Desulfovibrio is not always associated with adverse health effects in the Guangdong Gut Microbiome Project

Data acquisition and processing

Data acquisition and processing were applied as previously described for the GGMP (Chen et al., 2020). He et al. (2018) previously detailed and introduced GGMP. DNA was extracted from stool samples, and PCR amplification of the 16S rRNA V4 region and sequencing were conducted as described. Raw sequence data for the 16S rRNA gene are available from the European Nucleotide Archive at accession number PRJEB18535. These short reads were processed in the QIIME 2 framework using the Deblur denoising algorithm as previously described (Chen et al., 2020). A total of 6376 samples were remaining in the Deblur BIOM table for the follow-up analysis. Metadata for these samples can be found at https://www.nature.com/articles/s41591-018-0164-x. Deblur denoised sequences were assigned to bacterial features, which are almost equal to sub-OTUs (Wang et al., 2019). Taxonomic profiling of bacterial sub-OTUs was accomplished using the Greengenes reference database (version 13_8) as previously described (Chen et al., 2020).

Phylogenetic analysis

Complete 16S rRNA gene sequences of six DSV type strains and two Desulfobacter type strains were obtained from the National Center for Biotechnology Information (NCBI). The V4 region of the 16S rRNA gene sequence was amplified with the barcoded V4 primers used for the GGMP (forward primer: 5′-GTGYCAGCMGCCGCGGTAA-3′, reverse primer: 5′-GGACTACNVGGGTWTCTAAT-3′) (He et al., 2018; Walters et al., 2016). We constructed a phylogenetic tree based on the V4 region gene sequences using MEGA 5, including eight predominant features of DSV (defined as a feature detected in more than 1% of all samples), six DSV-type strains (D. piger ATCC29098, D. fairfieldensis ATCC70045, D. desulfurians ATCC27774, D. legallii strain H1, D. vulgaris DSM644, and D. intestinalis strain KMS2), and two Desulfobacter type strains (Desulfobacter postgatei DSM2034 and Desulfobacter vibrioformis B54). GenBank accession numbers for each type strain are available in Table S1. The phylogenetic tree was constructed using MEGA (version 5.05) by neighbor-joining (ref). The test of phylogeny was performed using the bootstrap method, and the number of bootstrap replications was 1000. The gaps/missing data treatment was a complete deletion. Sequence alignment was calculated using the BLAST search at NCBI (https://blast.ncbi.nlm.nih.gov/Blast.cgi).

To further show the phylogenetic relationship of DSV features, we also constructed a tree based on the gene sequences of eight predominant features of DSV, and their best matches. This tree includes the V4 region gene sequences of eight predominant features of DSV, and the sequences of twenty strains. The parameters and model for this tree were same as above.

Biostatistics analysis

To reveal the associations of DSV and predominant genera (defined as a genus with mean relative abundance above one percent) and the associations between predominant DSV features, we conducted a co-occurrence network analysis in R, as previously described (Chen et al., 2020). Spearman’s correlations were applied to each pair with FDR correction. Only significant correlations (FDR-adjusted P < 0.05) are shown.

Mann-Whitney U-tests were applied to compare DSV relative abundances in men and women, people with different BMI, normal waist and oversize waists, normal and elevated triglyceride (TG) levels, and normal and elevated uric acid (UA) levels. BMI and waist were classified on the basis of the Guidelines for Prevention and Control of Overweight and Obesity in Chinese adults. We classified TG levels on the basis of the Guidelines for the Prevention and Treatment of Dyslipidemia in Chinese adults. We classified UA levels according to the Chinese Guidelines for Diagnosis and Treatment of Hyperuricemia and Gout.

Kruskal-Wallis tests were performed to compute DSV relative abundance in seven Bristol stool types and at 14 geographical locations.

Correlations between Log₁₀ relative abundance of DSV and α-diversity indices and correlations between the number of predominant DSV features and α-diversity indices were calculated by Spearman’s rank correlation test by SigmaPlot 13.0. Correlation between Log₁₀ relative abundance of DSV and Log₁₀ relative abundance of Oscillospira was calculated by Spearman’s rank correlation test and visualized using “ggpointdensity” (version 0.1.0) package. Correlations between DSV (at the genus and sub-OTU levels) and host metadata were calculated by Spearman rank correlation test, and FDR correction was conducted to adjust all p-values. A two-tailed p-value less than 0.05 was considered to have statistical significance for all analyses.

Detection of Desulfovibrio in the GGMP samples

In total, we detected DSV in 3731 of 6376 gut microbiota. The mean relative abundance of the genus was 2‰. Seq14263 was the most prevalent. This sub-OTU was detected in 32% of all samples and accounted for 49% of the DSV-associated sequences (Table 1).

The mean relative abundance of eight predominant features ranged from 0.01‰ to 1.02‰ (Table 1 and Fig. 1A). Among the total samples, 2831 samples carried only one predominant DSV feature, and 814 samples carried two or more predominant features. The maximum number of predominant DSV features detected in a sample was four, and only two samples carried four predominant DSV features. Co-occurrence analysis at the sub-OTU level showed that Seq14263 was positively associated with Seq15128. Seq7611 was positively associated with Seq295. Seq12972 was not associated with other predominant DSV sub-OTUs (Fig. 1B).

