Abstract
The use of sulfur dioxide (SO2) as an antimicrobial in winemaking is a well-established, common practice. Although much is known about the antimicrobial effects of SO2 at single concentrations, little is known about its effects on microbial growth dynamics across a range of concentrations or when used in conjunction with yeast inoculation. Using high-throughput marker-gene sequencing, we investigated the cumulative impacts of yeast inoculation and SO2 treatments across a broad concentration range (0 to 150 mg/L SO2) on the bacterial and fungal communities in wine fermentations. Our results indicated a dose-dependent effect of SO2, with lactic acid bacteria and Gluconobacter proliferating in fermentations with <25 mg/L SO2, but other bacteria and fungi were unaffected by the SO2 addition. Microbial profiles stabilized at concentrations ≥25 mg/L SO2, and fermentation performance decreased at higher concentrations (100 to 150 mg/L SO2). Yeast inoculation alone conferred a stabilizing effect, reducing the bacterial growth observed in unsulfited fermentations, but this effect was not additive with an increase in SO2 concentrations.
Sulfur dioxide (sulfite or SO2) is commonly added to wine juices and musts (i.e., crushed grapes) prior to fermentation to reduce polyphenol oxidase activity and to inhibit the growth of sulfite-susceptible microbes (e.g., non-Saccharomyces yeasts) (Boulton et al. 1996). This antimicrobial effect of SO2 has been investigated with culture-based methods (Cocolin and Mills 2003, Constanti et al. 1998, Takahashi et al. 2014), quantitative PCR (qPCR) (Andorrà et al. 2008), and DGGE (Andorrà et al. 2008, Cocolin and Mills 2003, Takahashi et al. 2014). These studies indicated that addition of SO2 inhibits the growth of select organisms, particularly of non-Saccharomyces yeasts and of lactic acid bacteria (LAB).
Previous studies have tested SO2 effects on microbial growth when added to fermentations at a single concentration by comparing the resulting growth with that in unsulfited fermentations. However, little is known about the effects of SO2 across a large concentration range for a single juice. Furthermore, the methods used in these studies were too limited for comprehensively investigating the impacts of SO2 treatments on microbial diversity. qPCR can detect only those clades of microbes targeted with this method and thus cannot track the complete microbial consortium, while the other techniques have low sensitivity, are unsuitable for quantitative measurements, and have other biases (Bokulich et al. 2012a, Bokulich and Mills 2012). Culture-based methods are particularly unsuitable, as sulfites may induce a viable but nonculturable state resulting in diminished microbial growth on culture media (Cocolin and Mills 2003). Therefore, culture-based detection may be inadequate for tracking microbial changes induced by SO2 and other stressors. High-throughput marker-gene sequencing methods provide a more sensitive, quantitative, and comprehensive approach for investigating complex microbial communities in food systems (Bokulich and Mills 2012). Moreover, these methods have been previously used to comprehensively profile microbes in musts and wines and on winery surfaces (Bokulich et al. 2012b, 2013a, 2014).
To investigate the impact of SO2 treatment and yeast inoculation on wine fermentations, we used marker-gene sequencing to characterize the bacterial and fungal communities of sulfited and unsulfited wines over the course of a fermentation. The diversity of microbial populations was assessed in different SO2 concentrations ranging from 0 to 150 mg/L and in wines inoculated or uninoculated with Saccharomyces cerevisiae to determine the cumulative effects of inoculation and SO2 treatments on species diversity.
Materials and Methods
Experimental winemaking.
