Salmonids represent one of the most important fish groups in the aquaculture industry (FAO, 2016). However, the success and sustainability of salmonid aquaculture largely depend on disease control.

Flavobacterium psychrophilum, a bacterium belonging to the family Flavobacteriaceae, is considered one of the most important bacterial pathogens in freshwater salmonid aquaculture worldwide (Nematollahi et al., 2003). The two main clinical forms are rainbow trout fry syndrome (RTFS) and bacterial cold-water disease (BCWD) (Cipriano and Holt, 2005) that can both result in considerable economic losses (Antaya, 2008; Starliper, 2011). All salmonid fish, especially coho salmon (Oncorhynchus kisutch) and rainbow trout (Oncorhynchus mykiss) are susceptible to RTFS and BCWD, as well as ayu (Plecoglossus altivelis), a fish related to salmonids. These infections are associated with skin ulcers, necrotic myositis, septicemia as well as exophthalmia (Nematollahi et al., 2003). F. psychrophilum has occasionally been detected in non-salmonid fish, samples of water, sediments and biofilms from rivers receiving outlet water from infected fish farms (Madetoja et al., 2002; Starliper, 2011). Despite extensive research on vaccine development (Plant et al., 2009; Gómez et al., 2014) and the recent marketing authorization of a vaccine in Chile, treatment still essentially relies on antibiotic administration (Schmidt et al., 2000). Diversity among F. psychrophilum isolates was reported at different levels, and a variety of typing schemes (e.g., biotypes, serotypes, and genotypes) was proposed accordingly. However, the correlation between these different schemes (Madsen and Dalsgaard, 2000; Madetoja et al., 2001; Ngo et al., 2017) and the impact of the different types on the onset and severity of the infection remain poorly understood.

Gram-negative bacteria have a complex set of surface polysaccharides that include the lipopolysaccharide (LPS) as well as capsular polysaccharides (CPS) or exopolysaccharides (EPS) (Raetz and Whitfield, 2002). These structures are thought to play important roles in the pathogenicity, immune escape, serum resistance, inflammation, adhesion and biofilm formation (Finlay and McFadden, 2006). Serotyping is based on the immunogenicity of various surface-exposed bacterial structures (e.g., LPS, capsular polysaccharides, flagella, peptidoglycan, etc.) biosynthesized by genes belonging to the accessory genome. This represents an advantage over many molecular typing methods (e.g., 16S rRNA sequencing, ribotyping, MLST) that rely on core genome genes, which are not phenotypically relevant determinants. LPS is one of the most pro-inflammatory compounds of Gram-negative bacteria typically consisting of three regions: a hydrophobic domain known as lipid A, a non-repeating “core” oligosaccharide, and a distal polysaccharide (or O-antigen). The O-antigen greatly varies between and within species, providing the main basis for many serotyping schemes. The structure of the O-polysaccharide of F. psychrophilum strain CSF259-93 has been solved and found to be an unbranched polymer of trisaccharide repeating units (MacLean et al., 2001). A 70 kb region (FP1299 to FP1234) encompassing genes predicted to be involved in the biosynthesis, export, modification and polymerization of polysaccharides was previously identified in the genome of F. psychrophilum JIP 02/86 (Duchaud et al., 2007). Several substitutions of genes enclosed in this region were noticed by comparing the genomes of strains CSF259-93 and JIP 02/86 (Wiens et al., 2014).

Serological differences among F. psychrophilum isolates were noticed since early studies and different serotyping protocols were proposed, resulting in the description of a varying number of serotypes. Wakabayashi et al. (1994) described two distinct serotypes (O-1 and O-2) among a collection of Japanese isolates; a third serotype (O-3) was added in 1999 (Izumi and Wakabayashi, 1999) and a fourth one (O-4) in 2003 (Izumi et al., 2003). In Denmark, Lorenzen and Olesen (1997) proposed a serotyping scheme based on three serogroups (Fp, Th and Fp^T). Using these schemes, some degree of association between serotypes and host fish species was observed. For example, strong relationship between host and O serotypes was reported, O-1, O-2, and O-3 corresponding to isolates infecting Coho salmon, ayu, and rainbow trout, respectively (Izumi and Wakabayashi, 1999). Moreover, the co-existence of genetically and serologically diverse isolates within individual farms was detected in Finland (Madetoja et al., 2001) and in the United Kingdom (Ngo et al., 2017). Finally, Mata et al. (2002) identified as much as seven different serovars (1–7) and also reported obvious links with host fish species. This serological diversity might have important consequences for the selection of appropriate candidate strain(s) for vaccine development (Gómez et al., 2014 and references therein), selective breeding of salmonids for increased disease resistance (Leeds et al., 2010), follow-up studies, epidemiological surveillance, disease control (Ngo et al., 2017), as well as for a better understanding of virulence and host resistance traits.

In the present study, we used the serotyping method proposed by Lorenzen and Olesen (1997) together with genome-wide association studies to identify key molecular determinants of serotypes. Accordingly, a multiplex PCR-based serotyping scheme (mPCR) was developed and successfully applied to a collection of 244 bacterial isolates.

