The study was based on four L. interrogans serovar Hardjo strains: KR40, KR84, KR85, and N116 – each isolated from different host and region in Europe. Strain KR40 was obtained from a horse in Italy in 1997, KR84 from a wallaby in the United Kingdom in 1983, KR85 from a dog in the United Kingdom in 1982, and N116 from a cow in Belgium in 2016. All Leptospira isolates were originally classified on the basis of cross-agglutination tests using serogroup-specific hyperimmune rabbit sera in the microscopic agglutination test (MAT). After genome sequencing, isolate identity was subsequently confirmed using the core-genome multilocus sequence typing (cgMLST) scheme based on 545 highly conserved loci, developed by Guglielmini et al. (2019) and implemented on the Institut Pasteur webstite BIGSdb-Pasteur (https://bigsdb.pasteur.fr/).

To enable the preliminary identification of selected isolates as belonging to the L. interrogans serovar Hardjo type Prajitno, we performed restriction endonuclease analysis (REA) following established protocols. Genomic DNA was extracted from 4 L. interrogans isolates (KR40, KR84, KR85, and N116) using the QIAamp DNA Mini Kit (Qiagen) Genomic, according to the manufacturer’s instructions. For comparative purposes, genomic DNA from two Hardjo strains were also included: L. interrogans serovar Hardjo strain Hardjoprajitno and L. borgpetersenii serovar Hardjo strain KR39 (BioProject: PRJNA828006).

DNA samples were digested with the restriction enzyme EcoRI (Thermo Fisher Scientific), using 1 µg of DNA and 10 units of enzyme in a 20 µL reaction volume, incubated at 37 °C for 1 h. The resulting DNA fragments were separated by electrophoresis on a 0.6% agarose gel in 1× TBE buffer at 70 V for 21 h. Gels were stained with ethidium bromide and visualised under UV light using a GelDoc imaging system (G:BOX Chemi system (Syngene) equipped with GeneSys/GeneTools software). Banding patterns were analysed visually and compared with the reference strains following the method described by Ellis et al. (1991). Strains exhibiting REA profiles consistent with the Hardjoprajitno reference were considered to belong to the L. interrogans sv. Hardjo (type Prajitno) lineage.

Genomic DNA was isolated from bacterial cultures using the Genomic Mini AX Bacteria kit (A&A Biotechnology, Poland) and quantified using the NanoDrop 2000 and Qubit systems (Thermo Scientific). Genomic DNA fragmentation was verified by agarose gel electrophoresis. Illumina sequencing libraries for strains under investigation were prepared from 200 ng total DNA according to the manufacturer’s instructions with the Lotus DNA Library Prep Kit (Integrated DNA Technologies, Inc.). Sequencing was performed using single-end 75 bp runs on an Illumina NextSeq 550 platform with a minimum of 40x genome coverage. Oxford Nanopore sequencing libraries, due to objective reasons, were generated only for the KR40 and N116 strains, and were prepared with 3 µg of total DNA using the Ligation Sequencing Kit (SQK-LSK110; Oxford Nanopore Technologies, Oxford, UK) according to the manufacturer’s protocol. The libraries were loaded onto R10.3 flow cells and sequenced on the MinION Mk1C system (Oxford Nanopore Technologies). Basecalling was carried out with Guppy (v5.0.11+2b6dbff) using the super-accuracy model.

For KR84 and KR85 strains, Illumina raw sequencing reads were evaluated for quality using FastQC v0.11.9 (Andrews, 2010) and filtered using TrimGalore software v0.6.10 (Krueger, 2021). Filtering was done through the removal of adapter sequences and trimming of low-quality read ends. Filtered reads were assembled into contigs using Shovill v1.1.0 (https://github.com/tseemann/shovill), an ultra-fast implementation of the SPAdes v3.14.0 algorithm (Nurk et al., 2013). The resulting contigs were polished using Pilon v1.23 (Walker et al., 2014). Contigs statistics were retrieved using QUAST software v5.0.2 (Mikheenko et al., 2018). Scaffolding was carried out by CSAR-web (Chen and Lu, 2018), using the closest available complete genome for the analysis at the time of analysis, determined by 16S rRNA analysis: L. interrogans serovar Hardjo strain Norma (reference assembly: GCA_001293065.1). Genome sequences were annotated using the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) with the best-placed reference protein set (GeneMarkS-2+) v6.0. The assemblies were deposited in GenBank under BioProject PRJNA809530 (KR84 assembly: GCA_022436545.1; KR85 assembly: GCA_022436605.1).

