Ecological diversification reveals routes of pathogen emergence in endemic Vibrio vulnificus populations

Significance

Our study addresses one main question: What are the ecological and genetic factors that drive pathogen emergence? To date, systematic experimental approaches to address this phenomenon are very limited. Here, we implemented a holistic approach to dissect the ecological, genetic, and evolutionary drivers that foster the selection of virulence traits and pathogenic clones within an environmental population using Vibrio vulnificus, an aquatic bacterium that can cause a deadly septicemia in humans, as a model system. Our results suggest how ecosystems may generate selective pressures that facilitate the emergence of specific strains with pathogenic potential in a natural population and can be applied toward predictive frameworks to assess the risk of pathogen emergence from environmental sources.

Abstract

Pathogen emergence is a complex phenomenon that, despite its public health relevance, remains poorly understood. Vibrio vulnificus, an emergent human pathogen, can cause a deadly septicaemia with over 50% mortality rate. To date, the ecological drivers that lead to the emergence of clinical strains and the unique genetic traits that allow these clones to colonize the human host remain mostly unknown. We recently surveyed a large estuary in eastern Florida, where outbreaks of the disease frequently occur, and found endemic populations of the bacterium. We established two sampling sites and observed strong correlations between location and pathogenic potential. One site is significantly enriched with strains that belong to one phylogenomic cluster (C1) in which the majority of clinical strains belong. Interestingly, strains isolated from this site exhibit phenotypic traits associated with clinical outcomes, whereas strains from the second site belong to a cluster that rarely causes disease in humans (C2). Analyses of C1 genomes indicate unique genetic markers in the form of clinical-associated alleles with a potential role in virulence. Finally, metagenomic and physicochemical analyses of the sampling sites indicate that this marked cluster distribution and genetic traits are strongly associated with distinct biotic and abiotic factors (e.g., salinity, nutrients, or biodiversity), revealing how ecosystems generate selective pressures that facilitate the emergence of specific strains with pathogenic potential in a population. This knowledge can be applied to assess the risk of pathogen emergence from environmental sources and integrated toward the development of novel strategies for the prevention of future outbreaks.

The emergence of human pathogens is one of the most concerning public health topics of modern times (1–4). According to the World Health Organization, over 300 emerging infectious diseases have been reported in the 1940 to 2004 period, a trend that has continued steadily with recent outbreaks of Ebola in West Africa, Cholera in Yemen, and the global pandemic caused by COVID-19 (3–5). Even though classical molecular approaches have advanced our understanding of bacterial pathogenesis, to date, the genetic adaptations and ecological drivers that facilitate selected strains within a species to emerge as pathogens and successfully colonize the human host remain poorly understood. Given the magnitude and complexity of this urgent threat, it is critical to develop tractable organismal model systems and theoretical frameworks that allow us to dissect the molecular adaptations and environmental factors that lead to the emergence of such human pathogens.

Vibrio vulnificus, an emergent human pathogen, is one of the leading causes of non-Cholera, Vibrio-associated deaths globally (6). Despite being a natural inhabitant of estuarine, coastal, and brackish waters (7), this flesh-eating bacterium has gained particular notoriety as one of the fastest killing pathogens (8, 9). Humans are typically infected with V. vulnificus through ingestion of contaminated raw seafood or by direct exposure of open wounds to seawater (6). V. vulnificus infections often result in fulminant septicaemia with an alarming mortality rate exceeding 50% (6, 10–13). The bacterium is particularly lethal in some susceptible hosts, such as immunocompromised patients or those with alcohol-associated liver cirrhosis, diabetes mellitus, or hemochromatosis (14). The annual case counts of V. vulnificus infections have steadily increased over the past 20 y in the United States (15). An upsurge in its worldwide distribution over the past three decades, in correlation with climate change, has led to disease outbreaks in regions with no history of V. vulnificus infections (16–18). Furthermore, models predict this trend to continue resulting in a steady expansion of its geographical range and the subsequent increased risk of human infections (16, 19–21).

Based on a series of biochemical and phenotypic traits, V. vulnificus strains have been historically classified into three Biotypes (BT): BT1, which is mostly associated with human infections (22, 23), BT2, which is primarily pathogenic to eels (24, 25), and BT3, which is geographically restricted to Israel and possesses hybrid characteristics from BT1 and BT2 (26, 27). In contrast to Vibrio cholerae, in which all strains capable of causing cholera belong to a single clade, genomic comparisons of V. vulnificus reveal a more complex pattern in the distribution of its clinical strains (28–30). Phylogenomic analyses indicate that the population of V. vulnificus is composed of four distinct groups or clusters (Cluster 1 to 4), which largely overlap with the classical BT classification system (23, 26, 28, 31, 32). Our analyses indicate that the two largest clusters, C1 and C2, exhibit high genomic divergence and appear to be speciating (28), with clinical strains from BT1 predominantly belonging to C1 (22, 23), whereas strains from C2 are primarily associated with BT2 (6, 24, 25). C3 is highly clonal and fully overlaps with BT3, and the rare C4 contains only four nonclonal strains and belongs to BT1 (28, 31). Interestingly, despite patients showing conserved clinical symptoms, C1 clinical strains arise from different clades within the cluster, suggesting independent emergence events of this deadly pathogen (28, 31, 32). To date, the unique genetic traits that allow certain C1 strains to cause severe septicemia remain mostly unknown, posing a daunting public health risk as it hinders our ability to detect potentially pathogenic V. vulnificus (33).

Recently, using a combination of bioinformatic and phenotypic analyses that surveyed more than 100 strains of V. vulnificus, we determined that V. vulnificus C1 appears to be associated with a unique ecological lifestyle or ecotype (28). Nonetheless, to date, the ecological drivers that lead to the emergence of clinical V. vulnificus C1 and their pathogenic traits remain poorly understood. In order to start untangling the complex in-situ interactions between genotypes and the environment that underlie the emergence of clinical strains, in this study, we recently surveyed a large estuary in eastern Florida, the Indian River Lagoon (IRL), where outbreaks of the disease frequently occur (7, 34). We found endemic populations of V. vulnificus in the estuary and established two sampling locations to study the environmental dynamics of this bacterium in several natural reservoirs such as water, sediment, oysters, and cyanobacteria. Interestingly, the two sampling sites show major differences in the distribution of V. vulnificus clusters. One of them, Feller’s house (Site A), appears to be significantly enriched with C1 strains, whereas in the second sampling site, Shepard Park (Site B), we mostly recovered strains from C2. Genomic analyses of these strains indicate that, despite these major differences in distribution, high recombination rates as well as frequent exchange of mobile genetic elements and virulence factors between these V. vulnificus populations occur. Microdiversity analyses of these genomes revealed unique genomic markers among C1 strains in the form of clinical-associated alleles (CAAs) with a potential direct role in virulence. The isolated V. vulnificus strains are resistant to numerous commonly used antibiotics irrespective of cluster or site of isolation. However, phenotypic analyses indicate that strains from Site A exhibit traits associated with clinical outcomes, including the ability to resist serum and catabolize sialic acid, unlike those from Site B. Finally, metagenomic and physicochemical analyses of the sampling sites indicate that this marked cluster distribution is strongly associated with distinct biotic and abiotic factors (e.g., salinity, nutrients, or biodiversity) revealing how ecosystems might generate selective pressures that facilitate the emergence of specific strains in a population with pathogenic potential.

