Cysteine dependence in Lactobacillus iners constitutes a novel therapeutic target to modify the vaginal microbiota

Vaginal microbiota composition affects several important reproductive health outcomes. Lactobacillus crispatus-dominant bacterial communities have favorable associations whereas anaerobe-dominant communities deficient of lactobacilli are linked to poor outcomes, including bacterial vaginosis (BV). Lactobacillus iners, the most abundant vaginal species worldwide, has adverse associations compared to L. crispatus, but standard metronidazole treatment for BV promotes L. iners-dominance, likely contributing to post-treatment relapse. L. iners is under-studied because it fails to grow in standard Lactobacillus media in vitro. Here we trace this in vitro phenotype to a species-specific cysteine requirement associated with limitations in cysteine-related transport mechanisms and show that vaginal cysteine concentrations correlate with Lactobacillus abundance in vivo. We demonstrate that cystine uptake inhibitors selectively impede L. iners growth and that combining an inhibitor with metronidazole thus promotes L. crispatus dominance of defined BV-like communities. These findings identify a novel target for therapeutic vaginal microbiota modulation to improve reproductive health.

Diverse, anaerobe-dominant bacterial communities are associated with negative sequalae and are a hallmark of bacterial vaginosis (BV) 8,9 , a syndrome involving vaginal discharge and mucosal inflammation that affects up to 58% of women worldwide, with disproportionately high impact in sub-Saharan Africa 10 . By contrast, bacterial communities dominated by Lactobacillus crispatus and most other FGT Lactobacillus species are generally considered optimal for health 9 . One notable exception is Lactobacillus iners-dominant communities, which are associated with many of the same unfavorable outcomes as BV 1,7,11 and have higher risk of transitioning to BV and BVlike communities 3,[12][13][14][15] . L. iners is the most prevalent and abundant vaginal Lactobacillus species worldwide, but factors influencing the balance between L. iners and more health-associated lactobacilli are incompletely understood 1, [16][17][18][19] .
The close associations between vaginal microbiota and disease make microbiota-targeted therapies a key component of interventions to improve women's health and reproductive outcomes 8 . Therapeutic microbiota modulation has primarily been studied in relation to BV, but current BV treatments exhibit low efficacy and high recurrence rates. Over 50% of women with symptomatic BV who receive standard therapy with the antibiotic metronidazole (MTZ) experience recurrence by 12 months, while up to 80% of women with a prior history of recurrent BV experience relapse within 16 weeks [20][21][22] . MTZ typically shifts BV-associated microbiota towards L. iners-dominant communities [23][24][25][26][27] , which have high probability of transition back to BV-sufficient for this growth phenotype (Extended Data Fig. 1a-f; reagent details in Supplementary   Table 1 and 2). L-glutamine (L-Gln) supplementation was neither necessary nor sufficient, but augmented growth in presence of L-Cys, so subsequent experiments were performed using a base of L-Gln supplemented MRS broth ("MRSQ").
In order to assess species-level generalizability of the Cys-dependent growth phenotype, we established a large, geographically diverse L. iners isolate collection and genome catalog of strains from women both with and without BV. The isolate collection was assembled from culture repositories as well as novel US and South African strains isolated on L-Cys-supplemented MRS agar or blood agar. The associated genome catalog consists of 327 genomes from disparate human populations, including genomes from the novel isolates and "reference" isolate genomes retrieved from RefSeq 38,39 . To enhance catalog diversity and minimize risk of isolation bias, we also included culture-independent metagenome-assembled genomes (MAGs) generated from FGT shotgun metagenomic sequencing studies of U.S., South African, Italian, and Chinese populations 39 . A species-level phylogenetic reconstruction based on this catalog revealed that L.
iners genomic diversity was largely independent of geographic source, genome type, or clinical context (Fig. 1b). South African and US L. iners strains broadly representative of species diversity were selected for further experimental characterization ( Fig. 1b Fig. 1c), confirming Cys-dependent growth as a specieslevel phenotype.

Vaginal Lactobacillus species lack canonical Cys biosynthesis pathways
We next investigated whether L. iners' species-specific in vitro Cys-dependent growth phenotype was due to differences in Cys biosynthetic capacity as compared to other common FGT lactobacilli. Bacteria canonically synthesize Cys de novo from serine (Ser) via either the Cys synthase pathway (cysE, followed by cysK, cysM, or cysO) or reverse transsulfuration pathway (cbs or mccA, followed by mccB; Fig. 2a) 40 . We assessed for presence of these pathways in our catalog of 327 L. iners genomes, along with similarly constructed catalogs for L. crispatus (n = 508 genomes), L. gasseri (n = 216), L. jensenii (n = 137), and L. vaginalis (n = 30) (Extended Data Fig. 2a,b) 39 . These catalogs represent substantial expansion of known genomic diversity for all species, most notably for L. iners with a >10-fold increase in the number of genomes and a 73% increase in pan-genome size compared to prior RefSeq-derived genomes (Extended Data Fig.  2b,c). No genomes from any of the species were predicted to possess intact Cys synthase or reverse transsulfuration pathways based on gene annotations performed using eggNOG 41 , although interestingly all non-iners species had predicted mccB orthologs ( Fig. 2b and Supplementary Table   4). Absence of intact Cys synthase and reverse transsulfuration pathways in these species was further supported by results of BLAST 42  We phenotypically tested this result by measuring the ability of L. iners and L. crispatus to convert Ser into Cys in vitro. Bacteria were grown in MRSQ broth supplemented with isotopically labeled Ser ( 13 C-L-Ser) plus homocysteine or with labeled L-Cys ( 13 C-L-Cys; labeling scheme as in Fig. 2a). Bacteria were lysed and hydrolyzed, then cellular amino acid isotopic labeling patterns were analyzed via liquid chromatography-mass spectrometry (LC-MS). Comparison to bacteria grown without isotopic tracers revealed that both L. iners and L. crispatus took up labeled Ser but failed to convert it to Cys (Fig. 2c,d), consistent with genomic predictions. By contrast, both strains readily incorporated labeled Cys, supporting the genomic prediction that despite their different growth phenotypes in MRS broth, both utilize exogenous Cys.

