Phylogenetic and genomic analysis reveals high genomic openness and genetic diversity of Clostridium perfringens

Abstract

Clostridium perfringens is associated with a variety of diseases in both humans and animals. Recent advances in genomic sequencing make it timely to re-visit this important pathogen. Although the genome sequence of C. perfringens was first determined in 2002, large-scale comparative genomics with isolates of different origins is still lacking. In this study, we used whole-genome sequencing of 45 C . perfringens isolates with isolation time spanning an 80‐year period and performed comparative analysis of 173 genomes from worldwide strains. We also conducted phylogenetic lineage analysis and introduced an openness index (OI) to evaluate the openness of bacterial genomes. We classified all these genomes into five lineages and hypothesized that the origin of C. perfringens dates back to ~80 000 years ago. We showed that the pangenome of the 173 C . perfringens strains contained a total of 26 954 genes, while the core genome comprised 1020 genes, accounting for about a third of the genome of each isolate. We demonstrated that C. perfringens had the highest OI compared with 51 other bacterial species. Intact prophage sequences were found in nearly 70.0 % of C. perfringens genomes, while CRISPR sequences were found only in ~40.0 %. Plasmids were prevalent in C. perfringens isolates, and half of the virulence genes and antibiotic resistance genes (ARGs) identified in all the isolates could be found in plasmids. ARG-sharing network analysis showed that C. perfringens shared its 11 ARGs with 55 different bacterial species, and a high frequency of ARG transfer may have occurred between C. perfringens and species in the genera Streptococcus and Staphylococcus . Correlation analysis showed that the ARG number in C. perfringens strains increased with time, while the virulence gene number was relative stable. Our results, taken together with previous studies, revealed the high genome openness and genetic diversity of C. perfringens and provide a comprehensive view of the phylogeny, genomic features, virulence gene and ARG profiles of worldwide strains.

Data Summary

This study generated sequencing data for 45 C . perfringens isolates and used published data for 128 isolates. All sequence data have been upload to the National Center for Biotechnology Information under accession number PRJNA597712. Isolate metadata and accession numbers can be found in Table S1 (available in the online version of this article).

Impact Statement.

Clostridium perfringens is associated with a variety of human and animal enteric diseases, but large-scale genomic understating of C. perfringens strains of different origin is still lacking. In this study, we sequenced 45 C . perfringens isolates with isolation times spanning an 80-year period and performed comparative analysis of 173 genomes from different sources. We showed that C. perfringens is an ancient bacterium, the origin of which dates back ~80 000 years ago. We introduced an openness index (OI) to evaluate the openness of bacterial genomes and demonstrated that C. perfringens had the highest OI compared with other randomly selected bacterial species. We showed that the various gene contents in phages and plasmids contribute to the genetic diversity of this bacterium. We also revealed that the C. perfringens typing toxin genes are conserved among strains, while the number of antibiotic resistance genes in isolates increased over time, which may suggest that the bacterium has been subjected to antibiotic selection pressures.

Introduction

Clostridium perfringens is a Gram-positive pathogen which inhabits various natural environments as well as the gastrointestinal tracts of humans and animals [1]. This spore-forming bacterium is known to cause diverse human and animal systemic and intestinal diseases. In humans, C. perfringens causes diseases including gas gangrene [2], food poisoning [3], gastrointestinal disorders [4], and liver and kidney damage [5, 6]. As a commensal, C. perfringens has potentially co-evolved with humans for thousands of years, as evidenced by the identification of this bacterium in the gut of a >5000-year-old mummified Neolithic ‘Tyrolean Iceman’ [7]. In animals, especially in pigs and poultry, C. perfringens is the causative agent of gangrenous dermatitis [8], enterotoxaemia [9] and necrotic enteritis (NE) [10].

The pathogenicity of C. perfringens depends largely on the various types of extracellular toxins and enzymes they produce [11]. The toxin-based C. perfringens typing method indicated that different toxinotype strains had different host preference and were responsible for specific diseases. For example, type B strains are predominantly isolated from sheep suffering dysentery; type F (previously CPE-positive type A) strains are associated with human food poisoning; and type G (previously NetB-positive type A) strains comprise isolates causing necrotic enteritis in chickens [12]. Genes of these typing toxins have different genetic locations with the α-toxin encoding genes carried by chromosomes while the other typing toxin genes are carried by mobile genetic elements, commonly large conjugative plasmids [13].

The first genome of C. perfringens was sequenced in 2002, highlighting the genetic features of this bacterium with pronounced low G+C content, diverse toxin and degradative enzyme-encoding genes, the presence of anaerobic fermentation pathways and the lack of enzymes for amino acid biosynthesis [14]. Since then, several comparative genomics analyses for C. perfringens of either human or animal origins have been performed [15–19]. By analysing 56 strains, Kiu et al. [20] showed that C. perfringens has a large pangenome with only 12.6 % core genes present in the pangenome. The study also indicated that C. perfringens is relatively lacking in the CRISPR system compared with other bacteria [21], which may be a reason for its high genomic variation caused by horizontal gene transfer. Recently, a comparative genomic study that included 88 chicken-associated C. perfringens genomes revealed that virulence genes netB, pfoA, cpb2, tpeL and cna variants are closely associated with chicken NE-linked strains [22]. These studies provide useful information on the diversity and evolution of C. perfringens . However, large-scale genomic analysis concerning isolates from a range of hosts and conditions is still needed to understand the genomic features, toxinotype and antimicrobial resistance gene profile of C. perfringens [12].