For KR40 and N116 strains (hybrid assemblies) raw Illumina reads were qualified by FastQC for quality and trimmed using Flexbar v3.5.0 (Dodt et al., 2012), removing adapter sequences and trimming low-quality read ends. Nanopore reads were trimmed using Fastp v0.23.2 (Chen et al., 2018), and adapter sequences were removed using Porechop v0.2.4 (Wick et al., 2017) and filtered with Filtlong v0.2.1 (https://github.com/rrwick/Filtlong), setting a minimum nucleotide length of 500 bp. Low-quality read ends were trimmed using Fastp. Filtered Illumina and Nanopore reads were assembled de novo with a hybrid strategy using Unicycler v0.5.0 (Wick et al., 2017) with default parameters. The obtained assemblies were annotated using the NCBI Prokaryotic Genome Annotation Pipeline (PGAP) with best-placed reference protein set (GeneMarkS-2+) v6.1 (strain N116) and v6.2 (KR40). Assembly was deposited in the GenBank under the BioProject number PRJNA828004 (assembly: GCA_023158895.3) and PRJNA828002 (assembly: GCA_023515975.1) for KR40 and N116 respectively.

Genomic relatedness of analysed strains was compared with 19 publicly available L. interrogans genomes from the NCBI database. All genomes were selected based on complete assembly status, to ensure consistent genome quality and to avoid bias in downstream analyses. They represent a wide range of serovars, geographic regions, host species, and collection years (Supplementary Table S1). The only exception was made for 3 L. interrogans serovar Hardjo strains (str. KR84, KR85, and OV5) which were included despite being high-quality draft assemblies. This was due to the limited availability of complete genomes for serovar Hardjo in public databases at the time of study design. Excluding them would have significantly reduced the representativeness of the Hardjo group.

The relationship between the number of coding sequences (CDS) and total genomic length was assessed using Spearman’s rank correlation, due to non-normal distribution. The average nucleotide identity (ANI) was calculated using JSpecies with ANIm option (Richter et al., 2016). Variant calling analysis (VCA) was performed to identify point mutations and establish population and phylogenetic structure. Filtered sequences were mapped to the reference genome L. interrogans sv. Hardjo str. Norma (Ref Seq: GCF_001293065.1) using BWA v0.7.17 (Li, 2013). Variants were identified using Freebayes v1.3.6 (Garrison and Marth, 2012) and filtered by a quality score cut-off >30 and depth >10. Variants were classified as SNPs (single nucleotide polymorphisms), MNPs (multi-nucleotide polymorphisms), INDELs (insertions/deletions), or mixed variants, which comprised a combination of the above types. The effects of the variants (excluding upstream/downstream, 5′UTR/3′UTR, and intronic changes) were predicted with SnpEff (Cingolani, 2022). Additionally, each of the selected high-impact variants was manually examined to evaluate its phenotypic relevance by determining whether the nucleotide change altered the protein sequence or led to its complete absence. Population structure analysis was conducted with the ADMIXTURE programme (Alexander et al., 2009) for K values from 1 to 10, selecting the optimal K based on cross-validation error. Results were visualised in Clumpak (Kopelman et al., 2015). The phylogenetic tree was constructed using the IQ-TREE web server with the auto-detect best-fit model with 1,000 ultrafast bootstraps (Nguyen et al., 2015) and visualised with iTOL v5 (Letunic and Bork, 2021). Ortholog clusters of the entire set of protein sequences were established using OrthoFinder (v3.0.1) with default parameters (Emms and Kelly, 2019). To assess differences between Hardjo strains and non-Hardjo strains, genomes were divided into two groups: L. interrogans serovar Hardjo group (LiH) and L. interrogans serovar non-Hardjo group (Li-nH). Statistical differences between groups were assessed using Fisher’s Exact Test, which is appropriate for categorical data and small sample sizes. To account for multiple testing, the Bonferroni correction was applied by dividing 0.05 by the number of tests (separately for VCA analysis and ortholog clusters analysis) to establish the initial significance threshold. However, due to the limited sample size and the highly conservative nature of the Bonferroni method, which may lead to a high false-negative rate, we applied a relaxed threshold - filtering based on raw p-values <5 × 10⁻⁶ for VCA analysis and 1 × 10⁻⁴ for ortholog clusters analysis. All calculations were performed in R (v4.4.2).

