Introduction

Several Prevotella species inhabit the human gut, among which Prevotella copri is the most common with estimates indicating ~ 40% prevalence in the wider human population and relative abundances that exceed 50% in some individuals1. This species is more prevalent in non-Western populations likely due to its association with high fibre low fat diets2,3,4, and is often linked with desirable health measures such as reduced visceral fat and improved glucose metabolism5,6. However, there are conflicting reports that implicate P. copri in adverse conditions such as insulin resistance7, hypertension8 and persistent gut inflammation9. Recently, Tett and colleagues demonstrated that the existing P. copri lineage is comprised of at least four distinct species-level lineages each sharing less than 95% average nucleotide identity with one another1. Furthermore, the use of polysaccharides among P. copri isolates can markedly vary indicating distinct metabolic patterns within this lineage10. These inconsistencies between traditional taxonomic designations and species boundaries coupled with strain-specific metabolic diversity and dietary preferences could be reasons behind the conflicting observations related to P. copri and human health11, and more broadly underscores the need to characterise microbial diversity in the human microbiome to understand its function.

Other than P. copri, the human gut hosts several other Prevotella species such as P. stercorea12, P. rectalis13, P. histicola14 and some only known by genomes recovered from human gut metagenomes15,16,17,18. Of these, P. stercorea is likely the second most prevalent and relatively abundant species in the healthy human gut after P. copri based on healthy human gut metagenomes19 while much less is known about the distribution of other Prevotella species. For example, P. rectalis13 and P. ihumii20 were recently isolated from individuals with no apparent disease and to our knowledge have not been implicated in any health conditions. Furthermore, DNA-based gut microbiome analyses often mention unclassified Prevotella but fall short of pinpointing exact species21,22 likely due to uncharacterised diversity in this lineage.

Since P. copri are important members of the human gut microbiome due to their roles in human health as well as putative associations with several adverse conditions, we hypothesised that uncharacterised Prevotella species could also play important functions in the human gut as they are likely to share similar traits with P. copri. As such, an assessment of overall Prevotella diversity and distribution in the human gut will greatly facilitate understanding of their roles in health. Here, we examined the genomes of Prevotella species recovered from human gut metagenomes and show that the most prevalent and relatively abundant species are primarily related to P. copri and P. stercorea, and these species possess distinct gene repertoires likely reflecting adaptations to metabolic niches. Specifically, P. stercorea and related species lack several genes for xylan metabolism but have carbohydrate esterases that suggest they work cooperatively with P. copri in metabolising dietary fibre.

Methods

Obtaining Prevotella genomes from public repositories and gut metagenome data sets

This study was approved by the Joint Chinese University of Hong Kong New Territories East Cluster Clinical Research Ethics Committee (reference number 2016.707 and 2016.407) and was performed in accordance with relevant guidelines, regulations, and the Declaration of Helsinki. Written informed consent was obtained from all participants. Microbial genomes were downloaded from publicly available sources including the Genome Taxonomy Database (GTDB)23 R202 which contains RefSeq genomes and from several studies of human microbiome diversity15,17,18. In addition, we obtained Prevotella genomes from two previous studies examining gut microbiome diversity in Hong Kong24 and Yunnan25. For these two studies, metagenome sequences were quality filtered using Trimmomatic26 v0.39 to remove adapter sequences and low-quality regions. Quality filtered reads were then de novo assembled using MEGAHIT27 v1.2.9. Metagenome-assembled genomes (MAGs) were reconstructed from each metagenome assembly by mapping quality filtered reads to their respective assemblies using BWA MEM28 v0.7.17 and then using these coverage profiles as input into MetaBAT29 v2.10.2, v2.12.1 and MaxBin30 v2.2.7. The resulting MAGs were dereplicated using DAS Tool31 and checked for completeness and contamination using CheckM32 (Parks et al., 2015) v1.1.3. Taxonomy was inferred using GTDB-Tk33 v1.4.1 with the R95 database.

