Vineyard under-vine floor management alters soil microbial composition, while the fruit microbiome shows no corresponding shifts

The microbiome of a vineyard may play a critical role in fruit development, and consequently, may impact quality properties of grape and wine. Vineyard management approaches that have directly manipulated the microbiome of grape clusters have been studied, but little is known about how vineyard management practices that impact the soil microbial pool can influence this dynamic. We examined three under-vine soil management practices: 1) herbicide application, 2) soil cultivation (vegetation removal), and 3) natural vegetation (no vegetation removal) in a Riesling vineyard in New York over a three-year period. The microbiomes associated with soil and grapes were profiled using high-throughput sequencing of the bacterial 16 S rRNA gene and fungal ITS regions. Our results showed that soil bacterial composition under natural vegetation differs from that seen in glyphosate-maintained bare soil. Soil fungal composition under the natural vegetation treatment was distinct from other treatments. Although our study revealed soil microbiome shifts based on under-vine management, there were no corresponding changes in fruit-associated microbial composition. These results suggested that other vineyard management practices or environmental factors are more influential in shaping the grape-associated microbiome.


Results
Fungal communities cluster distinctly between soil and grapes. Fungal community profiles showed distinct clustering of samples derived from grapes, and soil collected under grapevines (Fig. 1). The Bray-Curtis distance metric was used to determine multivariate sample distances, which were visualized through an ordination of a principal coordinates analysis (PCoA). Axes 1 and 2 explained 69% of the variance in the data. The soil samples clustered together distinctly, and separately from grapes along the first PCoA axis, which explained 66% of the variance in the data (Fig. 1). Shannon diversity indices for OTUs differ between soil and grape samples ( Supplementary Fig. S1), but no diversity differences were observed among treatments within the sample type in each year.
Under-vine soil management impacted soil fungal community structure. We evaluated the impact of under-vine soil management on microbial community composition. The three-year average under-vine soil vegetation coverage rate for NV was greater than 70%, while coverage rates for CULT and GLY were less than 20% at veraison. PCoA plots with samples from each of the three years of the study (generated using the Bray-Curtis distance metric) showed that NV soil fungal communities differed from those of GLY and CULT treatments (Fig. 2a). Over the three years of the experiment, sample clustering was based primarily on vintage, with each vintage clustered, and then by treatment, where NV separated from GLY and CULT. However, no clustering pattern was detected among the CULT and GLY samples. Notably, the dissimilarities between NV and the other two soil treatments grew with time since groundcover establishment, suggesting possible intensification of the NV treatment effect over time. In 2015 and 2016, the soil samples were taken at two different vine phenological stages -bloom and harvest, which showed separation by PCoA ordination.
These observations were confirmed by statistical analysis. According to the three-year overall Permutational Multivariate Analysis of Variance (PERMANOVA), vintage and treatment effects were both significant (P < 0.001), while year-to-year climatic differences (R 2 = 0.159) explained more variation than treatment (R 2 = 0.114). The treatment effect was significant across all three years (p = 0.032 in 2014, p = 0.001 in 2015 and  (Table 1).
Unclassified fungal genera in soil samples ranged from around 10% to more than 25% relative abundance. However, analyses excluding the unidentified genera did not change the differentiation of NV samples from CULT and GLY samples on the ordination. The top five fungal genera found in the soil (excluding unclassified) were Verticillium, Nectria, Mortierella, Gibberella and Fusarium, based on average relative abundances across all soil samples (Fig. 3a). Fungal genera relative abundance differences were found in Gibberella, Neopestalotiopsis, Verticillium and an unclassified genus under Amphisphaeriaceae family, where NV soils contained fewer Gibberella (P < 0.005 in 2015) and more Verticillium (P < 0.05 in 2015 and 2016) compared to the other two treatments, and less Neopestalotiopsis (P < 0.05 in 2015 and 2016) and unclassified Amphisphaeriaceae (P < 0.05 in 2015 and 2016) relative to GLY soils. CULT soils had less Neopestalotiopsis (P < 0.05 in 2015 and 2016) compared to GLY soils (Fig. 3b). Among these genera, Neopestalotiopsis and Verticillium are found in the top five most important variables along with Monographella, Paraphaeosphaeria and unclassified genera under Nectriaceae in the Random Forest model for soil treatment prediction ( Supplementary Fig. S2).
