A syntenic pangenome of Gardnerella reveals novel plasmids and phage, taxonomic boundaries, and species-level stratification of metabolic and virulence potential Authors: Heather K. Bouzek, Martha A. Zepeda-Rivera, Sujatha Srinivasan, Elliot M. Lee, Susan M. Strenk, Dakota S. Jones, Elsa F. McMahon, Tina L. Fiedler, Marko Kostovski, Michael T. France, Jacques Ravel, David N. Fredricks, Christopher D. Johnston Published: 2026 ABSTRACT Gardnerella species are key drivers of bacterial vaginosis (BV), a prevalent condition affecting nearly one in three women of reproductive age and associated with adverse reproductive outcomes. Despite decades of study, progress in defining Gardnerella diversity has been hindered by inconsistent taxonomy and poor-quality genomic resources. Here we sequenced 392 Gardnerella isolates spanning asymptomatic and BV-associated microbiota and integrated this collection with all publicly available genomes to create a curated, high-quality reference set of 312 genomes. We resolved 21 genomic lineages encompassing 11 species and 15 subspecies using phylogenomics, average nucleotide identity (ANI), digital DNA–DNA hybridization (dDDH) and assigned each a provisional taxonomic name. Long-read assemblies enabled construction of a syntenic Gardnerella pangenome, revealing lineage-specific repertoires of virulence, metabolic, and defense, including variable sialidases (NanH), vaginolysin, and amino-acid biosynthetic pathways alongside conserved genomic organization. Comparative methylome profiling uncovered restriction-modification system diversity suggesting barriers to genetic exchange. Finally, we identified native cryptic plasmids in Gardnerella, overturning the assumption that the genus lacks plasmids. Together, these results establish a complete genomic and functional framework for Gardnerella, providing a reproducible foundation for mechanistic and translational studies of BV and a model for resolving taxonomy and functional stratification in other urogenital-associated bacteria. Introduction Bacterial vaginosis (BV) is a common and global condition characterized by an imbalance in the vaginal microbiota1–3, associated with significant health risks, including increased susceptibility to sexually transmitted infections4,5 and adverse pregnancy outcomes such as preterm birth and low birth weight6–8. While molecular methods have associated many bacterial species with BV, Gardnerella vaginalis was the first bacterium isolated from individuals affected with BV9, and was long thought to be the sole causative agent of this condition1,10,11 due to its high virulence potential3,12–20, its presence in 95–100% of BV cases11,21,22 and its ability to thrive in the vaginal environment10,19,20,23–25. Our previous research revealed that while Gardnerella vaginalis is correlated with a greater risk ratio for BV than other validly named Gardnerella species, no single Gardnerella species is specific for BV based on a cpn60 qPCR. In fact, the presence of a high Gardnerella species diversity was identified as a risk factor for the development of BV in our longitudinal study22. The presence of Gardnerella species in both BV-affected and asymptomatic individuals suggests the potential for both commensal and pathogenic lineages or states3. However, the genetic capabilities of individual Gardnerella lineages remain unclear, partially due to inconsistent taxonomic assignments and overly complex Gardnerella nomenclature19, despite laudable efforts from multiple research groups26–35. Further, many of the public Gardnerella genome assemblies are duplicated, contaminated, or incompletely assembled and did not pass our quality metrics for pangenome analysis. Thus, we sequenced over 392 Gardnerella isolates genomes from both asymptomatic individuals or those with BV and combined these with all available public genomes to generate a highly curated and deduplicated collection for analyses. We applied comprehensive comparative genomic analysis of this curated collection to taxonomically identify Gardnerella lineages and propose a single naming scheme that is reflective of their phylogenetic placement and historical context. Syntenic pangenome partitioning revealed lineage-specific repertoires of metabolic pathways, virulence-associated genes, and defense systems. Collectively, this study integrates pangenome analyses, genome-based taxonomy, and experimental validation to define species boundaries within Gardnerella, characterize lineage-specific metabolic and virulence-associated potential, and identify mobile genetic elements, including a cryptic plasmid that enables the development of a shuttle vector for the genetic engineering of Gardnerella. Results Gardnerella and Bifidobacterium are distinct genera We obtained genomes for 392 Gardnerella isolates (Fredricks and Ravel bacterial isolate collections) from asymptomatic individuals and those with BV. We curated these and additionally available public Gardnerella genomes (n = 271 within the NCBI database as of May 2025) using stringent quality metrics to remove duplicate and low-quality genomes (Methods, Supplementary Data 1). This resulted in a final curated set of 312 genomes for comparative genomic analyses (Supplementary Data 2). Analysis of the cpn60 gene (which encodes a 60 kDa chaperonin) currently provides the most reliable method for Gardnerella species typing, as these species are indiscernible by traditional 16S rRNA gene sequence typing36,37. BLASTn analysis of the cpn60 gene of these isolates supported their identification as Gardnerella species (Supplementary Data 2). Despite the classification of these genomes as Gardnerella against and within the NCBI database, initial speciation against the Genome Taxonomy Database (GTDB) assigned many of these genomes to Bifidobacterium (Supplementary Data 2). However, Bifidobacterium is known to be a polyphyletic group, consisting of at least six distinct clades38,39, whose members exhibit substantial genetic diversity. The separation of Gardnerella from Bifidobacterium has previously been demonstrated through comparisons of genome length, G + C percentage, and overall genome composition31,39. Consistent with theses prior observations, comparisons of genome length and G + C percentage between our Gardnerella genome set, Bifidobacterium, and closely related scardovial genera (e.g., Alloscardovia40, Neoscardovia41, Aeriscardovia42, Scardovia43) (n = 137 genomes; Supplementary Data 3) indicate that Gardnerella genomes are significantly shorter (p < 0.001, Wilcoxon post hoc test) and have lower G + C content (p < 0.001, Wilcoxon post hoc test) than other Bifidobacterium members (Supplementary Fig. 1a, b). Notably, Gardnerella shares a human-focused and pathogenic niche with Alloscardovia, Aeriscardovia and Scardovia and clusters more closely with the scardovial genera across these genomic features than with Bifidobacterium. Additionally, average amino acid identity (AAI) values between Gardnerella and Bifidobacterium fell below genus thresholds of 65%44 (Supplementary Fig. 1c). Together, these results confirmed that our final curated genome set comprises only Gardnerella species, further supporting previous evidence that Gardnerella and Bifidobacterium represent distinct genera within the Bifidobacteriaceae family. Identification and proposed revised nomenclature for Gardnerella lineages Given the importance of Gardnerella species diversity in BV21, we first assessed the number of lineages present in our genome collection. Gardnerella has been classified using various approaches30–33,45,46, but consistent naming has been a challenge for Gardnerella despite extensive efforts26–28,32 (Fig. 1). Early subtyping by Piot et al.26 relied on sugar hydrolysis phenotypes and later expanded to 17 biotypes27. Genomic methods such as restriction fragment length polymorphism28 and amplified ribosomal DNA restriction analysis (ARDRA)29 identified several distinct genotypes. PCR and whole-genome sequence comparisons introduced two new naming schemes including designations A–D37 or 1–430 which led to hybrid labels (C/1, B/2, A/4, and D/3)47. These species were then reframed as reassorted ecotypes31 incorporating the genomic context of the distinct types. With widespread whole-genome sequencing, additional standards based on ANI emerged, describing thirteen genomic species (Gsp)32, nine genomospecies (GS)33, or nine different genotypes (GGtypes)35. Subsequent formal naming of a subset of these species included G. vaginalis9, G. pickettii48, G. piotii32, G. leopoldii32, G. swidsinskii32, G. greenwoodii48, G. lacydonensis49, G. bretellae49, G. massiliensis49, and G. phocaeensis49 (Fig. 1). Despite the use of diverse classification schemes, phylogenetic analyses across studies consistently recover a deep bifurcation within Gardnerella, with the same two major lineages forming the basal split of the genus. While this division has been explicitly designated as “Set A” and “Set B” only by Ahmed et al., analogous groupings confirm the primary branching structure in multiple independent phylogenetic frameworks30–35,37,46,50,51.Fig. 1Historical context of Gardnerella nomenclature.a From top to bottom, the schematic shows the identification method by type, publication, first author, names given to Gardnerella or naming framework, date of publication, and resulting conclusions from naming schemes. b Gardnerella subtyping resulting from the publication in (a), where each column represents a subset of Gardnerella. Arrows indicate consistent lineages between studies where enough evidence exists to link the groups. The two right-most columns contain updated lineage assignments with corresponding species and subspecies nomenclature proposed within this study. Created in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8). Historical context of Gardnerella nomenclature. a From top to bottom, the schematic shows the identification method by type, publication, first author, names given to Gardnerella or naming framework, date of publication, and resulting conclusions from naming schemes. b Gardnerella subtyping resulting from the publication in (a), where each column represents a subset of Gardnerella. Arrows indicate consistent lineages between studies where enough evidence exists to link the groups. The two right-most columns contain updated lineage assignments with corresponding species and subspecies nomenclature proposed within this study. Created in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8). Consistent with this, cpn60 gene sequences and reference-free whole-genome phylogenies separated all genomes in our collection into two deeply divergent lineages (Fig. 2, Supplementary Fig. 2, Supplementary Data 4) here maintaining the Set A and Set B designation. To determine the number of Gardnerella species represented, we calculated pairwise genome-wide average nucleotide identity (ANI) values across all genomes and examined their distribution (Fig. 2). In accordance with prior large-scale analyses demonstrating a pronounced discontinuity in ANI values around