Host genotype and age shape the leaf and root microbiomes of a wild perennial plant

Bacteria living on and in leaves and roots influence many aspects of plant health, so the extent of a plant's genetic control over its microbiota is of great interest to crop breeders and evolutionary biologists. Laboratory-based studies, because they poorly simulate true environmental heterogeneity, may misestimate or totally miss the influence of certain host genes on the microbiome. Here we report a large-scale field experiment to disentangle the effects of genotype, environment, age and year of harvest on bacterial communities associated with leaves and roots of Boechera stricta (Brassicaceae), a perennial wild mustard. Host genetic control of the microbiome is evident in leaves but not roots, and varies substantially among sites. Microbiome composition also shifts as plants age. Furthermore, a large proportion of leaf bacterial groups are shared with roots, suggesting inoculation from soil. Our results demonstrate how genotype-by-environment interactions contribute to the complexity of microbiome assembly in natural environments.

"PlantDiv" = number of plant morphospecies present in each block (not counting B. stricta); "Veg" = percent vegetation cover of each block (estimated for each of 50 subblocks of area 10x10cm, then aggregated). All other variables describe chemical content of soils (units: pH = none, conductivity = umho/cm, all others = ppm). The bottom and top edges of the boxes mark the 25 th and 75 th percentiles (i.e., first and third quartiles). The horizontal line within the box denotes the median. Whiskers mark the range of the data excluding outliers that fell more than 1.5 times the interquartile range below the first quartile or above the third quartile (dots). (b) Principal components analysis of the environmental variables shown in (a).     Table  1. Bars depict one standard error of the mean.
Supplementary Figure 8 | Genotype, site, and genotype-by-site interactions shape glucosinolate (GLS) content. Least-squares mean values (plus or minus one standard error of the mean) are plotted for Total glucosinolate concentration and for BC-ratio, a measurement of glucosinolate quality. (a) In this study, BC-ratio was plastic among sites for roots but not leaves; (b) total root glucosinolate concentration was plastic among sites; (c) Genotype controls BC-ratio in leaves but not roots; (d) genotype effects on leaf BC-ratio are site-dependent. The fold change in abundance between 2-year-old and 4-year-old plants (estimated using NBMs as described in the main text) is plotted for the 29 "unnatural" OTUs in leaves and roots. Small, faded points indicate changes that were not significantly different from zero. Only 6 of these OTUs show the expected decrease in abundance with time since transplant, suggesting that some may in fact have been rare natural bacteria that were not observed in the relatively small number of wild samples, simply by chance; nevertheless we excluded them from other analyses as a conservative precaution. (b) Magnitudes of fold changes in abundance due to each source of variation are plotted for "natural" and "unnatural" OTUs. The sensitivity of the 29 unnatural OTUs to host genotype, age, and site is comparable to that shown by the natural OTUs. Note that whereas panel (a) only shows fold changes between 2-and 4year old plants, here fold changes between 2-and 3-year old plants and between 3-and 4-year old plants are also shown.

Supplementary Figure 10 | The root microbiomes of greenhouse-grown plants and potting soils differ strongly from those in natural environments. (a)
Principal coordinates analysis of weighted UniFrac distances reveals that despite extensive differences among field sites in bacterial community diversity and composition (see main manuscript Tables 1-2, Figs. 2-3), root communities in the three field sites are much more similar to each other than they are to root communities of plants grown in potting soil in the greenhouse. (b) Alpha diversity was much lower in potting soil (and in B. stricta roots grown in potting soil) than in bulk soils or roots at any of the field sites.
Statistics describe linear random-intercept models of unweighted UniFrac principal coordinates in leaves and roots. All p-values were adjusted for multiple comparisons using the sequential Bonferroni correction. Significance was assessed using Type III ANOVA with F-tests for fixed effects and likelihood ratio tests for random effects. Linear random-intercept models of glucosinolate profiles in leaves and roots. "BCratio," a metric of glucosinolate quality, was arcsine-square root transformed before analysis. Total [GLS] = total concentration of glucosinolates (μmol per mg dry tissue). Total [GLS] was square-root transformed before analysis. "HPLC batch" is a nuisance variable to control for noise attributed to differences among HPLC runs. Significance was assessed using Type III ANOVA with F-tests for fixed effects and likelihood ratio tests for random effects.

