Bacterial strains were selected from the culture collection of the National Reference Laboratory for Aquatic Animal Diseases (AQUACEN) of the Brazilian Ministry of Agriculture, Livestock and Food Supply. These S. agalactiae (n = 50), L. garvieae (n = 11), S. iniae (n = 47), and SDD (n = 23) strains were isolated during bacteriological analyses of outbreaks in Brazilian fish farms in different years and geographical locations (Table S1). The isolation of these microorganisms was performed on chilled fish that were sent to AQUACEN for diagnosis. Swabs from brains, kidneys or abscesses were aseptically sampled and streaked onto 5% sheep blood agar (SBA) for the isolation of bacterial pathogens. These plates were incubated at 28°C for 48 h. Finally, the identification of bacterial species was carried out as previously described (Mian et al., 2009; Netto et al., 2011; Figueiredo et al., 2012; Fukushima et al., 2017).

The isolates were thawed and streaked onto 5% SBA and were incubated at 28°C for 48 h. Isolates were incubated in a lysozyme solution at 37°C overnight. Bacterial DNA was extracted with a Maxwell 16 Tissue DNA purification kit (Promega, Madison, WI, USA) according to the manufacturer's instructions. The extracted DNA was quantified using a Nanodrop spectrophotometer (Thermo Scientific, Wilmington, DE, USA). The purity of the extracted DNA was determined using the absorbance ratio at 260/280 nm. Samples with ratio of 1.8 ± 0.5 were stored at −80°C until use.

The 16S rRNA gene was amplified by PCR with the universal primers B37 (5′-TAC GGY TAC CTT GTT ACG A-3′) and C70 (5′-AGA GTT TGA TYM TGGC-3′) and PCR amplicons were purified according to the method described by Fox et al. (1995) for all strains used in this work. The sequencing reactions were performed using a BigDye™ Terminator Cycle Sequencing Kit (Applied Biosystems, UK) and evaluated with an ABI 3,500 Genetic Analyzer (Life Technologies, USA). Forward and reverse sequencing products were used to generate contigs with the BioEdit software (Ibis Biosciences, Carlsbad, USA) version 7.2. Their identity was evaluated using the BLAST webserver (http://www.ncbi.nlm.nih.gov/BLAST) by checking against existing sequences in the nt/nr database. A similarity of ≥ 97% was considered as the same species in accordance with Nguyen et al. (2016) and Větrovský and Baldrian (2013).

All isolates were thawed and streaked onto 5% SBA and incubated at 28°C for 48 h. A fresh, single colony of each bacterial strain was spotted using a toothpick into a target steel plate. For each strain, 1 μl of formic acid (70%) and 1 μl of MALDI-TOF MS matrix, consisting of a saturated solution of α-cyano-4-hydroxycinnamic acid (HCCA) (Bruker Daltonics, Bremen, Germany), were applied to the spot and allowed to air-dry. Spectra were acquired using the FlexControl MicroFlex LT mass spectrometer (Bruker Daltonics) with a 60-Hz nitrogen laser, in which up to 240 laser shots are fired in spiral movements to collect 40 shot steps for each strain spot. Furthermore, parameters for mass range detection were defined to allow the identification from 1,960 to 20,137 m/z, where Ion source 1 v was 19.99 kv, Ion source 2 voltage was 18.24 kv and the lens voltage was 6.0 kv for data acquisition. Prior to measurements, calibration was preceded with a bacterial test standard (E. coli DH5 alpha; Bruker Daltonics). The Real Time (RT) identification score criteria used were those recommended by the manufacturer: score ≥ 2.000 indicates a species-level identification, score ≥1.700 and <2.000 indicates a genus-level identification, and a score <1.700 indicates no reliable identification. Comparisons between MALDI-TOF MS strain identifications and those of other techniques were performed with R software version 3.0.1 (R Core Team, 2013) with the agreement rates determined by the Kappa coefficient.