Serological differences among F. psychrophilum isolates were reported since the pioneering studies in the nineties and different serotyping schemes were proposed (Wakabayashi et al., 1994; Lorenzen and Olesen, 1997; Mata et al., 2002). However, the conventional protocols for the serotyping of F. psychrophilum isolates available so far suffer several drawbacks such as the choice of the appropriate scheme, the need to raise antisera from animals, and the bias of human interpretation of agglutination results. In addition, some strains display auto-agglutination, lack of reactivity, or cross-reaction with multiple antisera that make laborious reciprocal absorptions against heterologous strains necessary. Moreover, inconsistencies between the results of agglutination and ELISA have been reported (Lorenzen and Olesen, 1997). Hence, serotyping by conventional serological methods is costly, labor-intensive and requires significant technical expertise. Therefore, molecular serotyping using mPCR is a relevant alternative to traditional serotyping, though identification of molecular markers that determine the serotypes is a prerequisite in order to target them with DNA probes. Indeed, mPCR serotyping assays have been developed for a number of bacterial pathogens such as Haemophilus parasuis, Pasteurella multocida, Salmonella sp., Streptococcus suis, S. pneumoniae and S. agalactiae (Kong et al., 2005; Liu et al., 2013; Harper et al., 2015; Ma et al., 2016).

We took advantage of the availability of the whole-genome sequences of 34 F. psychrophilum isolates (Duchaud et al., in review) and of the previously published serotyping scheme of Lorenzen and Olesen (1997) to identify key molecular determinants of the serotypes. Focusing on strains FI056 and DK002 that are phylogenetically closely related but belong to two different serotypes (Fd and Th) (Figure 1), we identified the strain-specific genes FI056_50102 and DK002_320117 encompassed in the polysaccharide biosynthesis-encoding locus. Indeed, among the genes surrounding FI056_50102 or DK002_320117, some are predicted to be involved in the biosynthesis of the nucleotide-sugar precursors of O-antigen (rmlA and rmlB for the synthesis of dTDP-L-rhamnose; wbpM for the synthesis of UDP-N-Acetyl-D-quinovosamine and flnA, flnB and flnC for the synthesis of UDP-N-Acetyl-L-fucosamine) whereas others encode the sugar transferases required for the assembly of lipid-linked O-repeat units (wbuA, wbuB and wcgN, the latter encoding the initial undecaprenyl-phosphate sugar transferase) (Figure 2). Previous studies aiming at identifying F. psychrophilum antigens recognized by polyclonal rabbit antisera have been performed (Crump et al., 2001, 2003). LPS-containing O-polysaccharide and core-region lipo-oligosaccharide have reported to be immunological relevant structures targeted by these sera. It is therefore tempting to speculate that the O-polysaccharide, predicted to greatly vary within species, provides the antigenic specificity of the serotyping scheme used in this study. So far, the O-polysaccharide structure has been solved for only one F. psychrophilum strain (CSF259-93), identified as belonging to Type 2 by mPCR (MacLean et al., 2001). Extensive biochemical analyses will be required for each F. psychrophilum serogroup to formally identify the nature of the carbohydrates targeted by each anti-serum.

Genes FI056_50102/DK002_320117 were therefore promising candidate markers for serotype affiliation. Following a genome-wide association study using the 34 F. psychrophilum isolates, four Types were defined and Types 1 and 2 were firmly associated to Fp or Th serotypes (Table 2). Type 3 was not associated to an exclusive serotype in the scheme defined by Lorenzen and Olesen as Fp^T serotype was shared between Type 0 and Type 3 isolates. However, its strong association with ayu (Figure 4 and Supplementary Table S2) indicates that Type 3 likely corresponds to the O2 serotype defined by Izumi and Wakabayashi (1999). Remarkably, neither the three genes located at this position (FI056_50102, DK002_320117 and FPC840_340035) nor their encoded proteins display any primary sequence identity. Initially annotated as “hypothetical protein,” they are all predicted to be multipass transmembrane proteins, located in the inner membrane, with weak homologies to polysaccharide-biosynthesis proteins (e.g., O-antigen polymerase, O-antigen ligase) (Table 3). They likely perform the same general function (i.e., an unknown step in polysaccharide biosynthesis) but their activity should lead to structural variations such as the nature of the glycosidic bounds, additional sugar ramification(s), chain length, etc. As no obvious orthologous genes have been identified in strains belonging to Type 0, one might hypothesize that other multipass transmembrane proteins encompassed in polysaccharide biosynthesis-encoding locus could be functional homologs in these strains. Because of the phylogenetic incongruence of these genes, one must conclude that gene exchange by recombination took place at this locus. Indeed, F. psychrophilum was found to have a very high homologous recombination rate that contributes to population diversification (Nicolas et al., 2008; Vos and Didelot, 2009).