All wines were produced at the Robert Mondavi Institute for Food and Wine Science Winery (University of California, Davis). Yolo County Chardonnay grapes were harvested, crushed, and pressed according to standard winemaking procedures. The unfermented juice had a sugar content of 24.6 Brix, a pH of 3.63, a titratable acidity of 5.03 g/L, and a NOPA concentration of 147 mg/L. Approximately 19 L of juice was racked into 23-liter stainless steel fermentation tanks for experimental fermentations. Sulfur dioxide was added into each tank at nine different concentrations (0, 15, 20, 25, 35, 50, 75, 100, or 150 mg/L); tanks were left uninoculated to allow for native fermentation, and no nutrients were added. The starting nitrogen concentrations were 147 mg/L NOPA and 78 mg/L ammonia. One treatment received no SO2 and another an average SO2 concentration of 50 mg/L, and both were inoculated with S. cerevisiae strain EC1118 (rehydrated according to manufacturer’s instructions) at 1 g/3.8 L. All experimental treatments were performed in triplicate. Fermentations were allowed to proceed at ambient temperature for 21 days. Samples were aseptically collected with sterile serological pipettes from the middle of each fermenter after 1, 2, 3, 5, 7, 10, 14, and 21 days of fermentation. The samples were collected from unstirred fermentations to avoid introducing oxygen and thereby affecting fermentation patterns. Instead, we relied on the natural mixing in active fermentations and on biological replicates to control for any sampling bias. All samples were placed on ice and frozen immediately in a −20°C freezer for storage. DNA was extracted using the standard protocol for the ZR-96 Fecal DNA MiniPrep Kit (Zymo Research, Irvine, CA), with bead beating in a FastPrep-24 bead beater (MP Bio, Solon, OH), and stored at −20°C until further processing.
Fermentation progress.
Fermentation progress was tracked daily as Brix, measured with a hydrometer. All measurements were adjusted to correct for temperature. One representative set of fermentation data is presented in Figure 1A.
Sequencing library construction.
Amplification and sequencing was performed as described previously for bacterial (Bokulich et al. 2012b) and fungal communities (Bokulich and Mills 2013). Briefly, the V4 domain of bacterial 16S rRNA genes was amplified using primers F515 (5′–NNNNNNNNGTGTGCCAGCMGCCGCGGTAA–3′) and R806 (5′–GGACTACHVGGGTWTCTAAT–3′) (Caporaso et al. 2011), with the forward primer modified to contain a unique 8 nt barcode (corresponding to the italicized poly-N section in primer F515 above) and a 2 nt linker sequence (bold, underlined portion) at the 5′ terminus. PCR reactions contained 5 to 100 ng DNA template, 1 × GoTaq Green Master Mix (Promega, Madison, WI), 1 mM MgCl2, and 2 pmol of each primer. Reaction conditions consisted of an initial 94°C for 3 min; followed by 35 cycles of 94°C for 45 sec, 50°C for 60 sec, and 72°C for 90 sec; and a final extension of 72°C for 10 min. Fungal internal transcribed spacer (ITS) 1 loci were amplified with primers BITS (5′–NNNNNNNNCTACCTGCGGARGGATCA–3′) and B58S3 (5′–GAGATCCRTTGYTRAAAGTT–3′) (Bokulich and Mills 2013), with a unique 8 nt barcode (shown in italics in primer BITS) and linker sequence (bold, underlined) incorporated into each forward primer. PCR reactions contained 5 to 100 ng DNA template, 1 × GoTaq Green Master Mix (Promega), 1 mM MgCl2, and 2 pmol of each primer. Reaction conditions consisted of an initial heating step of 95°C for 2 min; followed by 40 cycles of 95°C for 30 sec, 55°C for 30 sec, and 72°C for 60 sec; and a final extension of 72°C for 5 min. Amplicons were combined into two separate, pooled samples (keeping bacterial and fungal amplicons separately) at roughly equal amplification-intensity ratios, purified using the Qiaquick spin kit (Qiagen, Valencia, CA), and submitted to the UC Davis Genome Center DNA Technologies Core for Illumina paired-end library preparation, cluster generation, and 250 bp paired-end sequencing on an Illumina MiSeq sequencer (Illumina, San Diego, CA).
Data analysis.
