MALDI-TOF MS and genomic analysis can make the difference in the clarification of canine brucellosis outbreaks

Brucellosis is one of the most common bacterial zoonoses worldwide affecting not only livestock and wildlife but also pets. Canine brucellosis is characterized by reproductive failure in dogs. Human Brucella canis infections are rarely reported but probably underestimated due to insufficient diagnostic surveillance. To improve diagnostics, we investigated dogs in a breeding kennel that showed clinical manifestations of brucellosis and revealed positive blood cultures. As an alternative to the time-consuming and hazardous classical identification procedures, a newly developed species-specific intact-cell matrix-assisted laser desorption/ionization–time of flight mass spectrometry analysis was applied, which allowed for rapid identification of B. canis and differentiation from closely related B. suis biovar 1. High-throughput sequencing and comparative genomics using single nucleotide polymorphism analysis clustered our isolates together with canine and human strains from various Central and South American countries in a distinct sub-lineage. Hence, molecular epidemiology clearly defined the outbreak cluster and demonstrated the endemic situation in South America. Our study illustrates that MALDI-TOF MS analysis using a validated in-house reference database facilitates rapid B. canis identification at species level. Additional whole genome sequencing provides more detailed outbreak information and leads to a deeper understanding of the epidemiology of canine brucellosis.


Results
Serological examination of dogs with reproductive failure. In 2013, the pregnancy of a female dog in a breeding kennel in São Paulo, Brazil, comprising 17 adult pugs, 2 males and 15 females, resulted in abortion. Another 5 pregnancies ended in abortion in the following four months (Fig. 1A). To clarify whether the observed abortions were a consequence of B. canis infections, all animals were clinically examined. While three dogs showed no clinical manifestations related to canine brucellosis, 14 dogs suffered from at least one brucellosis-like symptom (Fig. 1B). Vaginal discharge and general lymphadenopathy were the most common clinical manifestations, followed by abortion/stillbirth and infertility (defined as inability to get pregnant and produce viable offspring) (Fig. 1B).
To test whether the animals had developed an immune response against Brucella antigens, we collected serum samples from each dog at three consecutive time points ( Fig. 1A and Supplementary Table S1) and subjected them to classical serological tests, namely the Immunochromatographic Test (ICT), the Rapid Slide Agglutination Test (RSAT) and a B. canis IgG ELISA. In all 17 dogs, we detected antibodies towards rough Brucella antigens at each sampling date with at least one of the three conducted serological tests (Fig. 2). In several cases, serological results were equivocal, because the tests revealed contradictory results. For example, the serum from the bacteremic dog D09 did not consistently react against the rough antigen of Brucella in the ICT at every sampling date, but in RSAT and ELISA. At the same time, when 2-mercaptoethanol (2ME-RSAT) was applied, which inactivates IgM in the RSAT, we could not detect any antibody reaction in the serum of this dog (Fig. 2).

Phenotypic characterization of Brucella isolates.
To assess the infection status of the kennel and to check whether the disease symptoms were indeed a result of an infection with B. canis, blood samples were taken from each individual animal at three different time points and aliquots were plated on Brucella-selective agar after enrichment broth culture. Gram-negative, short rod-shaped, non-motile bacteria were isolated from 15 out of 17 dogs at least once. A total of 29 isolates from 51 blood samples were obtained ( Fig. 2 and Supplementary  Table S1) and identified by genus-specific PCR as Brucella. Subsequently, genomic DNA was extracted from the 22 blood samples with negative culture results. In three of these blood samples (13.6%) the IS711 marker gene of Brucella could be amplified by PCR ( Supplementary Fig. S1), which suggests Brucella infection in these dogs although the isolation of the bacterium had failed.