Vaginal Cys concentrations are lower in women with BV and strongly correlate with in vivo Lactobacillus abundance
Given the prediction that vaginal lactobacilli lack canonical Cys biosynthesis pathways, we investigated in vivo relationships between vaginal Cys concentrations, BV, and Lactobacillus colonization. We measured Cys concentrations in cervicovaginal lavage (CVL) fluid from 142 participants in a cohort of South African women aged 18-23 years 1,4,44 , 53 of whom were also evaluated for BV using the Nugent scoring method 45 . Cys concentrations differed strikingly by BV status (p = 6.3 x 10 -9 ), with significantly higher concentrations among women without BV compared to those with BV (Fig. 3a). To explore underlying microbiota composition in this cohort, we performed bacterial 16S rRNA gene sequencing of cervical swab samples from all participants.
We next examined correlations between Cys concentrations and relative abundance of individual taxa within the microbiota. Analysis of key CT-associated genera revealed that Lactobacillus relative abundance had a strong positive correlation with Cys concentrations (r = 0.6, p = 5.2 x 10 -15 by Spearman correlation), while Prevotella abundance had equally strong negative correlation (r = -0.57, p = 1.9 x 10 -13 ) and Gardnerella abundance was weakly negatively correlated to Cys concentrations (Fig. 3d). L. crispatus and L. iners each positively correlated with Cys when analyzed individually (Extended Data Fig. 3b). To assess whether lactobacilli were the only bacteria correlated with Cys, we calculated correlations for each taxon detected in ³50% of samples at either the genus or species level (Extended Fig 3c,d), adjusting for multiple comparisons. Lactobacillus (genus-level) and L. crispatus and L. iners (species-level) were the only taxa positively correlated with Cys (Fig. 3e,f; full statistical results in Supplementary Tables   8 & 9). Other taxa, including various BV-associated bacteria, showed no correlation or had significant negative correlations with Cys concentrations. Results were similar for the Cyscontaining peptides reduced glutathione (GSH) and cysteinylglycine (Extended Fig. 4 and Supplementary Tables 8 & 9). Thus, vaginal Cys concentration is higher among women without BV and preferentially correlates with abundance of both L. iners and L. crispatus, consistent with the hypothesis that Cys availability is important for Lactobacillus colonization success in vivo.

L. iners utilizes L-cystine but has limited capacity to exploit more complex L-Cys sources
The species-specific nature of in vitro Cys-dependence for L. iners appeared discordant with the finding that other Lactobacillus species also lack canonical Cys biosynthesis pathways and that abundance of both L. iners and L. crispatus in vivo correlates with high vaginal Cys concentrations.
We therefore hypothesized that L. iners' unique in vitro growth phenotype was due to restricted capacity to utilize exogenous Cys sources in MRS broth compared to other lactobacilli.
Alternatively, Cys supplementation might support L. iners growth by acting as a chemical reducing agent rather than a direct nutritional supplement. Measurement of concentrations of free molecular Cys and cystine (its oxidized counterpart) in MRSQ broth revealed that they were ≤4.6 µM and 0.7 µM respectively, below levels required for growth by Cys-auxotrophic E. coli strains 46 .
Supplementation with L-cystine produced similar L. iners growth as L-Cys supplementation (Fig   4a), even in presence of the oxidizing agent hydrogen peroxide (H2O2; Extended Fig 5b), demonstrating that L-cystine supports growth by acting as a direct nutritional supplement rather than causing chemical reduction of the media. Thus L. iners does not require a chemically reduced environment to grow when it has access to a bioavailable source of L-Cys (including L-cystine), but L-Cys bioavailability in un-supplemented MRSQ broth is inadequate for L. iners.
The ability of non-iners lactobacilli to grow in un-supplemented MRS broth implied presence of other Cys sources that L. iners failed to utilize (MRS contains yeast and beef extracts, which contribute protein-derived peptides in addition to monomeric amino acids). Adding chemical reducing agents to MRSQ broth substantially raised Cys and GSH concentrations (Extended Data Fig. 5a), demonstrating that most Cys and Cys-containing molecules in unsupplemented media exist as mixed disulfide compounds. Addition of chemical reducing agents other than L-Cys also supported L. iners growth (Extended Data Fig. 5b,c), but in contrast to Lcystine, the oxidized counterparts of these reducing agents did not. Thus MRS broth contains complex sources of L-Cys, including mixed disulfides, that L. iners has restricted capacity to utilize.