Raw fastq files were demultiplexed, quality-filtered, and analyzed using QIIME version 1.8.0 (Caporaso et al. 2010b). The 250 bp reads were truncated at any site after which more than three sequential bases receiving a quality score <Q10, and any read containing ambiguous base calls or barcode/primer errors was discarded, as were reads with <75% (of total read length) consecutive high-quality base calls (Bokulich et al. 2013b). Reverse primer sequences were trimmed from the ends of the ITS sequences after demultiplexing. Operational taxonomic units (OTUs) were clustered at 97% identity using the QIIME subsampled reference OTU-picking pipeline using UCLUST-ref (Edgar 2010) against either the Greengenes 16S rRNA gene database (May 2013 release) (McDonald et al. 2012) or the UNITE fungal ITS database (Abarenkov et al. 2010, Koljalg et al. 2005), modified as described previously (Bokulich and Mills 2013). OTUs were classified taxonomically against these same databases using the Ribosomal Database Project (RDP) classifier (Wang et al. 2007). Any OTUs composing <0.01% of the total sequences for each run were excluded from further analysis (Bokulich et al. 2013b). Bacterial 16S rRNA gene sequences were aligned with PyNAST (Caporaso et al. 2010a) against a reference alignment of the Greengenes core set (McDonald et al. 2012). From this alignment, chimeric sequences were identified and removed using ChimeraSlayer (Haas et al. 2011), and a phylogenic tree was reconstructed from the filtered alignment with FastTree (Price et al. 2010). Sequences failing the alignment or identified as chimeras were excluded from downstream analysis.
β-diversity (i.e., between-sample community dissimilarity) estimates were calculated within QIIME, using weighted UniFrac (Lozupone and Knight 2005) distance between samples for bacterial 16S rRNA reads (evenly sampled at 400 sequences per sample) and using Bray–Curtis dissimilarity for fungal ITS reads (evenly sampled at 100 reads per sample). Principal coordinates were computed from the resulting distance matrices to compress dimensionality into three-dimensional principal coordinate analysis (PCoA) plots, enabling visualization of sample relationships. To determine whether sample classifications (i.e., SO2 concentration, day of fermentation, and inoculation) contained differences in phylogenetic or OTU diversity, analysis of similarities (ANOSIM) (Clarke 1993) with 999 permutations was used to test the null hypothesis that sample groups were not statistically significantly different. For all classifications rejecting this null hypothesis, a Kruskal–Wallis test was used to determine which taxa differed significantly (with Bonferroni error correction) between sample groups receiving different SO2 concentrations. Paired Wilcoxon rank-sum tests were used to determine which taxa differed significantly between inoculated and uninoculated samples at paired time points and SO2 concentrations. Bacterial α-diversity (i.e., within-sample species richness) was calculated on the basis of observed OTU richness in samples evenly subsampled at 5,000 reads per sample. A three-way analysis of variance (ANOVA) was used to test which factors (i.e., SO2 concentration, day of fermentation, or inoculation) and which of their interactions significantly affected OTU richness.
Results
The aim of this study was to update and extend our knowledge of how SO2 addition impacts the microbial communities of wine fermentations. In particular, molecular techniques have not been previously used to assess the dose-dependent effects of SO2 on wine fermentations over a wide concentration range. Additionally, previous studies were limited by the detection limits in older culture-independent techniques in wine fermentations (Andorrà et al. 2008). The secondary aim was to update our current knowledge of SO2 effects on wine microbial communities by using marker-gene sequencing as a highly sensitive, culture-independent technique.
The initial goal was to assess whether different SO2 concentrations alter the overall microbial community composition of wine fermentations. We also wanted to determine how yeast inoculation may modify the fermentation profile and microbial diversity either alone or in combination with conventional addition of 50 mg/L SO2. β-diversity estimates are a useful way for quantifying and visualizing compositional differences between treatments on the basis of shared phylogenetic or species diversity between samples. The marker-gene sequencing indicated a temporal shift in bacterial β-diversity in a dose-dependent fashion (Figure 1B). Using weighted UniFrac PCoA, we found that all fermentations receiving >20 mg/L SO2 and both inoculated fermentations (not annotated in Figure 1B) gravitated around the point of origin along principal coordinate 1 (explaining 54.879% of the variation in bacterial phylogenetic diversity between the samples), which is also where all treatments clustered from days 1 to 5 of the fermentations. At day 7 and onward, fermentations receiving ≤20 mg/L SO2 deviated in bacterial diversity along this axis but not in the same direction: SO2-free fermentations migrated negatively, while fermentations with 15 mg/L or 20 mg/L SO2 migrated positively along principal coordinate 1. ANOSIM tests supported these observations, showing that SO2 concentration significantly affected bacterial diversity, according to both abundance-weighted and unweighted UniFrac (Table 1). Fungal populations did not show the same response to the SO2 treatments. Instead, day of fermentation was the only factor that significantly influenced fungal phylogenetic diversity (Table 1). Taken together, these results indicate that SO2 exerted the greatest effect on bacterial diversity, with 25 mg/L of SO2 being the minimum concentration required to stabilize bacterial profiles throughout the course of the fermentation. These results also indicate that inoculation with Saccharomyces was sufficient to stabilize these fermentations and to restore a normal fermentation profile, even without any SO2 addition.