The isolated bacteria were further characterized with phenotyping methods commonly used for the identification of the genus Brucella and subtyping of its species and biovars (Supplementary Table S2). The isolates did not require CO 2 for growth and could be cultured in ambient air (21% O 2 ) on blood agar as well as on Brucella Agar with or without serum. The bacteria showed oxidase, catalase and urease activity but no H 2 S production similar to the reference strain B. canis RM 6/66. Moreover, the isolates grew on thionine and basic fuchsin-supplemented Brucella Agar plates comparable to B. canis RM 6/66. Additional phage typing experiments revealed the same lysis patterns for the isolated bacteria (constantly lysed by phage R/C, but not by any other tested phage) and B. canis RM 6/66 ( Fig. 3 and Supplementary Table S3). The isolates did not agglutinate with anti-A and anti-M but with anti-R sera, specific for Brucella species expressing rough lipopolysaccharide (LPS). Likewise, LPS phenotyping with crystal violet staining revealed that all isolates expressed a rough LPS, which is characteristic of B. canis and B. ovis.
Taken together, the classical microbiological methods clearly diagnosed an infection of the dogs with Brucella and suggested B. canis as the disease-causing agent.  32 , but could be clearly distinguished from B. suis bv 1 (Fig. 4B), which is one of the frequently found Brucella species in dogs apart from B. canis. By visual comparison of the MALDI-TOF MS spectra we were able to identify additional discriminant biomarker signals for B. canis and B. suis besides the previously described biomarkers at m/z 5833 and m/z 9076 specific to canis-vs-suis bv 1 32 , corresponding to m/z 5834 ( Supplementary Fig. S3A) and m/z 9075 (Fig. 4C), respectively. Both peaks represent single charged molecules and we could also detect their corresponding double or triple charged ions at m/z 2915, m/z 4539 and m/z 3024 (Supplementary Fig. S3B and Fig. 4D,E), respectively. In particular, the peak signals at m/z 9109 (single charged) and m/z 4555 (double charged) occurred prominently in B. suis bv 1 but not in B. canis or B. suis bv 4.

MALDI-TOF MS-based identification of the canine
Brucella canis and B. suis bv 4 share a unique biomarker at m/z 7073 allowing for distinction from B. suis bv 1 (Fig. 4F) and other B. suis biovars ( Supplementary Fig. S4). Further comparison of spectra revealed many differences in peak intensities rather than a strict peak absence or presence. Hence, the peaks at m/z 7661 (single charged) and m/z 3830 (double charged) can be used for the discrimination of B. canis and B. suis bv 4 against B. suis bv 1 (Fig. 4G,H) as well as other B. suis biovars ( Supplementary Fig. S4). While an accurate and unambiguous distinction between B. canis and B. suis bv 4 by MALDI-TOF MS has been difficult in previous studies 32 , the isolates used in this study exhibited two signals at m/z 5900 and m/z 3926 that were mainly found in B. canis but not in B. suis bv 4 strains (Fig. 4I,J). Genomic and phylogenetic characterization of the Brucella canis isolates. Besides the serological, biochemical and MALDI-TOF MS-based identification tools described above, molecular methods like ribotyping, pulsed-field gel electrophoresis (PFGE), multilocus sequence typing (MLST) or multiple locus variable number of tandem repeats (MLVA) analysis are used for subtyping of Brucella spp. [35][36][37] and for the differentiation of B. canis subclades 38 . As a consequence of the high genetic homology among Brucella species 39 , we performed next-generation sequencing (NGS) analyses for a more sophisticated phylogenetic characterization of the Brucella isolates in our study. Analyzing the assembled DNA sequencing short reads clearly identified all isolated bacteria as B. canis with an average draft genome size of 3,295,525 bp and an average G + C content of 57.27% ( Fig. 5 and Supplementary Table S5). Excluding lower quality sequences, the applied short read sequencing technique allowed de novo genome assemblies with 24 to 57 contigs. Repetitive elements like IS711, ISBm2 and ISBm3, IS1953 and IS2020 insertion sequences or rRNA operons hampered the completion of contiguous, circular chromosome sequences, a typical drawback of Illumina and other next-generation sequencing techniques based on short read lengths 40 .
