Conditionality of Soil Microbial Mediation of Plant Phenotype

Purpose: While distinct soil microbiomes and individual soil microbial taxa can alter particular plant traits under highly controlled conditions, little is known about the role of particular microbial taxa and microbial functions within complex soil microbial communities for mediating plant phenotypes or if the strength of microbial mediation of plant phenotype varies among plant species or plant phenotypic traits. Examining how the plant phenotype spectrum is inuenced by the taxonomic and functional composition of complex soil microbial communities allows for a more accurate understanding of the biotic environmental drivers of plant phenotype. Methods: Using rhizosphere soil collected from eld sites, we conducted a microbiome transfer glasshouse experiment to test the hypothesis that the taxonomic and functional composition of different soil microbiomes would differentially shift growth, physiological or reproductive phenotypes of three Solidago species. Results: We found that soil microbiome inoculations inuenced Solidago growth traits more than physiological and reproductive traits. We found that root growth of one of the Solidago species was negatively correlated with 77% of the indicator bacterial and fungal taxa from one of the soil microbiome treatments. Conclusions: Soil microbial mediation of plant phenotype varies by plant traits, is not universal across plant species, and can be associated with a small number of microbial taxa. This study illustrates that specic microbial taxa within a soil microbiome are associated with shifts in plant phenotype by pinpointing important individual microbial taxa from complex eld soil microbial communities. and identify and/or in complex are plant phenotypes. Using associated with three phenotypically distinct Solidago species, we conducted a glasshouse experiment and inoculated three Solidago species in separate treatments of each eld-collected soil and microbiome. tested the following hypotheses: 1) Soil microbiome source inoculation will differentially alter phenotypes of three Solidago species; 2) Soil microbiome source is associated with distinct taxonomic and/or functional soil microbial communities; 3) Specic microbial taxa and/or microbial functions are associated with particular Solidago phenotypes. Shifts in Solidago phenotypes between microbiome source treatments would indicate that plant traits are inuenced by variation in microbial taxonomic and/or functional composition. Variation in response to microbiome treatments among traits would indicate conditional effects of microbial mediation of plant phenotype. Variation in response to microbiome treatments among Solidago species would indicate conditional effects of microbial mediation of plant phylogeny.


Introduction
In the past 10-15 years, numerous and diverse relationships discovered between plants and the soil microbiome have shifted the long-established paradigm of plant phenotype as the sole product of interactions between a plant's genes (G) and the abiotic environment (E) (i.e., G x E interactions; Clausen et al., 1948;Conner & Hartl, 2004) to that of a 'holobiont' interpretation (G x G x E interactions), in which microbes at the root-soil interface serve as a reservoir of additional genes and functions for the host plant ( Wagner et al., 2014) and reproductive traits such as fruit (Lau & Lennon, 2011 and ower production (Lau & Lennon, 2012).
While ndings from these research elds show that both isolated individual microbial taxa and diverse soil microbial communities can in uence plant function, pinpointing the important individual microbial taxa and functions within complex soil microbial communities remains a challenge. Identifying signi cant individuals or functions within complex microbial communities is crucial for advancing ecology of natural ecosystems because plants in natural landscapes interact simultaneously with a multitude of bene cial, benign, and pathogenic microbes (Morris et al., 2007;Putten et al., 2016;Zolla et al., 2013). Bene cial or deleterious effects from individual taxa may be enhanced or suppressed by interactions with other nearby microbial members. Examining how the taxonomic and functional composition of soil microbial communities affects plant phenotype will allow for a more accurate understanding of the surrounding biotic environmental drivers of plant phenotype.
