Foliar pathogens of California grasses infect multiple hosts: implications for grassland diversity

Pathogen infection is common in wild plants and animals, and may regulate their populations. If pathogens have narrow host ranges and increase with the density of their favored hosts, they may promote host species diversity by differentially limiting common species while giving rare species advantages. Yet because many pathogens infect multiple co-occurring hosts, they may not strongly respond to the relative abundance of a single host species. Are natural communities dominated by specialized pathogens with species-specific responses to host density or by pathogens with broad host ranges and limited responses to host density? The answer determines the potential for pathogens to promote host diversity, as hypothesized in many plant communities, or to have negligible or even negative effects on host diversity. We lack a systematic understanding of the impacts, identities, and host ranges of pathogens in natural communities. Here we characterize the community of fungal pathogens associated with symptomatic grass leaves and evaluate their host specificity and fitness impacts in a California grassland community of native and exotic grass species. We found that most of the commonly isolated fungal pathogens were multi-host and exhibited intermediate to low specialization. The amount of pathogen damage each host experienced was independent of the local relative abundance of each host species. Despite pathogen sharing among the host species, fungal communities slightly differed in composition across host species. Plants with high pathogen damage tended to have lower seed production but the relationship was weak, suggesting limited fitness impacts of pathogen damage. Moreover, seed production was not dependent on the local relative abundance of each plant species, suggesting that coexistence mechanisms may operate at larger spatial scales in this community. In sum, pathogens in this grassland community are multi-host and have small fitness impacts. As a result, foliar pathogens are unlikely to promote negative frequency-dependence, nor to promote plant species coexistence, in this system. Still, given that pathogen community composition differentiates across host species (effective specialization), some more subtle feedbacks between host relative abundance and pathogen community composition, damage, and fitness impacts are possible, which could in turn promote either coexistence or competitive exclusion.


Introduction 38
Pathogens are ubiquitous in ecological communities (Burdon 1993, Gilbert 2002, 39 Lafferty et al. 2008). Because they affect host demographic rates, pathogens are often expected 40 to regulate host species population growth (Burdon andChilvers 1982, Burdon 1982). Most 41 pathogens infect only a subset of the available host species, so their incidence, and by extension 42 their impacts, may be host-specific (Gilbert and Webb 2007, Beckstead et al. 2014, Parker et al. 43 2015. Host-specific population regulation can ultimately promote species diversity by 44 differentially suppressing species when they become common, thereby facilitating the 45 establishment of competing species and providing a relative advantage to rare species (Fig. 1). 46 This pathogen-mediated negative frequency-dependence, sometimes called the  hypothesis (Janzen 1970, Connell 1971, has growing support in diverse plant communities, 48 including tropical forest trees (Augspurger 1983 For each isolate, we scraped fungal mycelium from living culture, transferred it to a 182 microcentrifuge tube, and extracted genomic DNA using REDExtract-N-Amp Tissue PCR Kit 183 (Sigma-Aldrich, Inc.), following the manufacturer's protocol. We amplified and sequenced the 184 internal transcribed spacers (ITS) 1 and 2 and the 5.8S nuclear ribosomal gene using the primer 185 pairs ITS-1F and ITS-4 (Gardes and Bruns 1993). For PCR amplification, we used a T100 186 thermal cycler (Bio-Rad Laboratories, Inc., Hercules, CA) and thermal cycling conditions 187 following U'Ren et al. (2010). Following electrophoresis on a 1.5% agarose gel, we visualized 188 PCR products using GelRed™ (Biotium Inc., Hayward, CA) and sent to them MCLAB (San 189 Francisco, CA) for cleanup and bidirectional sequencing on an ABI 3730 XL sequencer. 