Another pertinent question was whether SO2 addition and inoculation influence α-diversity, or within-sample species richness. One hypothesis was that SO2 suppresses microbial diversity by preventing growth of certain taxonomic groups. Observed species counts increased in all groups during the first days of fermentation, peaking at day 10. In the SO2-free, uninoculated fermentations, however, mean bacterial diversity declined rapidly at day 10 and remained lower than those for the other treatments for the remainder of the observation period (Figure 1C). Inoculated, SO2-free fermentations did not show this same drop in diversity, and instead resembled the diversity in the fermentations with higher SO2 concentrations (data not shown). A three-way ANOVA indicated that day of fermentation, inoculation, the SO2 × day interaction, and the inoculation × day interaction all significantly affected bacterial species richness, reflecting the decline in observed species in the SO2-free, uninoculated treatment (Table 2). Thus, SO2-free fermentations displayed lower bacterial diversity, contrary to the anticipated results. This surprising result was most likely due to overgrowth of a few dominant species, causing less-abundant species to fall below the detection threshold.
Microbial heat maps were used to visualize the relative abundances of individual bacterial and fungal taxa in these fermentations across time and treatments (Figure 2). The heat maps show the relative ratios of different species of bacteria (left) or fungi (right). The sequencing method used detected all organisms with intact DNA, including DNA from viable, nonculturable, and nonviable but intact cells. Bacterial profiles showed a simple, immediately observable trend: Lactobacillus and Lactobacillaceae dominated the late fermentation of SO2-free, uninoculated fermentations, and Gluconobacter became significantly more abundant in the late fermentation of low-SO2 (≤20 mg/L), uninoculated fermentations (Kruskal–Wallis test, Bonferroni-corrected, p < 0.05). In the inoculated fermentations, Lactobacillus, Lactobacillaceae, and Gluconobacter were significantly suppressed compared with their levels in uninoculated samples matched by day and SO2 treatment (Wilcoxon sum-rank test, p < 0.05) (Figures 2 and 3). Otherwise, Erwinia and Enterobacteriaceae were the most abundant bacteria in the early stages of these treatments and throughout all other fermentations and were directly displaced by these other taxa in the low-SO2 treatments. Several other minor bacterial taxa were detected and became more abundant in the late fermentations (Figure 2), corresponding to the generally increasing trend in α-diversity (Figure 1C).
Fungal profiles were much more varied (Figure 2), but common temporal trends were observable across all treatments. Kluyveromyces marxianus dominated the early stages of almost all fermentations before being displaced by Hanseniaspora uvarum and then by S. cerevisiae. Many other yeasts and molds were detected variably across the different treatment groups and times, but no general trends were observed and no significant SO2 effects were detected. Most of these fungi were previously detected in grape must (Bokulich et al. 2014) and their presence in fermentations probably reflect the fungal communities on the grape rather than microbial changes related to the treatments. The inoculated fermentations generally showed dominance of Saccharomyces across most time points. For the uninoculated samples, Saccharomyces appeared dominant during the most active phase of fermentation, but it was sometimes displaced by non-Saccharomyces yeasts later in the fermentation. However, the relative abundance of Saccharomyces was not significantly affected by the SO2 treatments or by the inoculation because of large variation within replicate fermentations and because low sequencing coverage required removal of some samples prior to statistical analysis. We also note that in the late fermentation stages, Saccharomyces populations settle to the bottom of the fermentation vessels and may be less represented in dispersed populations given the challenges in mixing tank volumes. These data indicate that non-Saccharomyces yeasts may persist even in inoculated fermentations.