In this study, we performed whole-genome sequencing of 45 C . perfringens isolates preserved in our culture collection centre. These isolates, collected over a period of ~80 years, were originally isolated from at least 14 different animal hosts with the majority associated with chicken NE. We then downloaded 128 additional publicly available genomes in GenBank (as of December 2018), including isolates of human gas gangrene and food poisoning, and performed comparative analysis of the 173 C . perfringens genomes.

Methods

Isolate collection and metadata

By considering isolation time and host range, 45 C . perfringens isolates conserved in our culture collection centre were selected for genome sequencing (Table S1). All these isolates were of animal origin wand of long isolation age; the oldest strain, CP-40, was originally isolated from the former Soviet Union in 1937, while the most recent, CP-45, was isolated from Hebei province, China, in 2018. Before DNA extraction, the freeze-dried bacteria were inoculated in fluid thioglycollate medium broth (FTG; Beijing Land Bridge Technology) and then plated on tryptose-sulfite-cycloserine agar (TSC; Beijing Land Bridge Technology). A single colony was picked and incubated in brain heart infusion broth (BHI; Beijing Land Bridge Technology) at 37 °C for 24 h under anaerobic conditions.

For comparative genomics, all available C. perfringens genomes (as of December 2018) deposited in the National Center for Biotechnology Information (NCBI) (n=128) were retrieved. The year of isolation and the geographical origin of each of these isolates were taken from the published research papers or from the information provided by the submitters (Table S1).

DNA sequencing, assembly and annotation

For whole genome sequencing, the genomic DNA of 45 C . perfringens isolates was extracted by using a QIAamp DNA Mini Kit (Qiagen). We performed paired-end genome sequencing on the Illumina HiSeq 2500 with an genome coverage of 300× [coverage was calculated from the length of the genome (G), the number of reads (N) and the read length (L) as N×L/G]. Illumina HiSeq-generated sequencing reads were clipped for adaptors with Trimmomatic [23]. The sequence reads were assembled using Unicycler (v0.4.7) with default parameters [24], which employs SPAdes (v3.13.0) for genome assembly [25]. The genomes were annotated using Prokka (v1.13.3) [26].

Phylogenetic analysis and lineage assignment

A maximum-likelihood phylogenetic tree was generated based on the core genomes, using RAxML-NG (v0.9.0) with 500 bootstrap replicates [27]. Bayesian phylogenetic analysis was performed on the core genome using beast 2 (v2.5) [28] to infer the timescale of C. perfringens evolution. All unknown dates were set to year 0 [29]. Evolutionary rates and tree topologies were analysed using the general time-reversible (GTR) substitution models with gamma substitution models with gamma distributed among-site rate variation with four rate categories. We tested a strict molecular clock, which assumes the same evolutionary rate for all branches in the tree. A Yule tree prior was selected as the tree prior, which assumes a constant lineage birth rate for each branch in the tree. Convergence was assessed using Tracer (v1.7.1) [30], ensuring all relevant parameters reached an effective sample size (ESS) of over 200. A run of 1×10⁸ generations was sampled every 10 000 generations to ensure independent convergence of the chains. Lineage assignment for the 173 strains was made using RhierBAPS based on the core genome [31] with default parameters. For verification of lineage assignment, we applied three additional methods to confirm the current definition: principal-component analysis (PCA) was carried out with Jalview [32]; pairwise SNP distances (core genome SNPs) were calculated using the ape package in R (v3.5.3) [33]; pairwise fixation index (F _ST values, estimation of population separation) were calculated using Arlequin (v3.5) [34].

Openness index definition and comparison

We introduced an openness index (OI) to evaluate the openness of a bacterial strain, which was calculated according to equation 1:

$${\displaystyle \mathrm{O}\mathrm{p}\mathrm{e}\mathrm{n}\mathrm{n}\mathrm{e}\mathrm{s}\mathrm{s}\mathrm{i}\mathrm{n}\mathrm{d}\mathrm{e}\mathrm{x}(\mathrm{O}\mathrm{I})=\frac{{}^{\mathit{n}}\mathit{s}\mathit{t}\mathit{r}\mathit{a}\mathit{i}{\mathit{n}}^{\mathit{-}\mathit{n}}\mathit{c}\mathit{o}\mathit{r}\mathit{e}}{{}^{\mathit{n}}\mathit{s}\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}}$$ (1)