Genes and encoded proteins of interest were further searched in the UniProt database (Bateman et al., 2023) and the STRING database (Szklarczyk et al., 2023). Functional classification was established with the COG database using COGclassifier v1.0.5 (Tatusov et al., 2000). The CELLO programme v.2.5 (Yu et al., 2006), PSORT database v3.0.3 (Venus Lau et al., 2021), and SOSUI GramN programme v1.11 (Imai et al., 2008) were used to predict protein subcellular localisation with a majority voting strategy, as previously described (Techawiwattanaboon et al., 2023). Additionally, proteins that were predicted to be outer membrane, inner membrane or periplasmatic proteins by different programmes, without a majority voting strategy, were assigned to a newly created category labelled as ‘Membrane’. The SignalP 6.0 server (Teufel et al., 2022) was used to search for signal peptides. Proteins not detected by SignalP were further processed by the SecretomeP 2.0 (Bendtsen et al., 2005) to predict non-classical secreted proteins. To improve the prediction of spirochaetal lipoproteins, we additionally applied SpLiP tool (Setubal et al., 2006). All predictions were performed with default settings for Gram-negative bacteria. For genes and proteins lacking predicted information from the described procedures, we employed DeepFRI programme for functional annotation (Gligorijević et al., 2021). Genes and variants of interest were further categorized into those occurring outside the rfb locus and those within the rfb locus. The localisation of the rfb locus was determined based on the region spanning from the MarR regulatory gene to the DASS gene (Fouts et al., 2016; Ferreira et al., 2024). Probable genomic islands associated with horizontal gene transfer were predicted using the IslandViewer 4 platform (Bertelli et al., 2017). Gene localisation and COG categories were visualised using in-house R scripts (v4.4.2).

To confirm the hypothesis that genetic changes could impact the capacity of Hardjo and non-Hardjo strains of L. interrogans to form biofilm, we evaluated biofilm formation in vitro. Biofilm formation by L. interrogans (serovars Hardjo, Copenhageni, and Bratislava) was evaluated using a static microtiter plate assay according to the protocol of Thibeaux et al. (2020a). Bacteria were grown in EMJH medium at 30 °C under static conditions to the exponential phase, adjusted to a final density of 1 × 10⁶ cells/mL, and seeded into sterile, flat-bottom 96-well polystyrene plates (150 µL per well). Plates were incubated at 30 °C for 21 days without medium replacement to allow for mature biofilm formation. After incubation, planktonic cells were gently removed, and biofilms were fixed with 4% paraformaldehyde for 30 min. Wells were then washed with PBS, stained with 0.1% crystal violet for 10 min, and washed again. Stained biofilm was solubilised using 30% acetic acid, and absorbance was measured at 570 nm using a microplate reader. Biofilm assays were performed in three independent experiments, each with triplicate wells per condition. Statistical comparisons were based on biological replicates (n = 3 per group per time point). Statistical significance between serovars at each time point was assessed using the Kruskal-Wallis test, followed by pairwise Mann-Whitney U tests with Bonferroni correction for multiple comparisons. A p-value <0.05 was considered statistically significant.