the canonical prokaryotic boundary, most inter-lineage comparisons fell well below 95%, whereas closely related genomes clustered above 95% with an exceedingly small fraction of the sequence comparisons falling above 94% (Supplementary Fig. 3c). Because ANI and digital DNA–DNA hybridization (dDDH) values were strongly colinear across the dataset, dDDH did not provide additional resolution but independently supported the same groupings: all genomes that clustered together by ANI also exceeded the 70% dDDH species threshold (Supplementary Fig. 3a–d). Thus, species boundaries inferred from ANI were fully congruent with the universally accepted 70% dDDH. Integrating whole-genome phylogenies with ANI/dDDH clustering, we identified eleven Gardnerella species. Within five of these species, additional phylogenomic structure supported sub-species level differentiation, resulting in twenty-one distinct Gardnerella lineages (Fig. 2, Supplementary Data 4). These species-level assignments were congruent with the classification scheme proposed by Potter33. The Gardnerella species bretellae, lacydonensis, leopoldii, phocaeensis, pickettii, piotii, and swidsinskii were previously described as distinct species when integrating phenotypic characterization with genome-based analyses. In our dataset, these lineages likewise form well-supported monophyletic clades in whole-genome phylogenies and remain clearly differentiated from other Gardnerella taxa. However, when evaluated across the expanded genome collection as analyzed here, pairwise ANI and dDDH comparisons reveal that these lineages consistently fall within the genomic coherence of broader species-level clusters defined by ≥95% ANI and ≥70% dDDH thresholds. While they are phylogenetically structured and phenotypically distinguishable, they do not cross the currently accepted genomic discontinuity thresholds that would necessitate recognition as separate species under genome-based standards.Fig. 2Average nucleotide identity values between Gardnerella genomes.Average nucleotide identity values between Gardnerella genomes in a Set A and b Set B. Key for each panel indicates corresponding values to each color. White represents values below 90%. Genomes are organized based on a kSNP4-generated whole-genome reference-free phylogeny. Figure is a composite of the kSNP4 tree created in iTOL and the map sorted by the values in the kSNP4 tree as created using ComplexHeatmap in R. Final figure arranged and annotated in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8). Average nucleotide identity values between Gardnerella genomes. Average nucleotide identity values between Gardnerella genomes in a Set A and b Set B. Key for each panel indicates corresponding values to each color. White represents values below 90%. Genomes are organized based on a kSNP4-generated whole-genome reference-free phylogeny. Figure is a composite of the kSNP4 tree created in iTOL and the map sorted by the values in the kSNP4 tree as created using ComplexHeatmap in R. Final figure arranged and annotated in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8). Given the extensive history of overlapping and partially incompatible classification schemes for Gardnerella (Fig. 1), we next harmonized existing species and subspecies names and established a nomenclature across all identified lineages (Fig. 1, Fig. 2, Supplementary Data 2, Supplementary Data 4). Details on the rationale for each proposed nomenclatural revision, including species-specific functional attributes and ecological isolation sources, are provided in the Protologues. Our aim was to integrate previously used species and subspecies names to maintain a consistent framework. The historically recognized G. vaginalis corresponds to a subspecies-level clade within a broader species and is designated here as G. vaginalis subspecies vaginalis, preserving historical continuity while reflecting its phylogenetic placement (Fig. 1, Supplementary Data 2, Supplementary Data 4). Similarly, we propose distinct species names for lineages with well-supported genomic and phenotypic divergence, including G. bivia, G. hutchinsoni, and G. washingtoni, among others (Supplementary Fig. 3g). While we encourage the use of full species names, when possible, we also recognize that abbreviations may be useful within scientific publications and figures. To minimize ambiguity in referring to Gardnerella species in this paper, we therefore intentionally propose species names that can be abbreviated with unique two letter abbreviations (Gb, Gg, Gh, Gi, Gk, Gl, Gm, Gp, Gr, Gv and Gw), while individual subspecies were designated with discernable three-letter abbreviations (except for Gardnerella piotii subsp. pickettii and Gardnerella piotii subsp. piotii, which each require four letters, Gppk and Gppt) (Fig. 1, Supplementary Data 4). A syntenic pangenome for Gardnerella Having stringently identified Gardnerella species and subspecies lineages in our genome set and assigned names that adhered as closely to precedent, where possible, we were next able to investigate the genetic features associated with each lineage. Pangenome analysis provides a comprehensive view of the genetic repertoire within a group of bacteria by identifying genetic features that are well-conserved across most (>95%) genomes (persistent or core genome) versus those that are only present in a subset of genomes (cloud or accessory genome). While the core genome often contains genes involved in essential cellular processes52,53, analysis of the accessory genomes can provide key insights about genetic diversity, adaptation, and evolutionary relationships. For example, previous pangenome efforts in Gardnerella have expanded taxonomic groupings, identified metabolically essential genes, and uncovered virulence and niche adaptation features34,50,54. However, bacterial genes often function in co-regulated units or clusters involved in specific biological processes (e.g., operons). To identify putative operons within the Gardnerella pangenome, we applied a syntenic approach that accounts for gene order by using the Partitioned Pangenome Graph of Linked Neighbors (PPanGGOLiN) tool52 (Fig. 3, Supplementary Data 5). Briefly, in these analyses, the coding sequences of genes are grouped into nodes based on amino acid similarity and visualized as circular nodes, with node size reflecting the relative presence of that gene group across the genome set (Fig. 4a–c). Nodes representing neighboring genes are linked into syntenic units called modules (Supplementary Fig. 4). This analysis clearly distinguished between the core (persistent partition) and accessory genomes, and enabled identification of accessory genetic elements present at intermediate frequency (shell partitions) or low frequency (cloud partition) (Fig. 3). By leveraging modules and pangenome partitions, conserved arrangements of genes can be traced across multiple genomes and enables insight into the functional and evolutionary dynamics driving diversity at the genus, species, and strain levels.Fig. 3Gardnerella pangenome partitions.The tile plot displays the presence and absence of nodes in each partition where each row represents a node, and each column corresponds to a Gardnerella genome. Colors on the side correspond to core/persistent (orange), phage (red), shells (green), and cloud (blue) partitions. Genomes are ordered by kSNP4 phylogenies of Set A and Set B, with colors on dendrogram indicating lineage. Heatmap created by PPanGGoLiN, dendrogram was created in iTOL. Figure components were arranged and annotated in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8) and the final figure was made in Adobe Illustrator.Fig. 4Syntenic maps of the Gardnerella pangenome.a–c Visualizations of the PPanGGOLiN-generated Gardnerella pangenome. Each node represents a gene group, with edges connecting syntenic neighboring genes. Node size reflects the relative frequency of the gene group across Gardnerella genomes. Colors indicate core (orange) or shell (other colors) partitions. b, c The core and shell partitions are shown individually. d map of the presence and absence of the sialidase genes (nanH1-nanH4), type II α-amylase-pullulanase and vaginolysin as colored by the species and subspecies of Gardnerella. e Presence–absence bar chart showing the percentage of conserved Gardnerella vaginolysin and the sialidase genes across each Gardnerella set (gray). Raw presence/absence counts are shown above each bar. f Gene context maps for the sialidase loci. e, f were arranged and annotated in BioRender. Bouzek, H. (2026) (https://BioRender.com/gsklyn8). Figure made with Adobe Illustrator. Gardnerella pangenome partitions. The tile plot displays the presence and absence of nodes in each partition where each row represents a node, and each column corresponds to a Gardnerella genome. Colors on the side correspond to core/persistent (orange), phage (red), shells (green), and cloud (blue) partitions. Genomes are ordered by kSNP4 phylogenies of Set A and Set B, with colors on dendrogram indicating lineage. Heatmap created by PPanGGoLiN, dendrogram was created in iTOL. Figure components were arranged and annotated in BioRender (Bouzek, H. (2026) https://BioRender.com/gsklyn8) and the final figure was made in Adobe Illustrator. Syntenic maps of the Gardnerella pangenome. a–c Visualizations of the PPanGGOLiN-generated Gardnerella pangenome. Each node represents a gene group, with edges connecting syntenic neighboring genes. Node size reflects the relative frequency of the gene group across Gardnerella genomes. Colors indicate core (orange) or shell (other colors) partitions. b, c The core and shell partitions are shown individually. d map of the presence and absence of the sialidase genes (nanH1-nanH4), type II α-amylase-pullulanase and vaginolysin as colored by the species and subspecies of Gardnerella. e Presence–absence bar chart showing the percentage of conserved Gardnerella vaginolysin and the sialidase genes across each Gardnerella set (gray). Raw presence/absence counts are shown above each bar. f Gene context maps for the sialidase loci. e, f were arranged and annotated in BioRender. Bouzek, H. (2026) (https://BioRender.com/gsklyn8). Figure made with Adobe Illustrator. In our collection, the core genome of Gardnerella reflected approximately 16% of identified nodes, enriched in essential cellular functions including translation and ribosomal structure, transcription, RNA processing, signal transduction, energy production, and metabolism of nucleotides, amino acids, inorganic ions, and lipids (Fig. 3, Supplementary Data 5). Periodically, these conserved nodes are interrupted by shell-associated nodes (Fig. 4a; nodes in distinct colors, branching from orange nodes), which mark points of evolutionary divergence. These shell nodes often represent horizontal gene transfer events and are commonly associated with alternate evolutionary pathways. We noted node differences between shell partitions that were consistent with virulence, metabolism, and defense systems, warranting