Phylogenetic analysis showed that D. piger might be the most common and abundant DSV species in the GGMP. The V4 regions of Seq295, Seq5554, Seq1686, and Seq10119 were 100% identical to that of D. fairfieldensis ATCC70045, D. piger ATCC29098, D. desulfuricans ATCC27774, and D. intestinalis strain KMS2, respectively. The V4 region of Seq12972 was not very similar (<93%, Table S3) to that of the six DSV type strains (Fig. 2). All the predominant sub-OTUs of DSV showed low similarity (<94%) to D. vulgaris DSM644 (Table S3), suggesting that it may be an uncommon strain in the GGMP.

A phylogenetic tree that includes the eight predominant features of DSV and twenty strains of DSV also showed that Seq14263, Seq15128 and Seq5554 clustered with D. piger strains with good (80%) bootstrap support (Fig. S1), suggesting that D. piger might be the most common and abundant DSV species. D. fairfieldensis, D. desulfuricans, D. legallii, D. vulgaris, and D. intestinalis were less prevalent in the GGMP.

Microbial community profile link with Desulfovibrio

The relative abundance of the genus of DSV was positively correlated with the microbiota α-diversity indices. As the Log₁₀ relative abundance of the DSV genus increased, α-diversity also increased. The correlation between ascending α-diversity and DSV detection was positive when we examined α-diversity with Shannon, PD_whole_tree, and Observed sub-OTUs (Figs. 3A–3C). Spearman rank correlation analysis showed positive correlations between the number of predominant DSV features and Shannon, PD_whole_tree, and Observed sub-OTUs (Figs. 3D–3F).

The relative abundance of DSV was positively correlated with beneficial genera, and was negatively correlated with harmful genera. The relative abundance of DSV was positively associated with Oscillospira, Phascolarctobacterium, Prevotella, Coprococcus, Dialister, Ruminococcus, Akkermansia, Roseburia, Faecalibacterium, and Bacteroides and negatively associated with Streptococcus, Ralstonia, Sediminibacterium, Clostridium, SMB53, Escherichia, and Klebsiella (Fig. 4, Fig. S2). Correlation between DSV and Oscillospira was the strongest among all the correlations (Fig. 4, Table S4). Log₁₀ relative abundance of DSV was positively correlated with Log₁₀ relative abundance of Oscillospira in the GGMP in samples that DSV and Oscillospira can both be detected (Fig. S3A).

Host parameters linked to the detection of Desulfovibrio

DSV relative abundance did not vary significantly with gender (P = 0.1559, Fig. S4A). The mean relative abundance of DSV was similar across samples with different Bristol stool types (P = 0.4879), ranging from 1.3‰ (type 7) to 2.4‰ (type 5) (Fig. S4B). There was a clear difference in the relative abundance of DSV among different geographical locations (P < 0.0001, Fig. S4C). DSV were most abundant in subjects from the Nanshan, Shenzhen (G440305), and least abundant in samples from subjects in the Wuchuan, Zhanjiang (G440883). The mean relative abundance of DSV ranged from 1.2‰ to 3.1‰ (Fig. S4C).

DSV relative abundances in people with normal weight (18.5 ≤ BMI < 24) and low BMI (BMI < 18.5) were significantly more abundant than those in overweight (24 ≤ BMI < 28) and obese (BMI ≥ 28) (Fig. 5A). Women and men with normal waist sizes had higher levels of DSV than those with oversized waists (P = 0.0197 and P = 0.0015, respectively; Fig. 5B). The mean relative abundance of DSV was 2.2‰ vs. 2.0‰ in women and 2.1‰ vs 1.9‰ in men with normal and oversized waists, respectively (Fig. 5B). People with normal TG had higher levels of DSV than those with elevated TG (2.2‰ vs. 2.0‰, P = 0.0031, Fig. 5C). People with normal UA had higher levels of DSV than those with excessive UA (2.2‰vs 1.8‰, P = 0.0074, Fig. 5D).

At the genus level, DSV correlated negatively with host BMI, waist, TG, and UA. The mean relative abundances of DSV were 2.2‰ (low weight), 2.2‰ (normal weight), 2.0‰ (overweight), and 1.8‰ (obese). DSV genus and sleep time correlated positively (Fig. 5E). At the sub-OTU level, Seq14263 was negatively linked with BMI, waist size, TG, and UA and positively linked with systolic pressure (SBP), high-density lipoprotein (HDL), total cholesterol (TCHO), and sleep time; Seq12972 was positively linked with SBP; Seq7611 was negatively linked with diastolic pressure (DBP), SBP, HDL, and TCHO; Seq15128 was positively linked with HDL and sleep time; and Seq10119 was negatively linked with SBP (Fig. 5E, Table S5). Seq5554, Seq295, and Seq1686 were not correlated with these parameters. Low-density lipoprotein (LDL) was not correlated with DSV and these eight sub-OTUs (Table S5).