The goal of this study was to identify the strength of soil microbial mediation for different plant phenotypes, the consistency of these relationships among plant species, and identify speci c soil microbial taxa and/or functions in complex eld soil communities that are associated with particular plant phenotypes. Using eld soils associated with three phenotypically distinct Solidago species, we conducted a glasshouse experiment and inoculated three Solidago species in separate treatments of each eld-collected soil and microbiome. We tested the following hypotheses: 1) Soil microbiome source inoculation will differentially alter phenotypes of three Solidago species; 2) Soil microbiome source is associated with distinct taxonomic and/or functional soil microbial communities; 3) Speci c microbial taxa and/or microbial functions are associated with particular Solidago phenotypes. Shifts in Solidago phenotypes between microbiome source treatments would indicate that plant traits are in uenced by variation in microbial taxonomic and/or functional composition. Variation in response to microbiome treatments among traits would indicate conditional effects of microbial mediation of plant phenotype. Variation in response to microbiome treatments among Solidago species would indicate conditional effects of microbial mediation of plant phylogeny. Correlations between speci c microbial taxa and/or microbial functions and particular Solidago phenotypes would provide evidence for the importance of individual taxonomic or functional components within a microbiome for in uencing plant phenotype.

Study system
Solidago species are a model system for this study because they commonly occur across North America, with 120 species native to the United States (Semple, 2016) that grow in variable habitats, with different morphologies and phenotypes. We chose to use S. caesia, S. exicaulis, and S. gigantea in this study because they were the most abundant Solidago species found across our sampling range (northeastern TN) and vary in evolutionary history, leaf, stem, and ower morphology and habitat preference. Solidago caesia and S. exicaulis grow in woodlands and belong to the Glomeruli oraea subgroup of Solidago (Semple, 2016). Solidago gigantea grows in meadows and elds and belongs to the Triplinerviae subgroup (Semple, 2016). Furthermore, previous work has found evidence for the in uence of interspeci c and genotypic diversity on above-and belowground biomass of S. altissima and S. gigantea (Genung et al., 2012(Genung et al., , 2013, suggesting that some Solidago phenotypes may be mediated in part by modi cations of soil biota from neighboring Solidago species.

Preliminary eld surveys
To assess differences in plant phenotypes among the three Solidago species, we conducted eld surveys of three geographically distinct populations of each species, all located throughout northeastern Tennessee, U.S.A. In May 2017, we measured stem height, stem base diameter, speci c leaf area (SLA), and stomatal density of 15 randomly selected putative genotypes of each species (S. caesia, S. exicaulis, and S. gigantea) in northeastern TN for a total of 45 individuals per species (Fig. 1a). The eld survey con rmed that the three species vary in this suite of growth and physiological phenotypes (Table S1).

Soil collection and processing
To assess Hypothesis 2 that the soil microbiome sources have distinct taxonomic and/or functional microbial communities, we collected rhizosphere soil from each genotype in the eld surveys by collecting soil attached to the roots of each plant (Fig. 1b). We pooled individual soil samples by eld site to represent an average belowground microbiome of three soil sources (n = 3 sites per soil source). While we tried to collect soil microbes that were only associated with the rhizosphere soil of each plant species, it is likely that we also captured microbes that are representative of surrounding non-rhizosphere soil. Due to this fact and that some climatic and edaphic soil characteristics including mean annual temperature, soil organic matter content, and soil bulk density slightly varied among the three groups of Solidago species sites (Table S2), we refer to the three groups of sites as soil microbiome sources rather than soils associated with each Solidago species. Soil samples were transported to the laboratory on ice and stored at 0°C until analysis at the University of Tennessee, Knoxville, TN, U.S.A. A 2 g subsample of soil from each eld site was stored at -80°C for molecular analysis. We assessed the taxonomic community composition of the soils using high-throughput amplicon sequencing of the V3-V4 region of the 16S rRNA gene (16S) and the ITS2 region of the internal transcribed spacer gene regions for bacteria and fungi, respectively. Detailed methods are described in Methods S1 of the Supplementary Materials.