further investigation. Virulence factors and predicted pathogenicity across Gardnerella lineages Colonization of the vaginal niche by Gardnerella, and its transition to a pathogenic state, depend on mucin degradation, adherence to the vaginal epithelium, and the formation of structurally stable biofilms13,55. Canonical Gardnerella virulence factors implicated in these processes include exported sialidases and vaginolysin56, a cholesterol-dependent cytolysin. Among Gardnerella-associated sialidase isoforms, NanH2 and NanH3 account for the majority of extracellular sialidase activity, whereas NanH1 lacks a signal peptide and is therefore predicted to function intracellularly for sugar catabolism57. Consistent with this role, nanH158 is embedded within a conserved operon across genomes (Fig. 4f). In contrast, nanH2 and nanH459 are frequently located in proximity to this operon but are associated with mobile genetic elements, suggesting more recent paralogous diversification and horizontal acquisition. Notably, nanH3 is uniquely not operon-associated; instead, it is found as a solitary gene within defense-associated genomic regions, often adjacent to prophage elements, consistent with a distinct evolutionary trajectory and potential phage-linked mobility. Within our pangenome dataset, we identified lineage-specific nodes encoding vaginolysin, which contributes to host interaction by binding to and lysing epithelial cells of the vaginal mucosa55, and multiple enzymes involved in the stepwise degradation of mucin-associated glycans (Fig. 4d–e, Supplementary Data 5). These include secreted sialidases and additional carbohydrate-active enzymes, such as L-fucosidase60–62 and galactosidase27 that facilitate mucin catabolism when exported63. A Type II α-amylase-pullulanase protein was well conserved (present in 100% of Gardnerella genomes), in agreement with previous reports64 (Fig. 4d). Comparisons between Gardnerella Set A and Set B showed that the four sialidase genes (nanH1-nanH4) are significantly enriched in Set A genomes, while vaginolysin (vly) is significantly enriched in Set B genomes (Fig. 4e), highlighting lineage-dependent differences in metabolic capacity and the distribution of putative virulence-associated genes across Gardnerella. Predicted metabolic capabilities across Gardnerella lineages Given the observed difference in metabolism-related nodes, we bioinformatically predicted the biosynthesis of amino acids and the catabolism of small carbon sources across our Gardnerella genome dataset (Fig. 5a, f, Supplementary Data 6). The biosynthesis of glutamine, glycine, and proline was well conserved (>99% of genomes) while a high proportion of genomes (89%) harbored the capacity for cysteine biosynthesis (Fig. 5a). Asparagine biosynthesis was enriched for Set B versus Set A, but within Set A, 100% of the Gr strains carried the genetic pathway. Within Set B, asparagine biosynthesis was enriched for Gg (although depleted for Ggi) and the Gll lineage alone amongst other Gl (Fig. 5a, b). Methionine biosynthesis is also enriched throughout Set B, but absent in all of Set A. Within Set B methionine biosynthesis is notably lacking in Gb and Ggg strains (Fig. 5d). Threonine biosynthesis is much more prevalent in Set B as compared to Set A (Fig. 5a, e). Within Set A however, threonine synthesis is more likely to be found in Gv compared to other Set A lineages (Fig. 5a, e). Although cysteine biosynthesis is well conserved, this pathway is entirely absent in Gh genomes (Fig. 5a) and depleted in Gps as compared to other Gp genomes (Fig. 5c).Fig. 5Predictions for metabolic capabilities across Gardnerella lineages.Presence-absence map for GapMind-predicted pathways to a synthesize amino acids (AA) or f catabolize small carbon sources (SCS) where each row represents a Gardnerella genome. Genomes are ordered by kSNP4 phylogenies of Set A and Set B. Presence by proportion comparison of genomes for the b–e biosynthesis of a subset of amino acids or g–k catabolism of a subset of small carbon sources. Colors represent each lineage. Figure made with Adobe Illustrator. Predictions for metabolic capabilities across Gardnerella lineages. Presence-absence map for GapMind-predicted pathways to a synthesize amino acids (AA) or f catabolize small carbon sources (SCS) where each row represents a Gardnerella genome. Genomes are ordered by kSNP4 phylogenies of Set A and Set B. Presence by proportion comparison of genomes for the b–e biosynthesis of a subset of amino acids or g–k catabolism of a subset of small carbon sources. Colors represent each lineage. Figure made with Adobe Illustrator. Genes for the catabolism of ethanol appear to be a well-conserved feature of Gardnerella genomes (Fig. 5f). Genomes across Set A were additionally significantly enriched for the catabolism of alanine, gluconate, and serine (Fig. 5h, j, k). Notably, lineage-specific associations within Set A were observed for the catabolism of each of these small carbon sources. While gluconate catabolism is significantly enriched in Gv, alanine catabolism is significantly depleted in Gvl as compared to other Set A members (Fig. 5h, j). Serine catabolism was completely absent in Gr and (Fig. 5f) and was significantly depleted in Ggb and Gls compared to other lineages within their respective species (Fig. 5k). Interestingly, the catabolism of acetate and deoxyribose was lineage-specific and unique to Gb and Gv, respectively (Fig. 5f). Deoxyribose catabolism was enriched in the Gvv compared to other Gv subspecies (Fig. 5i). Our results indicate that Set A and Set B genomes are distinguished by their relative enrichment of pathways for catabolism of small carbon sources or synthesis of distinct amino acids with lineage-specific associations. Phage and antiphage defense mechanisms across Gardnerella genomes In our pangenome data set, we noted that multiple nodes were related to phage-associated proteins that encompassed two distinct Siphoviridae family Gardnerella prophage, including the previously described vB_Gva_AB151,65 and a novel phage we have tentatively designated as pGamma (Supplementary Fig. 5, Supplementary Data 7). This phage is from a distinct family from vB_Gva_AB1 within the Class Caudoviricetes in whole genome comparisons66. vB_Gva_AB1 is sometimes reported as several different prophages46,51, the discrete synteny within our PPanGGOLiN pangenome suggests that these variants represent at least one discrete taxon (Supplementary Fig. 5). It is notable that some strains appear to have remnant phage genes, although these appear to be non-syntenic and scattered throughout the genome (Supplementary Data 5). While both vB_Gva_AB1 and pGamma were identified in Set A and Set B genomes, vB_Gva_AB1 was significantly enriched in the Set B species Gl compared to pGamma (42.1% vs. 15.0%, OR = 0.242 [95% CI 0.126 - 0.467], BH-adjusted p < 0.001) and Gg (31.0% vs. 6.9%, OR = 0.165 [95% CI 0.032 - 0.847], BH-adjusted p = 0.0308) (Supplementary Fig. 6). To counteract phages, bacteria evolved an array of defense systems67, with the most widespread and well-characterized being restriction-modification (RM)68,69 and CRISPR-Cas mechanisms70. Across our Gardnerella genome dataset we identified the presence of sixty different systems, with an average of twelve defenses per genome (Fig. 6a, b, Supplementary Fig. 7, Supplementary Data 8). Of the defense systems with described mechanisms, dXTPase systems71 were well conserved (99% genomes). Type I RM systems were prevalent (56% of genomes) and did not differ significantly between Gardnerella sets. Set A genomes were significantly enriched for some Abi proteins (AbiD72 and AbiL73), CBASS defenses74,75, and DRT mechanisms76,77, while Set B showed enrichment for a larger number of defenses, including those with Abi-mechanisms (AbiE69, Gabija75, and Lamassu75,78), CRISPR-Cas (CRISPR arrays and Cas Type IE proteins), and Type II RM defenses (Supplementary Fig. 8). Across Gardnerella lineages, Gls showed an enrichment for DNA-modification system (DMS) proteins, while Gvv showed an enrichment for AbiL and DRT defenses (Supplementary Fig. 8).Fig. 6Defense systems, methylation motifs, and plasmid architectures across Gardnerella lineages.a–d Presence-absence maps for PADLOC-predicted defense mechanisms. Genomes are ordered by kSNP4 phylogenies of a, Set A and b, Set B. c Sunburst plot shows the sequence of each methyl-modified motif detected within each Gardnerella lineage. Size of section represents prevalence within each lineage. d Schematics depict plasmid annotations for pDNF01199S-01, p0756_546_1_1_BHK10-01, and pNR047-01. Full sequences for the plasmids are available in Supplementary Data 10. Sunburst plot was created in R. Plasmid schematics were generated in BioRender. Johnston, C. (2026) https://BioRender.com/er7gxuk. Figure made with Adobe Illustrator. Defense systems, methylation motifs, and plasmid architectures across Gardnerella lineages. a–d Presence-absence maps for PADLOC-predicted defense mechanisms. Genomes are ordered by kSNP4 phylogenies of a, Set A and b, Set B. c Sunburst plot shows the sequence of each methyl-modified motif detected within each Gardnerella lineage. Size of section represents prevalence within each lineage. d Schematics depict plasmid annotations for pDNF01199S-01, p0756_546_1_1_BHK10-01, and pNR047-01. Full sequences for the plasmids are available in Supplementary Data 10. Sunburst plot was created in R. Plasmid schematics were generated in BioRender. Johnston, C. (2026) https://BioRender.com/er7gxuk. Figure made with Adobe Illustrator. Although defense mechanisms evolved primarily to protect against bacteriophages, their activity extends against other horizontally transferred genetic elements, as well as plasmid vectors used during in vitro transformation. RM defenses, which are dependent upon detection and destruction of inappropriately methylated foreign DNA, are one of the most prevalent across bacteria. Further, it is notable that approaches to overcome these RM systems during genetic engineering have been shown to significantly improve transformation efficiency across multiple bacterial species79–81. Thus, we sought to characterize RM defenses across a subset of our Gardnerella genome dataset. We obtained and analyzed methylomes for 42 Gardnerella isolates across thirteen lineages within the Fredricks isolate collection (Supplementary Data 2). A total of 39 methylated motifs were detected and assigned to Type I (25 motifs), Type II (11 motifs), or Type III (3 motifs) RM systems (Fig. 6c, Supplementary Data 9). Identification of a native Gardnerella cryptic plasmid There is renewed interest in developing genetic tools for engineering Gardnerella species82,83. Because Gardnerella was long thought to lack mobile genetic elements, such as native plasmids, most efforts have focused on adapting non-Gardnerella genetic tools for use in this genus. Recent studies have been extensive and show promising results, achieving targeted chromosomal integrations and deletions using E. coli-based suicide vectors82, as well as