Three Reunionese strains of S. iniae were utilized in this study. The first was isolated in 2002 from a grouper (Variola louti). The second was isolated in 2014 from a striped large-eye bream (Gnathodentex aurolineatus) and the third, isolated in 2009 from diseased red drum (Sciaenops ocellatus). This last isolate was attributed as the cause of a fish kill in an offshore marine fish farm (Table 1). In addition, we utilized the type strain CIP 103769^T. Figure 1 pinpoints the location of the epizootics investigated on Reunion Island.

A further 85S. iniae genomes, (76 obtained from fish, two from dolphins, and seven from humans), representing all sequencing data available to date, were also used to assess the performance of the MLST for identifying origins of the three Reunionese strains. The sequences were available on Genbank (genome accession numbers in Table 1) or provided by co-authors (AB and OS). These isolates were temporally (isolated between 1976 and 2016) and geographically diverse, obtained from different countries including North and Central America, Australia, Asia and Israel (Table 1). The collection of isolates has been previously confirmed as belonging to S. iniae using the specific PCR protocol described by (Mata et al., 2004). Details regarding the origin of each isolate are provided in Table 1.

The discriminatory power of the MLST scheme was measured using the Simpson’s index (Hunter and Gaston, 1988). This index measures the probability for two randomly sampled strains to present different STs. 95% confidence interval for this index was calculated using the Comparing Partitions website.¹ The standardized index of association, I_A^S, was calculated both on the complete data set (89 strains) and on a subset representing each ST only once (13 STs) using LIAN 3.7 (Haubold and Hudson, 2000).² I_A^S is used to evaluate the role of recombination in population evolution. A value of zero corresponds to frequent recombination events while clonal populations are identified by an I_A^s value significantly differing from zero (Smith et al., 1993). LIAN 3.7 was also used to calculate the genetic diversity (H) at each locus, analyzed as a measure of their expected genetic variability. This value ranges from 0 (no diversity) to 1. The number of variable nucleotide sites, non-synonymous (d_N) and synonymous substitutions (d_S) as well as Z test of selection on coding regions were calculated using Mega7 (Kumar et al., 2016). The Z test evaluates the null hypothesis of strict neutrality (d_N =d_S) versus the alternative hypotheses of purifying (d_N <d_S) or positive (d_N >d_S) selection (Nei and Gojobori, 1986) and was performed on coding regions only, using 1,000 bootstrap replications and Jukes-Cantor adjustments.

DNA sequences of the alleles determined in this study for each locus were deposited on Genbank under the accession numbers MN327582 to MN327612. Accession numbers for each allele of the MLST are provided in Table 3. The scheme is curated and available to the public via pubMLST (Jolley et al., 2018) at: https://pubmlst.org/organisms/streptococcus-iniae/

A total combined sequence length of 4,010 nucleotides was obtained when concatenating all eight housekeeping loci. Fragment lengths ranged from 404bp (dnaN) to 582bp (rnhC). Considering all 89 strains, the number of alleles for a locus varied from three (mutX and yfhQ) to five (recD2 and mutL) and the number of polymorphic sites from four (dnaN, mutX and rnhC) to 15 (mutS). The genetic diversity, estimated by the H index, reached a minimum of 0.0447 (rnhC) and a maximum of 0.3639 (mutS). The main characteristics of genetic diversity for each locus are summarized in Table 2.

For most genes, synonymous substitutions (d_S) were more frequent than non-synonymous substitutions (d_N). Only mutS and rnhC had more non-synonymous substitutions than synonymous substitutions but this result was not statistically significant (Table 2). The higher number of synonymous substitutions over non-synonymous substitutions could indicate that these loci are under neutral or purifying selection. However, the Z test for purifying selection was not significant for any loci. LIAN analysis of linkage disequilibrium based on the eight housekeeping genes suggested that the bacterial population analyzed in this study is in linkage disequilibrium when considering all 89 strains (I_A^S=0.1348; p<0.01), or a subset consisting of the 13 unique STs (I_A^S=0.1839; p<0.01). The discriminatory power of this MLST estimated by the Simpson’s index was 0.761 with a confidence interval of 95% between 0.677 and 0.845.