Here, n _strain is the total number of genes in a specific strain; n _core is the number of core genes for the corresponding bacterial species; and n _strain − n _core represents the number of accessary genes in that strain. Therefore, OI reflects the ratio of accessory genes in a specific strain. From the equation, we can see that the n _core value determines the value of OI for a single strain, but the n _core in a bacterial species largely depends on the strain number used for building the core gene catalogue. So, for comparison of the OI among different bacterial species, it is essential to make sure the core gene analysis is based on using the same number of genomes in different species. To achieve this, we first selected all the species having a sufficient genome number (>200) in NCBI. Then, to ensure a high-quality genome was included in the following study, we randomly (using the ‘shuf’ command in Linux) selected 100 genomes with N50 over 50 kb and an average nucleotide identity (ANI) greater than 0.95 [35] calculated by pyani (v0.2.3) [36] among genomes for each species. Finally, 51 bacterial species (5100 genomes in total) met our filter demands and were compared with C. perfringens (100 genomes selected from 173 isolates using the same genome selection criteria as for other species). The core genome and pangenome prediction was carried out using the Roary software (v3.11.2) [37], and the core genome was defined by including both the hard-core and soft-core genes, i.e. genes present in more than 95.0 % of strains [38]. For each species, the rRNA operon copy number was retrieved from rrnDB (v5.5) [39]. The core gene and accessory gene functional annotations of C. perfringens were carried out using eggNOG (v4.5) (evolutional genealogy of genes: Non-supervised Orthologous Groups) platform [40].

In silico analysis of virulence factors, ARGs, prophages, CRISPR and plasmids

Virulence genes were identified by blast (The Basic Local Alignment Search Tool, v2.9.0+) [41] against the latest version of the Virulence Factor Database (VFDB) [42] with a cutoff value of ≥80 % identity and ≥50 % coverage. We added the nucleotide sequences of 11 genes to the core dataset of VFDB, as these genes are not included in the core dataset of VFDB but are important for C. perfringens (Table S2). To test the conservation of the toxins used for strain toxinotyping [12], we retrieved the plc, cpb, etx, iap, ibp, cpe and netB gene sequences from the 173 genomes, and expanded the dataset by adding the nucleotide sequences of these seven toxins from the UniProt database. Then, we curated the toxin nucleotide sequences dataset by CD-HIT (v4.6) [43] at 100.0 % identity to remove redundant sequences. A total of 158, 8, 21, 5, 6, 8 and 6 nucleotide sequences for these seven genes, respectively, were compared. KaKs_Calculator (v2.0) [44] was applied to calculate non-synonymous (K _a) and synonymous (K _s) substitution rates of nucleotide sequences for all the seven toxins.

Antibiotic resistance gene (ARG) analysis was performed by using the Comprehensive Antibiotic Resistance Database (CARD, v4.2.2), which is a manually curated resource containing high-quality reference data on the molecular basis of antibiotic resistance [45]. We also added the nucleotide sequences of four genes (bcrA, bcrB, bcrD and bcrR) encoding resistance to bacitracin [46] to the database (Table S3). ARG-sharing network analysis was performed according to our previous procedure, and if an ARG was shared between two bacterial species with ≥99.0 % nucleotide identity, the ARG was considered to have the potential to transfer between species [47]. For this analysis, the 18 ARGs (34 sequences) identified in C. perfringens were analysed by blast against 11 395 complete genomes downloaded from NCBI on 26 September 2018.

Prophage predictions were carried out using the PHAge Search Tool Enhanced Release (PHASTER) web server [48] to explore the ‘intact’, ‘questionable’ and ‘incomplete’ prophages. CRISPR (clustered regularly interspaced short Palindromic Repeats) predictions were performed with CASC (CASC Ain’t Simply CRT, v2.6) [49]. To explore the source of the CRISPR spacer sequence, we performed a BLASTn analysis against the NCBI Virus database (ncbi.nlm.nih.gov/labs/virus/vssi/#/) with a cutoff value of ≥90 % identity and ≥90 % coverage.

Complete plasmid sequences were retrieved from the complete genomes, and partial plasmid sequences from the draft genomes were predicted using PlasFlow (v1.1) [50] and ABRicate (v1.0.1, https://github.com/tseemann/abricate). For plasmid identification, the predicted plasmid sequences were analysed by blast against a custom plasmid database which was constructed based on a previous study [51], using best-hit approach with query coverage threshold ≥70 % and nucleotide identity ≥90 %. Virulence and antibiotic resistance genes carried by the plasmid sequences were analysed using the same procedure and cutoff for the whole genome analysis.

Statistical analysis

Spearman’s rank correlation coefficient was applied to test the correlation between two variables. A Wilcoxon rank-sum test was used to identify the difference. Fisher’s exact test was used to determine if there are non-random associations between two categorical variables. All statistical analyses were performed using R software.

Results

Strain information and sequence statistics

In the present study, we sequenced 45 C . perfringens and downloaded additional 128 genomes from NCBI. These strains, covering all the seven toxinotypes, and were from at least 14 species of animals as well as from humans and the environment. In addition, the strains we sequenced were isolated earlier (median year of isolation: 1985) than the 128 genomes downloaded from the NCBI (median year of isolation: 2008). Sequencing statistics can be found in Table S4. The estimated size of the genomes, as predicted from total contig size, ranged from 2 696 148 to 4 127 102 bp, with an average length of 3 382 459 bp.