190 We manually inspected and edited all reads. Bidirectional reads were automatically 191 assembled into a consensus sequence, using a minimum of 20% overlap and 85% sequence 192 similarity. We then clustered the 288 consensus sequences and two unidirectional sequences into 193 operational taxonomic units (OTUs) based on a minimum of 40% overlap and 90, 95, 97 and 194 99% sequence similarity. We used Sequencher (Gene Codes, Ann Arbor, MI) for sequence 195 editing and assembly and OTU designations. All sequence data will be submitted to GenBank 196 (accession numbers XXXX-XXXX). 197 Because percent sequence similarity for the ITS region varies within versus between 198 fungal species (as described in O' Brien et al. 2005) and given the limitations of GenBank (Kang 199 et al. 2010, U'Ren et al. 2010), we conducted phylogenetic analyses to improve our inference of 200 taxonomic placement. We partitioned OTUs with similar sequences into datasets (1-52 JRBP 201 isolates per dataset), and generated a phylogenetic tree for each dataset following the procedures 202 described in Higginbotham et al. (2014) and Spear (2017). Because not all sequences mapped 203 onto named fungal species, we assigned each operational species a unique species code, which 204 we list in parentheses after the lowest estimated taxonomic placement (Table S2). We treated 205 isolates belonging to a species complex as a single species. 206 207

Analyses of fungal community composition and host associations (Conditions 1A-B) 208
We conducted several analyses of the fungal community. First, we describe sampling 209 efficacy and fungal species richness and diversity. Second, we compare fungal communities 210 across host species. Third, we explicitly consider the frequency of shared pathogens from the 211 grass species' perspective. Finally, we evaluate host specialization from the fungal species' 212 perspective. 213 For the full fungal community dataset (N = 290), we report (i) sampling efficacy, (ii) the 214 triple-and quadrupletons to estimate the number of unseen species in the community and correct 221 for the negative bias associated with under-sampling in highly diverse assemblages (Chiu et al. 222 2014). We visualized the richness, evenness, and dominant species of the fungal community by 223 plotting species abundance versus species rank abundance (iv). Fungal species isolated ten or 224 more times were defined as abundant. We estimated diversity (v) based on Fisher's alpha, which 225 is robust to unequal sample sizes (Fisher et al. 1943, Magurran 2013, the Shannon index, which 226 is sensitive to sample size (Magurran 2013), and the effective number of species (Jost 2006 PERMANOVAs, and tests of homogeneity of variance, we considered each focal grass species-240 by-perennial density transect combination to be a distinct community. We excluded grass 241 species-by-transect communities for which fewer than three isolates were collected and B.
number of isolates (in total, 20 communities, 21 fungal species, 5 grass species, and 99 isolates 244 were included in the dataset). From the resulting raw data matrix, we created a matrix of pairwise 245 dissimilarities in fungal community composition for use with the NMDS, PERMANOVAs, and 246 homogeneity of variances test. The dissimilarity matrix was created via function vegdist with the 247 Chao method, which implements an abundance-based Jaccard index that was adjusted to 248 consider unseen species (Chao et al. 2005, Oksanen et al. 2016. 249 To evaluate pathogen sharing among grass species, we first characterized the observed 250 number of shared fungal species between pairs of grass species. Second, we estimated the 251 similarity of pairs of grass species (i.e., fungal community overlap) based on the relative 252 abundances of the fungal species for the eight grass species for which 10 or more isolates were 253 collected, using the Morisita-Horn index (200 bootstrap replicates). 