Serological characterisation indicated that the tested field isolates belonged to the Sejroe serogroup (data not shown). For the preliminary identification of isolates as members of the L. interrogans serovar Hardjo, we applied the restriction endonuclease method (REA). Six strains belonging to serovar Hardjo were included in the analysis. The results revealed that DNA isolated from KR84, KR85, KR40, and N116 strains (Figure 1, lanes 4, 5, 6, 7) showed similar restriction patterns, consistent with that of the reference genome L. interrogans sv. Hardjo str. Hardjoprajitno (Figure 1, lane 3). The pattern for L. interrogans sv. Hardjo differed from that of L. borgpetersenii sv. Hardjo (Figure 1, lane 2). A minor difference was observed in the absence of an extra high molecular weight band of approximately 9,416 bp (Figure 1, white arrow) in the KR84, KR85, and N116 isolates (Figure 1, lanes 4, 5, and 7) compared with strains KR40 and Hardjoprajitno (Figure 1, lanes 3 and 6). Consistent with these REA findings, cgMLST analysis ultimately confirmed the taxonomic assignment - all four isolates grouped in cluster 40 (clonal group 19) of the cgMLST scheme, together with other representatives of L. interrogans serogroup Sejroe serovar Hardjo. The curated records are publicly available in BIGSdb-Pasteur (https://bigsdb.pasteur.fr/leptospira) under IDs: 1234 (KR85), 1271 (N116), 1272 (KR40) and 1336 (KR84).

Sequencing statistics for L. interrogans strains under study are summarised in Supplementary Table S2. The generated sequencing reads provided genome coverage of approximately 42x for assemblies based solely on Illumina reads (strains KR84 and KR85) and over 114x for hybrid assemblies. Using hybrid assembly approaches, complete genomes consisting of five and four circular contigs were obtained for strains N116 and KR40, respectively. For strains KR84 and KR85, draft whole-genome sequences consisted of 177 and 186 contigs, respectively. After scaffolding against the genome of L. interrogans serovar Hardjo str. Norma, nine scaffolds were generated for KR84 and seven scaffolds for KR85. Hybrid assemblies could not be obtained for strains KR84 and KR85 in this study. The genome size and number of predicted genes suggest, however, a high level of the completeness of their genomes, even in non-hybrid assemblies. The high sequencing coverage enabled a nearly complete representation of the genomic content, with only homologous or repetitive regions potentially preventing full assembly into continuous, circularised sequences (str. KR84, KR85).

Table 1 presents the quality control parameters for the four studied strains of L. interrogans, while Table 2 summarises their genome annotation parameters. The annotated genomes displayed a similar number of detected genes, with an average of 3,930 and an interquartile range (IQR) of 111.5; the gene count was slightly higher in the hybrid assemblies. On average, 96% of the genes were protein-coding. Each genome contained between 3 (for whole genome sequences of KR84 and KR85) and 5 complete rRNAs (for complete genome sequences for genomes KR40 and N116), 37 tRNAs, and approximately 156 pseudogenes (IQR: 7.75). The genomic G + C content ranged from 34.96% to 35.05%.

Genetic analysis was conducted on the four genomes obtained during this study and on an additional 19 genomes of L. interrogans, selected from the publicly available NCBI database (Supplementary Table S1). The genomic sequence length ranged from 4,630,592 to 5,349,767 bp with an average GC content of 35.01%. The total number of genes per genome ranged from 3,762 to 5,312. The number of coding sequences had a positive Spearman’s correlation of 0.734 (p = 6.73 × 10⁻⁵) with the genomic length.