At day 21, the uninoculated fermentations with 0, 15, 20, or 25 mg/L SO2 had Brix values of 2.2, 0.6, −0.3, and −0.3 respectively. The fermentations with SO2 concentrations >25 mg/L ranged in Brix value from −1.0 to −1.5 on day 21, and the two inoculated fermentations had Brix values of −1.6 for 0 mg/L SO2 and −1.9 for 50 mg/L SO2. The 0 mg/L SO2 sample displayed higher levels of Lactobacillus and Lactobacillaceae, and the 15 and 20 mg/L samples displayed higher Gluconobacter levels. The results with these fermentations show that the presence of these organisms may affect fermentation completion or that whatever other factors may be affecting fermentation completion leads to an increase in the relative contributions of these classes of organisms to the overall microbial profile.
Discussion
The antimicrobial effect of SO2 in winemaking is well established and their use is standard practice. However, previous studies have not tested these effects over a range of SO2 concentrations in wine fermentations and have used older, less sensitive techniques for measuring microbial abundance and diversity than current methods (Bokulich and Mills 2012). The goal of the current study was to investigate the influence of SO2 on both bacterial and fungal communities in wine fermentations over a range of SO2 concentrations.
Our results indicated that SO2 addition affects wine microbial communities in a dose-dependent manner. In this study, 25 mg/L appeared to be the ideal concentration at which microbial communities stabilized. Below this concentration, LAB and Gluconobacter bloomed during the fermentation. Above this concentration, no further microbial stabilization was achieved and, indeed, a trend was observed for increasingly longer fermentation times. The optimal SO2 concentration in any wine will depend on other factors, however, including pH and the presence of sulfite-binding compounds (Boulton et al. 1996). Furthermore, in this experiment, inoculation with S. cerevisiae offered a similar level of stabilization, restricting the growth of LAB and acetic acid bacteria (AAB). Thus, yeast inoculation confers some protection and—while these pilot-scale experiments may not scale to commercial production conditions—may on its own offer a reasonable level of microbial stabilization during fermentation of nonsulfited wines. However, long-term stability was not tested, and inoculation alone may not provide a robust level of detection beyond active fermentation.
The inhibitory effects of the SO2 treatment and of S. cerevisiae inoculation on LAB growth observed here corroborates the observations by Andorra et al. (2008). However, contrary to the results of the current study, these authors found no effect of SO2 on AAB. As that study and the current one were both performed in pilot wine fermentations but in different global regions, this discrepancy may merely reflect differences in the strain characteristics and population sizes of regional (Bokulich et al. 2014) or winery-resident AAB (Bokulich et al. 2013a). Sulfite resistance has been investigated in individual strains of Acetobacter and Brettanomyces (du Toit et al. 2005), but differences in SO2 resistance among strains are largely unexplored despite their spoilage potential in wine fermentations. In our study, we did not see evidence of Acetobacter, suggesting winery populations of this organism may be low. The main AAB species observed were those most commonly found on grape surfaces.
An important confirmation from this study is that the antimicrobial impact of SO2 appeared to be limited to LAB and AAB. Using marker-gene sequencing, we expected to observe inhibitory effects on some of the rarer microbes that have only recently been detected and cultured from wine with this highly sensitive technique (Bokulich et al. 2012b, 2014). Although a number of such microbes were detected at low abundance in the fermentations, including Halomonas, Delftia, Virgibacillus, and many other bacteria and yeasts, SO2 had no effect on their detection. The role of these microbes in wine fermentations has yet to be investigated, but the lack of significant changes in their populations among the SO2 treatments is reassuring, because it suggests that these organisms pose little spoilage risk in low-sulfite wines—if indeed they are even capable of actively growing during wine fermentation. With respect to SO2 concentration, none of the concentrations completely eliminated the presence of the non-Saccharomyces yeasts or significantly altered yeast populations.
