Clostridium perfringens is a Gram-positive, spore-forming, anaerobic, rod-shaped bacterium, known as an important causative agent of foodborne and non-foodborne gastroenteritis (Grass et al., 2013). The ability of this bacterium to form resistant spores contributes to its survival in many environmental niches, including soil, sewage, foods, and the intestinal microbiota of humans and animals (Xiao, 2014; Li et al., 2016). C. perfringens can cause necrotic enteritis, enterotoxemia, and gastroenteritis in animals (Keyburn et al., 2008; Uzal et al., 2015). It is also a foodborne pathogen that causes diarrhea and abdominal pain in humans (Scallan et al., 2011). C. perfringens is commonly found as a contaminant in meat and poultry products, as well as in vegetables and crops. This bacterium is an important pathogen involved in foodborne outbreaks (FBOs) and ranks as the fourth most common causative agent of foodborne illness due to bacterial toxins in France and Europe, with more than 1,500 human cases each year (EFSA, 2016). The pathogenicity of this bacterium is largely attributable to its ability to produce a variety of virulence factors. These virulence factors include well-characterized pathogenic toxins and hydrolytic enzymes. Their roles in FBO remain to be defined but their presence is an aggravating factor (Li et al., 2016). Toxin production varies among C. perfringens strains and is the basis for the classification system, which is based on the production of four major toxins, namely CPA, CPB, ETX, and ITX that divide C. perfringens strains into toxinotypes A to E (McClane et al., 2013). This classification was recently revised to include two additional toxinotypes, i.e., C. perfringens toxinotype F strains producing enterotoxin (CPE) but not CPB, ETX or ITX, and C. perfringens toxinotype G strains producing NetB (Rood et al., 2018) (Table 1). The number of characterized virulence factors is constantly increasing, with more than 20 toxins and hydrolytic enzymes identified to date (Li et al., 2016; Kiu and Hall, 2018). However, there is considerable variability in the toxin armamentarium, and a single strain cannot produce all these virulence factors (Freedman et al., 2016). The virulence factors of C. perfringens can be classified functionally as membrane-damaging enzymes, pore-forming toxins, intracellular toxins, and hydrolytic enzymes (Revitt-Mills et al., 2015). Genes encoding these virulence factors may be located on the chromosome, and on the large plasmid (Freedman et al., 2016). Chromosomal virulence factors include the cpa gene (encoding α-toxin), colA (κ-toxin), pfoA (θ-toxin), and nagH (hyaluronidase or μ-toxin). These genes are located on the variable region of the chromosome (Chalmers et al., 2008a). The nanH, nanI, and nanJ genes encoding different sialidases, and cadA encoding ν-toxin are located on a conserved region of the chromosome (Obana and Nakamura, 2011). The genes cpb (encoding CPB-toxin), cpb2 (CPB2-toxin), etx (ETX-toxin), ia/ib (ITX-toxin), netB (NetB-toxin), tpeL (TpeL-toxin), ureABC (ureases), cpd (δ-toxin), lam (λ-toxin), and becA/becB encoding the binary enterotoxin for C. perfringens are located on large plasmids of variable size ranging from 65 to 110 kb (Li et al., 2013). Recently, partial genome sequencing of the canine isolate revealed three novel putative toxin genes encoding proteins related to the pore-forming leucocidin/hemolysin family. These putative toxin genes were designated netE, netF, and netG (Mehdizadeh Gohari et al., 2015). netE and netF are located on a large conjugative plasmid, and netG is located on another large plasmid (Mehdizadeh Gohari et al., 2017). The C. perfringens enterotoxin gene (cpe) can be located on either the chromosome or on the plasmid (Gao and McClane, 2012). A strain harboring both a chromosomal and plasmid-borne cpe gene has not yet been observed (Freedman et al., 2016). C. perfringens enterotoxin (CPE) is one of the toxins associated with sporulation and has been implicated in FBOs and non-foodborne gastroenteritis illnesses (Lukinmaa et al., 2002).

