Wine terroir and the soil microbiome: an amplicon sequencing–based assessment of the Barossa Valley and its sub-regions

Soil is an important factor that contributes to the uniqueness of a wine produced by vines grown in specific conditions. Recent data shows that the composition, diversity and function of soil microbial communities may play important roles in determining wine quality and indirectly affect its economic value. Here, we evaluated the impact of environmental variables on the soil microbiomes of 22 Barossa Valley vineyard sites based on the 16S rRNA gene hypervariable region 4. In this study, we report that environmental heterogeneity (soil plant-available P content, elevation, rainfall, temperature, spacing between row and spacing between vine) caused more microbial dissimilarity than geographic distance. Vineyards located in cooler and wetter regions showed lower beta diversity and a higher ratio of dominant taxa. Differences in microbial community composition were significantly associated with differences in fruit traits and in wine chemical and metabolomic profiles, highlighting the potential influence of microbial communities on the phenotype of grapevines. Our results suggest that environmental factors affect wine terroir, and this may be mediated by changes in microbial structure, thus providing a basic understanding of how growing conditions affect interactions between plants and their soil microbiomes.

Soil is an important factor that contributes to the uniqueness of a wine produced by vines grown in 22 specific conditions. Recent data shows that the composition, diversity and function of soil microbial 23 communities may play important roles in determining wine quality and indirectly affect its economic 24 value. Here, we evaluated the impact of environmental variables on the soil microbiomes of 22 Barossa 25 Valley vineyard sites based on the 16S rRNA gene hypervariable region 4. In this study, we report that 26 environmental heterogeneity (soil plant-available P content, elevation, rainfall, temperature, spacing 27 between row and spacing between vine) caused more microbial dissimilarity than geographic distance. 28 Vineyards located in cooler and wetter regions showed lower beta diversity and a higher ratio of 29 dominant taxa. Differences in microbial community composition were significantly associated with 30 differences in fruit traits and in wine chemical and metabolomic profiles, highlighting the potential 31 influence of microbial communities on the phenotype of grapevines. Our results suggest that 32 environmental factors affect wine terroir, and this may be mediated by changes in microbial structure, 33 thus providing a basic understanding of how growing conditions affect interactions between plants and 34 their soil microbiomes. 35 36 1 Introduction 37 Wine price differs considerably depending on its quality (e.g., flavor, color and typicity), which is 38 largely determined by the interactions between the grape and the growing conditions including climate, 39 soil, topography, agricultural management, and the wine making process ( However, microbial assemblage function is intrinsically difficult to measure and define because of its 68 highly changeable nature (Nannipieri et al., 2003). Additionally, due to the complex interactions 69 between soil microbes, the influence of certain microbial communities can be substituted by other 70 microorganisms with the same ecological function (Nannipieri et  In order to answer these questions, we undertook a soil microbiome survey in an iconic wine region, 80 the Barossa in South Australia. The Barossa has a winemaking history of over 160 years and because  81  of its importance as a growing region, has been chosen as a model to investigate terroir previously (  82  Wolf et  to the laboratory on the same day of collection. Soil samples from the same row were thoroughly mixed 104 to obtain three samples per vineyard, and a total of 66 samples across the study. Coarse debris was 105 removed from each soil sample using a 2mm sieve, and each sample was then divided into three sub-106 samples (approximately 850 cm3 each). The first subsample (approx. 20 g) was used for determination 107 of soil gravimetric moisture content. The second was air-dried until a constant mass was achieved and 108 used for analysis of soil texture, pH, electrical conductivity, and plant-available (Colwell) P 109 (phosphorus), as described previously (Cavagnaro, 2015)). The third soil subsample was stored at -110 80˚C for DNA extraction and downstream genomic analysis (see below). 111

