While wine fermentation has long been known to involve complex microbial communities, the composition and role of bacteria other than a select set of malolactic LAB has often been assumed negligible or else detrimental. We chose to study Dolce wine fermentation because, based on previous studies of the same vintages, these fermentations contain complex, unusual yeast communities 
, and previous DGGE surveys of the Dolce fermentation 
predicted that the bacterial communities would exhibit similar diversity.
Paired-end sequencing of V4 and V5 16S rDNA amplicons was performed in two separate runs on the Illumina GAIIx. The 5′ and 3′ reads were analyzed separately, as QIIME 
currently does not handle paired-end data. Additionally, others have shown that concatenating paired-end reads does not necessarily improve taxonomic depth and phylogenetic analyses 
, so we handled 5′ and 3′ reads independently, with the purpose of comparing their efficacy as single-direction reads. Raw read counts and quality filtration statistics for each run are presented in Table S3
. Alpha rarefaction plots of observed species (Figure S1
, left) were constructed to determine that adequate sequence coverage was obtained to reliably describe the full diversity present in these samples. Samples exhibiting largely inadequate sequencing depth, as indicated by a non-asymptotic rarefaction curve, were removed prior to further analyses. Additionally, the PD metric (Figure S1
, right), which measures the complete phylogenetic distance represented within a community 
, demonstrated an apparently greater level of phylogenetic diversity in V4 reads compared to V5 reads.
BAS of Dolce fermentation samples identified a range of minor microbiota, many of which, to our knowledge, have not been reported in wine previously (, Figure S2
). In general, bacterial community structure exhibited little change across the fermentation, except for a gradual reduction of Proteobacteria
and increase of Firmicutes
over time. In both vintages, Rhodospirillales
, and Gluconoacetobacter
) were the most dominant bacteria detected, with secondary populations of Lactobacillales
. Fluctuating, minor populations (some as high as 10% of total sequences detected but typically <1%) of Chryseobacterium
), and Pseudomonas
, and Acinetobacter
) were also observed at different times, particularly in the 2009 vintage. For a complete list of genera detected, see Tables S4
, and S7.
Bacterial community structure determined by sequencing of the V4 (Panels A,B) and V5 (Panels C,D) domain of 16S rRNA.
Sequencing of both the V4 and V5 regions provided similar views of community structure in these wines, albeit with different degrees of evenness. The V5 region displayed a global dominance of Acetobacteriaceae
, while V4 data suggest that Firmicutes
and other Proteobacteria
represent a larger relative portion of the microbiota in these samples. Both resulted in similar taxonomic assignments, with slight differences at the genus level. The V4, for example, produced a higher number of genus-level assignments meeting threshold criteria for select Proteobacteria
) and Firmicutes
, particularly Lactobacillaceae.
were particularly disparate in assignment as the V5 data had only taxa assigned as “other Lactobacillales
". Taxonomic assignments from V5 sequences, however, were less sensitive to truncation, such that sequences truncated to <100 nt due to low-quality base calls were still assigned to order- and family-level taxonomic ranks, whereas truncated V4 sequences of equivalent length were assignable only at the phylum and class levels. For both regions, the 5′ read was slightly more taxonomically informative than the 3′ read (as these were analyzed separately in QIIME) but generally revealed the same community structure. The V4 3′ reads, in particular, selectively achieved shallower taxonomic resolution, most readily observed as the assignment of the dominant OTU as Proteobacteria
(as opposed to Gluconobacter
] by the V4 5′ and V5 reads; , Figure S2
), but with assignment of LAB comparable to that of other reads.
Most OTUs could be resolved to family-level, and many to genus, in spite of the short read length employed (150 bp). However, many Lactobacillales
could not be further discriminated. As this is the most important bacterial order to wine fermentation (both for spoilage potential and malolactic activity), sequencing data were augmented by and compared to LAB-TRFLP 
, which can identify most LAB to species (). In both vintages, V5 sequencing identified Lactococcus
as the most dominant genus, with secondary populations of Leuconostoc
and minor populations of Streptococcus
(). This was roughly mirrored by V4 sequencing as a dominance by Leuconostoc
with a significant population of Lactococcus
(). LAB-TRFLP presents a very different picture of the fermentation, particularly from the V5 reads (). Lactobacillus kunkeei
spp. dominated the early and late fermentations, respectively, while Lactococcus lactis
, Lactococcus raffinolactis
, Weissella minor
, Lactobacillus sakei
, and Pediococcus
were all detected as minor populations. The V4 reads, exhibiting increased relative abundances of Lactobacillaceae
and decreased abundance of Lactococcus
, were more comparable to LAB-TRFLP than the V5 reads.
Lactobacillales community of Dolce fermentation revealed by LAB-TRFLP (Panel A), V4 5′ read (Panel B) and V5 3′ read (Panel C).
In order to view relationships among samples based on differences in phylogenetic diversity, principle coordinates (PC) were calculated from jackknifed UniFrac distances 
between samples and used to construct three-dimensional principal coordinate analysis (PCoA) plots (). Samples cluster by batch based on weighted UniFrac distance (), with clear separation of inoculated and uninoculated samples (). Samples did not cluster based on age (weeks of fermentation), implying that the overall phylogenetic diversity changes little during the course of the fermentation. Cluster separation was less distinct based on V5 sequence data (data not shown). A P test 
confirmed significant differences in genetic diversity between each batch (Bonferroni-corrected p
<0.001) and between inoculated/uninoculated batches (Bonferroni-corrected p
<0.001) with 1000 Monte Carlo iterations. Based on this significant result, we used ANOVA to test category-specific differences in abundance among OTUs assigned to the V4 5′ sequences. Ten OTUs demonstrate significant differences (false discovery rate-corrected p
<0.05) between inoculated and uninoculated groups (). A G test of independence verified that these differences were not significantly related to presence/absence of any OTU between groups (false discovery rate-corrected p
>0.05). To visualize relationships among these significant OTUs and sample types, we constructed a PCoA biplot plotting significant OTUs (as loadings) in relation to samples (). This plot was constructed from the weighted UniFrac PCoA of V4 5′ reads (), but OTUs are given coordinates in addition to samples in order to show how OTUs correlate with samples along the principle coordinates. OTU coordinates are indicated by grey orbs with size as a function of relative abundance, and labeled according to ID in . Most of the OTUs appear to associate more strongly with uninoculated samples, especially Gluconobacter
. Two OTUs, Zymobacter
, appear to be more associated with inoculated samples.
Inoculation and batch direct bacterial diversity of Dolce fermentations.
ANOVA Significance of Inoculation-based Differences in Bacterial Taxa.
Considering the high bacterial diversity exhibited in these samples based on BAS data, including species not previously found in wine, it was questioned whether some sequences represent residual DNA or spores from nonviable, plant-associated bacteria, or even sequencing artifact. In particular, the persistence of Sphingomonas and Alphaproteobacteria other than Acetobacteriaceae was surprising. Thus, an attempt was made to culture these low-abundance species from a finished fermentation using enrichment culture. To target this genus, we used sphingomonas broth containing 5% ethanol and erythromycin under microaerobic conditions to select for species competent under alcoholic conditions and to prevent growth of aerobic bacteria, primarily Acetobacteriaceae and spore-forming Bacillaceae. Under these conditions, two isolates were obtained representing the two prevailing colony morphotypes observed on sphingomonas agar plates. The closest matches identified by 16S rDNA sequencing were Methylobacterium populi and Sphingomonas pseudosanguinis. Surprisingly, neither of these isolates was capable of growth at wine-like conditions when cultured in high-ethanol, low-pH media (data not shown).