The Standardized method NF EN ISO 7937 is the conventional method for detecting C. perfringens strains in foods by the enumeration of bacterial colonies on specific agar media (Anonymous, 2005). With this robust method, it is possible to detect and enumerate viable C. perfringens in food samples and to obtain isolates that can be characterized further. This reference method therefore makes it possible to isolate the bacterium but does not provide any phenotypic information distinguishing the strains from one another, particularly with regards to their virulence (Law et al., 2015). While the detection of CPE in the stools of patients is a marker for implicating C. perfringens as the etiological agent in foodborne illnesses, CPE is produced in the intestine during sporulation of vegetative cells and is consequently absent from the contaminated food (Lin and Labbe, 2003). Moreover, isolation of cpe-positive C. perfringens in suspected foods and from stools of affected individuals may be accompanied by the presence of cpe-negative C. perfringens from the environment and microbiota (Scallan et al., 2011). This situation makes toxin detection hazardous to link isolates from patients to food vehicles (Lindström et al., 2011). The common assumption that the intestinal pathogenicity of C. perfringens relies on a specific toxin was recently challenged by findings suggesting that additional virulence factors may be involved in cytotoxicity (Yasugi et al., 2015). This prompted us to explore the virulence potential of C. perfringens, taking into account the more recently described virulence factors and phylogenomic inference presenting high discrimination power.

To improve epidemiologic knowledge about C. perfringens involved in FBOs, we developed a new real-time PCR typing technique targeting 17 genes encoding 15 virulence factors. The method was applied to 141 C. perfringens FBO strains isolated in the Paris region between 2013 and 2017. The virulence gene profiles obtained by this method would make it possible to describe strain diversity with regards to the cpe gene among these outbreaks, and robust phylogenomic analysis based on whole genome sequencing (WGS) data (Felten et al., 2017) would enable us to determine phylogenomic diversity at a large genomic scale, as recently emphasized (Kiu et al., 2017).

While many food pathogens have been sequenced and studied extensively, limited genome sequence data for C. perfringens are available in public databases, and very few genomic studies have been conducted on food isolates (Shimizu et al., 2002; Myers et al., 2006; Mehdizadeh Gohari et al., 2017; Lacey et al., 2018). To our knowledge, the largest genomic study conducted on 56 C. perfringens genomes contained only three foodborne strains (Kiu et al., 2017). Genomic comparison previously conducted on 12 strains from different toxinotypes, isolated from different sources, highlighted a lack of features differentiating toxinotype A from the other isolates, indicating that toxinotype A can easily shift to other types by acquisition of toxin-encoding plasmids (Hassan et al., 2015). A recent genomic study conducted on necrotic enteritis-causing isolates of C. perfringens demonstrated that pathogenicity does not correlate with coregenome content, but rather with accessory gene content located on the chromosome and plasmids (Lacey et al., 2018). To better understand the diversity among C. perfringens implicated in FBOs, we performed whole genome analysis on 58 C. perfringens strains selected on the basis of epidemiologic information and diversity of virulence gene profiles.

C. perfringens has been recorded as the fourth most common cause of FBOs in France (Santé Publique France, 2016). FBO caused by C. perfringens results from the ingestion of foods that are contaminated with spore or vegetative cells (Brynestad et al., 1997). The true human burden of C. perfringens is likely much higher as many outbreaks are never investigated and many illnesses occur sporadically (Vaillant et al., 2005). Our study shows that FBOs were commonly due to meat and meat product foods (52%, 22/42), which is consistent with the epidemiology of C. perfringens (Wen and McClane, 2004; Grass et al., 2013). The contamination of meats may occur through contact of carcasses with feces, as well as via cross-contamination by other foods or contaminated surfaces during slaughtering. Although most C. perfringens FBOs are attributed to meat and poultry products (Grass et al., 2013), our study shows that in eight cases C. perfringens was the only pathogen isolated from vegetable dishes. This unexpected finding is likely linked to the ubiquitous nature of C. perfringens, which can be isolated in the environment and the soil (Uzal et al., 2014). Improper holding temperatures and incomplete cooking of foods are recognized as major factors contributing to the development of C. perfringens FBOs (Sarker et al., 2000a). Moreover, vegetables may initially be contaminated by spores of C. perfringens (Xiao et al., 2012), while the cooking temperature of vegetable dishes is too low to destroy spores (Talukdar et al., 2016). C. perfringens can grow at temperatures from 20 to 53°C and spores can survive high temperatures (up to 95°C for 1 h) (McClane et al., 2013). We observed that certain virulence gene profiles (V, VII, X, see Table 4) are prevalent in our collection. It would be of interest to measure the heat resistance of a selection of isolates to determine if this phenotypic trait can explain the prevalence of specific genomic profiles in food.