Detailed genome sequence comparisons revealed a high clonality, with only seven SNPs among the B. canis isolates (Supplementary Table S6). Apart from isolate I16, which harbored mutations at two different positions, five B. canis isolates (I5, I8, I19, I22, I23) differed only by one nucleotide within their genome sequences, which strongly indicates a clonal origin of these bacteria. In four cases, nucleotide variations were identified in Brucella isolates from dogs at later sampling dates (Supplementary Tables S1 and S6). Two SNPs, A2198519T in I8 and C642523T in I23, may be the result of post-isolation, lab-acquired mutations or an artefact of the low sequencing coverage (5 ×) for these nucleotide positions. To create an improved consensus genome, sequence reads of four high-quality draft genomes without SNP deviations among the Brucella isolates were concatenated by SPAdes de novo assembly. The resulting B. canis BfR-SPBR-consensus genome consists of 23 contigs (≥ 500 bp) Figure 2. Blood culture and serological test results of dogs from a kennel with a suspected outbreak of canine brucellosis. To determine whether the 17 dogs from the kennel under study were infected with Brucella, whole blood samples were incubated in selective medium (BC). Additionally, dogs were serologically tested using an immunochromatographic test (ICT), the rapid slide agglutination test (RSAT), the 2-mercapthoetanol rapid slide agglutination test (2ME-RSAT) and an enzyme-linked immunosorbent assay (ELISA). Positive test results are shown in dark and negative ones in light color. A few tests gave inconclusive results (i) or were not conducted (nc); ID: identification code of the dogs; G: gender (M: male, F: female); S1, S2, S3: sampling dates in March, August and November 2014, respectively; * no serum sample available. Sequence variants between the B. canis consensus genome described here and the B. canis ATCC 23365 genome were determined by applying the SNP calling strategies of BioNumerics and Mauve/ParSNP. The Bio-Numerics read mapping approach revealed 179 positions with high sequence read coverage and five positions with low sequence read coverage SNPs, whereas the analysis with Mauve identified 183 SNPs in the assemblybased approach. Analyses with these two different bioinformatics tools identified 180 SNP positions in common that distinguish the Brucella isolates found in dogs of the kennel from the reference strain Brucella ATCC 23365 (Fig. 5). Besides various SNPs in non-coding regions of B. canis BfR-SPBR-consensus, 57 and 47 SNPs were identified in coding regions of chromosome 1 and chromosome 2, respectively.
To illustrate the relationship of B. canis BfR-SPBR-consensus with previously sequenced strains, we performed a core genome multi-alignment with ParSNP using all publicly available B. canis draft or complete genome sequences and included the B. suis bv 4 reference strain 40 as the outgroup (Fig. 6). This analysis clearly placed the B. canis BfR-SPBR-consensus in a clade together with other Brazilian B. canis isolates but also with strains from neighboring countries. The most closely related strains to B. canis BfR-SPBR-consensus were B. canis 10469 and B. canis 07-2859-6070, also isolated from kennels in São Paulo in 2005 and 1998 41 , respectively, whereas the genome sequences of B. canis strains from Chile and Colombia revealed more genetic variations. Interestingly, the human B. canis CNGB 1324 isolate from Argentina clustered within the group of Brazilian B. canis isolates from dogs (Fig. 6).  www.nature.com/scientificreports/ Detailed comparison of the SNP distribution in B. canis BfR-SPBR-consensus and other South American B. canis strains found several conserved missense and synonymous mutations in coding regions, which were absent in B. canis ATCC 23365 (Fig. 7). SNPs leading to stop-loss mutations in genes were conserved in most of the South American B. canis isolates as well (Table 1). In contrast, SNPs resulting in stop-gain mutations were less frequent and occurred especially in B. canis strains from Colombia and Chile (Table 1). The described stop-loss and stop-gain mutations affected genes with physiological functions like nutrient utilization but also potential classical virulence traits such as an autotransporter protein of the type V secretion family, including adhesins, invasins, toxins and proteases, found in various bacterial pathogens 42 .