L. iners lacks Cys-related transport mechanisms possessed by other vaginal lactobacilli and its growth is selectively inhibited by cystine uptake inhibitors
We hypothesized that L. iners's restricted ability to use exogenous Cys sources might reflect a limited repertoire of transport mechanisms for Cys and Cys-containing molecules in comparison to other lactobacilli. Cys transporters are not well-characterized in bacteria compared to transporters for other amino acids 47 . Uptake of Cys and Cys-containing molecules is described as occurring through multiple known or putative mechanisms. These include the well-characterized cystine transport systems TcyABC, TcyJKLMN, and TcyP 48 , a putative Cys ABC transporter associated with the Cys-binding protein CjaA 47,49 , and the heterodimeric ABC transporter CydDC encoded by the redox-regulating locus cydABCD 50 , a conserved system that exports both Cys and 13 GSH to the periplasm in E. coli 50 and is proposed to perform glutathione uptake in lactobacilli 51 .
Analysis of our expanded Lactobacillus genome catalogs annotated using eggNOG revealed that genomes of all non-iners species encoded cydABCD, while L. crispatus, L. vaginalis, and some L.
gasseri encoded one or more predicted cjaA orthologs and L. jensenii, L. gasseri, and L. vaginalis each encoded one or more cystine transport systems ( Fig. 4b and Supplementary Table 4). Lowaffinity Cys uptake can also occur via the branched-chain amino acid importer LivJ/LivKHMGF 47 , but no species possessed livKHMGF orthologs except 2 L. iners MAGs (out of 327 genomes), nor did any species possess the full GsiABCD glutathione transport system (Extended Data Fig 6a and Supplementary L. iners has thus far proved refractory to genetic manipulation and experimental tools to perform genetic screens are lacking 34 , so we used phenotypic approaches to characterize the breadth of its Cys-related transport mechanisms. Isotopic tracing (Fig. 4c) revealed that L. iners incorporated labeled L-cystine ( 13 C2-L-cystine) at levels similar to L. crispatus but failed to take up labeled GSH ( 13 C-GSH; synthesis detailed in Extended Data Fig. 7), consistent with predicted absence of GSH transport activity ( Fig. 4b and Extended Data Fig. 6a). We hypothesized that if L.
iners differed from other lactobacilli in relying on a more limited repertoire of transport mechanisms, its growth might be uniquely impacted by uptake inhibition. We therefore tested effects of the known cystine uptake inhibitors S-methyl-L-cysteine (SMC) and seleno-DL-cystine (SDLC) 48 on L. iners growth in L-Cys-supplemented MRSQ broth. SMC caused species-specific, dose-dependent growth inhibition of a diverse collection of L. iners strains at concentrations of 32-64 mM, while SDLC caused selective inhibition at a concentration of 0.25 mM (Fig. 4d,e and Extended Data Fig. 6b,c). Inhibitor potency is known to vary for different transporters, but these concentrations appear higher for SMC and slightly higher for SDLC than reported inhibitory concentrations for the TcyJKLMN and TcyP transporters in Bacillus subtilis 48 . To confirm that growth inhibition was not an artefact specific to MRS media, we assessed inhibition in NYCIII broth, a serum-containing, nutrient-rich medium that supports growth of diverse fastidious FGT bacteria 53 . Both SMC and SDLC inhibited L. iners growth in NYCIII broth, but not growth of L. crispatus (Fig. 4f). They also selectively inhibited L. iners growth in MRSQ broth treated with chemical reducing agents, supporting the conclusion that reducing agents promote L. iners growth by enhancing Cys bioavailability (Extended Data Fig. 6d). Collectively these results confirm the genomic prediction that L. iners possesses a uniquely restricted repertoire of Cys-related transport mechanisms, rendering it susceptible to selective growth inhibition.

SMC provides selective advantage to L. crispatus in direct competition with L. iners
To assess the functional significance of L. iners growth inhibition by cystine uptake inhibitors, we queried whether an inhibitor could shift the balance between L. iners and L. crispatus in mixed culture competition assays. We tested pairwise strain combinations of L. crispatus and L. iners in L-Cys-supplemented MRSQ broth at varying SMC concentrations, then measured ratios of L. iners to L. crispatus in the mixed cultures by 16S rRNA gene sequencing. Experimental controls were included to confirm expected growth patterns and absence of significant contamination during sample processing (Extended Data Fig. 8a-c). SMC significantly suppressed L. iners relative to L. crispatus in a dose-dependent fashion (representative experiments in Fig. 5a), confirming that it offers selective advantage to L. crispatus during direct competition.