Results and Discussion

Gene Marker, thiF, Can Detect V. vulnificus and Distinguish between Clusters.

Before initiating our sampling protocol, we looked for specific markers to rapidly screen environmental samples on a large scale. Specifically, we needed reliable genetic markers that could 1) detect specifically V. vulnificus, 2) accurately characterize them based on their cluster, and 3) discriminate between clonal and nonclonal strains. The hemolysin gene vvhA, typically used to detect V. vulnificus, although species specific, is limited in its potential to distinctly classify strains into clusters or discriminate between nonclonal strains (35–37) (Fig. 1A and SI Appendix, Fig. S1A). Other approaches, such as multilocus sequence typing (MLST), although effective in characterizing strains, require the PCR amplification, assembly, and concatenation of several housekeeping genes (32, 37–40), which is resource intensive and impractical for the rapid screening of V. vulnificus in environmental samples. In order to identify potential markers that meet all the requirements mentioned, we compared all available C1 and C2 genomes in public databases through pangenome analysis. The number of gene families shared was 978 (accounting for ∼22% average number of genes in a V. vulnificus genome), which we consider the core genome. We performed individual phylogenetic trees for these gene clusters and identified, across both chromosomes, a total of 47 genes that clearly differentiated C1 and C2 clusters. We compared these genes, using representative strains from all clusters (SI Appendix, Table S1), based on percentage of sequence identity, against CMCP6, a reference V. vulnificus strain. We singled out the genes that had the highest percentage identity with strains from C1 but the least identity with those from C2 and vice versa and compared them against vvhA. We finally selected a total of six candidate genes (yycF, pfeS, acuB, yqhD, uvrY, and thiF), three from each chromosome, as potential markers (SI Appendix, Table S2). Although all six candidate marker genes clearly differentiate C1 V. vulnificus strains from C2 (SI Appendix, Fig. S1A), the response regulator uvrY, and the sulfur carrier protein adenylyltransferase thiF, had the maximum resolution in distinguishing all four clusters (C1 to C4) as well as individual strains within each cluster, which serves as a proxy for discrimination of clonal populations (Fig. 1A). Additionally, the relative distances of the four clusters in the phylogenetic tree of thiF most accurately correspond to the evolutionary tree of V. vulnificus built using single nucleotide polymorphisms and average nucleotide identities of all known V. vulnificus strains (28). Upon testing the species specificity of thiF with Vibrio parahaemolyticus RMID2210633 or V. cholerae O395, thiF was found to be specific to V. vulnificus (SI Appendix, Fig. S1D). Thus, thiF has the potential to 1) detect V. vulnificus strains, 2) separate them by clusters, and 3) discriminate between clonal and nonclonal strains based on their whole genome. Furthermore, the concatenation of all six genes had at least twice the resolution and discriminatory power to differentiate all four clusters compared with vvhA, making it an accurate set of genes for MLST analyses of V. vulnificus strains (Fig. 1A).

Fig. 1.

Isolation of V. vulnificus from Eastern Florida. (A) Maximum likelihood phylogenetic tree of hemolysin gene, vvhA, sulfur carrier protein adenylyltransferase, thiF, and the concatenation of the six candidate genes (yycF, pfeS, acuB, yqhD, uvrY, and thiF) for representative strains from all four clusters. Members of the same cluster (C1 to C4) are indicated with the same color. Trees are unrooted and drawn to scale. Branch lengths indicate number of substitutions per site. (B) Map of Florida indicating the sampling sites: Feller’s house and Shepard Park. (C) Maximum likelihood phylogenetic tree of V. vulnificus isolates based on thiF. Branches containing members that belong to the same cluster are indicated with the same color: green for C1 representatives and blue for C2. The names of the strains are colored in relation to the location from which they originate. The colored circles represent where they were isolated from, and the red stars represent those strains that have been sequenced.

Detection of V. vulnificus along the IRL.

The IRL (Eastern Florida, United States) is one of the most biodiverse estuaries spanning an expansive geographic range with contrasting environments in Florida, where outbreaks of the disease frequently occur (Fig. 1B) (7, 34, 41). We recently surveyed this large estuary, and we established two sampling sites at environmentally distinctive locations along the IRL (Fig. 1B). We collected samples in three sampling events (15 November 2018, 24 July 2019, and 22 August 2019), including biotic reservoirs such as oysters and cyanobacteria. V. vulnificus was isolated by sequential plating of the enriched populations on CHROMagar Vibrio (CaV) and thiosulfate citrate bile salts sucrose (TCBS), as described in the Materials and Methods (42). From a total of 1,856 colonies screened, only 245 were identified as potential V. vulnificus isolates based on the chromogenic plating method. An overall higher proportion of V. vulnificus was detected at Site B (SI Appendix, Fig. S1B). At Site A, the distribution of V. vulnificus was found to be highest in oysters (45.3%) and water (43.4%) in contrast to sediments, which contain on average only 11.32% (SI Appendix, Fig. S1B). Furthermore, a higher proportion of V. vulnificus was observed during the summer at both sites (SI Appendix, Fig. S1B; 96.3% at Site A, 91.7% at Site B), likely as a consequence of increased water temperatures (>20 °C). The 245 potential V. vulnificus isolates were further confirmed using the novel gene marker thiF. PCR amplification of the thiF gene yielded 141 confirmed V. vulnificus isolates. We sequenced these PCR products and constructed a phylogenetic tree to determine cluster affiliation. To minimize further examination of strains of clonal origin that might have proliferated during enrichment, we only analyzed one strain within a group if 1) the thiF alignment looked identical within the group, 2) the strains came from the same replicate and fraction, and 3) they were isolated during the same sampling event. As a result, 87 out of the 141 confirmed V. vulnificus isolates were selected for further analyses (39 isolates from Site A and 48 from Site B) (Fig. 1C). Strikingly, phylogenetic analysis using gene marker thiF showed that most isolates from Site A belong to C1 (97.4%, 38/39), whereas the majority of isolates from Site B belong to C2 (87.5%. 42/48) (Fig. 1C). This clear ecological separation between the two clusters provides an ideal framework to examine evolutionary processes underlying the emergence of pathogenic traits within a population and a platform to understand how ecosystems generate pressures that facilitate the selection of strains with pathogenic potential. In order to address this, we first dissect the genomic determinants and population structure of these environmental V. vulnificus strains, assess their pathogenic potential, and finally link these results with environmental factors (abiotic and biotic) associated with their marked cluster distribution.