We performed all amplicon sequence processing using the DADA2 platform (Callahan et al., 2016). For 16S sequences, primers were removed prior to the DADA2 pipeline using the cutadapt function in conda. Samples were normalized for sampling depth with a variance stabilizing transformation with the DESeq2 package (Love et al., 2014). We chose this method over the common practice of rarefaction because rarefaction results in loss of data by using the lowest sampling depth and it in ates variances across samples (McMurdie & Holmes, 2014). Taxonomy of ASVs was assigned using the RDP (Wang et al., 2007) and UNITE (Abarenkov et al., 2010) databases for bacteria and fungi, respectively. After processing, we had 16,245 bacterial and 2,565 fungal ASVs, respectively. Additionally, we assigned fungal ASVs to functional guilds using the FUNGuild database (Nguyen et al., 2016). For analyses, we assigned taxa to one of seven broad functional guilds: arbuscular mycorrhizal fungi, ectomycorrhizal fungi, ericoid mycorrhizal fungi, endophytic fungi, plant pathogenic fungi, saprotrophic fungi, and "other." We considered only FUNGuild assignments with a con dence ranking of "highly probable" or "probable." Unassigned taxa were excluded from further guild-based analyses. Of the 2,565 fungal ASVs, 1,741 were assigned to a fungal guild. Of those assigned, we used the 1,328 ASVs that had a con dence ranking of "highly probable" or "probable." We assessed functional community composition with shotgun metagenomic sequencing, as detailed in Methods S1 of the Supplementary Materials. Sequences retrieved from shotgun metagenomic sequencing were assigned to KEGG (Kyoto Encyclopedia of Genes and Genomes) ortholog numbers using the MG-RAST online annotation tool. KEGG orthologs assign genes to microbial complexes, functional sets, and metabolic pathways and are a common tool used to describe Moon Nurseries, Minnesota Native Landscapes). Seeds were refrigerated at 4°C prior to sowing, and then were sown by population into a commercial peat moss-based, non-mycorrhizal potting mix (Premier Promix BX, containing perlite, vermiculite, and limestone). A subset of Solidago seeds did not withstand surface sterilization trials, so we did not surface sterilize the seeds used in the experiment. While it is possible that any seed-borne microbes may have impacted plant phenotype, all plants were grown in all soil treatments and exposed to the same glasshouse conditions, such that any effect of seed-borne microbes on plant phenotype should be equally distributed across treatment categories.
After approximately three weeks of growth, 54 similar-sized seedlings of each population were individually transplanted into half-gallon circular pots into soil inoculum treatments which consisted of factorial combinations of microbiome source (Microbiome source 1 vs. Microbiome source 2 vs. Microbiome source 3) (Fig. 1c). Furthermore, since soils from each eld site of each microbiome source were kept separate, seeds were planted into three sites of Microbiome source 1, three sites of Microbiome source 2, and three sites of Microbiome source 3. Each pot was inoculated with 2 teaspoons of eld soil (< 1% of the total pot volume) to reduce effects of variation in soil nutrients on plant phenotypic responses (Troelstra et al., 2001). In total, 243 pots were established: 3 Solidago species x 3 seed populations x 3 microbiome sources (Microbiome source 1, Microbiome source 2, Microbiome source 3) x 3 eld soil sites x 3 replicates = 243 total pots). Pots were randomly positioned in the glasshouse based on random number assignments. All plants were treated monthly for thrips and white ies throughout the experiment (0.5 tsp/gal Avid 0.15 EC insecticide, 0.5 tsp/gal AzaGuard insecticide). Plants were equally watered from above, as needed (approximately 4 days/week), and allowed to grow for 5 months in a glasshouse at the University of Tennessee.