Combining SMC with MTZ enhances L. crispatus dominance in mock BV-like communities
To assess whether cystine uptake inhibitors could augment BV therapy by preferentially promoting L. crispatus expansion, we investigated SMC's effects in combination with MTZ on mock BVlike bacterial communities grown in a rich nutritional milieu. Multiple South African L. crispatus isolates unexpectedly failed to grow in NYCIII broth, so we developed a novel serum-containing formulation named "S-broth" that supported the South African L. crispatus strains as well as diverse BV-associated bacteria (Extended Data Fig. 9a). Using S-broth, we tested effects of SMC and MTZ in pure cultures of L. crispatus, L. iners, G. vaginalis, Prevotella bivia, and Atopobium (Fannyhessea) vaginae. Neither MTZ nor SMC blocked L. crispatus growth, whereas SMC inhibited L. iners and MTZ suppressed both G. vaginalis and P. bivia (Fig. 5b). Our A. vaginae strain was relatively MTZ-resistant (a known phenotype of some strains from this species), but interestingly susceptibility was enhanced by addition of SMC.
We cultured defined communities of these species in S-broth and assessed whether adding SMC to MTZ enhanced L. crispatus community dominance. Each community contained a different pairwise combination of L. crispatus and L. iners strains (input ratios as in Extended Data Fig. 9b). We used 16S sequencing to quantify mock community composition, including technical controls to confirm input strain identity and absence of significant contamination during processing (Extended Data Fig. 9c,d). Cultivation in media without inhibitors produced diverse mixtures of all 5 species (representative examples in Fig. 5c). As hypothesized, SMC alone diminished L. iners relative abundance, MTZ alone suppressed BV-associated species, and MTZ combined with SMC preferentially favored L. crispatus. To quantify the competitive advantage for L. crispatus, we calculated ratios between L. crispatus relative abundance and the summed abundances of all other taxa (Fig. 5d). The combination of MTZ and SMC increased this ratio significantly more than MTZ alone, supporting the concept that cystine uptake inhibitors could complement MTZ in promoting L. crispatus dominance during BV treatment.
iners' biology has remained poorly characterized because it fails to grow in standard Lactobacillus MRS media, limiting laboratory-based experiments and resulting in a paucity of isolates and genomes 18,28 . Here we resolve this unique growth defect by identifying a species-specific requirement for exogenous Cys supplementation in L. iners, which we show corresponds to a uniquely limited repertoire of mechanisms for transporting exogenous Cys and Cys-containing molecules compared to other FGT lactobacilli. Employing functional assays along with a novel collection of isolates and >1200 Lactobacillus genomes derived from diverse geographical and clinical contexts, we extend and experimentally confirm predictions from smaller-scale genomic studies suggesting FGT lactobacilli lack Cys biosynthesis pathways 58 . We further demonstrate that vaginal Cys concentrations strongly correlate with Lactobacillus relative abundance in vivo in a South African cohort. However, compared to other lactobacilli, L. iners has a restricted repertoire of transport mechanisms for Cys and Cys-containing molecules, rendering it selectively susceptible to cystine uptake inhibitors. Combining a cystine uptake inhibitor with MTZ (a standard antibiotic used for BV treatment) enhances L. crispatus dominance in mock, BV-like bacterial communities by suppressing growth of L. iners. These advances establish an essential foundation for future characterization of L. iners' biology, diversity, and effects on human health -including its role in predisposition to BV and other adverse outcomes 1,7,11,12,14,15 -while simultaneously identifying a novel target for therapeutic vaginal microbiota modulation.
The close in vivo correlation between Cys and L. iners and L. crispatus abundance is consistent with the hypothesis that Cys availability may influence FGT microbiota composition.
A smaller US-based case-control study comparing metabolomes of women with and without BV found that vaginal Cys concentrations positively correlated with L. crispatus and L. jensenii and negatively correlated with various BV-associated species 59 . Neither strength nor statistical significance of individual species-metabolite correlations were reported in that study, but it appeared to show a weaker correlation between L. iners and Cys than the correlation we identify here 59 , which may reflect the different study designs or may indicate differences in species abundance and prevalence between the two study populations. Since FGT lactobacilli appear to lack de novo Cys biosynthetic capacity, we hypothesize that the high Cys concentrations observed in Lactobacillus-dominant states are likely to be primarily host-derived. Cys's positive correlation with lactobacilli and negative correlation with many BV-associated bacteria could thus be due to mechanisms including preferential Lactobacillus colonization of hosts with high mucosal Cys secretion, induction of host Cys secretion by lactobacilli, and/or degradation of Cys by BVassociated bacteria 59 as a means to outcompete lactobacilli. However other metabolites in addition to Cys also likely influence microbiota composition. Further investigation of FGT microbiome-metabolome relationships and in vitro exploration of bacterial metabolism and interactions with host epithelium will be required to fully assess these possibilities.
A striking aspect of our findings is the paucity of transport mechanisms for Cys and Cyscontaining molecules in L. iners compared to other FGT lactobacilli. Unlike other amino acids, transporters for Cys are not well characterized in bacteria due to technical challenges inherent to redox chemistry that complicate Cys quantification and efforts to distinguish between transport of reduced Cys and transport of its oxidized counterpart, cystine 47 . The relatively few functional studies of putative bacterial Cys transporters have focused predominantly on Gram-negative bacteria from the phylum Proteobacteria rather than lactobacilli or related Gram-positive species 47,49 . In contrast to Cys transporters, several bacterial cystine transporters are better characterized, including among Gram-positive species 48 . We found that genomes of non-iners FGT Lactobacillus species were each predicted to encode multiple, mechanistically distinct known or putative transporters for Cys, cystine, and/or Cys-containing molecules, but our L. iners genome analysis failed to identify predicted transporters for any of these molecules. However both L-Cys and L-cystine supported L. iners growth, and isotopic tracing experiments demonstrated uptake when L. iners was grown in presence of either labeled L-Cys or L-cystine, indicating existence of one or more currently unrecognized uptake mechanisms. Genetic tools to definitively identify the unknown transporter(s) do not currently exist for L. iners 34 , but our finding that L.. iners growth is selectively inhibited by known cystine uptake inhibitors provides further evidence that it lacks the multiple, mechanistically distinct Cys-related transport mechanisms present in other FGT Lactobacillus species.
In addition to elucidating L. iners biology and function within the microbiota, our results have important implications for BV treatment. Standard BV therapy with MTZ often provides initial relief, but disease recurs in up to 75% of cases [20][21][22] . High post-treatment recurrence rates may be partially explained by the fact that MTZ typically promotes L. iners-dominant FGT bacterial communities [23][24][25][26][27] , which are more prone than L. crispatus-dominant communities to transition back to BV-like states 3,13-15 . In a recent study of longitudinal FGT microbiota dynamics in a South African population, we demonstrated that L. iners-dominant communities constitute a "gateway" for transition to more diverse, BV-like communities. Modelling of microbiota transition probabilities predicted that interventions which specifically shift transition probabilities in favor of L. crispatus-dominance over L. iners-dominance would have the greatest effect in increasing the fraction of women with L. crispatus-dominant communities 15 . Therapies to promote L. crispatus dominance during BV treatment are under active investigation. In a randomized trial of a live L. crispatus biotherapeutic administered after MTZ treatment, BV recurred less frequently among biotherapeutic recipients than placebo recipients 21 . However recurrence rates in the treatment arm remained high (30% recurrence within 3 months) and post-treatment microbiota composition was not reported 21 . A small Israeli study without a control group reported achieving L. crispatus-dominance and long-term remission in four of five women treated for refractory BV with MTZ plus vaginal microbiome transplants from healthy donors 60 , but most patients required multiple rounds of therapy. Interestingly, that study did not report detecting L. iners in any donors or recipients -a notable difference from typical FGT microbiota composition in North and South America, Europe, Asia, Africa, and Australia 1,2,24-27, [53][54][55][56][57]59,3,61,4,5,11,12,16,17,23 . Thus live biotherapeutic approaches for BV show promise but appear unlikely to be a panacea. In this study, we show that combining MTZ with a novel growth inhibitor for L. iners was superior to MTZ alone in promoting L. crispatus community dominance of defined, BV-like bacterial communities.
The specific L. iners inhibitors identified here have potential limitations for direct therapeutic use 20 in humans including low potency (SMC) and potential host toxicity (SDLC), but show proof of principle for a novel therapeutic approach. Further development of this approach as an adjunct to antibiotics has potential to improve therapies for adverse microbiota-linked reproductive health conditions worldwide, including for women in low and middle income countries, where the burden of these adverse outcomes is highest.