Vineyard physical characterization 112
In this study, the climate was characterized on the basis of rainfall and temperature. The influence of 113 topography was studied through elevation above sea level and vineyard orientation. Soil texture was 114 determined following (Giddings, 2015). Soil pH and electrical conductivity were determined on a 1:5 115 soil/water mixture and then measured using pH/salinity meter (WP-81 Conductivity-Salinity-pH-mV 116 Meter, v6.0, TPS Pty Ltd). Plant-available phosphorus was extracted and measured using Colwell P 117 method (Rayment and Higginson, 1992) (Table S1). The remaining soil, topographic and climatic data 118 was obtained from the Barossa Grounds project (Robinson and Sandercock, 2014), while vineyard 119 management information was collected from participating growers (Table S2). 120

Fruit and wine chemical analysis 121
Fruit juice pH and total acidity (TA) was measured using an autotitrator ( The modified Somers assay was used to determine; wine colour density (WCD), SO2-corrected WCD, 135 degree of anthocyanin ionisation, phenolic substances and anthocyanins (in mg/L) (Table S4). 136 Non-targeted metabolomic analysis of the wine samples was performed using LC-MS/MS. The 137 metabolites were isolated from bottled wine samples using solid-phase extraction (SPE) with 138 Phenomenex Strata-X 33 um 85Å polymeric reverse-phase 60mg/3mL cartridges. A 2 mL aliquot of 139 each sample was evaporated to dryness under nitrogen at 30°C. SPE conditions are presented in Table  140 S5. A pooled mix of all samples was prepared and used to monitor instrument performance. The 141 analysis was performed on an Agilent 1200SL HPLC coupled to a Bruker microTOF-Q II in ESI 142 negative mode. The operating conditions are described in To identify the association of environmental variables and grape and wine properties (Table S1-S7)  204 with soil bacterial microbiome, bacterial community dissimilarities were visualized with non-metric 205 multidimensional scaling (nMDS) plots. Variables were fitted to the ordination plots using the function 206 envfit in the package Vegan version 2.5-2 (Oksanen et al., 2013) implemented in R version 3.5.0 (Team, 207 2013). Spearman's rank correlation coefficients were measured between individual taxon abundance 208 and fruit and wine traits using the function rcorr in the package Hmisc. Grape traits included those 209 from sensory, basic chemistry analyses, while wine traits included basic chemistry, wine fermentation 210 products and amino acids concentration. Those traits and taxa with a significant (p-value <0.05) 211 correlation coefficient larger than 0.80 or lower than -0.80 were deemed as significantly associated. 212 To identify which variables are important in explaining the composition of the soil microbial 213 community, we performed distance-based redundancy analysis (dbRDA), a form of multivariate 214 multiple regression that we performed directly on a Bray-Curtis dissimilarity matrix of OTUs using 215 the ADONIS function in Vegan. We used automatic model building using the function step in R. The 216 step function uses Akaike's Information Criterion (AIC) in model choice, which is based on the 217 goodness of fit. The model building proceeds by steps until the 'best' fit is identified. If two predictor 218 variables were highly correlated (>0.85) one, typically that which was more difficult to measure, was 219 removed as well as variables with missing replicates (Variables included in the automatic model 220 building are marked with * in Tables S2-S8 Results 226