reproducible transformation with a Bifidobacterium-E. coli shuttle vector83. During genome sequencing of the Fredricks clinical isolate collection, we observed an independent circular contig, with high coverage, forming a second replicon within the otherwise closed genome sequence of Gppk strain DNF01199S (Fig. 6d, Supplementary Data 10). Annotation of this contig showed three coding sequences for hypothetical proteins and a single putative replication initiation protein (Supplementary Data 11). Subsequent plasmid isolation and validation confirmed the episomal nature of this contig, marking pDNF01199S-01 a cryptic plasmid identified within the Gardnerella genus. Motivated by this discovery, we assessed whether additional cryptic plasmids were present in our Gardnerella pangenome. We reasoned that the current limited knowledge of such plasmids reflects their rarity. As such, we focused on the cloud partition of the accessory genome, which highlights strain-specific genetic factors and collectively comprises 47.23% of the Gardnerella pangenome. Using structural sequence insights from pDNF01199S-01, we identified two additional putative plasmid sequences within the cloud partition: one in Gvv strain 0756_546_1_1_BHK1084 (p0756_546_1_1_BHK10-01), which was isolated from cervicovaginal mucus in South Africa, and another in Gk strain NR04723 (pNR047-01), isolated from a vaginal swab in Nairobi, Kenya. All three plasmids are approximately 4 kb in size, contain coding sequences for putative hypothetical proteins, and share a highly similar replication initiation protein (84.48–95.26% similar), suggesting a potentially conserved replication mechanism for cryptic plasmids within the Gardnerella genus (Fig. 6d, Supplementary Data 10, 11). Given that pDNF01199S-01 replicates stably in Gardnerella, we next sought to engineer a Gardnerella-E. coli shuttle vector suitable for transformation of additional Gardnerella isolates. Analysis of our pangenome revealed that 22% of Gardnerella genomes encode tetM, including Gppk DNF01199S, consistent with previous reports on variable tetracycline resistance phenotypes across Gardnerella strains85. We therefore assessed the growth of both tetM+ and tetM- isolates in the presence of supplemented tetracycline. As predicted, tetM+ isolates were resistant to tetracycline, whereas tetM- isolates were susceptible, confirming that the tetM cassette confers tetracycline resistance in vitro. This supports the use of tetM as a selectable marker for this vector. To enable the propagation in E. coli strains, we combined pDNF01199S-01, the tetM cloned from DNF01199S, and an E. coli minicircle plasmid backbone (pMC) into a single parental plasmid (herein designated pG01-pp) (Supplementary Fig. 9a–c). As previously reported86, we induced and purified the pDNF01199S-tetM-MC minicircle vector (pG01-MC). We initially attempted to transform this minicircle into the tetM-strain, ATCC 14018T (Gvv), but no transformants were recovered (Supplementary Fig. 9d). The strain ATCC 14018 T encodes a restriction-modification (RM) system which modifies the “GGCC” motif87 and has been shown to limit transformation efficiency82. Thus, we used commercially available HaeIII methyltransferase (MTase) to methyl-modify these sites in pG01-MC prior to transformation. Electroporation with methylated pG01-MC successfully transformed Gvv ATCC 14018T, but at a very low efficiency (two transformants per µg of DNA). Further optimization of the protocol increased efficiency to 685 transformants per µg of DNA by transforming with 0.5 µg total HaeIII-treated vector on competent cells grown with 100 mM NaCl to an OD600 of 0.32 and washed with 10% glycerol (Supplementary Fig. 9d). A detailed, step-by-step version of the optimized protocol is available online at (10.17504/protocols.io.j8nlk1o91g5r/v1). Collectively, these results show that pG01-pp can function as a replicative shuttle vector derived from a native Gardnerella cryptic plasmid. Together with recent work by the Dillard and Hill groups82,83, these efforts should establish a robust platform for more mechanistic investigations of Gardnerella biology and pathogenesis. Discussion Taxonomic inconsistencies and fluctuating nomenclature have made it inherently challenging to delineate the roles of Gardnerella lineages in health and disease. In this study, we integrate whole-genome analyses to establish a refined and internally consistent taxonomic framework for the Gardnerella genus. Using a curated set of 312 distinct, high-quality genomes, we define clear lineage boundaries and propose an updated Gardnerella nomenclature that reflects genome-scale divergence while preserving continuity with existing naming conventions. Although nomenclatural revisions, particularly for well-established organisms such as ‘G. vaginalis,’ (updated here to G. vaginalis subsp. vaginalis), can often be contentious, applying established genomic criteria for species and subspecies delineation is warranted to prevent further taxonomic ambiguity and this misinterpretation of functional studies. While single-locus markers such as cpn60, which are widely used for Gardnerella typing, can distinguish between Gardnerella species, they lack sufficient resolution for subspecies assignments. Accurate subspecies assignments require whole-genome approaches including phylogenomic inference and ANI or digital DNA-DNA comparisons. As additional Gardnerella lineages are isolated, whole-genome sequencing efforts should allow for continuous integration into this framework, supporting a stable and extensible taxonomy. Our genome-based analyses also intersect with a broader, long-standing debate regarding the placement of Gardnerella within the family Bifidobacteriaceae. Although, marker gene frameworks such as the GTDB place Gardnerella within Bifidobacterium, the two groups differ markedly in genome size, G + C content, metabolic repertoire, virulence-associated gene content, mobile genetic element composition, and ecological specialization. Notably, Gardnerella genomes are substantially smaller and lack many biosynthetic and metabolic pathways retained in larger Bifidobacterium genomes. Such genome reduction is widely interpreted as a hallmark of specialization, reflecting adaptation to host-associated niches where broad metabolic versatility is unnecessary. In this context, reduced genome size represents ecological specialization rather than diminished biological significance. Together, these genome-scale differences suggest that conserved marker phylogeny alone may be insufficient to define biologically meaningful genus boundaries within this family. A parsimonious interpretation is that Bifidobacterium, as currently circumscribed, encompasses multiple genomically and ecologically distinct genera. While a formal genus-level revision of Bifidobacteriaceae lies beyond the scope of this study, our data provide additional genome-scale evidence supporting the continued recognition of Gardnerella as a distinct genus and contribute to an evidence base that may inform future ICNP-compliant taxonomic restructuring. With this refined taxonomic and nomenclature framework, we probed for genetic features and putative operons associated with each lineage by applying a syntenic pangenome approach. These analyses linked Gardnerella lineages to distinct repertoires of virulence-associated factors, metabolic pathways, and genetic defense systems, highlighting functional diversity across the genus. Notably, most Gardnerella gene content (84%) is species- or strain-variable with many differences related to virulence-associated functions, metabolism, and genetic defense. Gardnerella Set A genomes (Gr, Gp, and Gv) were enriched for sialidases and exhibited increased potential for the catabolism of small carbon sources. We further noted lineage-specific enrichment of established virulence-associated factors, including exported sialidases (nanH2 in Gr and Gppt; nanH3 in Gppk and Gppt) and vaginolysin (enriched in Gvv, Gvl, Gps and Gppk). The restricted distribution of these factors suggests functional specialization among lineages that may contribute to mucin degradation and epithelial barrier disruption within the vaginal microbiome. In contrast, Set B genomes (Gb, Gc, Gg, Gi, Gl, Gk, Gm, and Gw) were enriched in putative virulence-associated proteins (e.g., predicted surface-associated proteins, secretion system components and factors implicated in host interaction), exhibited greater capacity for amino acid biosynthesis, showed a higher propensity to harbor the vG_gva_AB1 prophage, and encoded a broader repertoire of genetic defense mechanisms. Collectively, these features are consistent with distinct colonization strategies among Set B Gardnerella lineages, including enhanced persistence on epithelial surfaces, flexibility under nutrient-limited or variable resource conditions, and adaptation to competitive, phage-rich, polymicrobial niches. Mechanistic studies to assess the roles of Gardnerella genetic factors, including those identified in this study, have been hindered by limited genetic tools available for Gardnerella. In fact, it has previously been reported that Gardnerella species lack mobile genetic elements, such as plasmids30. However, our extensive genome sequencing efforts reveal a Gardnerella cryptic plasmid naturally occurring within a single cultivated strain, along with additional homologous sequences in publicly available genomes. Through the identification of this cryptic plasmid (pDNF01199S), we developed a replicative Gardnerella-E. coli shuttle vector (pG01-pp). Coupled with recent advances in Gardnerella species engineering82,83, this discovery of extrachromosomal and self-replicative genetic elements within the Gardnerella genus should facilitate the development of advanced genetic tools. Combined, these foundational studies seek to advance genetic tools for mechanistic studies of Gardnerella species. Collectively, overcoming genetic intractability within Gardnerella will greatly advance our ability to investigate the role of this genus in health and disease states. Although these interpretations are based on comparative genomic potential and do not directly imply in vivo activity or clinical outcomes, taken together they support the view of Gardnerella as an ecologically diverse genus composed of lineages with distinct metabolic capacities, host-interaction mechanisms, and mobile genetic repertoires. Different Gardnerella seem to occupy overlapping but not identical niches within their human hosts. Some lineages may be better adapted to persistence and competitive growth in the vaginal microbiome and urinary tract, whereas others may be more consistent with transient or context-dependent colonization. Methods Study design and participants Vaginal bacterial isolates were collected from human study participants both with and without BV in Seattle, Washington, USA22,88 or from the Ravel isolate collection, harboring strains from multiple geographic locations, including USA and Africa (Supplementary Data 2). Bacterial isolates