease severity did 341 predict fungal community composition, but accounted for a small amount of variation (Figure 4, 342 PerMANOVA, R 2 = 0.08, P = 0.043). To test if any parasite symptoms were associated with a 343 more homogeneous fungal community, we quantified beta diversity within each symptom type 344 as the distances to the centroid from all measured Bray-Curtis dissimilarities among leaf 345 segments of that symptom type. There was no effect of symptom type on the homogeneity of 346 fungal community composition ( Figure S9, ANOVA, P = 0.693). Finally, considering fine-scale 347 variation in the fungal microbiome, within leaves symptomatic of any of the three parasite 348 feeding strategies, the symptomatic leaf segments differed from the asymptomatic leaf segments 349 in the relative abundance of multiple fungal genera (supplementary material). 350

Discussion 351
This study used a trait-based analysis of multiple co-occurring fungal parasites in a field 352 population of a grass host to evaluate how parasites alter the fungal microbiome of the host. 353 Microbiome diversity and composition were associated distinctly with symptoms of parasites 354 with a necrotrophic feeding strategy, and not with other parasites. These results are consistent 355 with a hypothesis based on niche modification: parasites with traits that more strongly impact the 356 host environment and available niches within the host also more strongly impact the host 357

microbiome. 358
In niche modification, a species changes the types of niches available within a site and, 359 consequently, the identities of species that can colonize the community (Lewontin, 1983;360 Fukami, 2015). Among parasites infecting plant leaves, only parasites with a necrotrophic 361 feeding strategy create necrotic host tissue throughout their entire infection process (Glazebrook, 362 2005;Suzuki & Sasaki, 2019). We therefore expected necrotrophs to be particularly strong niche 363 modifiers that impact the host environment and consequently, the host fungal community. 364 Indeed, symptoms of necrotrophic parasites were associated with fungal communities of lower 365 diversity relative to asymptomatic leaves, while symptoms of two other types of parasite feeding 366 strategies (biotrophs and hemibiotrophs) were not. 367 Our hypothesis that necrotrophic parasites are particularly strong niche modifiers was 368 further supported by analysis of fungal community composition. The composition of the fungal 369 communities of leaf segments exhibiting symptoms of necrotrophic parasites differed from that 370 of asymptomatic leaves, while there was no clear difference in fungal composition between leaf 371 segments exhibiting symptoms of biotrophic or hemibiotrophic parasites and that of 372 asymptomatic leaves. We hypothesize that this shift in composition resulted from necrotrophic 373 parasite infections making the host environment more suitable for saprobes (Suzuki & Sasaki, 374 2019). 375 Fungal community composition was only weakly associated with necrotrophic parasite 376 disease severity, and fungal diversity did not have a negative correlation with disease severity. 377 These results suggest that fungal community composition and diversity do not change 378 progressively as a parasite grows within a leaf. The parasite may instead disrupt the host 379 environment, and consequently, the fungal microbiome, upon initial infection. Such microbiome 380 disruption upon initial infection is consistent with evidence from at least one other system; in 381 experimental inoculations of frogs with Batrachochytrium dendrobatidis, microbiome diversity 382 declined upon infection and had no relationship with pathogen load (Jani and Briggs, 2014). 383 While we expected necrotrophs to act as particularly strong niche modifiers, we expected 384 hemibiotrophs to modify their environment as well, given that they create necrotic tissue in the 385 latter part of the infection process (Glazebrook, 2005;Suzuki & Sasaki, 2019). However, we 386 found contrasting results between necrotrophic and hemibiotrophic parasites; the fungal 387 communities of leaf segments exhibiting symptoms of hemibiotrophic parasites had no clear 388 differences in diversity and composition compared to those of asymptomatic leaf segments. 389 While both hemibiotrophs and necrotrophs ultimately require killing host cells, they differ in 390 how they initially interact with host tissue. Our results therefore suggest that the initial infection 391 by a necrotrophic parasite is the key stage in which diversity of the microbiome declines. This is 392 consistent with a weakly supported positive correlation between disease severity and fungal taxa 393 diversity that we observed, as diversity was lowest when necrotrophic parasite disease severity 394 was low (i.e. change in diversity occurred early in the infection process). 395 While we interrogated relationships between parasites and the fungal microbiome, there 396 is growing evidence that bacterial and fungal microbiota associate with different factors (Elhady 397 et al., 2017;Bergelson, Mittelstrass, & Horton, 2019). For example, while we found that a 398 biotrophic parasite had no relationship with fungal microbiome diversity, recent work 399 investigating the bacterial microbiome of wheat found that leaves infected with a parasite 400 infecting as a biotroph had higher bacterial diversity than uninfected leaves (Seybold et al., 401 2019). Such differences between the fungal and bacterial microbiome may result from their 402 differences in generation times and abundances within a host. For more complete understanding 403 of parasite-microbiome associations, studies that integrate surveys of the bacterial and fungal 404 communities will be essential (Porras-alfaro & Bayman, 2011;Laforest-Lapointe & Arrieta, 405 2018). 406 In studies of plant, human, and other animal diseases, increasing numbers of studies are 407 characterizing how microbial communities associate with specific parasites and progression of 408 disease (Cho & Blaser, 2012;Jani & Briggs, 2014a;Jakuschkin et al., 2016;Lebreton et al., 409 2019). Here, we propose that functional traits can be used to explain variation among parasites in 410 their associations with host microbiota. Trait-based approaches have played an important role in 411 plant functional ecology (Adler et al., 2014;Cadotte, 2017). More recently, Zanne et al., 2019 412 advocated complementing traditional and genomic approaches to fungal functional ecology with 413 trait-based approaches. Among traits of parasites, they discuss how parasite feeding strategy 414 (what they refer to as nutritional strategy) can help predict how parasites interact with their 415 abiotic and biotic environment. Integrating that trait-based prediction with the concept of niche 416 modification, we suggest that parasite feeding strategy can determine whether the parasite is a 417 strong niche modifier, and thus explain relationships between parasites and host-associated 418 microbiota. 419