Genomic Determinants of V. vulnificus Emergence.

Ecological preferences of V. vulnificus clusters.

To investigate the genomic determinants that potentially drive the ecological niche preferences of the clusters, we selected several strains for genome sequencing to obtain a proportionate representation of each cluster, reservoir, fraction, host, and date of isolation. This resulted in a total of 27 V. vulnificus isolates sequenced (SI Appendix, Table S3): 13 from Site A (2 sediment, 6 oyster, and 5 water isolates, 1 of which belonged to C2) and 14 from Site B (4 sediment, 4 cyanobacteria, and 6 water isolates, including 2 from C1). For a robust phylogenomic association, we included 74 dereplicated V. vulnificus genomes (e.g., genomes >99% average nucleotide identity [ANI]) currently available in public databases. We used both phylogenomic trees and ANI-based clustering of both chromosomes separately to group the genomes into the previously defined clusters [C1 to C4 (28)] (SI Appendix, Figs. S2A and S3). Based on these results, we decided to use the ANI of chromosome I as a reference for taxonomic classification since coverage is high (>70%), even among the most divergent clusters (C1 and C2). Interestingly, we found evidence of mixing or transfer of chromosomes between clusters of V. vulnificus. For instance, while chromosome I from FORC_037, an environmental strain isolated from soft-shell clam, had an ANI >98% with members of C2 and ca. 95% with C1, for chromosome II was the other way around (SI Appendix, Fig. S2 A and B).

Whole genome phylogeny confirmed the marked differences in the distribution of V. vulnificus clusters obtained with thiF gene, corroborating the enrichment of C1 strains in Site A (Fig. 2A), except for the strain IRLE0015 that together with NV22 clustered closely to BT3 strains from the Israel outbreak (Fig. 2A). As previously mentioned, we selected one nonclonal strain from Site A that belonged to C2 (IRLA0043) and two from Site B belonging to C1 (IRLE0056 and IRLE0004). These gave us the opportunity to investigate the presence of potential genomic determinants specifically associated with each site, that is, whether C1 and C2 strains from Site A have a unique pool of genes that is absent in strains from Site B irrespective of cluster. The common part of the pangenome of all C2 strains from Site B was subtracted from the genome of the IRLA0043 strain, the only one in this cluster isolated from Site A. More than 500 genes were specific to this strain, and apart from the capsule glycosylation genes, we found a second cluster of genes (rtxB-rtxD-rtxE) encoding a type I secretion system (T1SS) with a high similarity (99%) to several strains of Vibrio coralliilyticus. Specifically, this system appears to be associated with excretion of an enterotoxin (Efa-1/LifA) (43). Within these specific genes, we also found a second type VI secretion system (T6SS) (28) and an integrative conjugative element. On the other hand, C1 strains from Site B (IRLE0004 and IRLE0056) had only 200 unique genes compared to C1 strains from Site A. Among the specific genes of IRLE0004, we found a gene cluster conferring the ability to utilize tetrathionate as an electron acceptor, a common sulfur compound present in most soils (44); interestingly, this strain has been isolated from sediment. The ability to utilize tetrathionate has been associated with virulence in Salmonella enterica by providing a growth advantage to the bacterium in the inflamed gut (45). The functional annotation associated with the specific part of IRLE0056 was limited to the use of rhamnose, several toxin-antitoxin systems, and the gene encoding the HipA involved in dormancy (46). Although it highlights the presence in the environment of some virulence factors that can be easily shared between the two clusters, our analysis did not identify any specific genomic determinants that may explain the differential distribution of these strains.

Fig. 2.

Phylogenomic and population structure of V. vulnificus. (A) Maximum likelihood phylogenomic tree of V. vulnificus strains obtained in this study (highlighted in red) together with all available reference genomes using core genome of chromosome I. Branches containing members that belong to the same cluster (C1 to C4) (ANI > 97%) are indicated with the same color. The color chart of the circles of the plot indicates the isolated source and the host of the corresponding strains. Gray box shows the 15 recombinant populations detected among all strains. The orange box highlights the strains belonging to subpopulation 15. (B) Schematic representation of the CPS genomic island. Color-coded arrows show locations of important genomic features. Variable regions 1 and 2 are highlighted in blue and green, respectively.

Ecologically meaningful populations of V. vulnificus.

Despite the marked environmental preferences and genomic divergence between C1 and C2 clusters, our recent in silico studies indicate frequent exchange of mobile genetic elements (28). Here, we have the opportunity to study potential recombination in natural V. vulnificus populations in an endemic area. Recombination is particularly worrisome as novel practices such as aquaculture can lead to the emergence of hybrid strains, as evidenced by a deadly outbreak in Israel caused by an entirely new cluster (C3) (27) and the presence of a C3-like strain isolated in this study (IRLE0015) (Fig. 2A). To evaluate this phenomenon, we used an approach for assessing recent recombination events that enables the delineation of ecologically relevant populations, that is, groups with the potential to exchange genetic material (47). Our analyses revealed the presence of 15 major recombining populations. Some of these populations coincide with the cluster classification indicative of high intracluster recombination e.g., C3 and C4 (Fig. 2A). However, C2 is made up of 12 populations. Eleven of them formed by a single member and therefore indicating that there are no recombination events that connect these strains with the rest of the cluster (48). Interestingly, all members of C1 form a single population (P15) with the majority of C2 representatives indicating that, despite divergence (ca. 95% ANI), these clusters are connected by recent recombination events (Fig. 2A).