ANI analysis provided a comparative genomic relatedness between various serovars and strains of L. interrogans (Supplementary Figure S1). The results from the JSpecies analysis revealed a high ANI percentage, ranging from 98.15% to 99.98% between different strains in our study. Serovar Hardjo displayed ANI values close to 100% among its strains. Lower ANI values, as shown in the heatmap in blue, suggest increased genetic distance and greater diversity, especially within the L. interrogans sv. Canicola and sv. Manilae, signifying potential genomic variability or divergence from other serovars.

Population structure analysis and phylogenetic reconstruction were performed based on the output of variant calling analysis (VCA). Population structure analysis (Figure 2A), conducted using the ADMIXTURE programme, indicated an optimal K value of 5 (log-likelihood = −344427.89, CV error: 0.77). The analysis separated all sv. Hardjo strains from other serovars. Since traditional phylogenetic markers such as 16S rRNA lack the resolution to distinguish between closely related serovars within the same species, we employed SNP-based analysis to achieve a higher level of phylogenetic resolution for the analysed L. interrogans strains. A phylogenetic tree was constructed using IqTree with TVM + F substitution model and 1,000 ultrafast bootstrap option (Figure 2B). We obtained strong bootstrap support, particularly for internal nodes connecting major clades. The phylogenetic tree, with some exceptions, revealed serovar-specific branches. The topology of the tree suggest that serovar Hardjo is the most closely related to members of the Icterohaemorrhagiae serogroup. Among all Hardjo representatives, short phylogenetic branches indicate limited genetic diversity.

For the purposes of this study, VCA using L. interrogans sv. Hardjo str. Norma as the reference genome was chosen. Approximately 99.4%–99.9% of the Hardjo reads were successfully aligned to the reference genome, whereas alignment rates for reads from other serovars varied from 87.9% for sv. Canicola to 96.6% for sv. Copenhageni.

Table 3 provides a summary of all high-quality variants detected across the dataset, as well as the subset of statistically significant differences between LiH and Li-nH groups. In total, the analysis revealed 99,656 genetic variants and 99,771 predicted effects on gene and their encoded proteins. Out of these, 10,266 variants and 10,724 effects were identified as statistically significant as differentiating between the groups, including 6,299 in the coding region and 4,425 in the non-coding region. Statistically significant variants were primarily located outside the rfb locus, although some conserved segments within this region also exhibited polymorphisms. Among genetic variants, SNPs were the most abundant. Overall, the ratio of non-synonymous to synonymous changes (dN/dS) was 0.4336, and was similar for statistically significant differences between the Hardjo and non-Hardjo groups, with a value of 0.4256.

We identified four statistically significant high-impact SNPs with predicted effects on the corresponding protein sequences. One of them, was located in the G436_RS22205 gene (p = 8.7 × 10⁻¹²), involving a cytosine (C) to guanine (G) substitution at position 3,936,506 on chromosome I (Figure 3A). This mutation introduces a premature stop codon, resulting in the absence of the encoded protein (WP_002080799.1) in non-Hardjo serovars. According to UniProt, the protein is annotated as a mobile element protein (A0A0M4NZF1), showing 100% identity with a PF07600 domain-containing sequence. Two additional premature stop codons were detected with lower statistical significance (p = 2.8 × 10⁻⁶). The first was located at position 3,455,697, involving a cytosine (C) to thymine (T) substitution in gene G436_RS23575 (Figure 3B), affecting a hypothetical protein (WP_002069478.1), suggested by DeepFRI to localise in the cytoplasm and potentially participate in general metabolic processes. The second occurred at position 3,670,902, involving a cytosine (C) to thymine (T) substitution in gene G436_RS15655 (Figure 3C), encoding the CRISPR-associated helicase Cas3 (WP_002189459.1), which appears as a pseudogene or truncated protein in Li-nH group. Additionally, we identified a SNP in G436_RS14945 (Figure 3D), where an adenine (A) to guanine (G) substitution at position 3,497,928 alters a start codon from ATG to GTG in non-Hardjo strains, affecting translation initiation for the thiazole kinase ThiM (WP_002188296.1; p = 1.0 × 10⁻⁹).