Phylogenetic analysis reveals five lineages of C. perfringens

To investigate the phylogenetic relationship of C. perfringens strains, we performed core-genome phylogenetic analysis of 173 genomes (including 45 genomes we sequenced in this study and 128 genomes downloaded from NCBI) and estimated their divergence dates using the beast 2 software package [28]. The 173 isolates were from at least 19 different hosts and 19 geographical locations on five continents (Table S1). The majority of the strains were from animal sources (n=133), followed by human sources (n=21) and environmental sources (n=15).

To analyse the phylogeny of the 173 isolates, we compared the topology of two trees: a maximum-likelihood phylogenetic tree and a Bayesian-based tree. These two trees shared almost identical topologies, except regarding strains An68 and An185 (Figs S1 and S2). According to divergence dates of the phylogenetic tree from beast 2 and the assignment from hierBAPS, we assigned all the isolates into five major lineages (Lineage I to Lineage V) and two sub-lineages within Lineage V (Fig. 1a, Table S5). We combined the two sub-lineages within Lineage V into a major lineage considering the complexity and the divergence dates inferred from the Bayesian-based tree (Table S5). PCA showed clear separation of the four lineages (Lineages I, III, IV and V). As there is only one strain in Lineage II, which is blurred with Lineage I, no clear boundary can be found between Lineage I and Lineage II (Fig. 1b). All F _ST values between the lineages were well above 0.25, indicating that these lineages are clearly separated (Fig. 1c). The pairwise genetic distance between pairs of isolates within the lineages was significantly lower than that between lineages (Wilcoxon rank-sum test, P<0.001, Fig. 1d). Taken together, these results suggest that the current lineage definition is reasonable and reflects the true genetic relatedness of the C. perfringens strains.

Fig. 1.

Phylogenetic assignment of C. perfringens isolates and validation. (a) Bayesian-dated phylogeny of C. perfringens . We classified the 173 C . perfringens genomes into five lineages according to their phylogenetic relationships. The five lineages are labelled with different colours. Information about location, host and diseases are shown to the right of the phylogenetic tree. (b) PCA of core genome sequences in 173 strains. (c) Pairwise F _ST values between the four lineages. (d) Pairwise SNP distances within and between isolates of the four lineages.

Lineage I encompassed six strains, representing the oldest C. perfringens clade as shown in the phylogenetic tree. Coincidentally, the C. perfringens strain Tumat, isolated from the mummified remains of an ancient puppy found in Siberian permafrost, was placed in this lineage. Three isolates, CBA7123, Tumat and CP-33 in this lineage were from Asia, and the other three, PBS5, PBD1 and PC5, were from Oceania. Lineage II contained only one strain, MJR7757A, which was isolated from a human in the USA. Strains from Lineages III and IV showed the clearest eco-epidemiological relationships. Lineage III included 29 strains, the majority of which were from North America and Europe and were genetically closely related to each other. Lineage IV included 22 strains, and interestingly, except for CP-12 and CP-35, the remaining 20 strains were isolated from European countries, and may have expanded from the same clonal population. Lineage V was the largest one, comprising 115 strains of various geographical and host origins, which can be regarded as the modern C. perfringens lineage (Table S6). Eco-epidemiological relationships of some strains can be found in the sub-clades in Lineage V. For example, strains JFP771, JFP978, JFP981, JFP834, JFP833 and JP55 isolated from dogs or horses in 2000–2011 in Canada and the USA were probably from the same clonal population.

Based on the Bayesian phylogenetic inference, we tentatively hypothesized that the C. perfringens species originally emerged in Asia and Oceania ~80 000 years ago, and subsequently spread to America and Europe ~60 000 years ago. Thereafter, C. perfringens spread back to Asia ~50 000 years ago, and it became worldwide in the last 40 000 years (Fig. S2).

C. perfringens has an extremely open genome

Previous result suggested that C. perfringens has a high pangenome variation, suggesting a high plasticity of the C. perfringens genome [20]; however, the plasticity/openness of C. perfringens was not compared with other bacterial species [52]. We showed that the pangenome of the 173 C . perfringens strains contained a total of 26 954 genes, while the core genome (1020 genes, 3.8 %) makes up about one-third of the genome of each isolate and the rest is of variable gene content (Fig. 2a, b). The accessory genome harboured ten-fold more genes in 15 eggNOG function categories than the core genome (Fig. S3a). Genes associated with ‘[V] Defence mechanisms’ and ‘[N] Cell motility’ were clearly more enriched in the accessory genome than in the core genome (603 vs six genes and 241 vs four genes, respectively). We then compared the accessory genome and core genome sizes of C. perfringens with 51 other species (Table S7) and introduced an OI to evaluate the genome openness (see Materials and Methods). We showed that C. perfringens displayed the highest value of OI among the studied bacterial species, ranging from 0.43 to 0.66 (median 0.54), suggesting the accessory genes accounted for a large proportion of each C. perfringens genome (Fig. 2a–c). The human pathogens Bordetella pertussis and Mycobacterium tuberculosis showed the lowest OI; these two bacteria have relatively conserved genomes which mainly involve small genetic changes, such as SNPs, but rarely acquired new genes [53, 54].