To identify possible S. iniae strains and to enhance the S. dysgalactiae discrimination at the subspecies-level in a MALDI Biotyper, Main Spectra Profiles (MSPs) were created with reference strains for each species. Fresh colonies of the S. iniae SI23 strain and the SDD SD64, SD92 and SD142 strains were extracted according Alatoom et al. (2011). Briefly, the strains were collected from the agar and added to 300 μl of distilled water, followed by the addition of 900 μl of ethanol. Two rounds of centrifugation for 2 min at 13,000 rpm and the complete removal of supernatant was necessary to obtain dried pellets. The pellets were suspended in 50 μl of formic acid (70%) and vortexed. Finally, 50 μl of acetonitrile was added and the mixtures were centrifuged for 2 min at 13,000 rpm. For assays, one microliter of the supernatant was spotted eight times onto a steel target. Directly after air-drying, each spot was overlaid with 1 μl of HCCA matrix. Each spot was measured three times with the same protocol/parameters described in the section above. The obtained spectra were closely analyzed in the FlexAnalysis software (Bruker Daltonics) to assess the high level of reproducibility. Finally, the spectra of each strain were uploaded to the MALDI Biotyper software version 3 (Bruker Daltonics) and assembled to generate a Main Spectra Profile (MSP) for the strains using the BioTyper MSP creation standard method. All steps were done according to the manufacturer's recommendations.

A figure illustrating the SD64 spectra was generated using R software version 3.0.1 (R Core Team, 2013), using data exported from the FlexAnalysis software (Bruker Daltonics). In addition, in order to compare the custom MSPs with the MSP preloaded on the Bruker MSP library, the BioTyper software version 3.0 (Bruker Daltonics) was used to perform a dendrogram analysis. The parameters used were distance measure = “correlation,” linkage = “average,” maximum number of top level nodes = “0,” score oriented dendrogram “enabled,” score threshold values for a single organism = “300,” and score threshold values for a related organism = “0.”

The SDD strains that had subspecies suggested by Costa et al. (2014) were inferred by a BLAST comparison of the 16S rRNA and sodA genes, and the MALDI Biotyper (Bruker Daltonics) analysis suggested a closer relationship with S. dysgalactiae subsp. equisimilis (SDE). In addition to the 16S sequencing described above, a Next-Generation Sequence (NGS) experiment was performed. Three strains (SD64, SD92, and SD142) with different pulse-field gel electrophoresis (PFGE) profiles described in previous work from our group (Costa et al., 2014) were sequenced. DNA from the SDD strain was isolated from an overnight culture using a Maxwell 16 tissue DNA purification kit using the Maxwell 16 system (both from Promega). Sequencing was conducted on the Ion Torrent Personal Genome Machine sequencing system (Life Technologies) using a 200 bp fragment library kit, according to the manufacturer's recommendations. The barcodes of the raw data were removed using an in-house script (https://github.com/aquacen/fast_sample), and assembly was performed using SPAdes v3.9.1 (Nurk et al., 2013).

SDD taxonomic classification was determined using the Jensen and Kilian (2012) method, where the analysis of the phylogenetic relationship of seven housekeeping genes (map, pfl, ppaC, pyk, rpoB, sodA, and tuf) through a multilocus sequence analysis (MLSA) represent an improved basis for the identification of clinically important streptococci. The concatenated sequence of these housekeeping genes is used to establish differences between species that allow a more accurate identification within the pyogenic group of streptococci. The sequence of the draft genome of SDD ATCC 27957 is available on GenBank (Accession number: CM001076) and together with the genes of 30 streptococci strains submitted with the work of Jensen and Kilian (2012) were downloaded (Accession numbers: map: JN632385 to JN632479; pfl: JN632290 to JN632384; ppaC: JN632195 to JN632289; pyk: JN632100 to JN632194; rpoB: JN632005 to JN632099; sodA: JN631910 to JN632004; tuf : JN631815 to JN631909).

To extract the sequences of the corresponding housekeeping genes, a homology search for each of the seven genes in the SD64, SD92, and SD142 strains was performed using the BLAST webserver (http://www.ncbi.nlm.nih.gov/BLAST), with contigs generated by assembly software. The same strategy was performed with the SDD ATCC 27957 strain. All genes for each strain were concatenated in the following order: map-pfI-ppaC-pyk-rpoB-sodA-tuf. Alignment and phylogeny analyses were performed using MEGA6 (Tamura et al., 2013), with the Kimura-2 model parameters, using the Minimum Evolution algorithm, and a bootstrap of 1,000 replications.

The sequences of the 16S rRNA PCR products, which were generated with the aforementioned forward and reverse primers, were comprised in contigs for each strain. The mean lengths of the contigs were 1,514 ± 12, 1,537 ± 14, 1,519 ± 15, and 1,515 ± 17 bp for S. agalactiae, L. garvieae, S. iniae, and SDD, respectively. The contigs from each strain were used as queries for the BLAST webserver, and a percentage value of the similarities for L. garvieae was between 98 and 100, whereas S. agalactiae, S. iniae and SDD varied between 97 and 100. For the SDD strains, it was not possible make identification at the subspecies-level. For each SDD isolate there were results referring to the SDE and SDD with the same percentage value of identity that referred to the same query coverage.