Under-vine soil bacterial community structure was impacted by floor management practice.
The sequencing reads generated from the 2014 samples contained unexpectedly high amounts of short reads, whereas the sample sequences were comparatively low. Although the quality of the remaining reads was sufficient  for a within-year comparison, we decided to present the data year-by-year instead of the three-year overall analysis, due to the dramatic difference in read depth relative to 2015 and 2016 samples. Although the samples did not seem to cluster based on treatments on PCoA plots using UniFrac distance metrics (Fig. 4), the treatment effect was significant in year 2014 (p = 0.042) and 2016 (p = 0.013) according to PERMANOVA (Table 1). In fact, paired-PERMANOVA further revealed that the bacterial community structure among the treatments was different in 2014, where NV differed from GLY (p = 0.026) and CULT (p = 0.033), and 2016, where NV differed from GLY (P = 0.036) ( Table 2). Grape-associated bacterial community structure was not further examined due to low yield of bacterial DNA resulting in low PCR amplification.
Under-vine soil management did not impact fungal communities on grapes. Grape samples were collected at commercial harvest in each year. Over 71% of the variance in grape fungal community structure was explained by the first two PCoA axes, but the grape samples were not structured as a function of under-vine soil treatments (Fig. 2b). PERMANOVA and paired PERMANOVA were used to confirm that no community composition differences were found among treatments. The three-year overall PERMANOVA showed that the year-to-year differences were the only significant effects (Table 1).
Unclassified genera accounted for 5 to more than 30% of the relative abundance in grape samples. The top five fungal genera with the highest average relative abundance of the three years in the grape samples were Sporobolomyces, Aureobasidium, Rhodosporidium, Penicillium, and Entyloma. The fungal genera that differed in relative abundance in soil were not found to differ in relative abundance in grapes. Differences in relative abundance in grape-associated fungal genera were found in Penicillium, Sporobolomyces and unidentified genera across the years. The fungal genus Penicillium was found only in the 2014 grape samples, which was 16.6% in relative abundance, and Sporobolomyces was highest in relative abundance in 2015 (p < 0.05) and lowest in 2016 in grape samples (p < 0.01), and the unidentified genera relative abundance in 2016 was higher than that in 2014 and 2015 (p < 0.0001) ( Supplementary Fig. S3). The differences in these fungal genera may account for the separation of the grape samples by vintage on the PCoA plot. The grape-specific (not found in soil) fungal genera detected included Coprinellus, Ischnoderma, Mycosphaerella, Occultifur, Pestalotiopsis, and Tilletiopsis. Many yeast genera commonly found in abundance in grapes, such as Candida, Pichia, Debaryomyces, Lipomyces, Kluyveromyces, and Issatchenkia, were not found or were not abundant (<1% in relative abundance) in this study.

Discussion
The link between soil microbiome composition and regional wine characteristics has been recently studied 10,15 , leading to greater interest in the role of microbes in fruit and wine composition 4 . Our multi-year experiment examined whether different under-vine soil management practices could alter grape-associated microbial composition. While a previous study suggested that soil management in the vineyard can impact soil microbial assemblages 25 and that grapevine aerial organ-associated bacterial OTUs likely originated from soil 23 , we hypothesized that implementing different under-vine soil management practices would not only alter soil microbial composition, but that the grape-associated microbiomes would reflect these changes. In our study, changes in the fungal community of the soil, due to adopting different under-vine soil management practices, did not extend to the grapes, leading us to reject our hypothesis for our study site.