A suite of plant phenotypes was measured during and post-experiment. Stem height and stem diameter were measured every two weeks for the rst two months of growth, then at 13 weeks and at the termination of the experiment at 20 weeks. Relative growth rates were calculated from these data. For each individual plant, timing of ower bud formation (hereafter referred to as " ower bud break") and owering were monitored with daily surveys by recording the day of the appearance of the rst distinguishable ower bud and rst open ower, respectively. Prior to termination of the experiment, an average of four healthy and mature leaves were randomly selected per plant, scanned using WinFOLIA software (Regent Instruments Inc.), oven-dried at 70°C for 72 hours (Pérez-Harguindeguy et al., 2016), and weighed to calculate speci c leaf area (cm 2 /g) (SLA). After ve months of growth and regular watering, each individual was harvested and separated into shoot and root biomass and in orescence biomass. Shoot and root tissue was weighed after 48 hours of oven-drying at 60°C. Prior to drying, roots were carefully rinsed over 2 and 0.5 mm sieves to remove lingering soil and collect all ne roots.

Statistical Analyses
In the eld survey, we analyzed differences in Solidago phenotypes using linear mixed-effects models with the lmer function in the lme4 package (Bates et al., 2014). We built separate mixed-effects models for each phenotype (stem height, stem diameter, SLA, and stomatal density) using Solidago species as the xed effect and population as the random effect. When necessary, all data were transformed to conform to normality before analysis. To test Hypothesis 1 that phenotypes of each Solidago species differ when grown in soils inoculated with microbial communities associated with a different microbiome source, we built linear mixed effects models with the lmer function in the lme4 package. First, to identify traits most important to growth, physiology, and reproduction and to reduce Type I error, we tested for correlations between the ten phenotypes measured from the glasshouse experiment (relative growth rate in stem height, stem diameter at maturity, shoot biomass, root biomass, total biomass, root to shoot ratio, SLA, ower bud break, days to ower, in orescence biomass) using the cor.test function. We chose to exclude stem diameter, total biomass, and root to shoot ratio from the analysis because they were all signi cantly correlated with two other growth phenotypes, shoot and root biomass (Table S5). We also chose to exclude days to ower and in orescence biomass from the analysis because the experiment ended before the majority of S. gigantea individuals owered. Relative growth rate, shoot biomass, root biomass, SLA, and ower bud break were included in the analysis.
Multiple models were used to assess Hypothesis 1. Separate models were built for the ve phenotypes (relative growth rate, shoot biomass, root biomass, SLA, and timing of ower bud formation). When necessary, all data was transformed to conform to normality before analysis. First, to test that differences in soil microbial community composition have a general effect on plant phenotypes regardless of plant species, we built linear mixed effects models with microbiome source as a xed effect and Solidago species, seed population, and eld soil site as random effects. Then to identify interspeci c variation in response to microbial community composition we separated the dataset by each Solidago species and built individual linear mixed effects models for each Solidago species with microbiome source as a xed effect and seed population and eld soil site as random effects. For all models, we used the Anova function to calculate ANOVA tables using Type II sums of squares, with signi cance assessed for each xed effect using Wald X 2 statistics. If any of the xed effects were signi cant, we conducted post hoc Tukey contrasts using the TukeyHSD function.
To test Hypothesis 2 that each microbiome source is associated with distinct taxonomic and/or functional soil microbial communities, we took multiple approaches. First, we assessed microbial diversity across microbiome source by calculating hill numbers based on ASV counts and unique KEGG identities using the hill_div function in the hilldiv package (Alberdi & Gilbert, 2019). Hill numbers serve as effective numbers of diversity that provide more intuitive estimates of diversity compared to traditional diversity indices based on entropy (Chao et al., 2014). We calculated hill numbers for all orders of diversity at q = 0, q = 1, and q = 2, and tested for signi cant differences in hill numbers between microbiome source at each order of diversity using the div_test function in the hilldiv package. A diversity order q = 0 provides raw richness by weighting rare taxa the same as abundant taxa and thus not accounting for species' abundances. A diversity order q = 1 weights ASVs by their abundance but without disproportionately favoring abundant taxa. A diversity order q = 2 overweighs abundant ASVs.