The capsular polysaccharide (CPS) cluster is an essential virulence factor of V. vulnificus (49). Our previous analyses suggest that recombination may be a major evolutionary mechanism leading to the high diversity of the CPS cluster (28). Thus, we investigated the genomic diversity of the CPS between both clusters in these natural populations. Strain IRLA0152 (C1) isolated from the free-living fraction at Site A had a similar variant of the CPS found in an infected patient isolate (FDAARGOS_119) (Fig. 2B). One of the hypervariable parts of the CPS from the oyster isolate OH0023 was identical to that found in the reference clinical strain CMCP6, highlighting the environment as a reservoir of these essential virulence genes (Fig. 2B). Furthermore, certain CPS clusters are distributed in the population irrespective of cluster of origin and sampling location. Specifically, we found the same CPS in one C1 strain from Site B (IRLE0056) and three C2 strains, one of them from Site A (IRLE0043) and two from Site B (IRLE0062 and IRLE0057) (Fig. 2B). The only variation was a small insertion in IRLE0043 due to several insertion sequence (IS) elements, which suggests that this may be another mechanism that can introduce variability within the CPS cluster (Fig. 2B). Overall, our results indicate that despite the genomic divergence and their marked ecological differences, there is a wide recombination among the clusters in an endemic area such as the IRL, including the transfer of major virulence factors within their natural environment.

Pangenome analyses reveal genetic drivers associated with virulence emergence.

The majority of clinical V. vulnificus strains belong to C1, similarly to most strains isolated from Site A. To date, the specific genomic determinants that allow some C1 strains to successfully colonize humans remain mostly unknown. In order to elucidate genetic factors associated with the emergence of clinical V. vulnificus C1 from environmental gene pools and to determine whether C1 strains from Site A encoded clinical-associated traits, we compared genomes from strains isolated in this study against those from bona fide clinical C1 and nonpathogenic strains (50, 51). Specifically, we selected genomes from four distinct groups: 1) nine C1 strains isolated from Site A and 2) nine C2 strains from Site B together with 3) nine C1 strains that are bona fide clinical, that is, isolated from patients with septicaemia, as well as 4) nine nonpathogenic strains from C2, that is, isolated from environmental sources and susceptible to the bactericidal effect of serum and monocytes (50, 51). Microbial species diversity was analyzed via a Partitioned PanGenome Graph Of Linked Neighbors [PPanGGOLiN (52)]. The estimated size of the “persistent genome” (gene families present in almost all genomes) is similar for each individual group as well as for all the groups combined together, ca. 3,700 gene families (ca. 52% of the total genes families per genome). This is quite remarkable given the genomic divergence between groups (Fig. 3A). The proportion of gene families that formed the “shell genome” (gene families present in three to seven genomes) was only 1% of the total for both C1 groups and 2% for C2 groups. The remaining gene families present in low frequency (one to three genomes) were classified as the “cloud genome” (Fig. 3A). As predicted, the percentage of gene families assigned to functional categories (SEED subsystems database) for each pangenome partition varied significantly: from 64% assigned to the persistent genome to ca. 20% for the cloud and shell. The latter being typically associated with diverse environmental adaptations, including pathogenesis, which highlights the enormous genomic plasticity that remains to be addressed for these organisms.

Fig. 3.

Pangenome analysis of V. vulnificus strains. (A) Pangenome analysis for groups, 1) nine C1 strains isolated from Site A (C1 IRL), 2) nine C2 strains from Site B (C2 IRL), 3) nine reference C1 strains that are bona fide clinical (C1 Clinical), and 4) nine reference environmental strains from C2 (C2 environmental). The proportions of gene families in the persistent, cloud, and shell genome are highlighted in orange, green, and blue, respectively. (B) Schematic representation comparing the genomic island of the gene cluster involve in sialic acid catabolism. (C) Comparison of the ratio of nonsynonymous to synonymous substitutions (dN/dS ratio) between reference clinical strains and C1 strains isolated from Site A in the IRL. (D) Comparison of the dN/dS values of each individual strain versus the rest in the C1 IRL group for genes encoding these CAAs. Those with a value above the average have been highlighted in red.

Next, we compared the functional classifications of the gene coding sequences from the persistent genomes of the nine reference C1 clinical strains against the nine C1 strains analyzed from Site A. We found that both groups only differ in ∼2% of the total gene content of their persistent genome. Most of these differences were associated with the presence of genes belonging to the “Sialic Acid Metabolism” classification in the clinical C1 strains (Fig. 3B). This group of genes code for a complete tripartite ATP-independent periplasmic transport system (TRAP) involved in the transport of sialic acid for the enzymes responsible for its catabolism (N-acetylneuraminate lyase, N-acetylmannosamine kinase, and N-acetylmannosamine-6-phosphate 2-epimerase) as well as a sialic acid mutarotase (YjhT family) and sialic acid utilization regulator, RpiR family (53). The ability to scavenge, decorate their surface, and utilize sialic acid as a carbon source is an important virulence factor for pathogenic and opportunistic bacteria, including V. vulnificus (54–57). Using the C1 clinical reference genome CMCP6, we found that the complete cluster was located in a genomic island on chromosome II (Fig. 3B). The same gene cluster can be found in other Vibrio species (ca. 70% BLASTN identity) such as V. cholerae O1, Vibrio mimicus, or Vibrio anguillarum; however, unlike V. vulnificus, in these species, the cluster was flanked by insertion sequence elements.