Clinical cohorts and samples
This study includes novel specimens from participants in the Females Rising through Education, Support, and Health (FRESH) cohort, an ongoing prospective observational study based in Umlazi, HIV-uninfected FRESH participants (including the above 53 participants) who were sampled consecutively between May and October 2017. Nugent scoring for BV assessment among these study participants was performed only sporadically until a protocol change that instituted universal Nugent scoring in September 2017. We therefore restricted analysis of BV-metabolite associations to samples collected under the universal testing protocol to minimize risk of bias. One participant was excluded from analysis of vaginal metabolite concentrations based on the pre-specified criterion of low 16S sequence read count (<10,000 reads). No a priori sample size calculation was performed in relation to analysis of microbiota-metabolite correlations because this is (to our knowledge) the first analysis of FGT microbiota-metabolite relationships in a sub-Saharan African cohort and we therefore lacked sufficient information upon which to make informed a priori estimates about metabolite distributions for power calculations. Metagenomic whole genome shotgun sequencing (WGS) for metagenomic genome assemblies was performed on samples from the above 143 participants and on samples from additional FRESH study participants for whom 16S rRNA gene-based microbiome profiling has previously been reported 1,4 . Primary bacterial isolates from a subset of study participants were generated from cryopreserved swabs as described below.
Additional metagenomics data and bacterial isolates were generated by the Vaginal Microbiome Research Consortium (VMRC; https://vmrc4health.org/) as detailed more fully elsewhere 39,62 . These included samples originating from two longitudinal studies of US women of reproductive age. One cohort enrolled pregnant women and collected vaginal swab specimens during pregnancy and in the post-partum period 13 . The second cohort enrolled non-pregnant women and collected vaginal swabs longitudinally over 10 weeks 25 .

Bacterial culture conditions and additives for bacterial growth media
Bacterial cultivation was performed at 35-37°C under anaerobic conditions using an AS-580 anaerobic chamber (Anaerobe Systems) in an atmosphere of 5% carbon dioxide, 5% hydrogen, and 90% nitrogen (Airgas Ò ). All media and other culture reagents were pre-reduced (deoxygenated) by placement overnight in the anaerobic chamber prior to use. Nutritional additives or inhibitors (reagents listed in Supplementary Table 1; additives prepared as in Supplementary Table 2) were added to autoclave-sterilized broth media after cooling to room temperature (broth media), then broth was re-sterilized by passage through a 0.22 µm vacuum filter before being transferred to the anaerobic chamber to pre-reduce (deoxygenate) for use in experiments. For testing of nutrient pools (Extended Data Figure 1), a single concentrated stock comprising all pool constituents was added to media as per Supplementary Table 2 rather than stocks of each additive being added separately. In contrast to other additives, stocks of hydrogen peroxide (H2O2) and metronidazole (MTZ) were added to pre-reduced broth media as freshly prepared, filter-sterilized stock solutions immediately prior to inoculation of bacterial cultures due to potential volatility or chemical instability. For preparation of solid media, filter-sterilized additives were added to autoclave-sterilized agar media after cooling to 50°C, mixed using a magnetic stir/hot plate, then used to pour agar plates.