Second, we created Bray-Curtis distance matrices for microbial taxonomic and functional composition of the nine eld soils. To assess variation in community composition of bacteria, fungi, and KEGGs across microbiome source, we conducted PERMANOVA analysis with 9,999 permutations using the adonis function in the vegan package (Oksanen et al., 2019). Prior to conducting PERMANOVA we con rmed homogeneity of dispersion across microbiome source with the betadisper function in the vegan package. We then performed a distance-based redundancy analysis (db-RDA) using the dbrda function in the vegan package to assign variation in composition of bacteria, fungi, and KEGGs to microbiome source and geographic location. We conducted three individual db-RDAs for bacteria, fungi, and KEGG composition. We used the anova.cca function in the vegan package to assess the cumulative signi cance of microbiome source and geographic location on community composition. We partitioned the variation in composition with respect to microbiome source and geographic location using the varpart function in the vegan package. To visualize composition of bacteria, fungi, and KEGGs among soil origin, we used principal coordinate analysis (PCoA) for ordination based on the Bray-Curtis distance matrices.
We To test Hypothesis 3 that speci c microbes and/or microbial functions are associated with particular Solidago phenotypes, we assessed the effect of variation in microbial indicator taxa composition on Solidago phenotypes that responded to microbiome source treatment. Since no KEGG identities were identi ed as indicators across the three microbiome sources, subsequent analyses were conducted only with bacterial and fungal indicator taxa. Using a db-RDA, we assigned variation in composition of bacterial and fungal indicator taxa to the three microbiome sources and geographic location. We then extracted the axes scores from the db-RDA model. For each phenotype, we built a linear model that included the two axes (CAP1, CAP2) from the db-RDA model as xed effects. A signi cant relationship between db-RDA axes and plant phenotypes would indicate that differences in the community of bacterial and fungal indicator taxa associated with each microbiome source are associated with shifts in plant phenotype. To pinpoint individual bacterial and fungal indicator taxa that may be associated with particular plant phenotypes, we built linear models to test for correlations between the relative abundance of each bacterial and fungal indicator taxon and each phenotype that showed signi cant responses to the axes of variation from the indicator species db-RDA model.
All analyses were performed in R (R Core, 2020

Results
Plant phenotype responses to soil microbiome sources (glasshouse experiment) While the three Solidago species overall varied signi cantly in relative growth rate, shoot and root biomass, and ower bud break, only root biomass differed by microbiome source (χ 2 = 6.14, p = 0.04) ( Table 1). Among all three Solidago species, there was 29% greater root biomass production when plants were grown in inoculum from microbiome source 1 relative to microbiome source 3 (Tukey post hoc: p = 0.05) (Fig. 2a). In partial support of Hypothesis 1, the species-speci c models showed that phenotypic responses to microbiome source inoculum varied by Solidago species and by phenotype. Solidago caesia shoot biomass differed among microbiome source treatments, whereas no S. exicaulis or S. gigantea phenotypes differed among microbiome source treatments (Table 2). Solidago caesia produced 8.9% more shoot biomass when grown in inoculum from microbiome source 2 relative to microbiome source 3 (Tukey post hoc: p = 0.09) (Fig. 2b), indicating that different soil microbiomes can shift plant traits.  Community composition among soil microbiome sources Across the three soil microbiome sources, we identi ed over 16,000 bacterial and 2,500 fungal ASVs. Taxonomic and functional diversity of soil microbial communities did not vary by microbiome source at any order of diversity (Tables S3,  S4). In partial support of Hypothesis 2, whole microbiomes did not differ in taxonomic or genetic pathway composition among the microbiome sources, but distinct indicator taxa were identi ed for each microbiome source. PERMANOVA analysis revealed that taxonomic and functional composition of soil microbial communities did not vary signi cantly by Indicator species analysis identi ed signi cant bacteria and fungi indicator taxa for each microbiome source. In total, 77 bacterial ASVs (out of 16,245 detected; 0.5%) and eight fungal ASVs (out of 2,565 detected; 0.3%) were identi ed as indicator taxa among the three microbiome sources (Fig. 4, Tables S6, S7). Twenty-nine bacterial ASVs were uniquely shared among microbiome sources 1 and 2, whereas microbiome source 3 uniquely shared only six bacterial ASVs with either microbiome source 1 or 2. Fungal guilds were assigned to approximately 68% of the fungal ASVs. Of those assigned to a guild, approximately 76% had a con dence ranking of "probable" or "highly probable." Out of the ve fungal guilds (arbuscular mycorrhizal fungi, ectomycorrhizal fungi, ericoid mycorrhizal fungi, endophytic fungi, and plant pathogenic fungi), ericoid mycorrhizal fungi (χ 2 = 11.29, p = 0.004) and endophytic fungi (χ 2 = 20.14, p < 0.0001) differed signi cantly among the microbiome sources (Table S8). Both microbiome sources 1 and 2 had approximately 1.5-and 2fold greater abundance of ericoid mycorrhizal fungi and endophytic fungi, respectively, than microbiome source 3 (Fig. 5d, e).