Inferring Prevotella phylogeny

In addition to completeness and contamination estimates from CheckM and taxonomies from GTDB-Tk, all MAGs were scanned using Kraken234 v2.1.1 with the StandardPlusPF precompiled database (v20210517) to assess the fraction of each genome consistent with taxonomy inferred from GTDB-Tk to avoid including chimeric genome assemblies. Genomes that scored > 90% completeness and < 5% contamination based on CheckM, inferred as Prevotella by GTDB-Tk and had > 70% of their genomes classified as Prevotella by Kraken2 were selected for phylogenetic inference and genome annotation. Identical genomes were dereplicated based on a threshold of 99% average nucleotide identity (ANI) using CoverM v0.6.0 (https://github.com/wwood/CoverM). The genomes were also dereplicated at 95% ANI to obtain a set of unique species. Bootstrapped phylogenetic tree (1000 bootstraps) was inferred using IQ-TREE235 v2.1.2 based on a concatenated alignment of 120 phylogenetically informative single copy bacterial marker genes generated by GTDB-Tk for both the set of 99% and 95% ANI genomes.

Estimating Prevotella prevalence and relative abundance in diverse human populations

To estimate the prevalence and abundance of Prevotella in the human population, we downloaded publicly available human gut metagenomes (Table S1) from countries including Austria36, Denmark, Spain37, France38, Sweden39, China8,40, Japan41, El Salvador, Peru42, Fiji43, Israel44, Mongolia45, Tanzania46,47 and USA48,49. These data were quality filtered using Trimmomatic v0.39 and then mapped to the species dereplicated set of Prevotella genomes (95% ANI) using CoverM v0.6.0 which produces relative abundance estimates of each genome in each cohort based on coverage and number of mapped reads (see https://github.com/wwood/CoverM/#calculation-methods). Prevalence was calculated as the number of metagenomes in which Prevotella was detected divided by the total number of metagenomes in each cohort. For surveys with case–control cohorts (e.g. colorectal cancer in Japan41), only non-disease subjects were included in this study.

Annotating genes

We used Panaroo50 v1.2.8 to determine gene orthologues shared among the Prevotella genomes. Orthologues were assigned Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthologs (KOs) and Carbohydrate active enzymes (CAZy) families by searching against a database of Hidden Markov Models implemented in EnrichM (https://github.com/geronimp/enrichM) v0.5.0. Distribution of KOs and CAZy families among the MAGs were visualised using non-metric multidimensional scaling (NMDS) ordinations on Jaccard distances using the Vegan package51 v2.5.7 in R v4.0.3.

Results

Phylogenetic diversity of human gut associated Prevotella species

To assess the extent of Prevotella diversity in the human gut, we first downloaded publicly available collections of MAGs derived from human gut metagenomes15,17,18. Additionally, we obtained MAGs from gut metagenome surveys of Hong Kong24 and Yunnan populations25 as we were interested in the distribution of Prevotella in these locales. From these data sets, we selected MAGs classified as members of the Prevotella genus based on GTDB taxonomy. After quality filtering and dereplication at 99% ANI to remove identical genomes, 3943 MAGs remained of which more than 60% were comprised of five highly represented species including P. copri (1579 genomes), P. sp003447235 (416 genomes), P. sp000434975 (185 genomes), P. stercorea (174 genomes) and P. sp002265625 (164 genomes) (Tables S2 and Table S3).

To establish phylogenetic relationships among the 3943 MAGs representing Prevotella species detected in the guts of the wider human population, we downloaded an additional 222 NCBI RefSeq Prevotella genomes held in the GTDB database and incorporated these with the set of 3943 MAGs. Phylogenetic trees were then inferred based on a concatenated alignment of 120 bacterial single copy marker genes to produce a bootstrapped consensus tree (Fig. 1). Within the lineage consisting of P. copri species, MAGs from the Asian continent were phylogenetically closer to one another compared with those from Western geographies (USA, Canada, Europe) possibly reflecting adaptations to local populations. These patterns were also observed in P. stercorea in which 128 out of 174 genomes were derived from Asian populations. Conversely, species such as P. pallens (24 of 33), P. salivae (18 of 23), and P. bivia (41 of 47) were largely represented by MAGs from Canada and USA. As for MAGs derived from gut metagenomes representing rural populations (e.g. Tanzania, Fiji, El Salvador), they were primarily classified as several not yet formally characterised species including sp002299635, sp002265625, sp000434975 and sp900543975. These observations indicate that numerous Prevotella species are found in human guts (at least 25 when considering those represented by > 20 MAGs for confidence; Table S2), many of which have not been formally characterised especially those from rural populations.