consistent with Gardnerella morphotypes on Gram stain were subjected to 16S rRNA and/or cpn60 gene amplicon sequencing to confirm phylogenetic placement in the Gardnerella genus24. Phylogenetically diverse Gardnerella species were selected for whole-genome sequencing. High molecular weight DNA was extracted from each isolate using the MasterPure DNA purification kit (Lucigen; Middleton, WI), with two phenol/chloroform cleanups prior to DNA precipitation where needed. DNA extraction was quantified on a TapeStation 2200 instrument run with a Genomic DNA tape (Agilent Technologies; Santa Clara, CA). Genome sequencing and assembly Sequencing on the Pacific Biosciences SeqI/II Platforms A total of forty-two strains from the Fredricks collection, and seventeen from the Ravel collection, were selected for single-molecule real-time (SMRT) long-read sequencing using PacBio (SeqI or SeqII, Pacific Biosciences; Menlo Park, CA) at the Fred Hutch Cancer Center or the Genomic Resource Center of the University Maryland School of Medicine. Sequencing libraries were prepared using the SMRTBell Template Prep Kit 1.0/2.0 and were selected for a 3–5 kb size cutoff using a BluePippin (Sage Science; Beverly, MA). Sequencing reads were processed via Microbial Assembly software within the Pacific Biosciences’ SMRTAnalysis pipeline (v.9.0.0.92188). Sequencing on the Illumina HiSeq 4000 Platform 332 isolates were selected for Illumina sequencing on the HiSeq4000 platform. Short read libraries were prepared using the KAPA HyperPlus Kit (Kapa Biosystems/Roche; Willmington, MA) with KAPA Single-Indexed Adapter Kit Set B. 35 μL of genomic DNA was used as input, and libraries were prepared following the manufacturer’s protocol with modifications based on concentration. For samples with 0.5 or 0.2 ng input DNA, the fragmentation enzyme was diluted 1:2 or 1:5 with water. All samples were fragmented at 37 °C for 5 min. Adapter concentrations varied according to the input, and the adapter ligation was carried out overnight at 4 °C for all samples. The post-ligation cleanup was performed with 0.8x Ampure XP beads (Beckman Coulter; Indianapolis IN) and 20 μL of sample was used in library amplification. Post-amplification cleanup was performed with 1x Ampure XP beads; libraries with remaining adapter dimer peaks were cleaned a second time. The final elution was in 25 μL of nuclease-free water. Libraries were run on a TapeStation instrument with a D1000 tape (Agilent Technologies; Santa Clara, CA) to assess quality and concentration. Libraries were sequenced on an Illumina HiSeq 4000 instrument using the 150 bp paired-end protocol. Reads were assembled using Unicycler. Publicly available genomes from the National Center for Biotechnology Information (NCBI) We used publicly available genomes both to provide a direct reference for previous studies and to supplement our own sequencing efforts. We searched GenBank89 assembly databases for the genus Gardnerella and family Bifidobacteraceae (without genus assignment). Using the Type (Strain) Genome Server (TYGS)90 to confirm the genus of these genomes, we discovered that several of the unassigned Bifidobacteraceae bacteria in NCBI were Gardnerella and included them in our analysis. Bifidobacteraceae (without genus assignment) were included if they were collected as part of the same NCBI BioProject which added 20 Gardnerella that were not typed further than the Bifidobacteraceae family. Additionally, we included three sequences (350_GVAG, 476_GVAG, and 842_GVAG) that were initially assigned to Gardnerella but were later renamed to Alloscardovia to verify this placement. The initial number of potential genomes was n = 649. We removed genomes that were assigned outside of Gardnerella. Draft genomes with more than 250 contigs, modified in the laboratory, and those with truncated or missing cpn60 sequences were removed due to inherent unreliability. Bifidobacterium strains were collected and curated in much the same manner as the Gardnerella strains. A total of 98 strains from all known species of the Bifidobacterium genus and the closely related scardovial genera (e.g., Alloscardovia, Aeriscardovia, Bombiscardovia, Pseudoscardovia, and Scardovia) were downloaded from NCBI. Whenever possible, the type strains were collected. Sequences were assessed using CheckM, TYGS, and ANI via cutoffs similar to those used for the Gardnerella strains. Bifidobacteriaceae groups were assigned according to Alessandri et al.91. Genome quality and comparison assessments Genome completeness and contamination for all sequenced and publicly available Gardnerella genomes retrieved from NCBI were assessed using CheckM (v 1.0.18)92. All genomes were annotated via RAST (v 1.073)93. Genomes with contamination greater than 1.5% or completeness below 90% were excluded from downstream analyses. Redundant genome sequences were identified based on pairwise ANI. Genomes sharing >99.5% ANI with other genomes in the collection were considered duplicates and a single representative genome was retained for each duplicate group (Supplementary Data 1). Pairwise ANI metrics were calculated separately for Set A and Set B genomes using FastANI (v1.34)94 and visualized using the R packages ComplexHeatmap (v 2.27.0)95. Digital DNA-DNA hybridization (dDDH)96 values were calculated via the Genome-to-Genome Distance Calculator (v 3.0)97 and visualized in R (v 4.5.1 -− 2025-06-13 -- “Great Square Root”). AAI values for Bifidobacteriaceae type strains were determined using pairwise comparisons of FastAAI (version 2)44. PPanGGOLiN Gardnerella pangenome We used Partitioned PanGenome Graph of Linked Neighbors (PPanGGOLiN v 2.2.0)52 on our curated genome set to graph and cluster the genes from all strains, find regions of genomic plasticity and synteny75, and separate the pangenome gene clusters into core, shell, and cloud partitions, with clustering set to 50% coverage and 50% identity. EggNOG97, InterPro (accessed October 2025)98, and KEGG Mapper (v 5.2)99, were run separately on each genome to gather and confirm the CAZyme100, Pfam101, and KEGG (all accessed October 2025), designations of the gene clusters. COG categories102 were assigned based on the consensus of these designations. We added the annotation “Unclassified Metabolism” to capture gene clusters that are clearly related to metabolism, but the exact metabolite is not defined or could be of multiple types. Phylogenomic analyses Barrnap was used to find and extract 16S rRNA sequences for submission to NCBI99. Nucleotide sequences for the cpn60 gene were extracted from RAST-annotated genome files and MUSCLE100-aligned to generate a maximum-likelihood dendrogram using MEGA (v11)101. The resulting tree was visualized and annotated using iTOL (v 7.3)102. kSNP4 (v 4.1)103 was used to construct a reference-free whole-genome phylogenetic tree for Gardnerella Set A (kmer = 17, FCK = 0.167) and Gardnerella Set B (kmer = 17, FCK = 0.797) genomes, separately. Bombiscardovia bombi strain DSM19703 was included to root both dendrograms. The resulting parsimony trees were visualized in iTOL102.GapMind (accessed September 2025)104–106 metabolism predictions and PADLOC (v 2.0.0)107,108 defense mechanism predictions were annotated where appropriate. Bacteriophage detection was performed by VIBRANT (v. 1.2.1)109, PHageR (accessed Feb 2026)110, and the use of phage-annotated nodes to find syntenic modules of phages in PPanGGOLiN. These methods were combined as no one method found all bacteriophage. Phage sequences are mostly as found by VIBRANT, when VIBRANT failed to find the phage sequence they were determined from PPanGGOLiN. Phages were typed by aligning the large terminase sequences from each intact with MAFFT (v7.526) and then creating a maximum likelihood tree with IQTree2 (v. 2.3.6)111. Assessment of greater taxonomic grouping was determined using VICTOR (https://ggdc.dsmz.de/victor.php accessed March 2026)66. Isolation of pDNF01199S-01 plasmid Gardnerella piotii subsp. pickettii strain DNF01199S was grown anaerobically in New York City III (NYCIII) liquid media in an AS-580 anaerobic chamber (Anaerobe Systems, Morgan Hill, CA) in a gas mixture of 5% CO2, 5% H2, and 90% N2 at 37 °C. A QIAamp DNA Miniprep kit (Qiagen, Hilden, Germany) was used to isolate pDNF01199S-01 from a 5 mL culture of the bacteria. Isolated DNA was digested with XhoI (New England Biolabs, Ipswich, MA) and visualized by gel electrophoresis to confirm plasmid presence. Gardnerella tetracycline resistance testing Gardnerella strains were grown anaerobically in an AS-580 anaerobic chamber (Anaerobe Systems, Morgan Hill, CA) in a gas mixture of 5% CO2, 5% H2, and 90% N2 at 37 °C. Bacteria were recovered from frozen stocks on NYCIII agar plates and subcultured twice in 2 mL of NYCIII liquid media. Culture purity was verified by Gram stain. The final cultures were normalized to 0.5 McFarland turbidity using fresh NYCIII broth and 25 µL was spotted onto an NYCIII agar plate containing 16 µg/mL tetracycline and an antibiotic-free NYCIII agar plate. 25 µL of Lactobacillus reuteri CF48-3A was spotted onto each tetracycline plate as a tetracycline-resistant control103. Tetracycline resistance was determined according to Clinical and Laboratory Standards Institute for antimicrobial susceptibility testing protocols, and any growth on tetracycline NYCIII agar plates was considered resistant to tetracycline. pMC-pDNF01199S-tetM construction and minicircle preparation pDNF01199S-01 and tetM were both cloned from gDNA isolated from strain DNF01199S. Gene fragments were assembled with pMC by Gibson assembly104. The parental construct was transformed into E. coli ZYCY10P3S2T (System Biosciences, Palo Alto, CA) by heat shocking according to the SBI Minicircle DNA Technology user manual. Transformant colonies were selected by growing on Luria-Burtani (LB) agar plates containing 50 µg/mL Kanamycin. A successfully transformed colony was chosen and stocked, then induction of minicircles was performed according to the SBI Minicircle DNA Technology user manual by growing the bacteria in induction media containing arabinose. pDNF01199S-tetM minicircle DNA was then purified from the induced E. coli using a Qiagen Plasmid Midi Kit (Qiagen, Hilden, Germany). The sequence of purified plasmid was confirmed against the construct sequence by whole-plasmid sequencing performed by Plasmidsaurus (Plasmidsaurus, Eugene, OR). “GGCC” sites on the plasmid were protected by incubating 1 mg of the DNA with 1 µL of HaeIII MTase (New England Biolabs, Ipswich, MA) at 37 °C for 4hrs in the presence of 160 µM S-Adenosyl methionine (New England Biolabs, Ipswich, MA). Following enzyme treatment, DNA was purified with Monarch PCR & DNA Cleanup Kit (New England Biolabs, Ipswich, MA). Preparation of competent Gardnerella cells Gardnerella vaginalis subsp. vaginalis ATCC 14018T was grown anaerobically in an AS-580 anaerobic chamber (Anaerobe Systems, Morgan Hill, CA) in a gas mixture of 5% CO2, 5% H2, and 90% N2 at 37 °C. Cells were recovered from frozen stocks on NYCIII agar plates and subcultured into NYCIII liquid media. 