Symptoms due to foodborne C. perfringens are similar to those caused by other foodborne pathogens, especially B. cereus diarrheal strains (Tewari and Abdullah, 2015). Moreover, current methods for investigation of C. perfringens in FBOs are based on enumeration of characteristic colonies in specific media, followed by biochemical confirmation. These methods are useful to collect strains but do not provide any typing information. As mentioned above, the revised typing scheme classifies strains producing CPE as type F. However, understanding the role and virulence of C. perfringens strains isolated in the context of FBOs requires us to describe the genetic diversity of the implicated strains and to develop robust and novel typing methods. In the present work, we developed a new PCR-based technique on a comprehensive set of 17 genes encoding putative virulence and toxin factors. This new PCR-based technique was used to characterize the diversity of a collection of 141 strains implicated in 42 FBOs.

The cpa, colA, nanH, and cadA genes encoding α-toxin, κ-toxin, sialidase (i.e., a small intracellular sialidase with 43 kDa of molecular weight) and deoxyribonuclease were found to be conserved in all isolates analyzed. Previous genomic studies support this result (Goossens et al., 2017; Kim et al., 2017; Kiu et al., 2017). Consequently, these genes cannot be considered relevant markers to discriminate FBO isolates. Conversely, four genes were not identified in our collection. Consequently, our typing scheme was reduced to nine genes and was used to characterize the virulence profile of a collection of 141 C. perfringens strains implicated in 42 FBOs. Fifteen different profiles were recognized (Table 4), demonstrating the discriminatory power of this method. The epidemiological relationships between C. perfringens of different origins, including isolates from FBOs, have been investigated previously (Jost et al., 2006; Afshari et al., 2016). The previously published articles focusing on FBO isolates and animals concluded that isolates from the same outbreak have a similar pattern, while genetic diversity is high in non-outbreak isolates selected randomly (Johansson et al., 2006; Chalmers et al., 2008b; Fohler et al., 2016). Our results challenge this view and demonstrate that a range of virulence gene profiles can be found within the same FBO.

A more thorough examination of C. perfringens diversity revealed interesting patterns. Except for profile I, the presence of cpe (p < 0.05) was associated with a low number (5–7) of other virulence genes. More precisely, a strong association was observed between the presence of cpe and the lack of pfoA and nanI (r = -0.67, p < 2.2 × 10^-16). This observation is consistent with previous studies showing that most cpe-positive C. perfringens strains lack the pfoA gene (Deguchi et al., 2009). The presence of the pfoA and nanI genes was found to be significantly associated (61/66; p < 2.2 × 10^-16). The most predominant profiles in cpe-negative strains were the profiles X, with 40 of the 61 strains, and VIII, with seven of the 61 strains. In these two profiles, the pfoA gene was associated with sialidases and/or hyaluronidase. These toxins could contribute to the overall virulence of the bacterium (Revitt-Mills et al., 2015). Previous in vitro studies have demonstrated a synergic effect of perfringolysin O and α-toxin on the cytotoxicity of non-foodborne strains (Awad et al., 2001); the association between pfoA and nanI could have toxicologic consequences that would be worth studying in an appropriate experimental system.

Profiles IV, V, and VII were predominant and were found in 18 of the 29 FBOs with cpe-positive profiles. In addition, profile VII carries cpe as the sole toxin-encoding gene. These results confirm that CPE is a strong contributor to foodborne illness due to C. perfringens (Sarker et al., 2000b).

Two to three different profiles of virulence genes were identified in three FBOs where C. perfringens was the only isolated pathogen, demonstrating that Clostridium FBOs can be associated with heterogeneous contamination of food. This important result complicates the epidemiology of clostridial FBOs, and the contribution of the different gene profiles to the development of symptoms needs to be investigated further. A direct consequence is that pathogenic strains co-infecting foods with non-pathogenic strains can be overlooked. It is therefore important to couple a PCR-based method targeting genes with the NF EN ISO 7937 standard method and characterize more than five strains isolated in the same FBO.