Fig. 2.

Openness of the C. perfringens genome. (a) Proportions of accessary genes and core genes in the pangenome. (b) Accumulation of the pan genes and core genes with increasing genome number. (c) Comparison of the C. perfringens genome OI with 51 other bacterial species. (d) Spearman's correlation between bacterial genome OI and GC contents of each species. (e) Spearman's correlation between median values of the bacterial genome OI and median values of 16S rRNA gene copy number.

To understand factors associated with the openness of the C. perfringens genome, we explored the potential connection of OI with bacterial genome GC content and 16S rRNA gene copy number, which positively correlates with bacterial reproductive rate [55]. Interestingly, using the same dataset above, we found that the OI values were negatively correlated with the bacterial genome GC content (Spearman’s correlation, P=0.004) (Fig. 2d), and were positively with the median value of 16S rRNA gene copy number (P=0.048) (Fig. 2e). Collectively, these results indicated that C. perfringens has a highly open genome with diverse functions encoded in its accessory genes; the low GC content and rapid growth speed (high 16S rRNA gene copy number) may be closely related to its high genome plasticity.

The interaction between prophages/phages and C. perfringens contributes to genetic variations

We next investigated the prevalence of prophages and CRISPR systems in the C. perfringens genomes. A total of 527 phages including ‘incomplete’, ‘questionable’ and ‘intact’ were found in the 173 genomes, and the average length of the phage sequences was ~26.7 kb (Fig. S4a, b). Phage sequences were found in each C. perfringens isolate; as many as eight phage fragments were detected in a single genome (Fig. S4c). Sequences belonging to Clostridium phage vB CpeS-CP51 were the most abundant in C. perfringens (Fig. S4d). A total of 18 intact prophages were found among the 173 genomes, and nearly 70.0 % of isolates (119 genomes) contained at least one intact prophage (Table S8). The most common prophage in C. perfringens was Clostridium phage vB CpeS-CP51 (40/173, 23.1 %), followed by Clostridium phage phiSM101 (26/173, 15.0 %) and Clostridium phiCT19406C (22/173, 12.7 %). The different C. perfringens lineages have different prophage profiles (Fig. 3a, Table S9): Isolates from Lineage III had a higher prevalence of Clostridium phage PhiS63 (55.2 %, 16/29), while Lineage IV had Lactobacillus phage LLKu (50.0 %, 11/22) and Lineage V had Clostridium phage vB CpeS-CP51 (33.9 %, 39/115). The prophages in C. perfringens contained genes encoding various functions that were classified into 19 of the 25 eggNOG functional categories; these gene functions accounted for 14.1 % of the total gene functions in the C. perfringens accessory genome (Fig. S3b). Interestingly, four toxin genes, cloSI, cpe, nanH and plc, were found to be carried by at least four intact prophages in 16 isolates (Fig. 3b): the cloSI and cpe genes each in one isolate, the plc gene in six isolates, and the nanH gene in eight isolates.

Fig. 3.

Prophages in the C. perfringens genome and the toxins carried by the prophages. (a) Heatmap visualizing the prophage profile of 173 C . perfringens genomes. Cells in purple and white indicate the presence and absence of prophages, respectively. (b) Toxin genes carried by prophages in 16 isolates.

Further analysis showed that 73 C . perfringens genomes (Fig. S5), 42.2 % of the total, harboured CRISPR systems. CRISPR systems were less common in strains from animals (35.3 %) compared with those from humans and the environment (57.1 and 73.3 %, respectively) (Table S10), and strains from Lineage III showed the lowest prevalence of CRISPR systems (3.4 %, 1/29) (Table S11). The spacer sequences in these CRISPR systems of 43 isolates matched 12 specific phage genomes. The most frequently encountered phage genome sequence in the CRISPR spacers was Clostridium phage 39-O (36 isolates), followed by Clostridium phage phiCP13O (30 isolates), Clostridium phage PhiS63 (15 isolates) and Clostridium phage phiSM101 (15 isolates), suggesting that these four phages are more common invaders of C. perfringens . Among the 43 isolates with CRISPR spacers matched to specific phage sequences, 40 isolates contained no corresponding intact phages in their genomes, suggesting that the CRISPR systems in C. perfringens worked efficiently to protect against phage infection. The exception was the three isolates CP-40, CP-43 and ATCC 3626 from sheep, which had their CRISPR spacers matched to Clostridium phage phiSM101 genomes, although the complete Clostridium phage phiSM101 sequences were still found in their genomes.

Taken together, these results suggested that the frequent interaction between prophages/phages and C. perfringens is an important factor impacting the genetic variations within isolates.