Growth characteristics and nutritional requirements distinguishing L. iners from other
Lactobacillus species were investigated using Lactobacillus MRS broth. We tested two different commercial MRS formulations (BD Difcoä and Hardy Criterion). The BD Difcoä-formulated media is referred to as "MRS" while the Hardy Criterion-formulated media is referred to as "HMRS" throughout this paper. The broth was prepared by autoclaving according to manufacturer instructions and allowed to cool to room temperature. We confirmed that L. crispatus, L. jensenii, and L. gasseri grew rapidly and robustly in both formulations whereas L. iners failed to grow even after >10 days in HMRS and grew only after a prolonged delay of ~48-72 hours in BD Difcoä- NYCIII broth (ATCC medium 1685) was prepared using a slightly modified version of the standard ATCC protocol. Pre-media consisted of 4 g/L HEPES, 15 g/L Proteose Peptone No.3 and 5 g/L Sodium Chloride in 875 mL distilled water, was pH-adjusted to 7.3 and autoclaved. Prior to use, complete NYCIII broth was prepared from autoclaved, cooled pre-media by adding dextrose (3g/45mL) at 7.5% v/v, yeast extract solution at 2.5% v/v, and heat inactivated horse serum at 10% v/v, then sterilized by passage through a 0.22 µm vacuum filter. Where indicated, NYCIII broth was supplemented with IsoVitaleX™ (2% v/v), Vitamin K1-Hemin Solution (5% v/v), and/or Tween-80 (1 g/L); volumes of distilled water in the pre-media were decreased accordingly to ensure all other components were present at standard concentrations in the final solution.
"S-broth" pre-media consisting of 37 g/L BHI Broth, yeast extract (powder) 10 g/L, and dextrose 1 g/L was brought to a boil in 880 mL distilled water, then autoclaved (121°C for 15 min) and cooled, followed by addition of fetal bovine serum 5% v/v, Vitamin K1-Hemin Solution 5% v/v, and IsoVitaleX™ Enrichment 2% v/v to a final volume of 1L. The complete broth was then  Table 3).
Genomes corresponding to these strains were obtained from RefSeq. Information about the BV status of the samples from which these isolates and other isolate genomes were derived was obtained (when available) from metadata accompanying the isolate genome entries in RefSeq or the Genomes OnLine Database (GOLD) or from associated information from the reference strain repositories from which they were obtained. Additional novel US-based strains were obtained from Jacques Ravel and the VMRC, isolated from the sources detailed above 39 , and non-iners Lactobacillus strains previously isolated from a clinical cohort in Seattle, Washington, (a kind gift from Dr. Jeanne Marrazzo 63 ) were genome-sequenced and analyzed for genomic content as described below, but not otherwise studied experimentally.

Nucleic acid extraction
Total nucleic acids from cervicovaginal swab samples were extracted with a phenol-chloroform method, which includes a bead beating process to disrupt bacteria as previously described 4 .
Genomic DNA (gDNA) from bacterial isolates or mock communities cultured in vitro was extracted using a plate-based protocol that included a bead beating process and combined phenolchloroform isolation with Qiagen QIAamp 96 DNA QIAcube HT kit (Qiagen) procedures. and incubating for 10 min at 55°C. Tagmented DNA fragments were PCR-amplified using KAPA high fidelity library amplification reagents and primers incorporating Illumina adaptor sequences and sample barcodes. Products were pooled, purified with magnetic beads, and paired-end sequenced on an Illumina NextSeq with a 300-cycle kit.

Genome and meta-genome sequence processing and assembly
We constructed Lactobacillus genome catalogs from reference isolate genomes as well as from genome sequences of novel bacterial isolates and metagenomically assembled genomes (MAGs) assembled from culture-independent shotgun metagenomic sequencing of genital tract samples, described in greater detail elsewhere 39 . Briefly, we retrieved all reference genomes annotated as
Genomes of novel FGT isolates from the South African FRESH cohort (isolated as described above) and of non-iners Lactobacillus strains isolated from a US-based cohort 63 were sequenced as described above at the Ragon Institute. Additional US FGT Lactobacillus isolate genomes were provided by the VMRC from the sources detailed above 39 . A total of >1000 FGT shotgun metagenomic samples from studies in South Africa, the United States of America, Italy, and China were used to generate MAGS 39 . Sequence reads from isolate genomes and shotgun metagenomic samples were trimmed and filtered to high quality reads. Human reads from shotgun metagenomic samples were removed by mapping to GRCh38 (Genbank accession GCA_000001405.15). We assembled genomes and MAGs from high quality reads, then binned contigs and removed contamination. Bin completeness, contamination, and strain heterogeneity metrics were determined, then genomes and MAGs were assigned quality scores based on criteria from the Genome Standards Consortium 65 . Species genome bins (SGBs) were determined based on 95% pairwise absolute nucleotide identity (ANI).