The 89 strains analyzed in this study were grouped into 13 different sequence types (STs) and two clonal complexes (CCs; Figure 2). From the 13 STs, three (ST1, ST5 and ST12) were represented by a single isolate. ST1 corresponds to one of the non-fish hosts, an isolate obtained from a dolphin (Inia geoffrensis) in 1978. This ST was the most genetically distant and shared no alleles with any of the other 12 STs (Table 1 and Figure 2). ST5 was obtained from a giant snakehead (Channa micropeltes) isolated in Thailand in 1983 and ST12 from a red drum (Sciaenops ocellatus) in an offshore aquaculture facility in Reunion Island. The other STs comprised between two and 41 isolates each. CC1 was represented by sixty-eight isolates (76,40% of included isolates) and CC2 by nine isolates (23.60% of included isolates). Within CC1, 41 isolates (46.07% of the collection) belonged to ST4, which is considered by goeBURST analysis as the ancestral type or founder of CC1. As such, ST4 is the most common S. iniae ST in our collection and has been isolated from multiple geographical origins (United States, China, Taiwan, Australia, Reunion Island) over a large period of time (1976 to 2016). This ST has been found associated in our study to different fish species and a dolphin. CC2 comprised two distinct STs: ST10 (seven isolates) and ST9 (two isolates). ST10 is represented by vaccine escape strains sampled in Australian barramundi farms in 2009, considered as hypermutators (Barnes and Silayeva, 2016) whereas ST9 included the only Canadian isolate from fish and one of the two Thai isolates, again obtained from fish.

When the CC definition was relaxed (6/8 shared alleles), all STs except ST1 formed a single network (Figure 2) indicating close evolutionary relationships between the different STs. However, some STs seemed to be specific of a geographical area, such as ST7, which is found only in Queensland. Further, ST12 and ST13 have a limited geographical distribution (Reunion Island and Israel, respectively) and could be considered as descendants of ST11, which included earlier freshwater aquaculture isolates from these regions (trout and tilapia), as well as the first isolate involved in a reef epizootic in 2002 in Reunion Island. Altogether, despite a strong bias toward a few fish species, no clear host specificity could be detected, with isolates retrieved from mammals sharing STs isolated from fish as well (ST2, ST3, ST4, ST6). For instance, ST2 and ST3 were represented by fish and human isolates from North America, ST4 by fish and dolphin isolates, while ST6 included bat and ornamental fish isolates (Table 1).

Considering the 89 strains tested within our study, the MLST scheme produced a Simpson’s Index of Diversity (SID) of 0.761, i.e., two strains randomly sampled in our collection have an average probability of 0.761 to display different STs. The genetic diversity estimated by the SID is well below 0.90, considered as the desirable threshold to interpret typing results with confidence (Hunter and Gaston, 1988). The observed lack of diversity in the S. iniae strains assed in this study is in keeping with other studies using MLST and indicates that it may be hard to find sufficient for highly-descriminatory MLST variation in S. iniae core genome genes, despite the remarkable variability in pangenome including virulence and antigenicity regions such as capsule biosynthesis genes (Lowe et al., 2007; Heath et al., 2016; Zeng et al., 2016). As a result of this, one major advantage of the use of MLST over say Multiple-Locus Variable number tandem repeat Analysis (MLVA) for example, is the use of these slowly evolving housekeeping genes (Jenke et al., 2011). This allows researchers to more reliably deduce long-term evolutionary events which may occur. However, that said, MLVA might still have higher discriminatory power and be useful to distinguish closely related serotypes or to monitor vaccine escape strains as and when needed.