Barossa Valley soil bacteria community composition 235
After quality filtering of the raw sequencing results, an average of 130,949 paired sequences remained 236 per sample. Of these an average of 86,835 paired-end sequences per sample (66.3%) could be joined 237 using bbmerge (Table S9). 238 Both bacterial and archaeal DNA was detected in all soil samples. A total of 98.9% of sequences were 239 classifiable at the phylum level ( Figure 1A) and 95.2% at the genus level. Of those classifiable at the 240 phylum level, 96.5% were assigned to one of nine dominant groups (relative abundance ≥1.0%) in the 241 samples namely: Actinobacteria (26.9%), Proteobacteria (26.7%), Acidobacteria (12.0%), 242 Planctomycetes (6.2%), Chloroflexi (5.6%), Firmicutes (5.3%), Gemmatimonadetes (3.9%), 243 Bacteroidetes (3.5%), Verrucomicrobia (2.5%) ( Figure 1A). The only dominant Archaea group was 244 Crenarchaeota (4.0%). The overall dominant Bacteria and Archaea groups were consistently present in 245 the six regions, but at different ratios ( Figure 1A). The phylogenetic inference of microbiome 246 composition differences between sub-regions showed three clusters with Central and Northern 247 Grounds, and Eden Valley and Western Ridge sharing the more similar microbial profiles ( Figure 1A). 248 The number of observed OTUs ( Figure 1B) showed significant differences (t-test: p-value < 0.05) 249 between the OTU rich sub-regions (Northern and Central Grounds) and the relatively OTU poor sub-250 regions (Eden Valley and Western Ridge) (Table S10). Similarly, the Chao1 metric showed that 251 Northern and Central Grounds presented higher levels of OTU richness while Eden Valley and Western 252 Ridge had the lowest ( Figure 1C). Pairwise comparison of alpha diversity between sub-regions showed 253 significant differences (t-test, p-value < 0.05) between Northern Grounds and Eden Valley and Western 254 Ridge and between Central Grounds and Eden Valley and Western Ridge (Table S11). 255 Dissimilarities in microbial communities between samples (i.e. beta diversity) were calculated as 256 weighted and un-weighted UniFrac distances and both methods showed similar patterns, and so only 257 analyses based on weighted results are shown here. For the most part, the three replicates from within 258 a given vineyard were closely grouped on the ordination plot (Figure 2A), indicating that bacterial 259 communities were consistent within sites. Pairwise analysis of the differences between groups 260 (vineyards and sub-regions) showed that all vineyards and sub-regions are significantly different to 261 each other (Adonis, p-value < 0.001). Mantel test analysis of the association between microbiome 262 compositional differences and geographic distance, showed a small but significant correlation (rxy = 263 0.315; p-value = 0.0001) ( Figure 2B). 264 To further explore dissimilarities among and within regions, neighbor joining analysis was used to 265 cluster samples and to generate a similarity tree in QIIME. This information, along with a geographical 266 map of the regions and their locations, were combined using the GenGIS software package (Parks et  267 al., 2009). This approach showed a low level of clustering of vineyards according to their geographic 268 location ( Figure 2C). 269

Drivers of soil microbiome differentiation 270
Model selection was used to identify the combination of variables that explained the greatest variation 271 in the soil microbiome. This approach consistently selected soil plant-available phosphorus (P) and soil 272 texture as the main drivers (Model: p-value = 0.001) of soil microbiomes in the Barossa vineyards 273 tested ( Figure 3). Together, both variables explained 19.7% of the observed variability. Independent 274 pairwise analysis of UniFrac distances of vineyards grouped by these soil characteristics, showed that 275 microbial communities in clay soil types were significantly dissimilar from those in sandy soils 276 (PERMANOVA: p-value < 0.001, Figure 4A). Microbial communities in soils with high plant-277 available phosphorus (P > 30mg/kg) were also dissimilar from those with low plant available 278 phosphorous (PERMANOVA: p-value < 0.001, Figure 4C). Three and eight taxa were significantly 279 more abundant in clay and sandy soils respectively ( Figure 4B), while eight taxa were found 280 significantly associated with low plant available phosphorous content, and three associated high levels 281 of plant available phosphorous in soil ( Figure 4D). (family Conexibacteraceae), and the spacing between vines on the same row (family Haliangiaceae). 294

Terroir and vineyard soil microbiomes 295
Twenty four of the 75 grape and wine characteristics included in the study displayed a significant 296 correlation with the soil microbial community composition ( Table 2). The strongest associations 297 identified for each of the four groups of traits tested were: 50 berry weight and average color per berry 298 (basic berry properties); total anthocyanins and total phenolics (basic wine chemistry); Glycine and 299 Alanine (wine amino acids); and 2-phenyl ethyl ethanol and acetic acid (wine fermentation products). 300 Significant positive correlation (Spearman's >0.80, p-value <0.05) were identified between the 301 abundance of one taxon (order IS_44) and the average level of total phenolics mg/g berry weight 302 ( Figure S3A). Similarly, six wine traits showed positive correlations with the abundance of six 303 microbial taxa ( Figure S3B-E)