Given the frequent horizontal gene transfer in V. vulnificus populations, it is unlikely that presence/absence of genes or gene clusters is sufficient to explain the emergence of virulence traits that lead to clinical outcomes in this pathogen. Our previous investigations with V. cholerae suggest that allelic variations of core genes can be major drivers of virulence emergence (29). Thus, we evaluated the patterns of microdiversity of the persistent genome by estimating the ratio of nonsynonymous (dN) to synonymous (dS) substitution rates in pairwise genome comparison. We found six genes within the C1 clinical strains that showed a strong positive selection compared to the C1 IRL strains, which on average exhibited a strong purifying selection (Fig. 3C and SI Appendix, Table S4). In addition, average dN/dS values for these genes within C2 groups, both in the environmental references and the ones isolated from the IRL, also exhibited very low dN/dS values (SI Appendix, Table S4). The genes encoding these CAAs differ between clinical strains and are involved in virulence-associated processes and host-related nutrient metabolism (SI Appendix, Table S4). For instance, one of these genes encodes the outer membrane porin regulator OmpR, which regulates virulence in V. cholerae via aphB (58, 59). Another, encoding the subunit EntD, forms part of the enterobactin-synthetase enzyme complex, an iron acquisition system essential for virulence in Escherichia coli (60), and was proposed to play a role in the late stages of enterobactin biosynthesis in V. cholerae (61). The endonuclease vvn, identified as a periplasmic nuclease in V. vulnificus, prevents uptake of foreign DNA (62), thus hindering introduction of plasmids by transformation. Riboflavin synthase, ribE, catalyzes the final step in the biosynthesis of riboflavin or vitamin B2. Riboflavin is involved in a number of metabolic pathways [e.g., iron bioavailability and acquisition (63)] in many pathogens, including V. cholerae. Pyridoxal phosphate, PdxA, the catalytically active form of vitamin B₆, is an important cofactor for many enzymatic pathways involving breakdown of amino acids (64) and the sulfur transfer complex TusBCD TusB component. On average, these genes had lower dN/dS values in the C1 IRL strains in comparison to clinical C1; however, given that clinical V. vulnificus are endemic to this area, it is possible that some individual C1 IRL strains encode CAAs. To determine this, we analyzed their presence by identifying individual allelic variants that deviate from the average values (Fig. 3D). Interestingly, even though none of the alleles from C1 IRL stains were identical to those found in the clinical strains, each of them encoded at least one gene with a dN/dS above the average. Those ranged from strain OH0003 encoding one (tusB gene) to IRLA0186 that encodes four of them (ompR, ribE, entD, and pdxA) (Fig. 3D). Overall, our results demonstrate that 1) clinical strains encode unique CAAs and 2) allelic variants of these genes circulate in natural populations.

Assessment of Pathogenic Potential of V. vulnificus Strains.

In order to evaluate the pathogenic potential of IRL environmental strains and their association with phylogeny and location, we phenotypically tested their 1) antibiotic resistance profile, 2) survival in the presence of human serum, and 3) ability to use sialic acid as a sole carbon source. For these assays, we included V. vulnificus CMCP6 (clinical C1) and V. vulnificus SS108-A3A (environmental nonpathogenic C2) as bona fide reference strains. Furthermore, we constructed three isogenic mutant strains in the background of V. vulnificus CMCP6, where we deleted the genes encoding 1) the CPS transport protein Wza (∆wza), which has been shown to play a role in serum survival and capsule production (65), 2) N-acetylneuraminate lyase (∆nanA), the first enzyme in the catabolic pathway of sialic acid (54), and 3) the sialic acid TRAP transporter large permease (∆siaM), which is associated with sialic acid uptake and is also involved in serum resistance (66).

Antibiotic resistance.

First, we examined the antibiotic resistance profile of the IRL strains to determine whether there were patterns associated with the differential distribution of the clusters, as both sites have vastly different exposure to manmade perturbances, including antibiotics (67, 68). We tested several antibiotics recommended by the Centers for Disease Control and Prevention for the treatment of Vibrio spp. (69). While V. vulnificus CMCP6 showed resistance or intermediate resistance to virtually all the antibiotics tested (Fig. 4A), ∆wza, ∆nanA, and ∆siaM showed increased sensitivity to several of them compared to the wild type (Fig. 4A). The capsule typically confers resistance to antibiotics (70, 71); however, the mechanisms by which sialic acid catabolism and uptake are involved in antibiotic resistance remains to be elucidated. Most IRL strains are resistant to polymyxin B, gentamycin, sulfadiazine, and imipenem, a β-lactam antibiotic. In contrast, virtually no IRL strain was resistant to chloramphenicol or oxytetracycline (Fig. 4A). Seven strains from Site B exhibited intermediate resistance to nalidixic acid and/or trimethoprim, while only two of the isolates from Site A were resistant to these compounds. Strikingly, a C1 strain isolated from Site B (IRLE0004) showed varied resistance levels to all antibiotics tested with the exception of oxytetracycline. Interestingly, two C1 strains from Site A (IRLA0161 and IRLA0152) that belonged to the same clonal frame, that is, ANI > 99%, showed different antibiotic resistance patterns (Fig. 4A). Unlike IRLA0152, IRLA0161 is resistant to oxytetracycline, nalidixic acid, and trimethoprim. Genome analysis showed the presence of a 172 Kb plasmid in this strain, in which we identified a coding gene for a trimethoprim-resistant dihydrofolate reductase, DfrA family. Although the genes directly responsible for the other two resistances were not identified, we found several genes related to efflux pumps encoded in the same plasmid. It appears, from our analysis, that selective pressures at Site B, the site with most anthropogenic exposure, favor the emergence of antibiotic resistance, particularly to the folate inhibitor, trimethoprim, and the quinolone, nalidixic acid (Fig. 4A). Furthermore, the presence of resistant plasmids and their ease of transmission between the two clusters (28) increases the likelihood that strains from C1 acquire these genes through horizontal gene transfer.

Fig. 4.

Assessment of pathogenic potential of V. vulnificus IRL isolates. (A) Patterns of antibiotic resistance of 27 V. vulnificus isolates to commonly used 12 antibiotics. Red, resistant; pink, intermediate resistance; white, sensitive. (B) Serum resistance of V. vulnificus exposed to normal pooled human serum for 2 h and assessed for survival in terms of CFU/milliliters. Resistant strains, similar CFU/milliliters as input; sensitive strains, lower of CFU/milliliters than input; resistant and growth on serum, higher CFU/milliliters than input. (C) Ability to catabolize sialic acid assessed by growth of V. vulnificus isolates in M9 minimal media supplemented with NANA as the sole carbon source at salinities representing the two sampling locations. Growth was measured as a function of increased OD (OD595) of the cultures over time.

Serum resistance.

Some studies have previously reported the ability of clinical V. vulnificus strains to resist the bactericidal effect of serum, while most environmental strains tested being susceptible to it (50, 51). Given that serum resistance is an essential virulence trait for V. vulnificus pathogenesis, we analyzed the susceptibility of the IRL isolates to this primary host defense. As expected, the wild-type clinical C1 strain was resistant to serum, whereas the nonpathogenic C2 strain was sensitive to its bactericidal effect (3 to 4 log decreases in colony-forming units [CFUs]) (Fig. 4B). Only 3 out of 12 strains from Site A were sensitive to serum, whereas in Site B, we found the opposite pattern, with most of the strains (8 out of 14) being sensitive (Fig. 4B). These differences were strongly associated with cluster distribution and provided us with an opening to examine the possible genomic determinants that lead to serum resistance in V. vulnificus. We first compared the gene content between serum-resistant C1 strains (OH0023 and IRLA0152) against sensitive ones (OH0012 and IRLA0153). Among those unique genes in the resistant strains, we found several related to type I restriction-modification systems, capsule synthesis, and those involved in sialic acid metabolism. Subsequently, we analyzed the presence of the sialic acid cluster in the genomes of all IRL isolates in our study. We found that 12 out of 15 strains that were resistant to serum (8 Site A; 4 Site B) encoded the cluster, whereas only 1 out of 11 sensitive strains did (Fig. 4B). Given this clear association, we tested the serum resistance of ∆wza and the two sialic acid mutants, ∆nanA and ∆siaM. As expected, ∆wza was sensitive to serum. Interestingly, while ∆siaM exhibited a 2-log decrease in CFU compared to the wild type, ∆nanA was not affected by the bactericidal effect of serum, and the mechanism behind the difference in survival between these two mutants remains to be addressed.