Previous studies have shown that vineyard management alters grape and fermentation microbiome composition where systematic vineyard management practices or direct microbial management approaches were applied 17,19,20,29 . In one study, yeast dynamics during the spontaneous fermentation using grapes obtained from conventionally and non-conventionally managed vineyards differed 20 . Another study revealed that management practices applied directly onto grapes, such as pesticides, impacted grape-associated yeast diversity, which negatively correlated with the copper residuals found on the grapes 19 . Unlike these studies, our study did not directly manage the microbes on the grapes, but applied indirect changes to microbial composition in soils. Our study showed a link between under-vine management practices and soil bacterial and fungal composition, which confirmed results from previous studies that focused on soil bacterial 25 and fungal composition 22 . The results of this study further revealed no corresponding changes in the grape fungal microbiome, which does not dispute findings from another study 22 . The researchers reported that the juice fungal microbiome obtained from conventionally and biodynamically managed vineyards did not differ from each other, despite showing that fungal populations on the grape surface differed by vineyard management approaches.  In our study, under-vine soil treatment impacts on grape fungal composition could also be masked by factors such as climate, geological properties (e.g. soil type), management practices associated with cool climate viticulture (e.g. trellis system, fungal spray use and frequency), vineyard management history, and inter-row vineyard floor management. Among these factors, many are specific to the region, such as large vine size with tall trellis systems, frequent pesticide applications, and hilling soil up over the graft union in winter and down off of the graft union in the spring. In a broader sense, climatic conditions play a significant role in microbiome structure, which is shown in our study, with year-to-year climate differences being the most significant factor explaining variance in soil and grape fungal assemblages, which is consistent with a previous study 11 .
With weather variability increasing as a function of climate change, there is renewed interest in improving resilience of vines to environmental stress. Cover crops are known to improve soil health by retaining soil moisture, enhancing drainage, raising soil organic matter content, maintaining soil physical structure, and supporting soil microbial properties and processes [30][31][32][33][34] . Also, cover crops provide a prolific root zone (rhizosphere) that enriches for a diversity of microorganisms that perform many functions, such as mediating soil nutrient cycling, impacting plant growth and development, and influencing pathogen interactions 30,31,[35][36][37] . This may require long-term assessment, as no soil microbial diversity difference was observed between bare soil and vegetative soil over the course of three years. However, in our study, we did observe a lower relative abundance of Neopestalotiopsis and an unidentified genus under Amphisphaeriaceae in the soil with vegetation, which may possibly relate to grapevine trunk pathogenic species 38 . Since our study only examined short 16 S rRNA gene and ITS reads, we are not able to determine whether specific organisms we identify are pathogenic or not.
This study aimed to evaluate the role of management practices -specifically vineyard soil management -on the vineyard microbiome. We found that bare soil maintained by soil cultivation and herbicide led to soil bacterial and fungal communities that diverged from the non-cultivation natural vegetation treatment. The results indicate that vineyard microbiome could be susceptible to changes under different soil management practices; however, the spatial gap between soil and the fruiting zone, and the frequent pesticide applications, could impact the level of soil management effects. It also suggests that future studies on the movement of microorganisms from soil to grape would be key to understanding the role of vineyard soil management in shaping the microorganisms associated with grapes.
Despite previous findings on vineyard management effects on vineyard microbiomes, we show that altering soil microbial composition in the vineyard through under-vine management practices did not result in corresponding changes to the grape microbiome at our study site. The concept that soil microbial composition could be impacting fruit and wine composition should be examined in light of vineyard management practices that alter soil biotic components. Regional management practices such as ground cover management, height of the trellis system, and phytosanitation, that respectively modify soil conditions, transportation of microbes from soil to grapes, and grape-associated microbes, could have a significant role in shaping fruit and wine composition in vineyards.