In addition, three INDELs were identified with statistically significant frequency differences between LiH and Li-nH groups, each predicted to affect protein sequence. Two of them resulted in the complete loss of protein-coding sequence in the Li-nH group. The first, located in G436_RS23265 (Figure 4A), involves a thymine (T) insertion leading to the absence of protein WP_100224992.1 (p = 8.7 × 10⁻¹²). The second, in G436_RS23350 (Figure 4B), results from a guanine (G) insertion and frameshift, leading to disruption of WP_154643182.1 coding sequence in Li-nH (p = 1.0 × 10⁻⁹). Both proteins are hypothetical and predicted to localize to the cytoplasm, with putative roles in metabolic processes, as predicted by DeepFRI. The third INDEL, located in G436_RS18190 (Figure 4C), modifies the N-terminal region of the FecR family protein WP_029757821.1, shortening it from 389 to 385 amino acids in non-Hardjo strains (p = 8.7 × 10⁻¹²). FecR proteins are implicated in iron uptake via citrate transport, though the functional consequences of this truncation remain unclear.

We also identified statistically significant, high-impact mutations, including one start-loss variant, six stop-gain variants, 17 stop-loss variants, 25 frameshift variants, and 13 complex variants occurring in Hardjo pseudogenes (Supplementary Table S3). According to the RefSeq annotation, L. interrogans sv. Hardjo str. Norma has a total of 184 pseudogenes, which constitute 3.7% of all genes. A number of pseudogenes in each newly sequenced genome is shown in Table 2, and varies from 151 to 161.

In total, we identified 1,335 significant missense variants, 159 of which were located in pseudogenes. The remaining 1,176 variants were mapped to 954 distinct genes. Particular attention was given to genes encoding predicted outer membrane proteins (OMPs), due to their surface localisation and possible interaction with the host environment. To identify proteins with N-terminal signal peptides and signal peptidase cleavage sites, SignalP 6.0 was used. Among the 954 genes analysed, 191 were predicted to encode proteins with secretion signal sequences. Of these, 119 had secretory signal peptides transported by the Sec translocon and cleaved by Signal Peptidase I (SP); 66 had lipoprotein signal peptides transported by the Sec translocon and cleaved by Signal Peptidase II (LIPO); four contained Tat signal peptides transported by the Tat translocon and cleaved by Signal Peptidase I (TAT); one had a Tat lipoprotein signal peptide transported by the Tat translocon and cleaved by Signal Peptidase II (TATLIPO); and one had pilin and pilin-like signal peptides transported by the Sec translocon and cleaved by Signal Peptidase III (PilD/PibD) (PILIN). Due to the plasticity of spirochaetal lipobox sequences, SpLiP algorithm, a tool optimised for spirochaetes, was used. 55 proteins were predicted by SpLip as probable lipoproteins and 12 as possible lipoproteins. 80 proteins were predicted by SecretomeP to be exported through the non-classical protein secretion pathway. To further predict protein subcellular localisation, we applied the CELLO programme, the PSORT database, and SOSUI GramN programme applying a majority-voting strategy to enhance prediction accuracy. 29 proteins were predicted as ‘Extracellular’, 49 as ‘Outer membrane’, 130 as ‘Inner membrane’, 12 as ‘Periplasmic’, whereas 56 were assigned to general category ‘Membrane’ (Supplementary Table S4).

The OrthoFinder programme identified 4,326 orthologous groups across the analysed genomes. The pangenome of L. interrogans comprised 3,168 core orthologous gene clusters, defined as those present in at least 95% of the analysed genomes. 88 orthogroups were classified as unique to the Hardjo serovar (LiH). Conversely, 47 orthogroups were predominantly present in the non-Hardjo group (Li-nH) (Figure 5). A full list of LiH-specific proteins, along with predicted localisation in the genome, potential localisation in the cell compartment, potential prediction of non-classical protein secretion, and prediction of lipoproteins is in Table 4 (for genes located outside the rfb locus) and Supplementary Table S5 (for genes located within the rfb locus).