Sialic acid catabolism.

Sialic acid, besides playing an important role in host-pathogen interactions (54, 56), is critical for the interaction of several pathogenic Vibrios with some of their environmental reservoirs, such as Cyanobacteria, potentially linking different lifestyles of bacterial pathogens (72, 73). Both our pangenome and phenotypic analyses suggest that catabolism of this aminosugar appears to be an essential factor associated with clinical outcomes. In order to initially test our findings, we examined the ability of the IRL strains to utilize N-acetylneuraminic acid (NANA) as a sole carbon source. We tested their growth in M9 minimal media supplemented with NANA at two salinities reflective of the two sampling sites (1% and 3% NaCl; SI Appendix, Table S5). Neither the ∆nanA and ∆siaM mutants nor the IRL isolates that did not encode the sialic acid cluster were able to grow in these media. All strains from Site A that possessed the sialic acid cluster (8 of the 12) exhibited similar growth patterns to the clinical reference CMCP6 at both salinities. At Site B, only 6 of the 14 isolates were able to grow, all containing the sialic acid cluster (Fig. 4C).

Taken together, our genomic and phenotypic analyses of the IRL strains and their comparisons against clinical strains showed differential potential for pathogen emergence in these natural populations. For instance, strain IRLA0186 exhibits several traits that indicate its strong capability for emergence as a clinical strain, such as its ability to resist serum, catabolize sialic acid, and resist most of the antibiotics tested as well as encode variations in four of the six CAAs. On the other hand, OH0008, isolated from the same site IRLA0186 (ANI 98.3%), is sensitive to both serum and most of the antibiotics we tested but cannot grow on sialic acid and only encodes one allelic variation similar to CAAs, suggesting limited likelihood of pathogenic outcomes.

Environmental Factors Associated with Cluster Divergence.

Our analyses revealed distinct genomic and phenotypic signatures linked with the emergence of clinical-associated traits in environmental V. vulnificus. In order to uncover ecological drivers leading to the selection of these traits and the skewed distribution of V. vulnificus clusters, we investigated the abiotic and biotic parameters associated with each site. First, we measured several abiotic factors from the aquatic samples collected during strain isolation, such as temperature, dissolved oxygen, pH, dissolved organic matter, salinity, and phosphorous, among others (SI Appendix, Table S5). Next, water samples were sequentially filtered through 20-, 5-, and 0.22-μm pore size filters. DNA was obtained from the 0.22-μm filter that contain the free-living microbial fraction to analyze the microbial community structure (biotic factors) associated with each sampling site (Fig. 5A). We used principal coordinate analysis to examine possible correlations between cluster distribution and both abiotic (physicochemical parameters) and biotic factors (taxonomic classification from 16S ribosomal RNA [rRNA] gene metagenomic fragments) (Fig. 5B). The community structure from Site A is very similar to that found in marine environments where the main taxa were Cyanobacteria, SAR11, Bacteriodetes, Oceanospirillales, or "Candidatus Actinomarina" (Fig. 5A). In fact, salinity at this location was 29 ppm, which is slightly lower than seawater (35 ppm) (SI Appendix, Table S5). The percentage of 16S rRNA reads associated with the genus Vibrio accounted for a total of 1.8% of the total population (Fig. 5A). However, they are undetectable at Site B, where the salinity was much lower than in Site A (5 to 18 ppm), which are signatures of a brackish environment. We also found in Site B higher concentrations of phosphates, nitrates, and dissolved organic matter compared to Site A, which is likely due to runoffs from nearby Lake Okeechobee, which experiences an influx of fertilizers from nearby agricultural farms (SI Appendix, Table S5). These variations in environmental factors likely change the microbial community by predominantly low-salinity–adapted microbes, such as the genera Polynucleobacter and Limnohabitans within the family Burkholderiales or the Microtrichal and Frankial families within the order Actinobacteria (Fig. 5A). Microbial diversity, measured as Shannon index, indicated that diversity was higher in Site A than in Site B (Fig. 5C). These data suggest that C1 members prefer a more oligotrophic, marine-like environment with higher salinity and greater microbial diversity dominated by cyanobacteria, whereas C2 members appear to be better adapted to nutrient-rich brackish environments marked by the presence of several families of Actinobacteria (Fig. 5). Overall, our metagenomic and physicochemical analyses of the sampling sites indicate that the marked cluster distribution and genetic traits are strongly associated with distinct biotic and abiotic factors (e.g., salinity, nutrients, or biodiversity), revealing how ecosystems may generate selective pressures that facilitate the emergence of specific strains with pathogenic potential in a population.

Fig. 5.

Environmental factors associated with cluster divergence. (A) Taxonomic classification based on 16S rRNA gene fragments (raw reads) of the different metagenomes obtained from seawater 0.22-µm filter. Only those groups with abundance values larger than 1% in any of the metagenomes are shown. The size of the diameter of the circles indicates the percentage of the total reads for each taxon. (B) Principal coordinate analysis between physicochemical parameters and abundance of the different taxons based on 16S rRNA gene metagenomic fragments. (C) Box plots illustrating microbial community diversity measure using Shannon index.

Conclusions

Elucidating the factors associated with the emergence and spread of human pathogens is critical in order to develop tools to predict potential sources of disease outbreaks and to establish effective surveillance strategies. Pathogen emergence is a complex and multifactorial phenomenon that requires analytic methods and tools that can consider large amounts of and highly diverse data. Therefore, it is essential to develop tractable model systems that allow us to dissect the ecological, genetic, and evolutionary drivers that foster the selection of virulence traits and pathogenic clones within an environmental population. In this study, we used V. vulnificus, an emerging coastal pathogen that causes fatal sepsis, as a model system to investigate the genetic and ecological forces leading to pathogen emergence. The high genome plasticity of V. vulnificus paired with the unexpected outcomes associated with manmade environmental changes make this bacterium a major threat to human health for which no effective vaccines or therapeutic strategies are available (16, 28, 74). Here, we implemented a holistic approach that combines fields such as genomics, metagenomics, ecology, molecular biology, and bacterial pathogenesis to address this problem. Overall, we found a strong correlation between ecological factors (e.g., site of isolation, physicochemical parameters, and community structure) and pathogenic potential, as exemplified by skewed cluster distribution, and genetic and phenotypic traits associated with clinical outcomes.