Methods
Vineyard design. The experiment was conducted in a commercial vineyard on Howard gravelly loam soil located in Ovid, NY, USA for three consecutive years from 2014 to 2016. The vines, V. vinifera cultivar Riesling grafted onto 3309 C rootstock, were planted in 2001 with 2.13 m × 2.74 m intra-and inter-row spacing. The trellis system was cane pruned Scott-Henry system with 10 buds per cane on each of four canes. A complete randomized block design was applied to enable four replicates for each treatment, and the treatments were randomly assigned to the experimental units, which are one meter wide under-vine soil strips, within each block. Each experimental unit was across three rows with nine consecutive vines in a row. The grape and soil samples were collected from the middle three vines and the accordance under-vine 1 m × 5.8 m soil strip, in the middle row from each of the experimental unit where the other vines were served as guards for physical and spatial buffering. The vineyard canopy, pest-control and amendments were managed following standard commercial practice in the Finger Lakes region 39 by the professional vineyard crew.
Under-vine soil treatments. The experimental units were subjected to three different under-vine soil treatments in a one meter wide strip under vines including spot application of herbicide, in which the active ingredient was glyphosate, cultivation maintained bare soil, and natural vegetation, where weeds grew freely with periodic mowing to keep them out of the fruiting zone. Herbicide and cultivation bare soil strips were established following the commercial standard. In brief, 2% Roundup (Monsanto, MO, USA) was sprayed with electronic pumped spraying nozzle in rate about 3 kg a.i./ha. Cultivation was done by combining mechanical, rototiller to roughly 20 cm depth, and manual tillage, cultivation with hoes. Herbicide was applied on June 23 rd , July 9 th , July 18  Bioinformatic and statistical analysis. The raw sequences were processed and aligned following the protocol described in the Brazilian Microbiome Project 42 with some modifications 43 . Briefly, paired-end sequence merging, primer trimming, and singleton sequence removal were performed in Mothur v 1.36.1. Operational Taxonomic Units (OTU) were produced at 97% sequence similarity. Taxonomic classification of OTUs was performed in Mothur using the GreenGenes v.13.8 database for 16 S rRNA gene sequences and UNITE v. 7 database for ITS sequences. Suspected non-bacterial and non-fungal OTUs, including chlorophyll and mitochondria, were also removed in Mothur. All downstream data analysis was conducted in R version 3.3.3 with packages Vegan and Phyloseq. The microbial diversity was determined using Shannon Diversity Index using "diversity" function in package vegan. The β-diversity of the assemblage dissimilarity between samples were calculated with the Bray-Curtis distances for fungal community and weighted UniFrac distances for the bacterial community using package vegan. The dissimilarity matrices obtained were also used for Principal Coordinate Analysis (PCoA) plotting against the first two dimensions (highest variables explained). Multivariate dispersion analysis was performed to test the differences in variances among the treatments using command "betadisper" in package vegan where the β-diversities were obtained based on the Bray-Curtis distance metric for fungal community and UniFrac distances for the bacterial community. Permutational Multivariate Analysis of Variance (PERMANOVA) and paired PERMANOVA using "adonis" command in package vegan at 999 permutations and α = 0.05 were performed testing factors including year (for overall analysis only), stage (for soil samples in 2015 and 2016 only), and under-vine soil treatments. When three-year overall analysis was conducted, the year was positioned as a fixed effect with samples within each block in constrained permutation to account for the repeated measures. The overall PERMANOVA was not performed for soil 16 S data due to large differences in sequencing depth between 2014 and the other years (average 450 reads per sample in 2014, and 11648 reads per sample in 2015 and 2016). Paired-PERMANOVA was performed by subsetting the treatments and applying Bonferroni correction to the P-values. The relative abundance of selected fungal genera in the samples were compared using one-way analysis of variance (ANOVA) test followed by Tukey HSD performing in JMP Pro 12.0.1 (SAS Institute, NC, USA), with log transformations when needed under violations of normality.
Data availability. All of the data are provided fully in the result section within and supplementary data accompanying this paper.