Although we are unable to ascertain the exact origin of the S. iniae strains responsible for the epizootics on Reunion Island we are able to generate a number of hypotheses from our data. Indeed, the strains isolated during the different epizootics on Reunion Island belonged to different STs. This pattern does not support the emergence of a specific clone adapted to the local environment setup. Early strains isolated in 1996 (from a freshwater aquaculture facility) and 2002 (one of the reef fish epizootics) both belong to ST11, which also comprises three Israeli S. iniae strains isolated in 1989, 1998 and 2000 from Tilapia and rainbow trout aquaculture facilities (Table 1 and Figure 2). The strain isolated in 2009 on Reunion Island was obtained from diseased red drums in an offshore farm and belonged to ST12, itself a direct descendant of ST11. This suggests a common origin of these early Reunionese strains with Israeli aquaculture strains, followed by a local diversification. Even though these two countries are geographically distant, the pathogen could have been introduced to the Island via carrier fish import for aquaculture development on Reunion Island and/or aquarists. Israel was indeed, in the early 1990s, a major exporter of tilapia breeding lines for the growing tilapia industry worldwide (Popma and Lovshin, 1996). Unfortunately, no information regarding the origin of the diseased rainbow trout from the 1996 epizootic in Reunion, or of other fishes coexisting with them at that time in the farm, such as Tilapia, was available to confirm this hypothesis. The likely freshwater origin of the strain isolated during the 2002 epizootic is in keeping with the ecological plasticity of S. iniae, able to survive and adapt from a fresh water aquaculture facility around 900 meters above sea level to coastal environments, and across several host species. The example of ST8 and the probable spread of S. iniae in Australian barramundi, from freshwater aquaculture to sea cages tends to confirm that this hypothesis is plausible. Transmission of S. iniae from caged to wild fish has been suggested in the past (Colorni et al., 2002), nevertheless, to our knowledge, it is the first report of the possible spread of a pathogenic strain from freshwater aquaculture to reef fish. It is important to note that trout from the Reunionese facility affected by S. iniae in 1996 were later released into the rivers surrounding the farm in order to develop recreational fishing (Fédération de la pêche de La Réunion). As the first reef fish epizootic in 2002 originated from the same ST (ST11) as the one affecting trout from the freshwater aquaculture facility in 1996, this underlines the importance of developing robust and sensitive molecular tools to screen animals for pathogens before any release into the wild.

In contrast, the S. iniae strain involved during the last 2014 epizootic was identified as different from those isolated in 1996, 2002 and 2009 and matched the more ancestral and cosmopolitan strain, ST4. A double reversion of MutL and MutM would be needed to obtain the haplotype found in 2014 from the one found in 2009. This seems highly unlikely considering the very low mutation rates in DNA repair genes MutL and MutM (Denamur and Matic, 2006). As a consequence, we propose that this last epidemic results from the local emergence of a S. iniae lineage of cosmopolitan distribution.

Only 13 different STs belonging to two CCs were found within our collection of 89 strains. Moreover, a unique ST (ST4) representing 46.07% of our sample was found on all continents (North America, Australia, Asia). Clearly, the choice of critical functional genes as MLST loci will constrain diversification, as they are likely under strong purification selection. Consistently, the genetic diversity of housekeeping loci included into the scheme was low. However, this lack of variability might result from a sampling bias towards pathogenic strains from aquaculture. Indeed, it should be noted that further work on this topic including a greater range of samples may show a greater diversity in housekeeping genes. Of the 89 isolates examined, 62 are from only two species (tilapia and barramundi) and four locations. 48 originate from Australia and found in Barramundi, and 14 are from Tilapia (from Israel, United States and Hondura). That leaves only 13 strains originating from different fish species. We were also only able to map these to seven from human origin and two from dolphins. Therefore, the samples may be biased in both the host species and geographical distribution. We also did not include non-virulent strains or those solely associated with the environment as these were not available. It would be interesting to include more non-pathogenic strains in order to determine whether this lack of diversity is specific to pathogenic strains or common to environmental and/or avirulent strains, less subject to host immune selection. However, at the moment, it is still hard to isolate environmental strains. Indeed, confirmation of isolate identity as S. iniae by commercial bacterial identification kits is still problematic due to the biochemical profile being absent from databases supplied with the kits or variability of some atypical S. iniae strains (Roach et al., 2006). Moreover, most of the epizootics caused by this species occur in developing countries and remain underreported. Nevertheless, atypical strains isolated from mammals were almost always found in an ST also including strains isolated from aquaculture or ornamental fish. As such, increasing the number of host species may not necessarily increase S. iniae’s diversity and the results obtained in this study may constitute an informative baseline to be completed in the future to reassess the diversity of this pathogen on a broader host and geographical range.

