Grapevine bacterial communities across the Central Valley of California

Plant organs (compartments) host distinct microbiota which shift in response to variation in both development and climate. Grapevines are woody perennial crops that are clonally propagated and cultivated across vast geographic areas, and as such, their microbial communities may also reflect site-specific influences. These site-specific influences, and the microbial differences across site compose ‘terroir’, the environmental influence on wine produced in a given region. Commercial grapevines are typically composed of a genetically distinct root (rootstock) grafted to a shoot system (scion) which adds an additional layer of complexity. In order to understand spatial and temporal patterns of bacterial diversity in grafted grapevines, we used 16S rRNA metabarcoding to quantify soil and compartment microbiota (berries, leaves, and roots) for grafted grapevines in commercial vineyards across three counties in the Central Valley of California over two successive growing seasons. Community composition revealed compartment-specific dynamics. Roots assembled site-specific bacterial communities that reflect rootstock genotype and environment influences, whereas bacterial communities of leaves and berries displayed associations with time. These results provide further evidence of a microbial terroir within the grapevine root systems but also reveal that the microbiota of above-ground compartments are only weakly associated with the local microbiome in the Central Valley of California.

We sampled vines in three commercial vineyards located along a 177 kilometer north-south 129 transect running through Madera, Merced, and San Joaquin counties in central California ( Figure   130 to 'Freedom', '1103P, and 'Teleki 5C'). Six collection blocks were sampled at the San Joaquin 138 site (i.e., sets of 24 vines, at least 10 vines from the edge, across 2-3 rows; Figure 1A; Table S1). 139 In the other two vineyards, Madera and Merced, collection blocks for four of the six 140 scion/rootstock combinations were sampled as not all scion/rootstock combinations were present 141 at these vineyards ( Figure 1A; Table S1). In Madera, blocks were sampled for 'Cabernet  Figure S1). Sampling periods coincided with multiple developmental stages 152 for the vines, starting at early fruit formation to veraison and, for 'Chardonnay', early harvest.

153
Three compartments (roots, leaves, berries) were sampled from each vine. Roots were 154 collected at a depth of 20-30 cm using a sterile shovel. Three to five leaves approximately 8-12 155 cm in diameter were collected at roughly the middle position along a shoot and at a height of 1.5 156 m on the vine. Berries were collected as an intact cluster. We measured total soluble solids (sugar ( Figure 1B- used to remove contaminants using the prevalence-based detection method with a threshold of 0.5, 202 removing contaminant ASVs more prevalent in the negative controls than real samples. The data 203 set was filtered to remove singletons by retaining only ASVs present in five or more samples and 204 by removing samples with a read count less than 1,000. 206 All analyses were conducted within the R environment v4.1.0 (R Core Team 2021). We  forest classifier, the dataset was randomly split into a training set (80%) and a testing set (20%).

248
Optimal hyperparameters for each classifier were determined using a grid search over the number   (Table S5; P < 0.001) but not the second PC (Table S5) (Table S9; P = 0.041), no post-hoc comparisons were significant among sites. There was a significant interaction between rootstock and scion 320 (Table S9;    leaves were tightly clustered together on axis 3 ( Figure 4B). The bacterial composition of the root 338 compartment samples was relatively stable over the course of the growing seasons but was variable 339 within a site, particularly Merced, and across the sites ( Figure 4C; Table 2). 340 We examined the impact of experimental factors on the top ten phyla based off relative  Figure 4D). Post-hoc comparisons showed that 'Teleki 5C' drove the differences, 355 with higher relative abundances for Chloroflexi (mean = 1.9% increase), Planctomycetota (mean 356 = 1.87% increase), and Myxococcota (mean = 0.57% increase) and lower relative abundance for 357 Verrucomicrobiota (mean = 1.01% decrease). Compared to rootstock genotype, scion genotype 358 had a smaller impact on relative abundance of the top ten phyla for roots. Only four of the ten 359 phyla of the root compartment showed significant impacts for scion genotype. In Acidobacteria, 360 Chloroflexi, Planctomycetota, and Proteobacteria, we found scion cultivar explained less than 5% 361 of the variance ( Figure 4D). Post-hoc comparisons showed that 'Cabernet Sauvignon' had a larger 362 relative abundance of Acidobacteria, Chloroflexi, and Planctomycetota but the difference for each 363 phylum was <1% between scion genotypes, while Proteobacteria had a smaller relative abundance 364 (mean = 1.16% decrease).
Finally, we examined the influence of two aspects of time, collection year, and sugar 366 content of the berries (an analog for vine development within a season) on the relative abundance 367 of the top ten phyla of the root compartment ( Figure 4D). Collection year significantly impacted 368 six of ten phyla. Collection year had a particularly strong impact on the relative abundance of  Table 2). Taxonomic 381 barplots of the top ten phyla of shoot compartments showed fluctuations in relative abundance but 382 were non-specific to collection sites ( Figure S4). Using the same linear model framework as above,  Machine learning accurately predicts the root compartment but not berries or leaves 394 We used machine learning to identify the factors that were predictable across the 395 experimental design ( Figure S5) and to pinpoint the ASVs that aided the accuracy of those 396 predictions ( Figure S6-7). Overall, model accuracy was 68% ( Figure S5A); however, classifier 397 was near perfect when predicting root samples (F1 = 0.986; Table S11) but was less accurate when  Table S11). For collection year model accuracy was 83% ( Figure S5D) and an F1 score of 0.882 404 (Table S12). For scion genotype, model accuracy was 61% ( Figure S5E) and with an F1 score of 405 0.623 (Table S12). Finally, for sugar content, model wide accuracy was 71% ( Figure S5F) and an 406 F1 score of 0.795 (positive class was pre-ripening; Table S12).

407
When examining the phyla that were the most important in the classifier's predictions, we 408 found that across all factors many of the phyla have similar relative importance to their respective 409 classifier ( Figure S6). Proteobacteria was the most important phylum making between 39.7-54.6% of the relative importance to the classifier and Actinobacteria was the second most important 411 phylum with between 14-24.9% of the relative importance to the classifier ( Figure S6).  The goal of this study was to investigate factors influencing bacterial communities of 420 grapevine roots, leaves, and berries, including rootstock genotype, scion genotype, and vineyard 421 site, within growing seasons and over multiple years. We observed differences in soil texture, 422 elemental composition, and bacterial communities among vineyard sites sampled in this study. We 423 detected differences in bacterial composition of grapevine root compartments across sites; 424 however, site specific differences were less pronounced in the microbiota of the berries and leaves.

425
Both rootstock and scion genotype impacted composition and diversity of vine microbiota. Using 426 brix (berry sugar content) as a proxy for development, we observed only minor associations 427 between developmental stage and bacterial community composition of the berries and leaves. This 428 suggests that berry and leaf bacterial communities are likely seeded from tissues with mature 429 microbiomes and undergo largely stochastic changes in community composition across the season. 432 In the cultivation of wine grapes, the term terroir describes regional environmental factors, 433 including soil properties, geography, and climate, that influence characteristics of wine (Seguin    (Table S9 & S10). Despite this, we observed compartment specific seasonal    Bold values indicate significant results, P < 0.05.