254 To assess the specialization of the fungal species (Condition 1A) and the uniqueness of 255 each grass species' fungal community (Condition 1B), we visualized all interactions (links) with 256 a bipartite plot and then calculated species-level interaction specialization via the weighted 257 specialization index d'. The index illustrates the degree to which a species deviates from 258 expected, random interactions by comparing interaction frequencies to the overall availability of 259 potential partners (the marginal frequencies) (Blüthgen et al. 2006). As we did not sample the 260 grass species proportional to their relative abundances in JRBP, the total number of fungal 261 isolates cultivated from a grass species (marginal frequency) was not a suitable proxy for its 262 availability. Thus, to improve the accuracy of our calculations, we only included the five focal 263 have replicated the abiotic conditions that originally led to disease development in JRBP, we 290 tested multiple isolates of some of the non-singleton fungal species to avoid misclassifying a 291 fungal species as nonpathogenic. In Stanford greenhouses, we inoculated apparently healthy 292 grass leaves by securing colonized 2% ME agar plugs against leaves with small strips of 293 Parafilm (Sinclair and Dhingra 1995). Uncolonized agar plugs were used as negative paired 294 controls. We completed two rounds of inoculations. In round one, we inoculated each plant with 295 4-5 isolates and a negative control, with each treatment applied to a unique leaf (5-6 treated 296 leaves/plant; 9-12 leaves/treatment). In round two, we applied a single isolate or control 297 treatment to each individually-potted plant (4 unique leaves per plant and 5 plants per treatment, 298 totaling 20 leaves per treatment). Thirteen of the 99 isolates tested were tested in both rounds 299 (Table S3). After monitoring plants for up to one week, we assessed pathogenicity by censusing 300 leaves for obvious symptoms of disease (e.g., lesions or necrosis). To compare the proportion of 301 leaves with disease for a given fungal isolate versus its paired control (same plant species and 302 inoculation date) we used bias-reduced generalized linear models (brglm function; Kosmidis 303 2013), assuming binomial error distributions and probit link functions (Table S3). The analyses 304 excluded all leaves for which it was unclear if tissue damage was the result of disease, resulting 305 in 5 to 20 leaves per treatment (Table S3). We deemed an isolate pathogenic if a significantly 306 greater proportion of inoculated leaves had pathogen damage than paired control leaves (using a 307 = 0.05; Table S3). 308 309 Impacts of pathogen damage on per-capita seed production (Condition 2) competition, and pathogen impact, we harvested seeds from 350 of the marked grasses from the 313 perennial density transects. We measured per-capita seed production as a function of conspecific 314 and heterospecific density and percent leaf area damaged for Avena barbata, Bromus 315 hordeaceus, Bromus diandrus, Stipa pulchra, and Elymus glaucus. We did not measure seed 316 production of Phalaris aquatica because it occurs in monotypic stands with little variation in 317 relative abundance. 318 Some pathogen species may be more damaging, and in turn have higher fitness costs, to 319 their hosts than others. We addressed this issue by linking fungal species identity to the mean 320 leaf damage (N = 65 fungal isolates, representing 16 fungal species) and seed output (N = 56 321 fungal isolates, representing 12 fungal species) on marked plant individuals in the perennial 322 density transects. All had percent damage estimates in March and all but one had percent damage 323 estimates in April. 324 We used regression models to assess the extent to which pathogen damage correlated 325 with per-capita seed output. To compare variation across species, we standardized both seed 326 output and March and April percent damage by calculating z-scores (i.e., by subtracting the 327 within-species mean and dividing by the within-species standard deviation). By doing so, we 328 asked whether deviations from average pathogen damage correlated with deviations from 329 average seed production, which would suggest pathogen fitness impacts. We modeled the 330 relationships between standardized March or April damage estimates and standardized seed 331 production (we did not include both March and April damage together in the same model 332 because these were assessed on the same individuals with only a single seed output value). 