Alternatively, we suggest that the low diversity of housekeeping genes may be due to a recent speciation of S. iniae, as hypothesized for other streptococci such as S. thermophiles (Delorme et al., 2010) or S. agalactiae (Brochet et al., 2006), both also known fish pathogens in their own right (Pereira et al., 2010; Bowater et al., 2012). As far as S. agalactiae is concerned, the commonly used MLST of Jones et al. (2003) was not sufficiently resolutive to define the population structure of the pathogen in aquatic hosts, and new typing systems were developed, including combination of MLST (sequencing of up to 15 housekeeping genes), serotyping and presence/absence of virulence associated genes (Sørensen et al., 2010; Delannoy et al., 2013; Godoy et al., 2013). These combined approaches might allow resolving the main phylogenetic events, but are time consuming and costly. The low diversity observed herein may also result from a recent bottleneck history, reducing the size and diversity of S. iniae populations. In many instances, genetically monomorphic pathogens have undergone such bottlenecks, for example following a crucial genetic event, causing a change in ecological niche (Achtman, 2008).

The low diversity observed in the MLST seems associated to a great ecological flexibility of this pathogen. Our data confirms that S. iniae does not appear to have any major host specificity (Agnew and Barnes, 2007). Although most of our sampling originates from farmed fish, where it causes the highest mortalities, the few available STs obtained from infections in mammals (including humans) did not cluster together but rather with STs obtained from other fish isolates (Table 1; ST2, ST3), suggesting transmission from farmed fish to humans (Weinstein et al., 1997; Facklam et al., 2005). Genomic data indicate that a host jump might have been facilitated by rapid mutations (Silayeva et al., 2020). This ecological flexibility, underlined by the apparent lack of host specificity of S. iniae, is further emphasized by the ability of a particular clone to survive both in fresh and seawater. As already discussed, ST11 groups with early strains isolated from freshwater aquaculture in Israel and Reunion Island as well as the first strain responsible for a reef fish epizootic on Reunion Island. ST8 is another example of ST genotyped from barramundi freshwater farms in Western Australia (2004) and from sea cages in the Northern Territory (2005–2006; Table 1). This ST has been associated with important fish loss since 2004 in these two regions where S. iniae was not described before (Creeper and Buller, 2006). Although both Western Australia and Northern Territory strains were genotyped as ST8, Nawawi et al. (2009) showed that Northern Territory isolates were further characterized by a mutation on the lactate oxidase gene (lctO) and able to process lactate at a faster rate than the other strains (Nawawi et al., 2009). This change has been attributed to environmental influences of large tidal flows linked with increased swimming activity of the barramundi hosts in seawater, that have led to the evolution of the lctO gene variant, encoding a more efficient enzyme in these specific isolates (Nawawi et al., 2009). Although our MLST does not highlight this kind of rapid discrete mutation, it does enable retracing the probable common origin of isolates from different locations. Further, MLST enables the exploration of ecological flexibility of S. iniae, indicating the ability to switch to various hosts and habitats (freshwater vs. seawater).