Figure 1
figure 1

Phylogenetic tree representing phylogenetic relationships among Prevotella isolates and metagenome-assembled genomes (MAGs) recovered from human gut metagenomes. Five Paraprevotella genomes were used as outgroup to root the tree. MAGs included in this tree were downloaded from publicly available repositories15,17,18 and from healthy human gut metagenome surveys of Hong Kong24 and Yunnan populations25. Genomes were dereplicated at 99% average nucleotide identity (ANI) and a concatenated amino acid alignment consisting of 120 phylogenetically informative marker genes (from GTDB) was used to infer phylogenetic trees using IQTREE2. Taxonomy labels from GTDB are shown for species represented by at least 20 genomes (i.e., lower odds of spurious misassemblies or genome binning) (see Table S2 for genome counts); species boundaries are indicated by the alternating grey shades. Note that the GTDB-based taxonomy used here is not comparable to the four P. copri clades A–D from Tett et al.1- these four clades from Tett et al.1 are indicated separately in the figure. The outer colour strip indicates source country/continent of the respective genomes shaded by continent.

Uncharacterised Prevotella species are prevalent and relatively abundant in the human gut

Since many of these Prevotella MAGs are directly derived from human gut metagenomes and may not be identified using conventional microbial community profiling tools, we sought to estimate their prevalence and relative abundances in the human population by mapping metagenome reads. In total, we downloaded 4095 human gut metagenome data sets from non-disease cohorts representing populations from China, Japan, USA, Austria, Denmark, France, Spain, Israel, Sweden, El Salvador, Peru, Fiji, Mongolia, and Tanzania, and mapped quality filtered reads to the set of Prevotella MAGs dereplicated at the species level (95% ANI). As expected, total Prevotella relative abundance was highest in Fiji, a rural population data set (average 36.1% relative abundance), followed by Tangshan (China; 20.5%) and Mongolia (19.6%) (Fig. 2) (Table 1). They were also more widespread in rural communities such as Fiji (97.8% prevalence), El Salvador (94.7%) and Mongolia (91.8%) compared with urban cohorts in which prevalence ranged between 11 and 58% (Table 1). Among the Prevotella species identified above, P. copri was the most prevalent (average 30.7%) and relatively abundant overall (average 3.3%) (Table 2, Table S4). Other prevalent and relatively abundant species were primarily from two distinct lineages, one more closely related to P. copri and the second to P. stercorea (hereafter referred to as copri and stercorea groups) (Fig. 2) (Table S4). Collectively, these species other than P. copri account for an average 4.3% relative abundance which is comparable to the 3.2% for P. copri (Table S5). Like P. copri, most of them are more prevalent and relatively abundant in rural populations such as Fiji, Mongolia and El Salvador, possibly indicating a similar association with non-Western lifestyles and/or high fibre diets. In addition, we detected a few lineages at lower prevalence and relative abundances that are possibly more restricted in their distribution such as P. sp001275135 and P. sp003447235 (Table S4). For example, P. sp001275135 has an average prevalence of only 3.7% across all data sets included in this study, but its prevalence in France and Spain is 14.8% and 17.5%, respectively. Taken together, these observations indicate that the most common Prevotella species in the human gut are primarily comprised of P. copri, P. stercorea and several closely related lineages, although some species with more limited distributions may be prevalent in certain biogeographies.

Figure 2
figure 2

Estimated Prevotella relative abundances across several human gut metagenome data sets. Phylogenetic tree indicates relationships of the Prevotella genomes dereplicated at the species level (> 95% average nucleotide identity). Only genomes with reads mapped are included in this figure. Taxonomy labels are shown for species with at least 1% relative abundance (indicated with *) or at least 50% prevalence (indicated with ^) in any one data set. See Table S4 for relative abundance and prevalence values.

Table 1 Distribution of total Prevotella in the human gut.
Table 2 Most relatively abundant and prevalent Prevotella species in the human gut.