333 Models in which seed production was standardized by species but damage was not produced 334 similar results (shown in the R code accompanying the paper). The variance in seed production was not constant across the range of damage, so we used quantile regression to calculate 336 regression coefficients (slopes and intercepts) for the 25 th , 50 th , and 75 th percentiles of seed 337 production, using the function rq in the quantreg package in R (Koenker 2017  The strongest predictors of pathogen damage (based on model AIC) were species, 357 sampling date, and their interaction, although we also tested for interactive effects of host species 358 and plant community at the plot and sub-transect scale (Tables S4-S5). A. barbata and E. glaucus 359 had higher percent damage, particularly at the later sampling point (April). All species, except A. 360 barbata, had a higher proportion of leaves damaged in April than in March, suggesting that 361 pathogen damage accumulates through the growing season. Counter to Condition 1 for 362 pathogens to promote coexistence, focal species frequency within the plot did not significantly 363 affect pathogen damage (Fig. 2). The model that included that term had an AIC value only 364 slightly above the minimum AIC, indicating similar performance (the model that includes 365 frequency is shown in Table S4 for illustration). 366 367

Fungal pathogen identities and diversity (Conditions 1A-B) 368
We isolated fungi from 38% of the 772 foliar tissue pieces that we collected along 24 369 transects across Jasper Ridge Biological Preserve (JRBP). We successfully sequenced 290 of the 370 resulting 302 isolates (Table 1). Fungi were isolated from all nine grass species. However, the 371 grass species were unequally sampled and the isolation frequency (i.e., percent of tissue pieces 372 with growth) differed among the hosts (Table 1). 373 The accumulation curves for the fungal species and OTUs (based on phylogenetic 374 analyses and percent sequence similarity, respectively) were non-asymptotic, indicating 375 incomplete sampling and a diverse community (N = 290; Fig. S3 were abundant (22% were observed more than 10 times) (Fig. S4). The most commonly 384 observed genera included Pyrenophora, Ramularia, Alternaria, and Parastagonospora (Fig. 3  385 and Table S2). Using greenhouse-based inoculation experiments, we experimentally confirmed 386 the pathogenicity of 27% of the 99 isolates tested, representing 14 of the 35 fungal species tested, 387 eight of the nine common fungal species, and the most commonly observed genera in our survey 388 The observed and estimated lower bound of fungal species richness per grass species 392 (iChao1) ranged from one to 26 and from one to 43.27, respectively (  Fig. S5). 399 The majority (74%) of the 19 non-singleton fungal species infected multiple hosts (Fig.  400 3). On average, the multi-host fungi were isolated from four grass species. Two of the fungal 401 species, Pyrenophora lolii (A2) and Alternaria infectoria species-group (C1), were isolated from 402 seven of the nine grass species that we sampled (Fig. 3) Parastagonospora caricis (AM) (Fig. 3) The eight grasses species from which 10 or more isolates were collected shared at least 421 one and up to eight fungal pathogen species with every other grass species (median = 3; Table  422 S6; Fig. 3). As estimated by the Morisita-Horn index, the average estimated similarity between 423 pairs of host species was moderate (42%) (min = 5% for E. glaucus and P. aquatica; max = 424 100% for F. perennis and S. pulchra and for A. barbata and A. fatua; Table S6). Six of the grass 425 species shared their numerically dominant fungal species with one of the other grass species Two of the grass species had a relatively low fungal community similarity with the other 428 grass species in our study: E. glaucus, the second best-sampled grass species (N = 33; Morisita-429 Horn similarity index 5% to 11%), and P. aquatica (N = 18; Morisita-Horn similarity index 5% 430 to 70%; Table S6; Fig. 3). Congruously, the weighted specificity index d' for E. glaucus suggests 431 that the grass species hosted a relatively unique assemblage of fungal pathogens (d' = 0.665; 432 Table 1). 433 The fungal pathogen communities associated with the leaves of the five focal grass 434 species included in the PERMANOVA (N = 99 isolates) were significantly dissimilar (F 4,15 = 435 3.682, R 2 = 0.495, p = 0.001; Fig. 4). Specifically, (i) the fungal community associated with S. 436 pulchra (32 isolates) was significantly different from those of A. barbata (11 isolates) (P adj = 437 0.037), B. diandrus (16 isolates) (P adj = 0.037), E. glaucus (31 isolates) (P adj = 0.04), and P. 438 aquatica (9 isolates) (P adj = 0.037); (ii) the fungal community associated with P. aquatica (9 439 isolates) was significantly different from that of B. diandrus (16 isolates) (P adj = 0.05); and (iii) 440 the fungal community associated with B. diandrus (16 isolates) was significantly different from 441 that of A. barbata (11 isolates) (P adj = 0.05) (Fig. 4). The fungal communities of the grass species 442 had similar dispersions (F 4,15 = 0.035, p = 0.997; Fig. 4). 443 444