For each strain-spot, 1–3 spectra were expected, according to the manufacturer's instructions for quality assurance performed by MALDI Biotyper software of acquisition. For S. agalactiae, 64 spectra were acquired, whereas 11 spectra were acquired for L. garvieae. All strains for both species were identified at the species-level (score ≥ 2.000). The minimal and maximal scores for S. agalactiae were 2.083 and 2.377 (Table 1), respectively, and for L. garvieae were 2.081 and 2.218 (Table 2), respectively. For both species a perfect agreement (Kappa = 1; CI: 1.0–1.0; and p < 0.005) was observed between the 16S rRNA gene sequencing and MALDI-TOF MS techniques to identify the species.

A total of 52 spectra were obtained for the 47 strains. Identification of S. iniae was possible in ~53% of isolates at the genus-level (Table 3), and the minimal and maximal scores were 1.482 and 1.854, respectively, including 22 with no reliable identification. The genus-level was inferred by an approximation of the spectra with S. dysgalactiae (n = 7), S. equi (n = 1), and S. pyogenes (n = 17). The species identification agreement when comparing 16S rRNA gene sequencing and MALDI-TOF MS was poor (Kappa = 0.04; CI: −0.03 to 0.11; and p = 0.063).

To make possible the correct identification of S. iniae strains using the MALDI Biotyper, a custom MSP was created for this species (Figure 1; MSP available at http://www.renaqua.gov.br/aquacen-msp-si/). Twenty-four spectra were collected for one isolate (SI23) by the Biotyper RTC program. The spectra were analyzed in the FlexAnalysis software to identify a high level of reproducibility, and all spectra were used to create the MSP. A dendrogram generated in BioTyper software (Figure 2) shows the SI23 strain as a single leaf between the S. pyogenes and S. dysgalactiae clades. After the inclusion of the custom MSP of S. iniae, all the strains were identified at the species-level (Table 3), and the minimal and maximal score values were 2.013 and 2.426, respectively. A complete agreement between both tested techniques was observed (Kappa = 1; CI: 1.0–1.0; and p < 0.005) for species identification.

The identification of SDD isolates, using 25 spectra from 23 strains, was obtained by an approximation of S. dysgalactiae and SDE MSPs at the species-level. Minimal and maximal scores were 2.058 and 2.298, respectively. Of all the SDD strains, 13 were identified with proximity to the subspecies equisimilis, and in 10 strains, there was no discrimination of subspecies (Table 4). The agreement between techniques was perfect when considering the species-level (Kappa = 1; CI: 1.0–1.0; and p < 0.004), but when considering the subspecies-level the agreement was only fair (Kappa = 0.21; CI: −0.08−0.52; p = 0.075). This demonstrated that both techniques were unable to identify strains at the subspecies-level.

These strains, according to previous work of our group (Costa et al., 2014), are from SDD subspecies. Therefore, an NGS experiment was done to confirm the subspecies assignments. Contigs from the assembly of the strains SD64, SD92, and SD142 (data not shown) were used for a MLSA analysis. The three strains formed a clade with SDD from work of Jensen and Kilian (2012), confirming the classification of theses strains as SDD subspecies in accordance with the methodology used (Figure 3).

To improve the identification by the MALDI Biotyper, custom MSPs were created for SDD (Figure 4; MSP available at http://www.renaqua.gov.br/aquacen-msp-sdd/). Twenty-four spectra were collected for each isolate as described above and the spectra were analyzed in the FlexAnalysis, where all spectra were used to create the MSP. A dendrogram generated in BioTyper software (Figure 2) shows the SD64, SD92, and SD142 strains in an intra-species-specific clade of S. dysgalactiae spp. Figure 4 shows the common and exclusive peaks of custom MSPs and Bruker library MSPs, and, interestingly, the SDE DSM 23147^T shows 25 exclusive peaks.

After this inclusion (Table 4), all isolates matched with to the three included custom MSP for the three best matches (Table S2), with minimal and maximal scores of 2.277 and 2.579, respectively. The agreement between 16S rRNA gene sequencing and MALDI-TOF MS was poor (Kappa = 0.08, CI: −0.05−0.22; p = 0.050), considering that 16S rRNA gene sequencing was unable to identify subspecies, whereas with MALDI-TOF MS, they could be determined effectively. In contrast, considering the MLSA analysis, the agreement between this technique and MALDI-TOF MS was perfect (Kappa = 1; CI: 1.0–1.0; and p < 0.005).

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.