Virulence gene distribution and conservation of the typing toxin genes

To provide a comprehensive view on C. perfringens virulence, we first analysed the virulence gene profile of the 173 isolates (Fig. 4a). A total of 29 genes displaying high nucleotide identity with known virulence genes were found in all these strains. In general, all the strains were positive for the genes encoding the α-toxin (plc) and κ-toxin (colA) [56], and more than 50.0 % of the isolates harboured genes encoding α-clostripain (cloSI), β-toxin (cpb2), θ-toxin (pfo), hyaluronidase (nagH, nagI, nagJ, nagK, nagL), Zmp and sialidase (nanH, nanI and nanJ). Sixteen isolates, mainly from dogs and horses, possessing as many as 18 virulence genes, encoded the highest number of virulence genes among the 173 isolates.

Fig. 4.

Virulence genes in the C. perfringens genome. (a) Heatmap visualizing the virulence gene profile of 173 C . perfringens genomes. Cells in purple and white indicate the presence and absence of genes, respectively. (b) The virulence gene number in each lineage. (c) The identity of the nucleotide sequences of the seven toxinotyping genes. (d) K _a/K _s values of all the seven toxins.

We then calculated the virulence gene number in each lineage (Fig. 4b). We showed that Lineage III harboured more virulence genes than the other lineages (P<0.05, Wilcoxon rank-sum test), while Lineage IV contained the least number of virulence genes (P<0.05, Wilcoxon rank-sum test). Isolates from Lineage IV, mainly from Europe (Fig. 1), were lacking μ-toxin and sialidases, which may decrease bacterial attachment to host cells and their spread into deeper tissues [57]. Lineages III and IV showed a higher prevalence of the cpe gene, which is responsible for human food poisoning [58]. However, the cpe gene was not host-associated, as the cpe-positive strains in Lineages III and IV were not only of human, but also of animal and environmental origins. Interestingly, the netB gene was found in Lineage V strains, and all netB-positive isolates were associated with chicken NE (Tables S1 and S12), suggesting this gene’s essential role in causing chicken NE [59]. The rarely reported ebpC and srtC genes were solely contained in a Lineage V strain (SYD-NE41) isolated from chickens having chicken NE in Australia, which were probably acquired from Enterococcus faecalis by horizontal gene transfer [42].

According to a recent update, C. perfringens is classified into seven toxinotypes (types A to G) based on the combination of seven major toxins they produce [12]. We investigated the nucleotide sequence conservation of these typing toxins (Table S13). The results showed that the nucleotide sequences of the cpb, cpe, etx and netB genes all displayed an identity over 95.0 % (Fig. 4c). The iap and ibp genes were relatively less conserved, with an identity ~90.0 %. The overwhelming majority of plc genes showed more than 97.0 % sequence identity [60] with only one exception (85.8 %) which was from a swan C. perfringens isolate [61]. In addition, we showed that the K _a/K _s values of all the seven toxins were much lower than 1 (Fig. 4d), indicating purifying selection during evolution. These results suggested that the typing toxin genes in C. perfringens are highly conserved, which warrants using their genomic sequences for strain toxinotyping.

ARG profile of C. perfringens

To reveal the ARG distribution in C. perfringens genomes, we searched the 173 genomes for known ARGs collected in he CARD database and genes encoding resistance to bacitracin. Unlike its proficiency in producing toxins and extracellular enzymes, C. perfringens possessed fewer ARGs, with only 18 ARGs found in all these genomes (Fig. 5a). The mprF gene encoding multiple peptide resistance factor F was the most prevalent ARG in C. perfringens genomes (169/173, 97.7 %), probably indicating that it is an intrinsic ARG in this bacterium. The tetracycline resistance genes tetA(P) and tetB(P) also showed high prevalence, present in 65.3 and 35.2 % of the total strains, respectively. Macrolide resistance genes, bacitracin resistance genes and aminoglycoside resistance genes were less frequently found. Interestingly, strains from the modern lineage, Lineage V, harboured more ARGs than strains from other lineages: all the 18 ARGs can be found in Lineage V while only three ARGs were contained in the other four lineages (Fig. 5b). Geographically, strains from Asia and Africa contained more ARGs than those from Europe, North America and Oceania (Fig. 5c).

Fig. 5.

ARGs in the C. perfringens genome and their transfer network among different bacterial genomes. (a) Heatmap visualizing the ARG profile of 173 C . perfringens genomes. Cells in purple and white indicate the presence and absence of genes, respectively. (b) ARG number in each lineage. (c) ARG number in strains from different continents (AF: Africa; AS: Asia; EU: Europe; NA: North America; OA: Oceania). (d) The ARG transfer network between C. perfringens and other species. The cyan octagon represents C. perfringens ; nodes with a brown circle represent ARGs found in the genomes of C. perfringens ; nodes with squares represent bacterial species sharing (>99.0 % nucleotide identity) ARGs with C. perfringens (purple: species from Firmicutes ; pink: species from Proteobacteria ; green: species from Actinobacteria ).