Lactobacillus genome taxonomy assignment
We assigned taxonomy to SGBs with FGT Lactobacillus genomes based on presence of reference genomes with defined NCBI taxonomy, confirming that each contained reference genomes only from a single species. Of note, for both L. jensenii and L. gasseri, publicly available reference genomes classified in RefSeq as belonging to these species actually segregate into two separate SGBs per species, indicating that L. jensenii and L. gasseri (as traditionally defined) each comprise two distinct genomic species 39,66 . However we analyzed L. jensenii and L. gasseri as single species units in this study to correspond with prevailing paradigms in literature on 16S gene-based FGT microbiota profiling 67 .

Cervicovaginal Lactobacillus pan-genome construction and analysis
We constructed a cervicovaginal Lactobacillus pan-genome using genes from catalogs of L. iners, L. crispatus, L. jensenii, L. gasseri, and L. vaginalis genomes and MAGs. To maximize comprehensiveness of the pan-genome, we included genomes and MAGs classified as high-quality (>90% completeness and <5% contamination) or medium-quality (³50% completeness and <10% contamination) assemblies based on Genome Standards Consortium criteria 65 . Genes were identified within individual genomes by Prokka v1.14.5 68 , which predicts genes using Prodigal v2.6.3 69 . The resulting genes were clustered into a comprehensive pan-genome at 95% nucleotide identity using Roary v3.13.0 70 to generate a multi-fasta file containing gene sequences numbers, and GO (Gene Ontology) terms, followed by manual curation of the initial search results and BLAST-based sequence review for genes with unclear annotations. We then determined the number of genomes from each species predicted to encode each gene function of interest. Since the Genome Standards Consortium criteria for high-and medium-quality genomes and MAGs allow for low-level sequence contamination within assemblies 65 , we restricted gene presenceabsence analysis to gene sequences of interest that were detected in at least two genomes or MAGs from a species in order to exclude singleton contaminant sequences from our analysis. Importantly, including genomes and MAGs with completeness as low as 50% maximizes genome catalog diversity, thus increasing pan-genome size and sensitivity for detecting genes of interest within each species, but also results in a fraction of genomes and MAGs appearing to lack universally present genes due to incompleteness of assemblies.

Maximum likelihood phylogenetic distances and phylogenetic reconstruction
Phylogenetic reconstruction of L. iners genomes and MAGs (Fig. 1b) was performed using assemblies with >60% completeness and <5% contamination to maximize robustness of the analysis. fetchMG v1.0 72 was used to extract DNA sequences for each of forty single-copy universal bacterial marker genes from each genome (called using Prodigal 69 ). Genome assemblies containing fewer than 10 of the 40 universal marker genes were omitted from the subsequent alignment. A phylogenetic reconstruction was then produced using ETE3 v3.1.1 (parameters: "ete3 build -w clustalo_default-trimal-gappyout-none-none -m cog_85-alg_concat_default-fasttree_default") 73 , and the tree was visualized using iTOL v4 74 . An additional phylogenetic reconstruction encompassing other major FGT Lactobacillus species (Extended Data Fig. 2a) was performed using L. iners genomes and MAGs as well as genomes from L. crispatus, L. jensenii, L. gasseri, and L. vaginalis, filtered based on the same quality and completeness criteria.

Preparation of bacterial inocula for growth experiments
For initial experiments identifying cysteine and cystine as key nutrient requirements for L. iners growth (Fig. 1a,c, Fig 4a, and Extended Data Fig. 1), bacteria from frozen stocks were plated on solid media, incubated for 3 days, then suspended in sterile, pre-reduced Dulbecco's phosphate-buffered saline (PBS), adjusted to an OD600 of 0.3+/-0.05, and then inoculated into the indicated broth media formulations for measurement of growth kinetics. In subsequent growth experiments for Lactobacillus species, experimental bacterial inocula were prepared from liquid starter cultures in MRSQ broth with L-Cys (4 mM) that were incubated for 18-20 hours. The starter cultures were then pelleted by centrifugation for 10 min at 3716 x g, spent media was decanted, and bacteria were washed 2 times by re-resuspending in sterile, pre-reduced PBS followed by centrifugation to avoid carryover of nutrients from the original starter culture media. Washed bacteria were resuspended in PBS, adjusted to OD600 0.3 (± 0.05), then inoculated into experimental media at 3.5% (v/v), equating to bacterial titers ranging from ~1x10 5 to 1x10 6 colony-forming units (C.F.U.) per mL, depending on the experimental species and strain.

Growth kinetics quantification
We found that Lactobacillus growth kinetics in broth media were adversely affected by either

Inhibitor experiments and analysis
To test mono-culture growth inhibition, bacteria were cultured in media containing inhibitors at varying concentrations as indicated, including a reference (no-inhibitor) control. Since growth kinetics differed between species and strains, inhibition was determined for each strain at the first timepoint fulfilling the European Commission on Antimicrobial Susceptibility Testing (EUCAST) criterion of "definite turbidity" in the reference control (which we defined experimentally as OD600 >0.2), unless otherwise specified. At the selected timepoint for each strain, the median OD600 value for the reference control was set to a reference value of 100%, which was used to for each input mono-bacterial suspension as described above and used to calculate starting ratios within the mixed cultures. At 28 hours, cultures were harvested by centrifuging at 4700xg for 25 minutes at 4C, spent media was decanted, and pellets were frozen for later DNA extraction and analysis. Relative growth within mixed cultures was assessed by bacterial 16S rRNA gene sequencing as described below. Aliquots of mono-bacterial suspensions in the corresponding media type were cultured separately to confirm expected growth patterns (e.g. Fig. 5b and Extended Data Fig. 8a).