500 µL of culture was subcultured into 5 mL NYCIII media and grown overnight. 4 mL of this culture was inoculated into 40 mL of modified de Man-Rogosa-Sharpe media105 supplemented with 5 g/L glucose, 0.4 g/L L-cysteine HCl, 100 mL/L filter-sterilized horse serum, and 100 mM NaCl when stated. Once the target OD600 was reached, cells were harvested by centrifugation at 4 °C at 4052 × g for 10 min. Cells were washed in one of the following three ways: (1) Twice in 25 mL ice-cold maltose citrate buffer (0.2625 g/L citric acid and 0.5 M maltose at pH 5.8), (2) Twice in 25 mL ice-cold, autoclaved, 10% glycerol, or (3) Once in 25 mL ice-cold autoclaved water from a Milli-Q IQ7000 pure water system (Millipore Sigma, St. Louis, MO) followed by a second wash in 25 mL ice-cold maltose citrate buffer. Washed cells were then resuspended in 200 µL of maltose citrate buffer, mixed with 200 µL 80% glycerol, and stored at −80 °C in aliquots. Gardnerella electrotransformation and verification Competent cells were mixed with plasmid DNA on ice and transferred into pre-chilled 0.1 cm gap Gene Pulser/MicroPulser Electroporation Cuvettes (Bio-Rad, Hercules, CA). Electroporation was then performed on a Gene Pulser Xcell (Bio-Rad, Hercules, CA) at 25 µF, 200 Ω, and 2000 V. Immediately following electroporation, 950 µL of 37 °C pre-warmed NYCIII broth was added to the cells and they were transferred to an AS-580 anaerobic chamber (Anaerobe Systems, Morgan Hill, CA) in a gas mixture of 5% CO2, 5% H2, and 90% N2 to recover at 37 °C for 1 h. 100 µL–200 µL of transformed cells were then plated on NYCIII agar plates with and without 16 µg/mL tetracycline. Plates were incubated anaerobically at 37 °C for 48 h and transformed colonies on the tetracycline plates were counted to determine transformation efficiency. After the first successful transformation, transformant colonies were grown in 5 mL NYCIII media with 16 µg/mL tetracycline and pDNF01199S-tetM was recovered from them using a QIAamp DNA Miniprep kit (Qiagen, Hilden, Germany). The sequence of the purified plasmid was confirmed by sequencing performed by Plasmidsaurus (Plasmidsaurus, Eugene, OR). Following transformation of Gardnerella vaginalis subspecies regina strain DNF00354 with pDNF01199S-tetM, the transformation reaction was plated onto NYCIII agar plates with and without 16 µg/mL tetracycline. A small amount of growth from a transformant colony on the tetracycline plate or lawn growth from the control plate was resuspended in sterile water and colony PCR was performed to amplify and detect the tetM gene. One microliter of resuspended bacteria was added to a reaction mix of 12.5 µL GoTaq Green master mix (Promega, Madison, WI), 0.5 µL of each primer, and 6.5 µL H2O. PCR was carried out for 28 cycles with an annealing temperature of 60 °C and an extension time of 2.5 min. A detailed version of this protocol112 is available online at 10.17504/protocols.io.j8nlk1o91g5r/v1. Primer sequences were as follows (5’ to 3’): Forward: CGCGACCAAATATTGGTACATTATTACAGCTATTTTGTAATCACGTACTC Reverse: GAGACTGTCGTATTGATGTCACGGGCTTAATAGGTCTTATGTCGG Protologues Description of Gardnerella vaginalis subsp. vaginalis Gardner and Dukes9 (Approved Lists 1980) Bouzek et al. comb. nov Basonym: Haemophilus vaginalis Gardner and Dukes (Approved Lists 1980)9. Basonym: Haemophilus vaginalis Gardner and Dukes (Approved Lists 1980)9. Earlier homotypic synonym: Gardnerella vaginalis (Gardner and Dukes9) Greenwood and Pickett 1980113. Earlier homotypic synonym: Gardnerella vaginalis (Gardner and Dukes9) Greenwood and Pickett113. Gardnerella vaginalis subsp. vaginalis (va.gi.na’lis. L. fem. n. vagina, sheath, vagina; L. masc. adj. suff. -alis, suffix denoting pertaining to; N.L. fem. adj. vaginalis, pertaining to vagina, of the vagina). The description of this subspecies remains as given to Gardnerella vaginalis by Gardner and Dukes 19559 and Greenwood and Pickett 1980113 with the following additional information. Curated genomes of 79 Gardnerella vaginalis subsp. vaginalis cultivated isolates were used in this analysis, isolated from the vagina, endometrium, cervix, catheter urine, and placenta of humans from Bangladesh, Belgium, China, France, Russia, the United Kingdom, the United States, and Zambia (Supplementary Data 2). The mean genome length of Gardnerella vaginalis subsp. vaginalis is 1.68 Mb [range 1.58–1.78] and the median G + C content is 41.26% [range 40.99–41.82%]. The type strain for Gardnerella vaginalis subsp. vaginalis remains 594T of Gardner and Dukes (=ATCC 14018T = DSM 4944T = CCUG 3717T = JCM 11026T = NCTC 10915T = NCTC 10287T). Strains used in previous classifying publications: ATCC14019, ATCC49145, DNF01149, FDAARGOS296, FDAARGOS568, GH015, GH021, HMP9231, JCM11026, JCP7276, JCP7672, NR001, NR037, NR038, NR039, UGent09.01, UGent09.07, UGent25.49, UMB0032A, UMB0032B, UMB0061, UMB0233, UMB0298, UMB0386, UMB0768, UMB0770, UMB0775, WP023, 287, 377, 653, 1492, 3549624, 0284 V, 0286E, 0288E, 1-010-6, 1-014-4, 104-V2-10, 1388E, 18-4, 203-8, 23-12, 302000480-V1-10, 315-A, 7571-2, 772-1, and 825-1. Description of Gardnerella vaginalis subsp. lacydonensis (Allini Ntiguemassa et al.49) Bouzek et al. comb. nov Basonym: Gardnerella lacydonensis Allini Ntiguemassa et al.49 Basonym: Gardnerella lacydonensis Allini Ntiguemassa et al.49 Gardnerella vaginalis subsp. lacydonensis (la.cy.do.nen’sis. N.L. fem. adj. lacydonensis, from Lacydon, the name of the ancient port of Marseille, the French city where the strain was first described). The description of Gardnerella vaginalis subsp. lacydonensis remains as given to Gardnerella lacydonensis by Allini Ntiguemassa et al.49 with the following additional information. Our collection of curated genomes contains 15 Gardnerella vaginalis subsp. lacydonensis cultivated strains, isolated from the vagina, endometrium, and ectocervical mucosa of humans from France, Kenya, South Africa, the United States, and Zambia (Supplementary Data 2). The median genome length of these isolates is 1.66 Mb [range 1.62–1.72] and the median G + C content is 41.29% [range 41.11–41.42%]. The type strain of Gardnerella vaginalis subsp. lacydonensis remains Marseille-Q9181T (=CSUR Q9181T = CECT 31121T). Description of Gardnerella vaginalis subsp. regina subsp. nov Gardnerella vaginalis subsp. regina (re.gi’na. L. fem. n. regina, the queen, referring to its placement in the first validly published species of the Gardnerella genus). We analyzed 17 curated genomes from cultivated strains of Gardnerella vaginalis subsp. regina, isolated from the vagina of humans from Bangladesh, France, the United States, and Zambia (Supplementary Data 2). The median genome length of the Gardnerella vaginalis subsp. regina is 1.63 Mb [range 1.57–1.67], and the median G + C content is 41.41% [range 41.25–41.62%]. The isolate is a facultative anaerobe and has gram variable cells. Growth develops on Brucella agar in anaerobic conditions at 37 °C after 1–2 days. In a genome comparison of Gardnerella vaginalis subsp. regina strain DNF00622AT with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with its sister subspecies Gardnerella vaginalis subsp. vaginalis ATCC14018T (96.82%) and Gardnerella vaginalis subsp. lacydonensis Marseille-Q9181T (96.17%), followed by Gardnerella piotii subsp. piotii UGent 18.01T (89.24%) (Fig. 2). Within the genome, 83 CAZymes were identified, including α-mannosidase (GH125), β-galactosidase (GH2), β-mannosidase (GH130), Class II α-mannosidase (GH38), β-N-acetylglucosaminidase (GH20), α-L-fucosidase (GH29), and sialidase (GH33)114. KEGG analysis identified a total of 97 transporters, 13 secretion genes, and 365 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of cysteine, glutamine, glycine, and proline and catabolism of ethanol, gluconate, and serine (Supplementary Data 6). The type strain for Gardnerella vaginalis subsp. regina is DNF00622AT, isolated from the vaginal fluid of a woman positive for bacterial vaginosis in Seattle, Washington, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for DNF00622AT are PZ013896 and CP058395. Strains used in previous classifying publications: 1-004-3, 819-1, and 819-3. Description of Gardnerella piotii subsp. piotii (Vaneechoutte et al.32) Bouzek et al. comb. nov Basonym: Gardnerella piotii Vaneechoutte et al.32 Gardnerella piotii (pi.o’ti.i. N.L. gen. masc. n. piotii, of Piot, named after Peter Piot, a Belgian microbiologist who first attempted to distinguish different groups within G. vaginalis). The description of Gardnerella piotii subsp. piotii remains as given to Gardnerella piotii by Vaneechoutte et al.32 with the following additional information. There are 17 genomes of cultivated Gardnerella piotii subsp. piotii in our curated collection, isolated from the vagina and endometrium of humans from Bangladesh, Belgium, China, France, and the United States (Supplementary Data 2). The median genome length of these isolates is 1.54 Mb [range 1.50–1.59] and the median G + C content is 42.43% [range 42.22–42.61%]. The Gardnerella piotii subsp. piotii type strain remains UGent 18.01 T (=LMG 30818T = CCUG 72427T). Description of Gardnerella piotii subsp. pickettii (Sousa et al.48) Bouzek et al. comb. nov Basonym: Gardnerella pickettii Sousa et al.48 Gardnerella piotii subsp. pickettii (pic.ket’ti.i. N.L. gen. masc. n. pickettii, of Pickett, named after M. John Pickett, an American microbiologist who proposed the transfer of Haemophilus vaginalis to the genus Gardnerella). The description of Gardnerella piotii subsp. pickettii remains as given to Gardnerella pickettii by Sousa et al.48 with the following additional information. We included the curated genomes of 40 cultivated isolates of Gardnerella piotii subsp. pickettii in our analysis, isolated from the vagina, endometrium, ectocervical mucosa, urine, bladder and rectum of humans from Bangladesh, Canada, China, Kenya, Portugal, South Africa, the United States, and Zambia (Supplementary Data 2). The median genome length of Gardnerella piotii subsp. pickettii is 1.56 Mb [range 1.51–1.68] and the median G + C content is 42.32% [range 42.00–42.74%]. The type strain of Gardnerella piotii subsp. pickettii remains c17Ua_112T (=DSM 113414T = CCP 71T). Description of Gardnerella piotii subsp. sialicivorans subsp. nov Gardnerella piotii subsp. sialicivorans (si.a.li.ci.vo’rans. N.L. neut. n. acidum sialicum, sialic acid; L. pres. part. vorans, devouring; N.L. fem. part. adj. sialicivorans, sialic acid devouring). The subspecies Gardnerella piotii subsp. sialicivorans is named for the intact sialidase catabolism operon and one or more sialidase genes present in the genomes of Gardnerella piotii. Our collection of curated genomes includes three Gardnerella piotii subsp. sialicivorans cultivated strains, all isolated from the vagina of humans from the United States (Supplementary Data 2). The median genome length is 1.51 Mb [range 1.50–1.53], and the median G + C content is 42.48% [range 42.39–42.48%]. This isolate is a gram-positive coccus and produces pinpoint colonies (≤1 mm) on Brucella agar