Supplementary Note 1 | Genetic variation and plasticity of glucosinolate content could partially underlie patterns of microbiome variation.
For a subset of plants harvested in 2011, we also quantified glucosinolates (GLS), secondary chemicals produced by B. stricta to protect against predators 1 .
Because GLS are known to influence plant-associated bacterial communities and microbial pathogens 2-3 , variation in GLS caused by genetic effects or phenotypic plasticity could partially underlie patterns of microbiome variation. To test this hypothesis we measured "BC-ratio," (the proportion of aliphatic GLS derived from branched-chain amino acids), which is a measure of GLS quality that affects biological activity 1,4 ; and in roots we also measured absolute concentration of GLS. We modeled GLS using the same predictors that we used to model microbiome features except for Year Harvested, because all GLS data was collected in a single year (2011).
Leaf BC-ratio varied strongly among genotypes and also showed a genotype-bysite interaction (Supplementary Fig. 8, Supplementary Table 6), consistent with the hypothesis that GLS quality partially underlies differential abundance of leaf-associated bacteria between genotypes depending on site ( Fig. 6; Fig. 7). Leaf BC-ratio did not differ by site when averaged across all genotypes (Supplementary Table 6).
In the current study both absolute GLS concentration and BC-ratio in roots varied among gardens, suggesting that if root-associated microbes are sensitive to GLS (as reported in other systems 3 ), the strong differences in bacterial communities among sites ( Fig. 2) may have been caused not only by biogeographic patterns, but also by plasticity of GLS production by the plant (Supplementary Table 6, Supplementary Fig. 8).
Fifty μL of each extract was analyzed on an Agilent 1100-series high-performance liquid chromatography machine with a diode array detector and a Zorbax Eclipse XDB-C18 column (4.6 x 150 mm, 5-micron pore size; Agilent Technologies, Santa Clara, USA).
We separated glucosinolates using a water-acetonitrile (ACN) gradient at 40°C and a flow rate of 1mL/minute: [ACN] was held at 1.5% for 6 minutes then increased to 2.5% by minute 8, to 5% by minute 15, to 18% by minute 17, to 46% by minute 23, to 92% by minute 24, and then decreased to 1.5% by minute 29. We identified compounds based on UV absorption spectrum at 229 nm and retention time 5 . We calculated the absolute amount of each compound by multiplying the area under its HPLC peak by the known amount of sinigrin (0.05 μmol), and then dividing by the area under the sinigrin peak and by its relative response factor 6 . We calculated "BC-ratio" as the proportion of aliphatic glucosinolates derived from branched-chain amino acids 1,4 . For roots only we also calculated total glucosinolate concentration by summing absolute amounts of all compounds and dividing by the dry mass of the tissue sample. Because we did not weigh leaf samples, we could not standardize by tissue mass, and therefore we report only BCratio for rosettes. BC-ratio was arcsine-square root transformed and Total [GLS] was square-root transformed to improve homoscedasticity. We fit linear mixed models with fixed effects Site + Genotype + Site*Genotype + Age + Site*Age and random intercepts Block + Line + HPLC batch. We then assessed statistical significance of fixed predictors using Type III ANOVA with Satterthwaite's approximation of denominator degrees of freedom in the package lmerTest 7 , and of random effects using likelihood ratio tests.

Supplementary Note 2 | Age-related changes in root bacterial communities are
partially, but not completely, consistent with a hypothesis of ongoing succession after transplant from greenhouse.
In fall 2011, we planted a small greenhouse replicate experiment with a subset of the genetic lines (4 lines per genotype) used in the field experiment. Seedlings were planted into the same brands of standard greenhouse potting soil that were used to grow rosettes for the field experiment, and watered with tap water (see main Methods).
Although there is no guarantee that the resident potting-soil bacterial community (or the tap water community) was exactly the same in 2011 as in 2008 or 2009, we assume here that it is an approximate replication of the initial inoculum for roots of our experimental plants. Roots were harvested from 6-week-old rosettes-the same developmental stage as the field experimental plants at the time of transplant-and processed along with the samples from the field. We summarized root community composition using unconstrained principal coordinates analysis of weighted UniFrac distances between OTU tables (after applying the variance-stabilizing transformation) for all root samples for which data were available, as described in the main Methods.
The dissimilarity between roots growing in potting soil and roots transplanted into the field dwarfed the dissimilarity between the three field sites (Supplementary Figure 10 If community turnover after transplant from greenhouse to field is the only mechanism of the observed effects of host age on root microbiomes, then we would expect to observe the following patterns: (i) As plant age increases, the average root communities at the three field sites should become more different from each other, because they have had more time to diverge from the common potting-soil inoculum; (ii) As plant age increases, root communities of experimental plants should become more similar to those of local endogenous plants; and (iii) As plant age increases, root communities of plants in the field should become less similar to those of plants grown in potting soil in the greenhouse, which represent the original common inoculum of the field plants.
To test these predictions, we compared the means of the three major PCoA axes for each age group in each site. To control for other known sources of variation (including host genotype, year of observation, block within field site, and technical nuisance variables), we used least-squares means from the Age x Site interaction term in the statistical models described in the main text. Considering the effect of age independently in each site allowed for communities to change in different directions in PCoA space without obscuring results; for instance, root communities might diverge from the potting-soil inoculum in the direction of positive PCo1 at one site, but in the direction of negative PCo1 in another; yet both are becoming more distinct from the potting-soil control. From the least-squares means of each age group at each site, for each of PCo1, PCo2, and PCo3 we calculated (i) the variance among sites; (ii) the absolute value of the distance from the mean endogenous root community at each site; and (iii) the absolute value of the distance from the mean potting-soil root community.
Each of these values was regressed onto plant age; the slopes of these regressions describe the movement of root communities in PCoA space as host plants age.
We tested the significance of each of these slopes using permutation tests. Age  Figure 11b). These results suggest that the ongoing replacement of potting-soil inoculum by wild bacteria after transplant is not the sole cause of the observed changes in root community composition attributed to plant age.
Detailed methods for greenhouse sub-experiment. Surface-sterilized seeds (1-minute wash and vortex in 70% ethanol + 0.1% Triton X-100; vortex and 15-minute incubation in 10% bleach + 0.1% Triton X-100; three rinses with sterile diH 2 O) were plated on autoclaved filter paper in petri dishes moistened with sterile diH 2 O, stratified in the dark at 4°C for one week, and then allowed to germinate in a growth chamber (22°C, ambient humidity, 11-hour day length). One week after germination, seedlings were transplanted into one of five soils: standard greenhouse potting soil (as described in main Methods) or wild soil collected from the field site Mah, Mil, Par, or Sil. In this analysis we focus only on the plants grown in potting soil, because we are specifically interested in replicating the starting conditions for the plants that were later moved into the field. All seedlings grew for 6 weeks in a greenhouse with conditions as described in the main Methods. Plants were watered with tap water as needed. We harvested roots and removed soil particles using flame-sterilized utensils, and rinsed them with sterilized diH 2 O. These greenhouse-grown roots (total N=162) as well as samples of unplanted bulk soils (N=23) were then processed along with the samples from the field (see main Methods).