Impacts of pathogen damage on per-capita seed output (Condition 2) 445
The relationships between pathogen damage and seed production were generally 446 negative but highly variable, particularly at the low levels of pathogen damage that most of the 447 plants in the survey experienced (Fig. 5). Because the variance was higher at low pathogen 448 damage, we used quantile regression to estimate how the relationship between damage and seed production distribution. We found that increasing pathogen damage in April was associated with 451 reduced seed output at the 50 th percentile, but that the negative effect was not statistically 452 significant for the 25 th or 75 th percentiles (Table S7; Fig. 5). In other words, plants of average 453 seed production had significantly reduced seed output with increasing pathogen damage, but 454 both high-and low-seed output individuals did not have a significant response to pathogen 455 damage. The effect of March damage was weakly negative but not statistically significant for 456 any quantile (Fig. 5). Focal species frequency had no association with either seed output (Fig. 5) 457 or pathogen damage (Fig. 2). Much of the variation in seed production was not explained by 458 pathogen damage in either month. analysis for the z-score of seed production (calculated by host species) against fungal species 465 identity. None of the common fungal species or genera were notably associated with higher than 466 average damage or lower than average seed production (Fig. S6) did not strongly respond to host relative abundance (Fig. 2) in a way that would promote 486 negative frequency dependence (Condition 1). This is perhaps unsurprising given that the grass 487 species extensively shared foliar pathogens, and that most of the common pathogens infected 488 several species (Table 2;  Poaceae), pathogen damage may be decoupled from the relative abundance of a single host 491 species. It is also possible that pathogen damage responds to host relative abundance at the 492 regional, rather than local, scale (Mitchell et al. 2002). 493 Despite pathogen sharing among the grass species (Table S6; Fig. 3), fungal community 4 and S5). Moreover, several common fungi were relatively specialized (partially supporting 496 Condition 1A). Pyrenophora tritici-repentis (E) occurred mainly on the native perennial E. 497 glaucus and was the most common fungus infecting that species. The only other occurrence of 498 this pathogen in our surveys was on a single native S. pulchra individual located within an E. 499 glaucus stand, suggesting a potential pathogen spillover event. Moreover, the recently invading 500 perennial grass P. aquatica (http://jrbp.stanford.edu/content/oakmead-herbarium-arrivals-501 weeds#Arrival) had low pathogen damage and a relatively distinct pathogen community, 502 including two fungi exhibiting host preference, Pyrenophora cf. dactylidis (L) and 503 Parastagonospora caricis (AM). Taken together, these results suggest that although this 504 Northern California grassland fungal pathogen community is dominated by multi-host 505 pathogens, they vary in their affinity for each host species (Condition 1A) and pathogen 506 communities may be structured, at least in part, by host species identity (Condition 1B). 507 The low pathogen damage and relatively distinct pathogen community of P. aquatica 508 suggest that only a few of the resident pathogens have been able to make the ecological or 509 evolutionary jump to infect P. aquatica since its introduction. Exotics that are slow to 510 accumulate pathogens in their naturalized range tend to be more noxious (Mitchell and Power 511 2003). In JRBP, P. aquatica is an aggressive species that forms monotypic stands that suppress 512 native plant species. By contrast, the other exotic species are all annuals that co-occur with 513 native perennial species and have been naturalized in the area for over one hundred years, 514 potentially allowing more time for pathogen host switches to occur. 515 Although higher pathogen damage was weakly associated with lower seed production 516 (Fig. 5), foliar pathogens, which reduce photosynthetic function, did not appear to dramatically 517 affect fitness (Condition 2). Nonetheless, the ability to detect any relationship between percent damage and seed production is notable given the range of other sources of variation, including 519 individual size, competition with the surrounding community, and microhabitat suitability. 