after 3 days of anaerobic incubation at 37 °C. Colonies are gray, pleomorphic, and alpha-hemolytic. In a genome comparison of Gardnerella piotii subsp. sialicivorans strain C0013B2T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with members of its sister subspecies Gardnerella piotii subsp. pickettii c17Ua_112T (95.68%) and Gardnerella piotii subsp. piotii UGent 18.01T (94.66%), followed by Gardnerella rara C0101A1T (90.88%) (Fig. 2). Within the genome, 76 CAZymes were identified, including sialidase (GH33) and Class II α-mannosidase (GH38)114. KEGG analysis identified a total of 79 transporters, 13 secretion genes, and 355 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of glutamine, glycine, and proline, and catabolism of alanine, ethanol, and serine (Supplementary Data 6). The Gardnerella piotii subsp. sialicivorans type strain is C0013B2T, isolated from the vagina of an asymptomatic menstruating woman in Birmingham, Alabama, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for C0013B2T are PZ016854 and NZ_JBWPXA000000000.1. Strain used in previous classifying publications: 582. Description of Gardnerella rara sp. nov Gardnerella rara (ra’ra. L. fem. adj. rara, rare, uncommon, referring to its rare genitourinary colonization of humans). Our curated collection includes six genomes of Gardnerella rara from isolates cultivated from the vagina and endometrium of humans in Bangladesh, the United States, and Zambia (Supplementary Data 2). The median genome length of Gardnerella rara is 1.52 Mb [range 1.47–1.58] and the median G + C content is 43.23% [range 43.01–43.29%]. This isolate is a gram-positive bacillus and produces small colonies on Brucella agar after 2 days of anaerobic incubation at 37 °C. Colonies are gray, round, and alpha-hemolytic. In a genome comparison of Gardnerella rara strain C0101A1T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with Gardnerella piotii subsp. piotii UGent 18.01T at 91.38% (Fig. 2). Within the genome, 71 CAZymes were identified, including β-mannosidase (GH130), Class II α-mannosidase (GH38), and sialidase (GH33)114. KEGG analysis identified a total of 78 transporters, 13 secretion genes, and 344 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of glutamine, glycine, and proline, and catabolism of alanine and ethanol (Supplementary Data 6). The type strain of Gardnerella rara type is C0101A1T, isolated from the vagina of a woman positive for Chlamydia trachomatis in Baltimore, Maryland, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for C0101A1T are PZ016855 and NZ_JBVBJP000000000.1. Strains used in previous classifying publications: 1401E, GED7760B, and HUZ601048-2. Description of Gardnerella leopoldii subsp. leopoldii (Vaneechoutte et al.32) Bouzek et al. comb. nov Basonym: Gardnerella leopoldii Vaneechoutte et al.32 Gardnerella leopoldii (le.o.pol’di.i. N.L. gen. masc. n. leopoldii, of Leopold, named after Sidney Leopold, who described the first strains of the species). The description of this subspecies remains as given to Gardnerella leopoldii by Vaneechoutte et al.32 with the following additional information. There are 24 genomes of cultivated isolates of Gardnerella leopoldii subsp. leopoldii in our curated collection, isolated from the vagina and catheter urine of humans from Belgium, Canada, France, Kenya, and the United States (Supplementary Data 2). The median genome length is 1.53 Mb [range 1.37–1.67], and the median G + C content is 42.15% [range 42.07–42.34%]. The type strain of Gardnerella leopoldii subsp. leopoldii remains UGent 06.41 T (=LMG 30814T = CCUG 72425T). Description of Gardnerella leopoldii subsp. swidsinskii (Vaneechoutte et al.32) Bouzek et al. comb. nov Basonym: Gardnerella swidsinskii Vaneechoutte et al.32 Gardnerella swidsinskii (swid.sins’ki.i. N.L. gen. masc. n. swidsinskii, of Swidsinski, named after Alexander Swidsinski, a German microbiologist, who recognized that clue cells in Gram stains of vaginal swabs of women with BV are indicative for biofilm formation by Gardnerella). The description of this subspecies remains as provided by Vaneechoutte et al. for Gardnerella swidsinskii32 with the following additional information. We included 46 genomes of cultivated Gardnerella leopoldii subsp. swidsinskii in our curated collection, isolated from the vagina, endometrium, rectum, and catheter urine of humans from China, France, Russia, and the United States (Supplementary Data 2). The median genome length is 1.63 Mb [range 1.52–1.72], and the median G + C content is 41.93% [range 41.64–42.29%]. The type strain of Gardnerella leopoldii subsp. swidsinskii remains GS 9838-1T (=LMG 30812T = CCUG 72429T). Description of Gardnerella leopoldii subsp. aggregata subsp. nov Gardnerella leopoldii subsp. aggregata (ag.gre.ga’ta. L. fem. part. adj. aggregata, joined together, referring to the presence of genes related to cell aggregation). The genomes of Gardnerella leopoldii subsp. aggregata, as well as those of the overarching species, differ from other Gardnerella species in the presence of genes consistent with adherence and biofilm formation. There are five genomes of cultivated Gardnerella leopoldii subsp. aggregata in our curated collection, isolated from the vagina and ectocervical mucosa of humans from Bangladesh, South Africa the United States, and Zambia (Supplementary Data 2). The median genome length is 1.48 Mb [range 1.48–1.53], and the median G + C content is 42.19% [range 42.11–42.50%]. This isolate is a gram variable coccobacillus and produces pinpoint colonies (≤1 mm) on Brucella agar after 3 days of anaerobic incubation at 37 °C. Colonies are glossy white, round, convex, and alpha-hemolytic. In a genome comparison of Gardnerella leopoldii subsp. aggregata strain C0026F5T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with its sister subspecies Gardnerella leopoldii subsp. phocaeensis Marseille-Q9179T (96.09%), Gardnerella leopoldii subsp. leopoldii UGent 06.41T (95.58%), and Gardnerella leopoldii subsp. swidsinskii GS 9838-1T (94.70%), followed by Gardnerella greenwoodii subsp. bretellae Marseille-QA0894T (87.40%) (Fig. 2). Within the genome, 74 CAZymes were identified, including α-glucosidase (GH65), endo-β-galactosidase (GH16), and β-glucosidase (GH3)114. KEGG analysis identified a total of 84 transporters, 14 secretion genes, and 353 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of glutamine, glycine, methionine, proline, and threonine, and catabolism of ethanol (Supplementary Data 6). The type strain of Gardnerella leopoldii subsp. aggregata is C0026F5T, isolated from the vagina of an asymptomatic menstruating woman from Birmingham, Alabama, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for C0026F5T are PZ016853 and NZ_JBWPWP000000000.1. Description of Gardnerella leopoldii subsp. phocaeensis (Allini Ntiguemassa et al.49) Bouzek et al. comb. nov Basonym: Gardnerella phocaeensis Allini Ntiguemassa et al.49 Gardnerella phocaeensis (pho.cae.en’sis. L. fem. adj. phocaeensis, referring to Phocaea, the name of the Ionian Greek city where the founders of Marseille came from. The strain was isolated in Marseille). The description of Gardnerella leopoldii subsp. phocaeensis remains as given to Gardnerella phocaeensis by Allini Ntiguemassa et al.49 with the following additional information. There are eleven genomes of cultivated Gardnerella leopoldii subsp. phocaeensis in our curated collection, isolated from the vagina and endometrium of humans from Bangladesh, France, the United States, and Zambia (Supplementary Data 2). The median genome length of Gardnerella leopoldii subsp. phocaeensis is 1.54 Mb [range 1.52–1.66] and the median G + C content is 42.26% [range 42.03–42.43%]. The type strain of Gardnerella leopoldii subsp. phocaeensis remains Marseille-Q9179T (=CSUR Q9179T = CECT 31120T). Description of Gardnerella greenwoodii subsp. greenwoodii (Sousa et al.48) Bouzek et al. comb. nov Basonym: Gardnerella greenwoodii Sousa et al.48 Gardnerella greenwoodii (green.woo’di.i. N.L. gen. masc. n. greenwoodii, of Greenwood, named after James Greenwood, an American microbiologist who proposed the transfer of Haemophilus vaginalis to the genus Gardnerella). The description of this subspecies remains as given to Gardnerella greenwoodii by Sousa et al.48 with the following additional information. We included 13 genomes of Gardnerella greenwoodii subsp. greenwoodii in our curated collection, cultivated from the vagina, endometrium, and urine of humans from France, Portugal, the United Kingdom, and the United States (Supplementary Data 2). The median genome length of Gardnerella greenwoodii subsp. greenwoodii is 1.54 Mb [range 1.49–1.65] and the median G + C content is 43.40% [range 42.90–43.44%]. The type strain of Gardnerella greenwoodii subsp. greenwoodii remains c31Ua_26T (=DSM 113415T = CCP 72T). Description of Gardnerella greenwoodii subsp. intermedia subsp. nov Gardnerella greenwoodii subsp. intermedia (in.ter.me’di.a. L. fem. adj. intermedia, intermediate, between, referencing the shared possession of operons with both Gardnerella subgenera). Pangenome analysis revealed that Gardnerella greenwoodii subsp. intermedia, as well as other members of the species, shares biologically relevant operons with the Set A subgenus (Gardnerella vaginalis, Gardnerella piotii and Gardnerella rara) while also falling within the Set B subgenus, suggesting that the species is an intermediate between Set A and Set B. Our curated set of genomes contains two strains of Gardnerella greenwoodii subsp. intermedia, both cultivated from the vagina of humans from the United States (Supplementary Data 2). The lengths of these genomes are 1.53 and 1.55 Mb, and the G + C contents are 43.04% and 43.05%. Cells are gram variable. Optimal growth occurs within 2–3 days on Brucella agar in an anerobic atmosphere at 37 °C but can occur in aerobic environments. In a genome comparison of Gardnerella greenwoodii subsp. intermedia strain KA00603T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with sister subspecies Gardnerella greenwoodii subsp. bretellae Marseille-QA0894T (95.76%) and Gardnerella greenwoodii subsp. greenwoodii c31Ua_26T (95.04%), followed by Gardnerella kenyensis NR047T (88.32%) (Fig. 2). Within the genome, 66 CAZymes were identified, including β-glucosidase (GH3), and KEGG analysis identified a total of 85 transporters, 13 secretion genes, and 347 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of cysteine, glutamine, glycine, methionine, proline, and threonine, and catabolism of ethanol and serine (Supplementary Data 6). The type strain of Gardnerella greenwoodii subsp. intermedia is KA00603T isolated from the vaginal fluid of a woman positive for bacterial vaginosis in Seattle, WA, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for KA00603T are PZ013893 and CP058371. Strain used in previous classifying publications: 1500E. Description of Gardnerella greenwoodii subsp. bretellae (Allini Ntiguemassa et al.49) Bouzek et al. comb. nov Basonym: Gardnerella bretellae Allini Ntiguemassa et al.49 Gardnerella bretellae (bre.tel’lae. N.L. gen. fem. n. bretellae, honoring Florence Bretelle for her contribution to the description of vaginal microbiota). The description of Gardnerella greenwoodii subsp. bretellae remains as given to Gardnerella bretellae by Allini Ntiguemassa et al.49 with the following additional information. Our curated collection of genomes contains eleven strains of Gardnerella greenwoodii subsp. bretellae, cultivated from the vagina and ectocervical mucosa of humans from Canada, France, Kenya, South Africa, the United States, and Zambia (Supplementary Data 2). The median genome length is 1.52 Mb [range 1.50–1.64], and the median G + C content is 43.22% [range 43.05–43.50%]. The Gardnerella greenwoodii subsp. bretellae type strain remains Marseille-QA0894T (=CSUR QA0894T = CECT 31122T). Description of Gardnerella washingtoni sp. nov Gardnerella washingtoni (wa.shing.to’ni. N.L. gen. n. washingtoni, named in honor of Washington University in St. Louis, MO, USA, where the first two genomes of this species originated, and the University of Washington in Seattle, WA, USA, which is associated with the first complete genome of this species). There are five genomes of Gardnerella washingtoni in our curated collection. These isolates have been cultivated from the vagina and urine of humans in the United States (Supplementary Data 2). The median genome length of Gardnerella washingtoni is 1.60 Mb [range 1.56–1.65] and the median G + C content is 42.96% [range 42.89–43.11%]. Cells are Gram variable. The isolate is a facultative anaerobe that grows on Brucella agar in anaerobic conditions at 37 °C in 1–2 days. In a genome comparison of Gardnerella washingtoni strain DNF00536T with validly published Gardnerella type strains and the type strains proposed in this study, Gardnerella greenwoodii subsp. bretellae Marseille-QA0894T has the highest ANI value (87.40%) (Fig. 2). Within the genome, 69 CAZymes were identified, including endo-β-galactosidase (GH16), α-glucosidase (GH65), and β-glucosidase (GH3)114. KEGG analysis identified a total of 83 transporters, 13 secretion genes, and 348 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of cysteine, glutamine, glycine, methionine, and threonine, and catabolism of ethanol (Supplementary Data 6). The type strain of Gardnerella washingtoni is DNF00536T, isolated from the vaginal fluid of a woman positive for bacterial vaginosis in Seattle, WA, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for DNF00536T are PZ013897 and CP058398. Strains used in previous classifying publications: JCP8481A, JCP8481B, and PSS7772B. Description of Gardnerella bivia sp. nov Gardnerella bivia (bi’vi.a. L. fem. adj. bivia, having two ways, in reference to the genome containing genes for two major pathways for mucin degradation). Gardnerella bivia has genes associated with two major pathways for mucin degradation. There are six isolates of this species in our curated collection, cultivated from the vagina of humans in Bangladesh, the United States, and Zambia (Supplementary Data 2). The median genome length of Gardnerella bivia is 1.59 Mb [range 1.55–1.64] and the median G + C content is 38.08% [range 37.90–38.31%]. The isolate is a gram variable obligate anaerobe. Growth occurs on Brucella agar within 1–2 days in anaerobic conditions at 37 °C. In a genome comparison of Gardnerella bivia strain DNF00502T with validly published Gardnerella type strains and the type strains proposed in this study, Gardnerella hutchinsoni KA00225T has the highest ANI value (82.91%) (Fig. 2). Within the genome, 75 CAZymes were identified, including β-galactoside phosphorylase (GH112), α-L-fucosidase (GH29), and β-glucosidase (GH3)114. KEGG analysis identified a total of 74 transporters, 13 secretion genes, and 349 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of cysteine, glutamine, glycine, proline, and threonine, and catabolism of acetate, ethanol, and serine (Supplementary Data 6). The type strain of Gardnerella bivia is DNF00502T, isolated from the vaginal fluid of a woman positive for bacterial vaginosis in Seattle, WA, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for DNF00502T are PZ013890 and CP058399. Strain used in previous classifying publications: CMW7778B. Description of Gardnerella hutchinsoni sp. nov Gardnerella hutchinsoni (hutch.in.so’ni. N.L. gen. n. hutchinsoni, named in honor of Fred Hutchinson Cancer Center, where the isolate was first cultivated and the first complete genomes of this subspecies were sequenced). We included six genomes of Gardnerella hutchinsoni isolates in our curated collection, all cultivated from the vagina of humans in the United States and Zambia (Supplementary Data 2). The median genome length of Gardnerella hutchinsoni is 1.67 Mb [range 1.64–1.72] and the median G + C content is 40.64% [range 40.53–40.76%]. Cells are gram variable. Optimal growth occurs within 1–2 days on Brucella agar in an anerobic atmosphere at 37 °C but can occur in aerobic environments. In a genome comparison of Gardnerella hutchinsoni strain KA00225T with validly published Gardnerella type strains and the type strains proposed in this study, Gardnerella leopoldii subsp. aggregata C0026F5T has the highest ANI value (86.69%) (Fig. 2). Within the genome, 69 CAZymes were identified, including β-galactoside phosphorylase (GH112) and β-glucosidase (GH3)114. KEGG analysis identified a total of 70 transporters, 13 secretion genes, and 341 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of asparagine, glutamine, glycine, methionine, and proline, and catabolism of ethanol (Supplementary Data 6). The type strain of Gardnerella hutchinsoni is KA00225T, isolated from the vaginal fluid of a woman positive for bacterial vaginosis in Seattle, WA, USA. Accessions for the 16S rRNA gene sequence and whole genome sequence for KA00225T are PZ013891 and CP058376. Strains used in previous classifying publications: 3-005-3V and HUZ601990-0. Description of Candidatus Gardnerella kenyensis sp. nov Gardnerella kenyensis (ken.yen’sis. N.L. fem. adj. kenyensis, from Kenya, referring to the country of isolation of the type strain). All known strains of Candidatus Gardnerella kenyensis to date (NR002, NR021, and NR047) have been isolated from the vaginal fluid of a human in Nairobi, Kenya (Supplementary Data 2). Two of these are genomic duplicates (NR002 and NR047) and one (NR021) did not pass our genome quality threshold metrics. As such, only NR047T was included in our curated collection (Supplementary Data 2). The genome length of this isolate is 1.66 Mb, and its G + C content is 43.62%. In a genome comparison of Candidatus Gardnerella kenyensis strain NR047T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with Gardnerella insolita subsp. libera C0085C4T (90.59%) (Fig. 2). Within the genome, 66 CAZymes were identified, including sialidase (GH33), and KEGG analysis identified a total of 84 transporters, 13 secretion genes, and 340 enzyme114. Metabolism predictions using GapMind104–106 include biosynthesis of asparagine, cysteine, glutamine, glycine, methionine, proline, and threonine, and catabolism of ethanol (Supplementary Data 6). The type strain of Candidatus Gardnerella kenyensis is NR047T and the accession for the whole genome sequence is NQOH00000000.2. Description of Gardnerella insolita sp. nov Gardnerella insolita (in.so’li.ta. L. fem. adj. insolita, unusual, strange, uncommon, referring to the low number of species cultivated). To date, only three strains of Gardnerella insolita have been cultivated, all isolated from the vagina of humans in the United States and Zambia (Supplementary Data 2). The median genome length of Gardnerella insolita is 1.61 Mb [range 1.59–1.66] and the median G + C content is 43.60% [range 41.27–43.80%]. The Gardnerella insolita type strain is C0396A3T, isolated from the vagina of a pregnant woman at 19 weeks of gestation and who delivered preterm in Zambia. Accession for the 16S rRNA gene sequence for C0396A3T is PZ236487. Description of Gardnerella insolita subsp. insolita comb. nov Gardnerella insolita (in.so’li.ta. L. fem. adj. insolita, unusual, strange, uncommon, referring to the low number of species cultivated). There are two strains of Gardnerella insolita subsp. insolita in our curated genome collection. The genome lengths of these isolates are 1.66 and 1.61 Mb, and the G + C contents are 41.27% and 43.80%. This isolate is a gram-positive bacillus and produces pinpoint (≤1 mm) colonies on Brucella agar after 2 days of anaerobic incubation at 37 °C. Colonies are round, clear, and glossy. In a genome comparison of Gardnerella insolita subsp. insolita strain C0396A3T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with Gardnerella insolita subsp. libera C0085C4T (95.24%), followed by Gardnerella kenyensis NR047T (89.89%) (Fig. 2). Within the genome, 66 CAZymes were identified, including β-glucosidase (GH3)114. KEGG analysis identified a total of 85 transporters, 13 secretion genes, and 347 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of asparagine, glutamine, glycine, proline, and threonine, and catabolism of ethanol (Supplementary Data 6). Strain used in previous classifying publications: 1-004-1. Description of Gardnerella insolita subsp. libera subsp. nov Gardnerella insolita subsp. libera (li’be.ra. L. fem. adj. libera, independent, free, referring to the rarity of the subspecies). To date, only one strain of Gardnerella insolita subsp. libera has been cultivated, C0085C4T, from the vagina of a woman positive for Chlamydia trachomatis in Baltimore, Maryland, USA. Its genome length is 1.59 Mb, and its G + C content is 43.60%. This isolate is a gram variable coccobacillus and produces pinpoint colonies (≤1 mm) on Brucella agar after 2 days of anaerobic growth at 37˚C. Colonies are gray, glistening, round, and β-hemolytic. In a genome comparison of Gardnerella insolita subsp. libera strain C0085C4T with validly published Gardnerella type strains and the type strains proposed in this study, the highest ANI values correspond with Gardnerella insolita subsp. insolita C0396A3T (95.30%), followed by Candidatus Gardnerella kenyensis NR047T (90.82%) (Fig. 2). Within the genome, 70 CAZymes were identified, including Class II α-mannosidase and α-glucosidase114. KEGG analysis identified a total of 84 transporters, 12 secretion genes, and 342 enzymes114. Metabolism predictions using GapMind104–106 include biosynthesis of cysteine, glutamine, glycine, proline, and methionine, and catabolism of ethanol and serine (Supplementary Data 6). The type strain of Gardnerella insolita subsp. libera is C0085C4T. Accessions for the 16S rRNA gene sequence and whole genome sequence for C0085C4T are PZ016852 and JBWPJS000000000.1. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Description of Additional Supplementary Files Supplementary Fig. 1–9 Supplementary Data 1–11 Reporting Summary Transparent Peer Review file Description of Additional Supplementary Files Supplementary Fig. 1–9 Supplementary Data 1–11 Reporting Summary Transparent Peer Review file Source data Source Data Source Data