Another notable observation was the dominance of bacterial profiles by Erwinia and other Enterobacteriaceae. These bacteria are not considered typical participants in wine fermentations, because they have not been previously detected by nontargeted culture-based techniques. However, similarly high abundances of Erwinia and other Enterobacteria were detected in wine fermentations by culture-independent methods (Bokulich et al. 2012b, Nisiotou et al. 2011, Takahashi et al. 2014). These repeated findings that contradict those in culture-based surveys suggest either that these Enterobacteriaceae represent mostly dead or viable but nonculturable cells carried into the fermentations on the grapes (discussed below) or that nontargeted culture-based techniques are insufficient for capturing the full microbial diversity of some wine fermentations (Bokulich et al. 2012a), as others have demonstrated (Heard and Fleet 1986). In the current study, Erwinia displayed high relative abundance from the start of fermentation, suggesting that the former scenario is most likely. However, without absolute quantification, relative abundance data cannot confirm whether the Erwinia species represent a stable, growing, or a declining microbial population. The repeated detection of these bacteria highlights the need for future studies to determine whether wine Enterobacteriaceae are alive, viable but nonculturable, or dead—and if not the latter, what role they may play during wine fermentations.
Because the current study analyzed only microbial DNA for marker-gene sequencing, it remained unknown whether the microbial profiles described here were from active, viable populations. DNA detection techniques detect both living and dead cells, and other methods must be used to limit detection to only live cells (Bokulich et al. 2012a, Bokulich and Mills 2012). However, the dynamic changes observed in our microbial profiles during fermentation are unlikely to represent dead cells. For example, the successions of yeasts observed in all fermentations, the emergence of low-abundance bacterial organisms in the late fermentation stages, and the increase of LAB and AAB in low-sulfite wines all suggest microbial growth, and such succession patterns are typically observed in wine fermentations (Bokulich et al. 2012a).
Next-generation sequencing (NGS) techniques have revolutionized the study of microbial communities in wines, foods, and other complex microbial systems (Bokulich and Mills 2012). NGS also offers many promising applications relevant to commercial winemaking. As sequencing costs continue to decline, such practical uses become much more accessible to the wine industry. The current results demonstrated that sequencing detects and recapitulates well-defined links between problem fermentations and aberrant bacterial growth (Bisson 1999). These data suggest that NGS will be useful for assessing microbial changes in wine treatments from oak chips to fining, filtration, and beyond. NGS will also be useful for tracking the efficacy of mixed microbial inoculations, e.g., with non-Saccharomyces yeasts or in malolactic fermentation cultures. Under practical winemaking scenarios, such techniques could be applied to retroactively deduce (or rule out) microbial links to past problem fermentations, to product spoilage, or to specific quality parameters. NGS will thus support evidence-based decision-making for refining winemaking processes or for preventing future problems.
Conclusion
NGS was used to measure microbial population dynamics as a function of SO2 concentration with or without concomitant inoculation with a commercial wine strain, S. cerevisiae EC1118. Low levels of SO2 in uninoculated fermentations led to slow and protracted fermentations, which displayed higher levels of LAB and AAB competing with Saccharomyces. Initiation in these fermentations coincided with the development of high titers of Saccharomyces. The results in this study confirmed that LAB are more sensitive to SO2 than AAB. They also demonstrated that some AAB may be present during the fermentation and bloom later at lower SO2 concentrations. The fermentation data also suggested that the presence of AAB is associated with sluggish fermentations.
Acknowledgments
Acknowledgments: The authors thank Charles Brenneman, Chad Masarweh, and Morgan Lee for logistical and technical support during this study. Nicholas A. Bokulich was supported by an American Wine Society Educational Foundation Endowment Fund scholarship, an American Society for Enology and Viticulture scholarship, the Samuel Adams Scholarship Fund (awarded by the American Society of Brewing Chemists Foundation), and Grant Number T32-GM008799 from NIGMS-NIH during the completion of this work. This work was supported in part by funding from the American Vineyard Foundation and the National Tax Agency of Japan.
- Received August 2014.
- Revision received October 2014.
- Accepted October 2014.
- Published online January 2015
- ©2015 by the American Society for Enology and Viticulture