The analysis of orthogroups specific to Hardjo and non-Hardjo serovars, revealed distinct functional profiles, as classified by COG categories. In the LiH group, 36.36% (32/88) of sequences were assigned to COG functional categories, compared to 55.32% (26/47) in the Li-nH group (Figure 6). Among the categories, ‘Cell wall/membrane/envelope biogenesis’ displayed the most significant differences.

Regarding the predicted subcellular localisation, 70 proteins were classified as ‘Cytoplasmic’, 6 as ‘Inner Membrane’ and 1 as ‘Outer Membrane’. None were predicted by SignalP whereas seven were identified by SecretomeP as candidates for non-classical secretion. However, all were classified as cytoplasmic or of unknown localisation, mostly representing hypothetical or mobile element proteins, often secreted via type IV systems in Gram-negative bacteria (Waksman, 2019). One protein (WP_000808306.1) was predicted as probable lipoprotein by SpLip.

As rfb locus is known for its high variability across Leptospira serogroups and serovars, we examined the localisation of LiH-specific orthogroups with respect to this region and visualised their distribution in the reference genome of strain N116 (Figure 7). Subsequent analyses focused primarily on the LiH-specific genes located outside the rfb locus. Of the 88 LiH-specific genes, 59 were located outside the rfb locus (Table 4). 37 of them were annotated as hypothetical proteins The observed clustering of several genes prompted us to assess their presence within putative genomic islands (GIs) using IslandViewer 4. Three large genomic regions enriched in LiH-specific genes were identified. The first region (positions 2,280,254–2,289,105 bp in str. N116) included genes for eight hypothetical proteins, two mobile element proteins (WP_001185452.1, WP_001127181.1), and three proteins associated with toxin-antitoxin systems (WP_000044296.1, WP_000014356.1, WP_002188765.1). The GC content of this region (39.96%) is noticeably higher than the average genomic GC content of L. interrogans strain N116 (35.05%). A second region (positions 521,196–532,505 bp; GC content 30.29%) comprised fourteen hypothetical proteins, an IS1595 transposase (WP_002189216.1), and two Sce7726 and Sce7725 family proteins (WP_002189271.1 and WP_002189307.1, respectively). Third region (1,463,770–1,468,634 bp) contained mainly hypotethical proteins (GC content 30.98%). In these regions, several interspersed genes were not classified as unique to L. interrogans sv. Hardjo. Additionally, large intergenic distances between some of the genes suggest that these clusters are unlikely to form single operon. While the first region is located entirely on the same strand, the second and third region span both DNA strands. Notably, only the second region was predicted as a genomic island by IslandViewer 4, whereas the first region was not.

Beyond these defined regions, we also detected several LiH-specific genes that either formed small local groupings or were scattered throughout the genome without consistent clustering. Among them were genes encoding a toxin-antitoxin system component (WP_000881563.1), a Toll/Interleukin-1 receptor domain containing protein (WP_002187597.1), a guanylate cyclase domain-containing protein (WP_000288518.1), and a leucine-rich repeat domain-containing protein (WP_002188813.1), positioned near the gene encoding the flagellar motor switch protein FliG. Of note, IslandViewer 4 predicted that some of these individual genes are located within genomic islands (Table 4).

Crystal violet measurements revealed a clear time-dependent increase in biofilm biomass for all 3 L. interrogans serovars (Hardjo, Copenhageni, Bratislava) over the 21-day static incubation (Figure 8A). Serovar Hardjo formed significantly more biofilm than the other two serovars at each time point (p < 0.05), with absorbance values markedly higher by week 3. Serovar Copenhageni exhibited an intermediate biofilm production, whereas Bratislava consistently showed the lowest values. Microscopic examination mirrored these results: phase-contrast images (Figure 8B) demonstrated more extensive and confluent biofilm architecture for Hardjo, especially evident after 3 weeks, while Copenhageni and Bratislava produced progressively less dense microcolonies.