We then determined the potential recent ARG exchange between C. perfringens and other bacteria (3069 species in total, 11 395 genomes) using an ARG-sharing network analysis [47] (Fig. 5d, Table S14). In general, C. perfringens shared ARGs (99.0 % nucleotide identity) with a total of 55 different bacterial species, and 16 and six species in the genera Streptococcus and Staphylococcus , respectively, had at least one ARG shared with C. perfringens (Table S15). The erm(B) gene was the most frequently transferred ARG among C. perfringens and other bacteria. Highly similar genes to tetA(P), tet(44) and lnu(P) were not found in other bacterial species, which means that potential horizontal gene transfer events for these three ARGs have not been identified. These results indicate that although C. perfringens isolates contained a limited number of ARGs, it can readily exchange ARGs with other bacterial species, even phylogenetically distant species.

In silico analysis of virulence genes and ARGs carried by C. perfringens plasmids

Plasmids are important for the biology of C. perfringens , and interspecies plasmid transfer may enhance the pathogenicity and the tolerance to antibiotics of the recipients [13]. We then explored plasmids harboured in the 173 isolates and found that more than 90.0 % (157 out of 173) of the genomes contained plasmid sequences (Table S16). Annotation of these plasmid sequences showed that 14 out of 29 virulence genes and nine out of 18 ARGs identified in all the isolates could be found in the plasmids. Additionally, 81.0, 20.0, 93.5 and 93.1 % of isolates harbouring virulence genes cpe, netB, netE and netF, respectively, were found located in plasmids. For ARGs, 23.9 and 37.7 % of isolates containing tetA(P) and tetB(P), respectively, were located in plasmids (Fig. 6, Table S17). By using the custom plasmid database (Table S18), we identified 34 different known plasmids presented in 26 isolates (Table S16 and Fig. S6). Among these plasmids, pCP15_1 (n=5) was the most common, followed by pCP15_3 (n=3) (Table S19). Only two ARGs [tetA(P) and tetB(P)] and seven virulence genes (cna, cpb2, cpe, etx, netB, netE and netF) were present in these known plasmids, and the ARGs and virulence genes were not found to coexist in the same plasmid (Fig. S6).

Fig. 6.

Prediction of the locations of virulence and antibiotic resistance genes. Purple and pink cells indicate chromosome and plasmid locations, respectively.

Antibiotic resistance but not virulence of C. perfringens increased over time

To investigate the changes of antibiotic resistance and virulence of C. perfringens with time, we performed correlation analysis of ARG and virulence gene number with the isolation time of the strain (Fig. 7). The C. perfringens genomes used in this study allow us to detect such changes over an 80-year time span. The results showed that the number of ARGs in C. perfringens has increased significantly with the strains’ isolation time (Spearman correlation, P=0.001), while no significant change was found for the virulence genes (P=0.621), and no correlation between the ARG number and virulence gene number in specific isolates (P=0.457). When both ARG and virulence were taken into account, an increasing trend (marginally significant) of their total number over time was observed (P=0.079). These results reflect that although C. perfringens virulence has remained relative stable, its antibiotic resistance has increased steadily, which might be due to the selective pressures of antibiotic use in both animals and humans, especially over the last ~30 years (Fig. 1b).

Fig. 7.

The changes of ARGs and virulence factors (VFs) in C. perfringens over time. The correlation of ARG (a) and VF (b) number with strain isolation times. (c) The correlation between ARG number and VF number in different strains. Dot size is proportional to the number of strains. (d) Changes in the total number of ARGs and VFs in C. perfringens over time.

Discussion

As a human and animal pathogen that can cause various diseases, C. perfringens has gained increasing attention globally in recent years [12]. Revealing the evolutionary and genomic features of this bacterium will undoubtedly expand our understanding of its basic biological properties. In this study, we performed phylogenetic and genomic analysis of C. perfringens and provided a further view on the bacterium’s phylogeny, genome plasticity, virulence and antibiotic resistance. The newly sequenced strains in this study, especially those isolated longest ago, have enriched the genome collection in the databases, allowing us to address the evolution of C. perfringens from an historical perspective.

Previously, 56 C . perfringens genomes had been clustered into four major clades [20]. In the present study, based on the phylogenetic relationship of strains, we assigned the 173 isolates to five lineages. The major difference with previous studies was that isolates from the reported Clades 2 and 3 were assigned to Lineage V here, which we defined as the modern lineage of C. perfringens . Analysis of ARGs and virulence from the available isolates prompted us to suggest that Lineage V may be a more successful clade during the evolutionary process; the acquisition of higher antibiotic resistance and unique toxins has given this lineage an improved survival advantage. Based on a Bayesian phylogenetic analysis, we tentatively inferred that C. perfringens originated ~80 000 years ago and has prospered over the last 40 000 years. This long evolutionary history is comparable to the evolution of Mycobacterium tuberculosis which was inferred to have originated ~70 000 years ago [62], suggesting C. perfringens may be as old as this famous human pathogen. We should stress that this hypothesis may be biased by the sample size, and global isolates with different isolation times, locations and sources.