Human gut Prevotella species occupy distinct functional niches

Prevotella copri in the human gut are associated with high fibre non-Western diets as they possess extensive repertoires of carbohydrate active enzymes that allow this species to metabolise complex polysaccharides. We thus hypothesised that other gut Prevotella species are also adapted to carbohydrate-related metabolism although their exact gene content likely differs compared with P. copri, thus allowing them to utilise distinct fractions of dietary fibre. To identify differences in gene content, we first predicted protein coding genes in each genome and then compared predicted genes to the KEGG Orthology (KO) database to infer metabolic potential. NMDS ordinations of KO counts indicated that genomes of prevalent and relatively abundant species belonging to the two copri and stercorea groups identified above significantly differed in their gene content (Fig. S1), and this could be attributed to several pectin and hemicellulolytic enzymes detected in the copri group but missing from stercorea (Table S6). Hemicellulose is a polysaccharide polymer in plant cell walls and typically consists of sugars such as xylose, arabinose, mannose, galactose and rhamnose. The stercorea group either lacks or has fewer copies of genes such as xylanase (K01181), xylosidase (K01198), xylose symporters (K08138), mannosidase (K01218), pectinesterase (K01051), rhamnosidase (K05989) and endoglucanase (K01179) suggesting that stercorea species have reduced ability to utilise hemicelluloses compared with copri-related species. Instead, stercorea species may participate in breaking down oligosaccharides and glycoproteins based on detection of genes such as N-acetylglucosamine-6-phosphate deacetylase (K01443) and endo-beta-N-acetylglucosaminidase (K01227). In addition, stercorea species possess components of the glycine cleavage system and histidine utilisation system (hut) which could indicate a more extensive amino acid metabolism compared with the copri group. These carbohydrate utilisation patterns are also reflected in CAZy profiles indicating the stercorea group as non-hemicellulose degraders. CAZy families involved in hemicellulose degradation such as CE8 (pectin methylesterase), GH10 (β-1,4-xylanase), GH28 (polygalacturonase), GH30 (endo-xylanase), GH43_1 (β-xylosidase, a-l-arabinofuranosidase), GH67 (α-glucuronidase), GH115 (xylan α-1,2-glucuronidase) and GH141 (α-l-fucosidase, xylanase) were absent in nearly all stercorea group genomes, whereas the presence of CE3 (acetyl xylan esterase) and CE9 (N-acetylglucosamine 6-phosphate deacetylase) suggest a possible role in assisting with removal of ester modifications to facilitate carbohydrate degradation52 by P. copri or other gut bacteria with the necessary gene repertoire (Table S7) (Fig. 3). The presence of GH33 (sialidases) and GH123 (N-acetylgalactosaminidases) further suggest that the stercorea species could metabolise sialic acids derived from dietary (e.g. glycolylneuraminic acid in red meat53) and/or human glycans (e.g. acetylneuraminic acid) and are likely to utilise these substrates in addition to or instead of dietary fibre. Other Prevotella species such as sp001275135 prevalent in France and Spain possess hemicellulolytic CAZy families akin to copri, whereas sp002299635 and sp003447235 are more similar with stercorea in that they lack hemicellulolytic families. These similarities in CAZy content among species are supported by an NMDS of total CAZy counts showing two broad types of CAZy profiles in human gut Prevotella, one representing the copri group and the other the stercorea group with reduced capacity to utilise hemicelluloses (Fig. S2).

Figure 3
figure 3

Distribution of carbohydrate active enzyme (CAZy) families in human gut Prevotella (average copies per genome). Only genomes with > 1% relative abundance or > 50% prevalence as indicated in Fig. 2 are included. CAZy families were identified in the genomes using Hidden Markov Models. Asterisks indicate families with statistically different copy counts (FDR < 0.001 Kruskal Wallis test with false discovery rate adjustment) and with at least one copy in any one of the three groups (copri, stercorea or other). See Table S7 for copy numbers. CE carbohydrate esterase, GH glycoside hydrolase, GT glycosyl transferase, PL polysaccharide lyase.