Acknowledgements 420
We thank James Umbanhowar for helpful comments. We thank Anita Simha for help collecting 421

Data Accessibility Statement 431
Upon acceptance of the manuscript, all raw sequence data will be deposited in the National 432 Center for Biotechnology Information Sequence Read Archive, and all data sets used in this 433 study will be submitted to Dryad.  Table 1 Biology and ecology of the focal parasite feeding strategies 441 Figure 1. Necrotrophic symptoms were associated with foliar fungal communities that were less diverse. Panels show fungal diversity quantified using Hill numbers for observed species richness (N0), exponentiated Shannon entropy (N1), and inverse Simpson's diversity (N2). Letters mark differences in Hill's numbers evaluated with Tukey's HSD. Points are estimated marginal means with red bands indicating ± 95% confidence intervals. Figure 2. Leaf-associated fungal diversity and richness did not have a negative correlation with necrotrophic disease severity (percent leaf area exhibiting symptoms), suggesting that fungal community diversity does not change progressively as a parasite grows within a leaf. Panels show fungal diversity quantified using Hill numbers for observed species richness (N0), exponentiated Shannon entropy (N1), and inverse Simpson's diversity (N2). Each point represents a leaf segment. Lines represent best-fit regressions between disease severity and the diversity metric. Figure 3. Leaf segments with necrotrophic symptoms had foliar fungal communities that differed in taxonomic composition from asymptomatic leaf segments and leaf segments with symptoms of other parasite feeding strategies (PerMANOVA p<0.001, stress=0.18), su. Fungal taxonomic composition was quantified by the Bray-Curtis distance metric and is illustrated by non-metric multidimensional scaling (NMDS). Each point represents a leaf segment. As indicated by color, each leaf segment was either asymptomatic or symptomatic of one parasite feeding strategy (necrotroph, hemibiotroph, or biotroph). Figure 4. Severity of symptoms caused by necrotrophic parasites predicted fungal community composition, but only explained a modest amount of variation, suggesting that fungal community composition does not change progressively as a parasite grows within a leaf. Fungal taxonomic composition was quantified by the Bray-Curtis distance metric and is illustrated by non-metric multidimensional scaling (NMDS). Each point represents a leaf segment. Point size indicates percent leaf area symptomatic of necrotrophic parasites.