KEGG composition overall did not differ among the three microbiome sources, indicating functional redundancy among soil microbial communities. Of the 122 pathways identi ed, less than a quarter accounted for more than 1% of relative abundance of all KEGGs among the three origins (Table S9). Of this subset, 70% were pathways involved in metabolism of either energy (in the form of nitrogen, methane, sulfur, and oxidative phosphorylation), amino acids, carbohydrates, or lipids. The most abundant pathways across the three soil origins were two pathways for ATP-binding cassette (ABC) transporters, which accounted for 20% of the relative abundance. No KEGG pathways were detected as indicators among the microbiome sources.

Correlations between soil microbiome composition and plant phenotypes
In support of Hypothesis 3, individual microbial taxa were associated with speci c Solidago phenotypes. We only examined shoot biomass of S. caesia as it was the only phenotype that responded to microbiome source treatments. Axes of variation in composition of the bacteria indicator taxa were signi cantly correlated with S. caesia shoot biomass (Axis CAP1: F = 7.63, p = 0.03). Relative abundance of 77% (20 out of 26) of the bacterial and fungal indicator taxa of microbiome source 3 were signi cantly negatively correlated with S. caesia shoot biomass when S. caesia was grown in inoculum of microbiome source 3 (Fig. 6, Table S10). Although S. caesia produced more shoot biomass when grown in inoculum of microbiome source 2 compared to that of microbiome source 3 (Fig. 2b), none of the eight bacterial indicator taxa or the one fungal indicator taxon of microbiome source 2 were signi cantly positively correlated with S. caesia shoot biomass.

Discussion
Identifying ways in which the taxonomic and functional composition of the soil microbiome in uences plant phenotype is a central challenge for understanding the overall importance of complex soil microbial communities on plant function, as well as how changes to soil microbial communities may in turn affect plant function. While recent studies have explored the importance of both whole soil microbiomes and individual soil microbial taxa on particular plant phenotypes, it is also crucial to understand if and how particular taxa and functions within complex soil microbial communities in uence a broad spectrum of plant phenotypes and if these relationships are consistent across multiple plant species. In this study we compared the taxonomic and functional composition of rhizosphere soil microbial communities from three phenotypically distinct and naturally occurring Solidago species and sites (referred to as soil microbiome source above) and conducted a microbiome transfer glasshouse experiment to test for plant phenotypic shifts in response to eld soil inoculum. We found that soil microbiome taxonomic variation can shift some plant phenotypes and that this response varied by plant species, with some species more responsive to microbial taxonomic variation than others. Speci cally, we identi ed indicator bacteria and fungi associated with each microbiome source, some of which were correlated with shifts in plant growth responses. We found that microbiome source altered growth traits for only one of the three Solidago species. Lastly, we found that plant growth traits were more likely to be in uenced by variation in soil microbial communities than physiological or reproductive traits. Together, these ndings show that soil microbial mediation of plant phenotype 1) varies by plant traits, 2) is not consistent across plant species, and 3) can be in uenced, in part, by a small number of microbial taxa.