Taking advantage of this sequence divergence, we developed a mPCR assay able to discriminate strains belonging to Types 1, 2, 3 and 0. Our mPCR assay proved to be successful as all 244 strains tested were typable using this scheme (Supplementary Table S2). Among these strains, 28 were analyzed in the original paper of Lorenzen and Olesen (1997) and an outstanding correlation between the previously reported ELISA serotypes and the mPCR results was observed. The situation is more complex using agglutination assay. As a matter of fact, one should keep in mind that the antisera consist of polyclonal antibodies that were processed by multiple rounds of reciprocal absorptions against different heterologous strains. Therefore, cross-reactivity observed using slide-agglutination could be considered as a technical drawback, often noticed using such assays. Other variations in the locus (e.g., gene substitution, polymorphism or differences observed in other genes of the locus for some strains) may also impact the serotype and could explain the puzzling results that were obtained with the remaining cross-reacting isolates, especially since as much as seven different serotypes were reported for F. psychrophilum using a broader scheme (Mata et al., 2002). Using the latter scheme, strain FPC840 belongs to serovar 7 and using Izumi’s scheme it belongs to serovar O2. It is therefore tempting to speculate that Type-3 described in this paper actually corresponds to serovar 7 according to Mata and to serovar O2 according to Izumi. Another example is strain LVDI 5/1 that corresponds to serovar 5 using Mata’s scheme and to Fp^T using Lorenzen and Olesen’s (1997) scheme. An attempt to correlate all the previously published serotyping schemes with mPCR is proposed in Supplementary Table S3. Because Fp^T encompasses many strains displaying a poorer conservation of the locus together with a wider genomic diversity, one might suspect that they could also belong to some other serovars observed by Mata et al. (2002) or to other, yet undescribed serovars. One should conclude that the current serotyping schemes may not cover all existing serotypes of F. psychrophilum and that WGS is likely a promising and efficient approach to address this issue. Accordingly, our mPCR assay could be enriched by new targets allowing the detection of additional serogroups in the future.

Our results also revealed a striking association between mPCR-serotype and fish host species. All strains retrieved from coho salmon but one belong to Type-0. The only exception is the Chilean isolate MHC 1710K that belongs to Type 2. As previously proposed, Chilean fish farming practices have contributed to create overlapping host ranges as most fish farms often house mixed stocks of different salmonid species including coho salmon and rainbow trout (Avendaño-Herrera et al., 2009, 2014). The vast majority of strains retrieved from rainbow trout (119 among 151) belong to Types-1 and 2. However, 23 strains belong to Type-0 (20 from France, 2 from Switzerland, and one from Denmark) and 9 strains to Type-3 (8 from France and one from Israel). It is noteworthy that strains of serotypes Th and Fd were reported to be predominant in rainbow trout and associated to disease outbreaks (Lorenzen and Olesen, 1997; Dalsgaard and Madsen, 2000) and to tend to be more virulent than Fp^T isolates in experimental infection model using rainbow trout (Madsen and Dalsgaard, 2000; Madetoja et al., 2002). Considering that Types-1 and -2 correspond to Fd and Th serotypes, our results showed that Th and Fd are equally predominant (40%) in our rainbow trout set of isolates. Previous studies reported quite similar results, with serotype Th observed more frequently (60%) in the United Kingdom (Ngo et al., 2017) and Fd occurring more frequently in Finland (Madetoja et al., 2002). Altogether, the mPCR typing method represents a promising epidemiological tool as it allows determining if an isolate belongs to one of the two serotypes associated to disease outbreaks in European fish farms. Finally, all strains retrieved from ayu or their eggs belong to Type-3. It is therefore very tempting to speculate that the nature of the Type provides a selective advantage according to the host that is infected. Nevertheless, these associations might also reflect the underlying genetic structure of the population without necessarily involving a selective advantage linked to the serotype. Indeed, MLST analysis of F. psychrophilum population structure has also identified clear associations between particular clonal complexes and some fish host (Nicolas et al., 2008). Our identification of the genetic determinants of the serotypes provides a basis for future studies aiming at clarifying the links between serotypes, population structure, and virulence.

As previously observed for other bacterial pathogens, mPCR-based serotyping methods often perform better than traditional serotyping in terms of detection rate, accuracy, reproducibility and compatibility of the results between different laboratories (Harper et al., 2015; Ma et al., 2016). Therefore, our mPCR assay may be readily used in large-scale studies aiming at fish farming improvement. Indeed, this straightforward mPCR-based serotyping procedure might be very useful for a variety of non-academic purposes such as bacterial strain selection for vaccine development, selective breeding or epidemiological surveillance in fish farms.

TR: Substantial intellectual contribution throughout the study, data interpretation and manuscript preparation. EF-N, SC, and AL: Bacterial isolates management, DNA extraction and mPCR. ID and LM: Bacterial isolates management and serological tests. AC and PN: Participation in genomic data analysis and handling. TW and J-FB: Participation in data interpretation. ED: Substantial intellectual contribution throughout the study, gene mining, interpretation of data, manuscript preparation and responsible for acquisition of funding. All authors read and approved the final manuscript.

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.