Summarizing, our genetic analysis clearly reveals the clonality of the Brucella isolates from the diseased kennel in São Paulo and suggests a single-strain outbreak. Moreover, we identified SNPs in the South American B. canis isolates that may indicate diverse phenotypic properties of representatives within this subclade in spite of their close genetic relationship.

Discussion
Infections of dogs with B. canis and B. suis may be asymptomatic or lead to similar clinical manifestations in the infected animals like reproductive failures due to infertility, abortion, stillbirth, orchitis, epididymitis and prostatitis 10,43,44 . In recent years, not only dog-to-dog and pig-to-dog transmissions of B. canis and B. suis, respectively, have been reported, but both species have also been transmitted from dogs to humans. Hence, canine brucellosis mainly caused by B. canis but also by B. suis is an emerging zoonosis worldwide 18,26,27,[45][46][47][48] . Human infection may lead to various unspecific symptoms such as prolonged fever, shivering, chills, night sweats, weight loss, overall weakness or malaise, headache, enlarged lymph nodes, back pain, and arthralgia 10 . Therefore, fast and reliable laboratory methods for the identification of canine Brucella species and for epidemiological investigations on single strains are essential to prevent the spread of infection among dogs and transmission to humans.
For this purpose, we examined a potential brucellosis outbreak in a kennel of 17 dogs by comparing the traditional Brucella diagnostic tools with the latest proteomics-and genomics-based methods. Fourteen out of 17 animals exhibited clinical manifestations associated with canine brucellosis. One female and two male dogs were asymptomatic, which raised the question of whether these animals were subclinically infected (healthy carriers) or not infected. Subsequent blood cultures allowed us to isolate Brucella from two out of the three dogs confirming previous reports that infected dogs do not necessarily develop disease-specific symptoms 28,43 . Therefore, the isolation of bacteria is considered the gold standard method and the most reliable way to confirm canine brucellosis. Moreover, bacterial isolates are required for phenotypic characterization 49,50 in order to differentiate between the Brucella species potentially infecting dogs 8 . By performing the microbiological and biochemical tests commonly used for the classification of Brucella spp., we were able to identify all bacteria isolated from the kenneled dogs as B. canis. The phenotypic traits of our isolates were consistent with those of the reference strain B. canis RM6/66.
In about 40% of the analyzed blood samples, especially from later sampling dates, the isolation of bacteria was unsuccessful although Brucella could be recovered from preceding blood samples of respective animals. Obviously, diagnosis by culture may produce false-negative results, particularly in the later stage of infection when bacteremia ceases or becomes intermittent 51,52 . Although we were unable to isolate Brucella from the blood of several infected dogs at later sampling dates, we could still detect Brucella DNA by genus-specific PCR analysis, which further confirms that these dogs had been infected with Brucella. Our results verify previous observations that blood culture may be insufficient to diagnose canine brucellosis, especially since attempts to recover B. suis bv 1 from blood and urine had failed while bacteria could be isolated from the semen of dogs with clinical manifestations of brucellosis 53 .
Because of the drawbacks of blood culture for diagnostic purposes, serodiagnosis is commonly used to detect Brucella infections in animals. Serological approaches take advantage of seroconversion in infected hosts and detect host antibodies that react with Brucella antigens. These anti-Brucella antibodies can still be detected several years after the acute stage of infection 28,43,54 . Our serological screening of the 17 dogs using ICT, ELISA and RSAT suggested that all animals were infected with Brucella, even the two dogs with negative blood culture results at any sampling date. However, the results of the three serological tests were inconclusive in various cases, since some serum samples with positive ELISA results showed negative results with ICT or RSAT. Moreover, we observed negative serological test results in samples from which Brucella had been isolated. Hence, our analysis confirmed previous studies on the serological diagnosis of canine brucellosis demonstrating that although the detection of antibodies against rough Brucella species is widely used, misdiagnosis due to sensitivity and specificity failures may occur when these tests are applied 43,[54][55][56][57][58] . False-negative results can be a consequence of testing during the initial phase of infection prior to seroconversion, or of low levels of circulating antibodies in chronically infected dogs, after bacteremia has ceased 51,59 . Serological titers may also wax and wane during bacteremia leading to false-negative results 28 . False-positive results, for instance, may occur as a consequence of cross-reactivity of anti-Brucella antibodies directed against other pathogens. For B. canis harboring a rough LPS, cross-reactivity with Streptococcus, Staphylococcus, Bordetella, and Pseudomonas has been described, whereas the food-and waterborne pathogens Escherichia coli O157:H7, Yersinia enterocolitica O:9, Salmonella Typhimurium (group N; O:30), and Vibrio cholerae O1 induce immune responses that generate antibodies reacting against the smooth LPS of B. suis 43,[60][61][62] .