To further analyze the diversity of C. perfringens-associated FBOs, WGS of 58 isolates of C. perfringens was conducted to provide insight into the phylogenomic relationships and distribution of virulence factors among cpe-positive and cpe-negative strains. To the best of our knowledge, this study represents the first large genomic analysis focusing on C. perfringens food isolates, compared to other studies screening other sources and including few foodborne samples (Hassan et al., 2015; Kiu et al., 2017). The computed coregenome phylogenomic inference indicates that C. perfringens strains carrying the cpe gene on the chromosome have evolved independently from cpe-negative and plasmid carrying cpe strains. This argues for the possibility of its involvement in adaptation to specific niches, and supports previous research conducted on C. perfringens strains from different origins showing a clear partitioning of strains capable of inducing foodborne illness into lineages distinct from those of strains isolated from other environments (Hibberd et al., 2011). Previous studies (Miyamoto et al., 2002; Li et al., 2013) have localized the cpe gene near the dcm gene on both the pCPF4969 and pCPF5603 plasmids of type A isolates. Our investigation of the genetic environment of 16SBCL525, 17SBCL79, and 17SBCL85 is consistent with these previous studies and demonstrates that the dcm gene is proximal to the plasmid-borne cpe gene. The fact that these strains were phylogenetically clustered with cpe-negative strains supports the notion that the dcm region of plasmids represents a hot-spot for insertion of certain mobile genetic elements, including some carrying a cpe gene. An early study indicated that C. perfringens carrying the cpe gene on the chromosome is flanked by upstream and downstream IS1470 and, instead of a dcm gene, they carry the purine permease (uapC) and quinolinate phosphoribosyltransferase (nadC) genes on the chromosome (Brynestad et al., 1997). Our investigations support this assertion and, to our knowledge, this is the first study describing a diversity of genetic arrangements of the cpe locus in chromosomal cpe strains (Brynestad et al., 1997; Miyamoto et al., 2002; Li et al., 2010).

The coregenome phylogenomic reconstructions provide interesting epidemiologic information and indicate actual relationships between strains from different outbreaks. FBOs 2987, 2774, 2988, and 2773 were caused by closely related strains with regard to a limited average of pairwise distances between them (i.e., 7.0 SNPs) and common profiles of virulence genes. This similarity could be due to the same biotype that may reach food and cause FBOs. Due to wide biotype and ecological niches, genetically diverse strains of C. perfringens may be found in food matrices. Important epidemiologic findings have resulted from the high resolution of genome sequencing. For instance, WGS was used to demonstrate that many C. difficile cases could not be attributed to transmission from asymptomatic patients, but that there is high diversity of C. difficile in the environment (Tang et al., 2017). This approach could be used in our analysis. We have identified considerable genetic diversity within cpe-negative strains. The high genetic diversity may in part be explained by the broad distribution of this group in the environment. An earlier genomic comparison study conducted on three genomes of C. perfringens (ATCC13124, SM101, and strain 13) revealed considerable genomic diversity with more than 300 unique genomic islands identified (Myers et al., 2006). Our phylogenomic analysis supports this notion with regard to strains 16SBCL582 and 16SBCL584 from FBO 2273. However, a comparison of the virulence gene profiles of these distant strains (obtained by real-time PCR) indicates that they present the same virulence gene profile. This result suggests that the approaches based on the detection of virulence factors are not sufficient to efficiently discriminate C. perfringens strains.

Another interesting finding is that strains are grouped according to the distribution of virulence factors. In particular, the presence of the nanI, pfoA and nagH genes in cpe-negative strains is of interest, and it would be worthwhile to investigate the putative role of these virulence factors in cpe-negative disease.

The aim of our study was to characterize C. perfringens involved in FBOs from the Paris region between 2013 and 2017 by investigating profiles of virulence genes according to the presence or absence of 17 genes encoding 15 virulence factors screened by a new real-time PCR tool. High diversity was obtained between strains in different outbreaks. Heterogeneous profiles of virulence genes within the same outbreaks were highlighted for the first time, indicating that food may be contaminated with different C. perfringens strains that may or may not harbor the cpe gene. Our study emphasizes the importance of more detailed characterization of C. perfringens isolates involved in FBOs than simple enumeration on specific agar, as suggested by the EN ISO 7937 reference method and/or characterization of five random isolates.

We have shown that the reconstructed coregenome phylogenomic history of C. perfringens food isolates can identify two mains groups, both being able to contaminate food. Our research did not clearly associate food sources with a specific clade. However, future comparative genomic studies on strains of various origins and from different geographical areas should provide clues to understand C. perfringens pathogenicity and associations to ecological niches. Genomic studies combined with epidemiological data provide new leads to design toxicity assays that may renew our understanding of the molecular basis of C. perfringens pathogenicity, and ultimately identify new biomarkers to differentiate between pathogenic and non-pathogenic C. perfringens strains. Finally, it would be interesting to characterize C. perfringens strains implicated in the FBO with respect to their reservoir.

AMA conceived the study, designed the experiments, and wrote the manuscript. NR approved bioinformatics analyses, provided advice for statistical analyses, and revised the manuscript. AMA and KH performed all bioinformatics analyses. MLN contributed to the microbiologic analyses. SaD, PF, and SoD conceived the methodology and performed high throughput real-time PCR analyses. OF, J-AH, and M-YM conceived and piloted the project and revised and approved the manuscript for publication.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.