Correlations between soil bacterial communities and berry and wine parameters 411
Berry parameters were found to be significantly associated with both the composition and diversity of 412 soil microbiomes and with the abundance of single taxa. A total of six fruit traits correlated with 413 differences in bacterial community composition and diversity, while one fruit trait was found 414 significantly associated with the abundance of specific taxa. Soil microbiomes have previously been described as a contributor to the final sensory properties of 421 wines by affecting wine fermentation. Grape must microbiota was found to be correlated to regional 422 metabolite profiles and was suggested to be potential predictor for the abundance of wine metabolites 423 (Bokulich et al., 2016). Here we identified 19 wine traits correlated with differences in bacterial 424 community composition and diversity, and seven correlated with the abundance of specific taxa. 425 Vineyard soils may serve as a bacterial reservoir since bacterial communities associated with leaves, 426 flowers, and grapes share a greater proportion of taxa with soil communities than with each other 427 (Zarraonaindia et al., 2015). Unfortunately, the non-intervention nature of this research, the lack of 428 replicability and the use of commercially produced wines, preclude us from determining if the 429 relationships observed between vineyard soil microbiomes and fruit/wine traits are causal or simply 430 mere correlations. Each of these wines was made commercially by different producers so there is 431 potential for a certain level of winemaking effect. 432

Conclusion 433
Taken collectively our results show that geographic separation between vineyards contributes to 434 bacterial community dissimilarities at a much smaller scale than previously reported. Environmental 435 variables (e.g. climatic, topography, soil properties, and management practices) were the greatest 436 contributor to such differences. Particularly, we found that soil variables are the major shapers of 437 bacterial communities. Also, we show that variables highly affected by soil anthropogenisation (pH, 438 plant available Phosphorous) and agricultural management variables (plant age, planting density) have 439 strong correlations both with the community composition and diversity and the relative abundance of 440 individual taxa. Finally, our results provide an important starting point for future studies investigating 441 the potential influence of microbial communities on the metabolome of grapevines in general, and on 442 the definition of local Terroirs. It will also be important to study a wider range of soil physicochemical 443 properties, and vineyard floor vegetation, on the soil microbiome. 444

Conflict of Interest 445
The authors declare that the research was conducted in the absence of any commercial or financial 446 relationships that could be construed as a potential conflict of interest. Council through Centre of Excellence (CE1400008) and Future Fellowship (FT130100709) funding. 462

9
Acknowledgments 463 The authors would like to gratefully acknowledge the Barossa Grounds Project and in particular the 464 growers that allowed us to sample material from their properties and supplied information about their 465 vineyards and management strategies. Dr. Kendall R. Corbin contributed to soil sample collection. We 466 are thankful to Dr Hien To, Dr Steve Pederson and Dr Rick Tearle for assistance with data analysis. 467 468 10 Data Availability Statement 469 All the data and supporting information will be made available online. 470 1. Supplementary Figures S1-S3. 471 2. Supplementary Tables S1-S11. 472 The data that support the findings of this study are openly available in NCBI Sequence Read Archive 473 (accession number: PRJNA601984).     distances. Geographic distances were calculated from latitude/longitude coordinates using GenAlex 743 v6.5 geographic distance function implemented as Log(1+distances in Kilometres). The relationship 744 was tested using Mantel's correlation coefficient (rxy) with its probability estimate for significance (P) 745   wine (B-F) traits. Correlations were tested using Spearman's rank correlation coefficient with its 827 probability estimate for significance (P) and implemented using the function rcorr in the R package 828 Hmisc. Correlation coefficient and P values for each of the comparisons are included in each inset. 829