To examine whether diet may influence the distribution of gut Prevotella species, we assessed dietary intake information collected as part of the Yunnan gut microbiome study25 together with abundance values obtained from metagenome read mapping (see section above). The Yunnan survey was comprised of 881 individuals representing six ethnic groups in Yunnan province; 427 of the 881 were recruited from rural counties and another 454 from urban Kunming (capital of Yunnan province). A total of 676 individuals provided their past month dietary intake through a binary response questionnaire. We first established that diet primarily differed among the ethnic groups followed by rural/urban status (p < 0.05, PERMANOVA) (Table S8). While the Zang (Tibetan) diet was most distinct compared with other ethnic groups (Fig. S3A), total Prevotella relative abundance in the Zang was comparable to several ethnicities (11.7% vs. 10.9–15.5%) whereas the Dai (8.0%) and Han (3.9%) had significantly lower relative abundances (Fig. S3B). Across all ethnicities, only coffee consumption was associated with an increased Prevotella relative abundance (p < 0.05, generalised linear model with FDR adjustment) (Table S9) although coffee consumption was reported in only 19 individuals. Other foods in the questionnaire were not correlated with either total Prevotella relative abundance or with abundances of the two stercorea and copri lineages identified above. Similarly, the low relative abundances of Prevotella in the Dai and Han could not be attributed to any specific foods included in this questionnaire. These observations indicate that non-dietary influences (e.g. genetics, lifestyle) could play a role in gut Prevotella associations in the Yunnan cohort, although more in-depth and quantitative dietary information in addition to existing data provided in that Yunnan study may be required to resolve exact correlations between these factors and diet.

Discussion

The Prevotella genus is a lineage of interest as its member species are a common feature of the human gut microbiome and have been repeatedly implicated in health and disease. P. copri is often the primary focus of the Prevotella genus, but there is increasing incentive to catalogue and characterise other gut Prevotella species. For example, a gut microbiome survey of competitive cyclists revealed enrichment of P. copri and several unidentified Prevotella species correlated with exercise duration21. Another study of gut microbiome development in malnourished children reported that children who received a food formula designed to stimulate development of a healthy gut microbiota54 showed greater enrichment of P. copri and other unnamed Prevotella species compared with existing nutrition regimes22. To capture gut Prevotella diversity without relying on known marker genes (e.g. MetaPhlAn) or genomes deposited in RefSeq (Kraken2), our workflow in this study classifies metagenome reads against a large collection of human gut-derived Prevotella MAGs to better resolve and estimate abundances of Prevotella species in the human gut15,17,18.

We show that the most numerically abundant Prevotella species in the human gut are largely comprised of two distinct lineages, one closely related to and including P. copri while the other to P. stercorea. Both the copri and stercorea groups were more relatively abundant in rural non-Westernised human populations, suggesting that the presence of stercorea-related lineages is also driven by diet and lifestyle factors akin to P. copri2,3,4. Using the P. copri phylogenetic framework inferred by Tett et al.1, the copri group of species we identified here largely corresponded with the four clades circumscribed in that study and the potential for carbohydrate metabolism detected in this group also broadly mirrors that reported by Tett et al.1. While P. copri are known hemicellulose degraders55,56, compared with P. copri the stercorea group occupies a separate functional niche categorised by a lack of hemicellulolytic gene families. In addition, the stercorea group possesses carbohydrate esterase families that were comparatively much less common in the copri group. From these features, we posit that a possible role of the stercorea group is to remove ester-based modifications from dietary carbohydrates to facilitate hydrolysis by gut species with the necessary hydrolytic capacity52. Additionally, the stercorea group lineages possess sialidases. Sialic acids have been demonstrated to play roles in processes such as mediating interaction between immune cells, pathogen binding to human cells57 and inflammation in the gut58. The presence of sialidases in the stercorea group therefore suggests that stercorea-related bacteria have important functions that could impact human health.

The main limitation of this work is the lack of detailed lifestyle and dietary information needed to investigate whether distribution of the several gut Prevotella identified here is influenced by lifestyle and/or diet. Existing evidence indicate that P. copri is associated with dietary fibre intake59, but we did not observe such associations with the limited dietary data (yes or no responses) provided in the Yunnan study25. In addition, our metabolic inferences of the various Prevotella species are based on genome sequences and thus need to be validated by isolation and laboratory culture experiments. Nevertheless, since low fibre diets can result in poor gastrointestinal health60 and Prevotella are one of the major fibrolytic bacteria in human guts61, we foresee that there will be increased interest in exploiting Prevotella to potentially improve human health and nutrition. Our findings of distinct Prevotella species in the human population underscores the need to classify the human gut microbiome in greater resolution to understand and exploit the metabolic potential of gut microorganisms.