The layers of selection imposed by the different abiotic and biotic factors likely act as a major selective pressure driving the development of pathogenic features in V. vulnificus populations. From our analyses, there is a clear association between cluster distribution and abiotic (e.g., salinity or dissolved nutrients) and biotic (community structure, oysters, or cyanobacteria) factors. Give their relevance, investigating the association of V. vulnificus and the specific role of these and other abiotic factors and biotic reservoirs, such as protists (e.g., amoeba) and other metazoans (e.g., fish and crustaceans), in cluster selection will shed substantial light on the process of emergence of pathogenic traits in V. vulnificus.

Furthermore, each sampling site is exposed to different anthropogenic influences. For instance, Site A is located in a protected area with limited access in Cape Canaveral. On the other hand, Site B experiences nutrient overenrichment due to urbanization and agricultural expansion as well as other manmade contamination, such as fecal waste discharges. Given the drastic differences in the anthropogenic exposure between the two locations, it is likely that they play a role in cluster selection and distribution. It would be of interest for future studies to address the role of these anthropogenic disturbances in the emergence of pathogenic Vibrios.

Overall, our results indicate how ecosystems may generate selective pressures that facilitate the emergence and selection of specific strains within a population with pathogenic potential. Our study closely aligns with the One Health initiative (75) by 1) focusing on the connection between a disease agent and the environmental factors that lead to its emergence and 2) creating a combined approach to understand disease emergence from an integrated and tractable perspective. Our approach can serve to develop ecological and genetic markers for surveillance systems to predict sources of outbreaks or identify emergent human pathogens. Overall, we offer a general paradigm and methodology for studying and understanding disease emergence that can be naturally extended to other human pathogens.

Materials and Methods

Strains and Culture Conditions.

An extended version of the Materials and Methods can be found as part of SI Appendix. Strains of V. vulnificus (SI Appendix, Tables S1 and S3) were routinely cultured on Luria-Bertani (LB) agar plates supplemented with 2% NaCl (wt/vol; LB-2%), inoculated in LB-2% broth, and cultured for 16 h aerobically at 37 °C, unless otherwise specified. V. vulnificus strains CMCP6 and SS108-A3A were used as C1 clinical and C2 environmental controls, respectively, for all phenotypic assays. E. coli β2155, a diaminopimelic acid (DAP) auxotroph, was used for mutant construction and was cultured in LB supplemented with 0.3 mM DAP (LB-DAP).

Sampling Sites.

Samples were collected at two environmentally distinctive locations along the IRL (Eastern Florida, United States) in three sampling events. The first location, Fellers House Field Station (N28°54’25.315”; W80°49’15.017”; Northern IRL; Site A), is located within the federally protected Canaveral National Seashore. The second sampling site, Shepard Park, is located in Port St. Lucie (N27°11’48.864”; W80°15’33.172”: Southern IRL; Site B), which, due to urbanization and agricultural expansion, experiences nutrient overenrichment, leading to excessive macroalgal bloom (Fig. 1B) (76, 77).

Isolation of V. vulnificus from Environmental Sources.

Water samples.

V. vulnificus was isolated from water samples using a modified protocol from Huq et al. (42). The 500 mL of each sample was filtered successively through 20, 5, and 0.2-µm membrane filters (Sterlitech) to separate planktonic and free-living fractions. The filters were suspended in phosphate-buffered saline (PBS), pH 7.5, vortexed vigorously, and cultured in alkaline peptone water (APW) overnight at 37 °C.

Sediment samples.

V. vulnificus was isolated from sediment using a modified protocol from Schuster et al. (78). Samples were collected using a universal corer. Samples were suspended in PBS (1:1) and homogenized and enriched in APW.

Oyster samples.

Isolation of V. vulnificus from oysters was carried out by a protocol adopted and modified from the US Food and Drug Administration’s Bacteriological Analytical Manual for Vibrio (79). Briefly, oysters collected from Feller’s house were washed to remove sediment or dirt. Each oyster was individually shucked, homogenized in 30 mL PBS using the SCILOGEX D160 Homogenizer, and cultured in APW.

Cyanobacterial samples.

Cyanobacteria collected from Shepard Park were pelleted, supernatant removed, and cultured in APW. All samples were collected in triplicate. Enriched cultures in APW from water, sediment, oyster, and cyanobacteria samples were serially diluted and plated on CaV (CHROMagar), a Vibrio spp. selective agar. Turquoise blue colonies were further screened on TCBS (Sigma) agar plates on which V. vulnificus appear as green colonies. Colonies that appeared turquoise blue on CaV and green on TCBS were considered potential V. vulnificus isolates.

Verification of V. vulnificus Isolates.

Potential V. vulnificus IRL isolates were verified by PCR using primers for the thiF marker gene (SI Appendix, Table S2). PCR products of isolates positive for thiF were sequenced (GENEWIZ, AT, GA) to determine cluster affiliation. A number of diverse V. vulnificus isolates from both clusters and from each of the environmental reservoirs were selected for whole genome sequencing.

Genome sequencing.

Libraries of whole genomes were prepared using the Nextera DNA Flex Library Prep Kit from Illumina following the manufacturer’s instructions and sequenced using the Illumina iSeq100 Sequencing System. Sequenced genomes were analyzed using Illumina BaseSpace Sequence Hub. Reads obtained for each Biosample were assembled into contigs and scaffolds using the SPAdes Genome Assembler version 3.9.0 and Velvet de novo Assembly version 1.0.0.

Assembly, Gene Prediction, and Annotation.

Reads were trimmed using Trimmomatic v0.36 (80) and assembled de novo with SPAdes version 3.11.1 (81). ORFs from the assembled contigs were predicted using Prodigal version 2.6 (82). Transfer RNA and rRNA genes were predicted using tRNAscan‐SE version 1.4 (83), ssu‐align version 0.1.1 (84), and meta‐rna (85). Using DIAMOND (86), predicted proteins were compared against the National Center for Biotechnology Information nonredundant database, and they were compared against COG (87) and TIGFRAM (88) using HMMscan version 3.1b2 (89) for taxonomic and functional annotation.