The species identification of our Brucella strains as B. canis with traditional microbiological, serological, biochemical and phage typing methods required several working days. Therefore, one focus of our study was to compare MALDI-TOF MS, as a more reliable and faster identification and differentiation tool for Brucella spp., with the classical phenotyping tests. Previous studies have shown that the application of MALDI-TOF MS with the currently available public libraries is a highly sensitive and specific technique for genus identification 33 www.nature.com/scientificreports/ database containing reference spectra from closely related species, which, in our opinion, is a prerequisite for the reliable identification of B. canis and discrimination from B. suis. Our here presented weighted pattern matching approach overcomes this shortcoming, and its ability to identify and distinguish B. suis bv 1 from B. suis bv 4 and B. canis provides a valuable benefit for the diagnosis of canine brucellosis. The discriminatory power of MALDI-TOF MS can also be of public health importance because the transmission of B. suis has been reported not only from feral pigs and hares to dogs but also from feral pigs to domestic pigs and to humans 18,27,66,67 . The implementation of our optimized MALDI-TOF MS identification approach for B. canis and B. suis will speed up the specific diagnosis of canine brucellosis as described for the new diagnostic routines with other pathogens 68 . When we complemented our MALDI-TOF MS analysis with whole genome sequencing data of Brucella isolates from the different dogs of the kennel, we were able to determine that the diverse clinical presentations of the animals were indeed the result of a B. canis single-strain outbreak. A detailed SNP comparison of the consensus sequence derived from our B. canis isolates with published B. canis genome sequences clustered the outbreak clone with various strains from South America comprising a clade that was clearly distinguishable from B. canis isolates from other parts of the world 41 . The genetic proximity of the Brazilian and Argentinian B. canis strains indicates that the infection circulates in kennels by cross-border trade of dogs, which may finally lead to a high risk for public health due to the close contact of humans with dogs and the lack of surveillance for canine brucellosis. The close genetic relationship of certain B. canis strains isolated from dogs and humans hints to a yet underestimated zoonotic transmission of this pathogen. Such transmission was recently documented with the infection of a 3-year old child by the same B. canis strain found in the blood of the child's puppy 46 .
The close phylogenetic relationship of the South American B. canis strains based on the SNPtree analysis suggested that they represent a homogenous group with similar properties. However, our detailed and genome-wide analysis of single nucleotide mutations revealed that various SNPs affect coding regions either resulting in gene inactivation or activation of pseudogenes. Consequently, this genetic microdiversity might lead to more pronounced functional differences than their genetic homology may imply. Future studies have to address the extent to which such phenotypic differences exist and how they might affect the virulence of specific B. canis strains.  (6), SCL (7), Oliveri (8), CNGB 1172 (9) and ATCC 23365 (10). SNPs are shown as colored strokes in red (stop-loss variants), in light pink (stop-gain variants), orange (missense variants), blue (upstream or downstream variants) and green (synonymous variants). www.nature.com/scientificreports/ Our study supports previous notions 7 that B. canis is prevalent in Brazilian breeding kennels and can be readily spread. These cases highlight the need to assess potential public health risks associated with the trade of kennel dogs or the handling of hunting dogs. In this context, new surveillance and control measures are indispensable to protect animal and human health. The rapid identification of the pathogen involved in kennel outbreaks is an important step to prevent the spreading of canine brucellosis. When the infection is confirmed in a dog population, the detailed characterization of the Brucella strains is required to trace back the infection. Since the members of the genus Brucella show high DNA sequence identity and clonal evolution, high-resolution typing methods are necessary to perform epidemiological analyses. Our study illustrates that proteomics and genomics approaches are a necessary complement to the traditional tests used in Brucella diagnostics because these novel techniques may speed up the reliable diagnosis of canine brucellosis.