It has been reported that the core genome of C. perfringens is small, suggesting a high plasticity of the C. perfringens genome, but the openness of the C. perfringens genome has never been compared with other bacterial species. In fact, a few studies have been performed to investigate the openness of bacterial genomes among different species [63–65]. Here we were surprised to find that C. perfringens has the most open genome compared with 51 other bacterial species. We demonstrated that the OI of a bacterial genome was clearly associated with the GC content and 16S rRNA gene copy number in the genome. The rRNA operon copy number has been verified to correlate positively with the growth rate of bacteria. Therefore, the extremely low genome GC content (~28.0 %) [20] and relatively high growth speed (a generation time as short as ~7 min) [66] of C. perfringens may be major contributors to the high level openness of the C. perfringens genome. A highly open genome implies this bacterium’s good ability to acquire new genes from external environments and thus to cope with and adapt to hostile survival conditions easily.

Compared with the presence of intact prophages in C. perfringens genomes (nearly 70.0 % of isolates), we found that the prevalence of CRISPR systems in C. perfringens genomes is low (42.2 % isolates). As an acquired immune system in bacteria, the CRISPR system plays critical roles in defence against phage infection [49]. The low prevalence of CRISPR systems may indicate a low barrier to horizontal gene transfer events and probably a frequent interaction of phage/prophage with C. perfringens . Phage/prophage has been regarded as one of the most important factors driving the evolution of bacterial hosts [67]. Through horizontal gene transfer, phage/prophage can mediate the alteration of host physiological functions by changing their antibiotic resistance and virulence. It is noteworthy in our findings that four toxin genes were predicted to be carried by intact prophages in 16 C . perfringens isolates, especially the cpe gene which has been shown to exist in both genome and plasmid [68], but never in phage/prophage.

C. perfringens does not invade healthy cells but produces various toxins and enzymes that are responsible for the associated lesions and symptoms [1]. Remarkably, this bacterium is reported to secrete >20 degradative toxins which constitute its primary arsenal to initiate histotoxic pathogenesis in both humans and animals [13]. We have shown that although the genome of C. perfringens is highly plastic, most of the toxin genes, especially those used for strain toxinotyping, are conserved, which facilitates using PCR or genome sequencing for strain typing. The conservation of these toxin genes and the purifying selection (K _a/K _s <1) on them during evolution may imply their critical roles for the better survival of C. perfringens in different ecological niches. The majority of the virulence genes have no lineage preference, probably due to either their carriage by conjugative plasmids, providing them with the means to be readily transferred between isolates or their chromosomal carriage as intrinsic genes. However, the high prevalence of the food poisoning-related cpe gene in Lineage III and IV, and the presence of the NE-associated netB gene in only Lineage V warrants further investigation on the interaction between these two important toxin genes and their specific host genetic background.

Our analysis suggested that C. perfringens has the potential to exchange ARGs with various bacterial species, especially those from the genera Staphylococcus and Streptococcus . As species in these two genera have plentiful ARGs [69], it is possible that C. perfringens will acquire new ARGs from these bacteria in the future. We should note that the ARG-sharing network presented here does not mean direct ARG transfer events between C. perfringens and other bacterial species, as the strains we analysed have no direct contact with each other. However, the presence of the same ARG (more than 99 % sequence identity) in different species suggests these genes have the ability to transfer.

Plasmids are important mobile genetic elements which play critical roles in disseminating both virulence genes and ARGs. We showed that plasmid sequences were present in more than 90 % of C. perfringens isolates, suggesting the high prevalence or even frequent transfer of plasmids within this species. Virulence genes cpe, netB, netE and netF were frequently found carried by plasmids, which may indicate the important role of plasmids in spreading these important virulence factors. It should be noted that plasmids pJFP838C and pJFP55F both contained two virulence genes netE and netF, which deserves further investigation. The tetA(P) and tetB(P) genes were found in seven known plasmids, while they were still found in many strains having no predicted plasmid sequences, suggesting they were located in chromosomes or in novel plasmids in these strains. As we cannot accurately predict all the plasmid sequences from the draft genomes (only 17 complete genomes included in this study), the virulence gene and ARG contents in C. perfringens plasmids may have been underestimated, which will be further unravelled with more complete genomes sequenced.

In conclusion, we analysed 173 C . perfringens genomes from strains of worldwide origin. We assigned these genomes to five different lineages having different genomic features and demonstrated that C. perfringens has an extremely open genome, probably due to its low genome GC content and rapid growth speed. The gene contents in phages and plasmids contribute to the species' genetic diversity. The limitation of this study is that the isolates/genomes included are not a good global representation of the geographical, animal or environmental distribution of the organism because the availability of sequences is skewed by the specific studies and their different aims. As a result, the lineage features and distributions presented here may be biased by the limitation of currently available genomes. Therefore, larger scale sampling and sequencing efforts are continuously needed to fully understand this important human and animal enteric pathogen.