Soil microbial mediation of plant phenotype varies by plant phenotype and plant species
We found that soil microbial communities can shift some plant phenotypes, but that the strength of microbially-mediated phenotypic plasticity varies by plant phenotype and by plant species. Across all three Solidago species, plants grown in inocula from microbiome source 1 produced more root biomass compared to plants grown in inocula from microbiome source 3. While no S. exicaulis or S. gigantea phenotypes responded to microbiome source treatments, S. caesia produced more shoot biomass in microbiome source 2 inocula relative to microbiome source 3 inocula. Although we found that the amount of soil plant pathogenic fungi was similar across all three soil origins, we did nd that microbiome sources 1 and 2 had signi cantly greater amounts of endophytic and ericoid mycorrhizal fungi than microbiome source 3.
A difference in plant-bene cial fungi may in part account for the greater growth of S. caesia in microbial communities of microbiome source 2 relative to microbial communities of microbiome source 3.
Despite the breadth of research on plant-soil biota relationships, the role of soil microbial communities on plant phenotype is heavily limited by the scarcity of knowledge of these relationships on traits such as physiology and reproduction. indicating that not all plant phenotypes are in part mediated by soil microbes. However, further examination of soil microbial mediation across plant phenotypes is needed to reveal if the ndings here represent a general trend that microbial mediation is stronger for growth traits than for physiological or reproductive traits.
Soil microbial mediation of plant phenotype can be in uenced in part by small number of microbial taxa Despite similarity in overall microbiome composition among the three soil microbiome sources, we identi ed speci c bacterial and fungal indicator taxa of each microbiome source. In general, Proteobacteria taxa were more highly abundant in microbiome sources 1 and 2 than in microbiome source 3, whereas Acidobacteria taxa were more highly abundant in microbiome source 3 than in microbiome sources 1 and 2. Within the Proteobacteria phylum, indicator taxa of microbiome sources 1 and 2 spanned a larger diversity of taxonomic orders including Rhizobiales, Rhodospirillales, Burkholderiales, and Xanthomonadales relative to those of microbiome source 3 which comprised the orders of Rhizobiales and Myxococcales. These ndings highlight the importance of taxonomic resolution when assessing the role of the soil microbiome on plant phenotype as root-associated soil is known to contain some of the highest microbial biodiversity on Earth (Berendsen et al., 2012;Buée et al., 2009;Curtis et al., 2002). In this study, we identi ed over 16,000 bacterial ASVs and over 2,500 fungal ASVs among the three microbiome sources. These ndings suggest that identifying microbial differences among focal groups may require focusing on speci c indicator taxa that have high a liation with the plant species or eld site rather than overall microbiome composition.
We found that differences in indicator taxa among separate soil microbial communities may contribute to shifts in plant phenotype even though this relationship is likely not consistent across plant species. We found that Solidago caesia produced signi cantly less vegetative biomass in inocula of microbiome source 3 compared to inocula of microbiome source 2, and that the relative abundance of 77% of the indicator taxa of microbiome source 3 were correlated with decreases in S. caesia shoot production. These indicator bacteria included mostly members of the Acidobacteria and Actinobacteria phyla in addition to members of the Bacteriodetes, Firmicutes, and Proteobacteria phyla. Indicator fungi included members of the Ascomycota and Chytridiomycota phyla. Despite greater shoot production in microbiome source 2 inocula, none of the microbiome source 2 indicator taxa were correlated with positive shifts of S. caesia shoot production.
Although these indicator taxa account for a very small proportion of the total microbial communities identi ed in this study, it is notable that out of the thousands of ASVs identi ed in microbiome source 2, twenty bacterial and fungal ASVs explained on average 7% of the variation in root biomass when S. caesia was grown in inocula of microbiome source 2. This suggests that individual soil microbial taxa may be involved, in part, with mediating some plant traits. This study demonstrates the utility of this approach of examining correlations between individual microbes and plant traits.