Methods
Serodiagnosis of canine brucellosis. Blood sampling from dogs was performed according to a protocol approved by the Ethics Committee of the Faculty of Veterinary Medicine and Animal Science at the University of São Paulo, Brazil, under protocol CEUA 3113091015/2015, and in accordance with the relevant guidelines and regulations of the National Council for Control of Animal Experimentation (CONCEA). All blood samples were collected after mutual consensus between the owner of the kennel and the veterinarians responsible. Blood samples (~ 5 ml) were aseptically taken by cephalic or jugular venipuncture from dogs in a kennel located in São Paulo, Brazil. Half of the collected blood volume was mixed with sodium citrate as an anticoagulant. The remaining blood clotted at room temperature, was centrifuged and serum was stored at − 20 °C until use in serological tests. To test for anti-B. canis antibodies, we performed immunochromatographic tests (ICT; Rapid Canine Brucella Ab Test Kit, Bionote, Hwaseong, South Korea), ELISA (Novateinbio Kit, Cambridge, MA, USA) and rapid slide agglutination tests (RSAT; Canine Brucellosis Antibody Test Kit, D-TEC CB, Kansas City, MO, USA) with or without 2-mercaptoethanol (2ME) according to manufacturers' instructions. The 2ME-RSAT was only conducted on serum samples tested positive by RSAT.

Isolation of B. canis from blood culture and phenotypic characterization. Full blood samples
containing sodium citrate were used for culture as previously described 69 . Briefly, enrichment culture was performed in tryptose phosphate broth (Difco) including 5% fetal calf serum (FCS; Cultilab) at 37 °C for 30 days, followed by subcultures every four days on solid tryptose agar, also supplemented with 5% FCS. The isolated bacteria were primarily screened to confirm Brucella spp. colonies using the genus-specific PCRs targeting IS711 70 and bcsp31 71 . All Brucella isolates collected on the first and second sampling date (Supplementary Table S1) were further characterized using morphological, biochemical and metabolic tests, such as Gram, Stamp and crystal violet staining, CO 2 requirement, H 2 S production, oxidase and catalase activity, urea hydrolysis, agglutination with monospecific sera (anti-A, anti-M, and anti-R), dye sensitivity (basic fuchsin and thionine), bacterial motility (triphenyl tetrazolium chloride solid agar), and phage lysis (F1, F25, Tb, BK2, Iz, Wb, R/C) in order to identify and sub-differentiate Brucella spp. 30 .
Polymerase chain reaction (PCR) to detect Brucella DNA in canine blood samples. The whole blood samples were also used for direct detection of Brucella DNA by IS711 PCR when bacterial isolation was not successful. Briefly, the extraction of the bacterial DNA was based on mechanical and enzymatic pre-lysis of leukocytes, using zirconia/silica beads and lysozyme, respectively, followed by overnight enzymatic lysis using proteinase K and sodium dodecyl sulphate (SDS). The DNA was finally purified by phenol/chloroform and alcohol precipitation as described previously 71,72 . DNA gel electrophoresis images of the PCR products were edited using Adobe Photoshop CS5 (Adobe Systems, San Jose, CA, USA).

Matrix-assisted laser desorption/ionization-time of flight mass spectrometry (MALDI-TOF MS) to sub-differentiate Brucella species.