520 However, a fuller assessment would require measuring fitness effects across life stages in 521 experimental infections. Although some of the fungal species exhibited strong host affinity 522 (Condition 1A; Fig. 3; Table 2) and pathogen community composition subtly varied across plant 523 species (Condition 1B; Table 1 In addition to considering all plant demographic rates, it is necessary to investigate the 536 entire pathogen community. We focused on local lesion diseases of leaves and on culturable 537 foliar fungi, which may be disproportionately host generalist due to their necrotrophic lifestyle 538 and ability to grow on malt extract agar. Our study did not identify potentially important 539 biotrophic fungi, viruses, bacteria, or parasitic nematodes that may occur in roots or stems. 540 Biotrophic, obligate plant pathogens are generally expected to exhibit higher host specificity than 541 facultative pathogens (Gilbert 2005), making them more likely to respond to host density and 542 thereby maintain plant community diversity. Further, our study did not determine whether a 543 given multi-host pathogen exerts host-specific impacts (Condition 1C), which has been observed 544 in other systems (Sarmiento et al. 2017) and could contribute to the maintenance of local 545 diversity. 546 The broad host ranges and minimal demographic impacts of pathogens in this wild 547 grassland system contrast sharply with pathogen impacts on phylogenetically-related cultivated 548 agricultural grasses such as barley, wheat, and oats. The fungal pathogen species we encountered 549 are closely related to important cereal pathogens, such as Pyrenophora, Parastagonospora, and 550 The generalist strategy of broad host ranges and minimal host impacts is well suited to 562 persistence and spread in seasonal, high-density mixed species grasslands like our study system. Our holistic approach illustrated that foliar pathogens in this system are multi-host, do not 587 strongly respond to host species relative abundance, and have minor impacts on host fitness. 588 Together, these conditions make it unlikely that foliar pathogens promote coexistence via 589 negative frequency-dependence in this system. More broadly, this and previous studies suggest 590 that foliar pathogens may not maintain grassland diversity (Peters and Shaw 1996, Mitchell 591 2003; but see Allan et al. 2010). However, given conflicting results, more work is required to 592 clarify the role of foliar pathogens in plant species coexistence, particularly relative to other 593 potentially important factors. 594 Identifying the factors that shape the relative abundances and coexistence of species, 595 potentially including pathogens, is particularly important for grasslands. Grasslands are one of 596 the most diverse and widespread habitats in the West Coast of the U.S., yet they are also heavily 597 invaded, degraded, and exposed to variable and changing climate (Harpole et al. 2007 Table 2. Host specialization of the 17 non-singleton fungal species from symptomatic leaves of 798 five grass species with independent estimates of relative abundance in JRBP. Fungal species are 799 sorted by abundance from common to rare. Thirteen (76%) of the non-singleton fungal species 800 were isolated from multiple hosts. The specialization index d' represents the extent to which the 801 grass species attacked deviated from random given the availability of the grass species, varying 802 from 0 (perfect generalist) to 1 (perfect specialist) (Dormann 2011  host community diversity are: (1) pathogen damage must increase with host species relative 807 abundance, and (2) pathogen damage must reduce fitness, and in turn population growth. 808 Frequency-dependent pathogen damage (Condition 1) may occur because: (A) pathogen species 809 are host-specialized, (B) pathogen communities differentiate among host species, or (C) multi-810 host pathogens cause host-specific impacts. When these conditions are met, pathogens cause per-811 capita population growth rates to decline with relative abundance, stabilizing species 812 coexistence. If pathogens were removed, per-capita growth rates would decline less steeply with 813 relative abundance, making stable coexistence less likely. 814