Identifying individual microbial taxa that are associated with shifts in plant traits can pinpoint particular microbial taxa to target for further experiments that test causative mechanisms of plant trait variation. While the correlative relationships we identi ed in this study are context-speci c to these particular microbial taxa and plant species, the approach used here can be applied to any plant-microbial system.
While we did not assess microbial dormancy in this study, it likely plays a signi cant role in soil microbial mediation of plant phenotype. Dormancy, in which individuals undergo a temporary reduced state of metabolic activity, has long been hypothesized to be widespread among microorganisms because it allows them to cope with environmental variability, particularly when conditions are unfavorable (Lau & Lennon, 2011;Stevenson, 1977). Differentiating between active and dormant microbial taxa requires examining the active ribosomal RNA in addition to the total ribosomal DNA. Since we only used rDNA-based techniques in this study, our inferences are limited to microbial taxa that are potentially active.
However, despite the fact that the indicator taxa accounted for a very small amount of the diversity of each microbiome source, evidence suggests that rare taxa may confer particular importance within a microbiome. A previous study examining proportions of rRNA to rDNA in temperate lakes found that rare taxa had a higher probability of being metabolically active than common taxa (Jones & Lennon, 2010). Combined with the observation that soil microbiome diversity is primarily comprised of rare taxa (Elshahed et al., 2008), our ndings and those from Jones and Lennon (2010) highlight the importance of examining how less abundant rare taxa within soil microbial communities may in uence plant phenotype. Other microbial interactions within the soil environment also likely in uenced the plant trait variation we observed, such as differences in microbial growth rates, differences in decomposition via extracellular enzymes, and changes to the microbial communities due to conditioning from the plants. However, testing these mechanisms was outside of the scope of this study.
In this study, we found high similarity in soil microbial function (i.e. KEGG pathway composition) among the three microbiome sources, suggesting functional redundancy in which the absence of one or more microbial species does not greatly affect the functioning of the whole microbial community because the same functions are ful lled by many

Conclusions
Soil microbes represent a largely overlooked but often important biotic factor for in uencing plant phenotype. This is the rst study, to our knowledge, to examine how taxonomic and functional gene composition of complex soil microbial communities in uence a suite of multiple plant phenotypes across multiple plant species. Our study shows that soil microbiomes and speci c taxa within complex soil microbial communities can alter some plant phenotypes, but that not all plant species, even those belonging to the same genus, will respond to soil microbial communities in the same manner.
Thus, the belowground biotic environment is just one of a host of important biotic factors that can mediate plant phenotype, in addition to plant genetic background and abiotic environmental variation. While the ndings from this study are founded in ecology theory, identifying the nuances of relationships between soil microbes and plant phenotype has wide scale applications. Substantial efforts to engineer core rhizosphere microbiomes to optimize plant production signify the need to identify functional linkages between soil microbial communities and plants (Bakker et     Composition of (a) bacteria, (b) fungi, and (c) KEGG pathways among the three microbiome sources. Each data point represents a eld site.

Figure 4
Heatmap of relative abundance of the 77 bacterial ASVs produced from Indicator Species analysis across the three microbiome sources. Rows represent individual ASVs. Columns represent soil from individual eld sites. Taxa of each microbiome source with relative abundance 0.05 (5%) or greater are color coded by phylum. Figure 5 Mean abundance of (a) arbuscular mycorrhizal fungi (AMF), (b) ectomycorrhizal fungi (ECM), (c) plant pathogenic fungi, (d) ericoid mycorrhizal fungi, and (e) endophytic fungi. Emmeans are reported on the log scale. Data shown are pooled across samples (i.e. eld sites). Data that do not share letters are signi cantly different from one other (α < 0.05).