Brucella isolates were primarily inactivated in 75% ethanol for at least 90 min and stored at − 20 °C until use. Samples were prepared for mass spectrometry by ethanolformic acid extraction according to manufacturer's instructions before being spotted on a 96-spot steel plate target and covered with alpha-cyano-4-hydroxy-cinnamic acid (HCCA) matrix solution (Bruker Daltonics). Mass spectra were acquired using a Bruker Microflex LT MALDI-TOF MS system (Bruker Daltonics). Spectra were initially analyzed using the Bruker Biotyper 3.1 software with MSP library version MBT 7311 (7311 entries) and the Security-Relevant (SR) Database (104 entries). None of the commercial libraries contained reference spectra of B. canis. Therefore, we complemented the database with in-house mass spectra of various B. canis and B. suis strains (obtained from Karger et al. (2013) 32 and this study).
For each sample, a total of 24 spectra were measured, analyzed and visualized with the commercial Biotyper software as well as the open source statistical computing environment R v3.6.3 73 using the packages MALDIquant v1.19.3 74 and Tidyverse v1.3.0 75 .
Bruker Biotyper identification scores for each sample were calculated from a main spectrum (MSP) composed of the 20 best quality technical replicate spectra following manufacturer's instructions. Identification scores were interpreted accordingly, with scores ≥ 2.3 indicating highly probable species identification, scores between 2 and 2.3 indicating secure genus identification and probable species identification, scores between 1.7 and < 2 indicating probable genus identification and scores < 1.7 indicating no reliable identification.
Spectra processing with MALDIquant was performed similar to MSP creation by calculating an averaged spectrum of all technical replicate spectra of a sample. Averaged group spectra represent the mean of all available Brucella field isolates and reference strains for the respective species or biovar.  80 was applied with a fragment size of 200. Insertion sequences (IS) were identified by online BLASTn analysis against the IS database of the ISfinder website 81 . Inspection of the De Bruijn assembly graph was done with Bandage v1.0 82 . For molecular genotyping, in silico multi-locus variable number of tandem repeat analysis (MLVA-16) was calculated with a Python script 83 and in silico multi-locus sequence typing (MLST-21) was calculated with the mlst program by Torsten Seemann (https ://githu b.com/tseem ann/mlst).
Genetic relatedness between the Brazilian isolates and publicly available sequences was determined with ParSNP v1.0 84 . To this end, all available sequences including B. canis genomes in the PATRIC (www.patri cbrc. org) or NCBI (https ://www.ncbi.nlm.nih.gov/genom e/micro bes/) database as well as draft genomes assembled from read files retrieved from the Sequencing Read Archive (https ://www.ncbi.nlm.nih.gov/sra/) and the B. canis BfR-SPBR-consensus genome were collected. Maximum-Likelihood trees were calculated by FastTree2 85 . The origin of strains was extracted from genome metadata and integrated into the phylogeny with the EMBL interactive tree of life, iTOL v4 86 , for visualization.
The resulting SNP matrix was filtered for positions with a minimum coverage of five reads in total, with at least one SNP on the forward and the reverse strand. Due to the filter limitations of BioNumerics, SNP positions were extracted and manually curated in Microsoft Excel 2013 to differentiate between false base calls in the de novo assembly and SNPs at positions present in all 24 sequences or at positions with ambiguous base calls or insufficient coverage (N). Based on the final SNP matrix the pairwise distance between the isolates was calculated and clustering of isolates was performed using the Neighbor Joining algorithm in BioNumerics.
A comparison of B. canis BfR-SPBR-consensus against reference sequences including variant calling was performed in Mauve v2015-02-13. An identical SNP list was extracted using ParSNP v1.0 and hence could be used for SNP annotation by SnpEff 88 with the RefSeq annotation of reference GCF_000018525.1.
Nucleotide sequence accession numbers. The lllumina MiSeq raw sequence reads used to generate the draft B. canis BfR-SPBR-consensus genome sequence have been deposited in the database of the European Nucleotide Archive (ENA) under the accession numbers ERX4130724, ERX4130725, ERX4130726 and ERX4130727 within the BioProject ERP121782.