Phylogenomic Reconstructions.

The assembled contigs were assigned a chromosome by comparison to this group of reference genomes using Blastn (90). Genes were predicted using Prodigal (82) and clustered using the software MMsEqs (91). The resulting protein clusters that were present in all analyzed genomes were divided into two groups according to the chromosome they are encoded in, resulting in a group of 257 and 62 proteins for chromosomes 1 and 2, respectively. Protein clusters were then aligned with QuickProbs2 (92), trimmed with BGME (93), and concatenated. Finally, a phylogenetic tree was constructed using iqtree (94) with automatic model selection and 1,000 bootstrap replicates.

Genomic Pairwise Comparisons.

Reciprocal BLASTN and TBLASTXs searches between genomes were carried out, leading to the identification of regions of similarity, insertions, and rearrangements. ANI and coverage between pairs of genomes were calculated using PYANI software (95).

Pangenome and Recombination Analysis.

To analyze the gene family prevalence across all genomes, we used the software PPanGGOLiN to divide the gene families into persistent/shell/cloud partitions (52). The genes constituting each partition were then annotated against the SEED subsystem database (96) using DIAMOND (86), keeping all matches with E < 0.001 and alignment length > 0.5 for both subject and query. Finally, dN/dS values for the different protein partitions were obtained using the Orthologr package in R (97). The PopCOGenT pipeline (47) was used to define the recombinant populations based on gene flow between the different sequenced genomes.

Mutant Construction.

In-frame deletions of genes of interest, wza, nanA, and siaM, were constructed via homologous recombination (98) (primer list can be found in SI Appendix, Table S6). Briefly, two approximately 500 bp PCR fragments flanking the genes of interest were cloned into the sacB-counterselectable plasmid, pDS132, and electroporated into donor E. coli strain, β2155. The donor strains harboring the knockout vectors were conjugated with wild-type V. vulnificus CMCP6 on LB-DAP, and transconjugants were selected on LB-2% plates supplemented with chloramphenicol (Cm) (25 μg/mL). Cm^R exconjugant colonies were cultured in LB-2% without antibiotics, and serial dilutions were plated on LB-2% plates containing 10% (wt/vol) sucrose. Potential double-crossover deletion mutants were screened by PCR, and putative deletions were confirmed by DNA sequencing.

Antibiotic Resistance.

V. vulnificus isolates were examined for susceptibilities to the antibiotics highlighted in Fig. 4 at the highest concentrations in the breakpoint concentration range recommended by Clinical and Laboratory Standards Institute in M45-A (99–102) (SI Appendix, Supplementary Text). Briefly, individual colonies of each strain were transferred sequentially using sterile toothpicks onto LB-2% plates supplemented with respective antibiotics and incubated at 37 °C overnight. The diameter of the growth was measured, and resistance was defined as growth of at least 2 mm in the respective antibiotics. Strains exhibiting no growth were taken as sensitive, and any intermediate growth diameter was considered as intermediate resistance. Experiments were performed in three independent biological replicates.

Serum Resistance.

In vitro serum survival assay was adapted from Bogard and Oliver (103). Briefly, overnight cells were subcultured in LB-2% to obtain log-phase cells at an optical density (OD) of 0.15 to 0.25. Cells were then washed in PBS and inoculated at a 100-fold dilution into normal pooled human serum (Fisher Bioreagents) and incubated at 37 °C for 2 h. Resistance to serum was assessed by comparing the CFU/milliliters before and after exposure to serum. Experiments were performed in three independent biological replicates.

Sialic Acid Catabolism.

The ability to catabolize sialic acid by V. vulnificus isolates was assessed by growth in NANA, the predominant form of sialic acid in human cells, as the sole carbon source (57). Briefly, overnight cultures of each strain were washed and resuspended in M9 minimal media, and a 100-fold dilution of cells was made in M9 minimal medium supplemented with NANA (2 mg/mL) (Chem‐Impex International). The 200-µL aliquots of each sample were added per well to a 96-well microtiter plate and incubated at 37 °C with shaking. OD at 595 nm (OD595) was measured every hour for 24 h using a Tecan Sunrise microplate reader (Tecan US), and the results were evaluated using Magellan plate reader software. Growth assays were performed in triplicate across three independent biological replicates.

Measurement of Physicochemical Parameters.

Measurements of water temperature (°C), salinity (grams/liters), dissolved oxygen (%), pH, pressure (millimeters of mercury), dissolved organic matter, chlorophyll-a (micrograms/liters), and total algae (micrograms/liters) were made during the isolations. The measurements were recorded using a YSI EX02 sonde deployed at the sites at the time of sampling that was calibrated within 24 h prior to each sampling event. Water samples, collected in triplicates, were also examined for the concentration of phosphates (o-Phosphate–P, method 365.1), and nitrates (Nitrate–N, method 353.2; Ammonia–N, method 350.1) according to the standard protocols described by the US Environmental Protection Agency (104, 105). Briefly, collected water samples filtered through a 0.2-µm membrane filter were acidified to a pH < 2 with double distilled H₂SO₄ and stored at 4 °C until analysis. Samples were analyzed for nitrate + nitrite (NO3−), ammonium (NH4+), and ortho-phosphate (PO43−) on a SEAL AQ2 Automated Discrete Analyzer (Seal Analytical).

Metagenomic Analysis.

DNA extraction was performed from the 0.22-μm filter. Attached cells were disrupted using cetyltrimethylammonium bromide (CTAB) lysis buffer and glass beads followed by lysozyme treatment. The nucleic acids were then extracted using the phenol-chloroform extraction method (106). Metagenomes were sequenced using Illumina Hiseq-4000 (150 bp, paired-end read). To analyze the phylogenetic classification of the samples, candidate 16S rRNA gene sequences in the raw metagenomes were identified using USEARCH6 (107) (E-value < 10 to 5) against a database containing nonredundant 16S rRNA sequences downloaded from the Ribosomal Database Project (108). These sequences were then aligned to archaeal and bacterial 16S rRNA Hidden Markov Models (HMM) (109) using ssu-align to identify true sequences (84). Only hits to 16S rRNA sequences were then classified into a high-level taxon if the sequence identity was ≥80% and the alignment length ≥90 bp. Sequences failing these thresholds were discarded. Information on data availability can be found in SI Appendix, Supplementary Text.

Data Availability

Metagenomes and genomes data have been deposited in BioProjects (PRJNA694933 and PRJNA691530).