Culture-independent bacterial 16S gene amplification and Illumina MiSeq sequencing
The V4 region of the bacterial 16S rRNA gene was PCR-amplified following standard

Processing of Illumina MiSeq 16S gene sequences
Initial sequence demultiplexing was performed as previously described 61 , then dada2 version 1.6.0 76 was used to filter and trim reads, infer sequences, and assign initial taxonomy employing the RDP training database rdp_train_set_16.fa.gz (obtained from https://www.mothur.org/wiki/RDP_reference_files). Taxonomic assignments were refined and extended via extensive manual review (see Supplementary Table 7 for amplicon sequence variant (ASV) taxonomy). The denoised dada2 results with final taxonomic assignment were analyzed in R using phyloseq version 1.30.0 77 and custom R scripts (available at gitlab***).

16S rRNA gene-based microbiome analysis
For 16S-based microbiome profiling of clinical specimens, microbial communities were classified into four cervicotypes (CTs) as previously defined in a non-overlapping subset of participants from the FRESH cohort: CT1 includes communities with >50% relative abundance of non-iners For further sequence processing and analysis, amplicon sequence variants (ASVs) that could not be defined at least to the level of taxonomic class were pruned from the dataset and one sample was excluded due to sequence read count <10,000. The remaining 142 samples were rarefied without replacement to a uniform depth of 16,603 reads (the minimum read count among remaining samples), yielding a relative abundance limit of detection (L.D.) of 6.02 x 10 -5 . ASVs were collapsed at the species or genus level as indicated for further visualization and statistical analysis. In plots displaying taxon relative abundance using a logarithmic scale, a pseudocount equal to 0.5 x L.D. was added to assist visualization of any taxon with read count 0. Correlation analysis between bacterial relative abundances and cervicovaginal metabolite concentrations was restricted to taxa detected (³1 sequence read after rarefaction) in ³50% of samples (Extended Data   Fig. 7).

Sequence analysis for in vitro competition experiments and mock communities
For pairwise competitions between L. crispatus and L. iners, 16S rRNA gene sequences were generated, processed, and annotated as described above. Ratios of L. iners reads to L. crispatus reads in each sample were calculated. Significance of between-group differences for each mixture was determined by 1-way ANOVA. Significance of pairwise differences between the positive control condition (MRSQ + L-Cys without inhibitor) and each experimental condition was calculated using Dunnett's test and all significant pairwise comparisons were plotted (see Supplementary Table 10 for numeric p-values). Results of representative competition experiments are shown in Fig. 5a.
For mock BV-like community experiments, 16S rRNA gene sequences were generated, processed, and annotated as described above. Relative abundances of each experimental strain were determined for individual replicates and displayed in Fig. 5d. To assess L. crispatus enrichment, read counts of all experimental strains except L. crispatus were summed for each replicate sample, then ratios of L. crispatus to the summed taxa were calculated. Significance of between-group differences for each mixture was determined by 1-way ANOVA, and significance of pairwise comparisons were calculated using Tukey's test, with significance of selected pairwise comparisons plotted in Fig. 5e and shown in detail in Supplementary Table 11.

Measurement and analysis of metabolites in cervicovaginal lavage fluid
Concentrations of Cys, reduced glutathione (GSH, g-L-glutamyl-L-cysteinyl-glycine), and cysteinylglycine (Cys-Gly) were measured in CVL supernatants from the 143 FRESH cohort participants whose microbiome composition was concurrently profiled by 16S gene sequencing in For analysis, concentrations were volume-normalized and median values were adjusted to 1, then percent missingness (reflecting samples with analyte concentrations below the limit of detection, or L.D.) was calculated for each analyte. For analytes with missingness >0%, analyte limit of detection was inferred as equaling the lowest measured relative concentration in the cohort, then missing values were imputed at half the limit of detection. The Shapiro-Wilk test of normality was then performed on relative concentration values after log-transformation (not shown). In subsequent analysis of concentration differences between cervicotypes, differences were analyzed by 1-way ANOVA with post-hoc Tukey test for Cys (which had 0% missingness and was normally distributed, Fig. 3a,c) and by Kruskal-Wallis test with post-hoc Dunn's test for GSH and Cys-Gly (which had 21.7% and 7.0% missingness respectively and were therefore not normally distributed after imputation, Extended Data Fig. 4a-d). We calculated Spearman rank-order correlations between individual metabolites and bacterial taxa at both the genus and species levels, adjusting p-values for multiple comparisons where indicated using the Bonferroni method via the R stats package function p.adjust(). The R package DescTools was used to calculate confidence intervals for the Spearman correlation coefficients (r), including Bonferroni-corrected confidence levels at https://doi.org/10.13018/BMSE000906). This analysis confirmed the authenticity of the desired product and complete conversion of cysteine, and also revealed contamination with ADP, which was not removed by preparative HPLC (Extended Data Fig. 7). This material was used without further purification. Reduced labeled glutathione ( 13 C-GSH, 4 mM) for use in experiment measuring GSH uptake was produced from the oxidized form ( 13 C2-GSSG) by treating a stock solution of 81.6 mM labeled GSSG with TCEP at a 0.9:1 molar ratio to reduce the disulfide bonds prior to addition to MRSQ broth.

Quantification and isotopic analysis of amino acids
For isotopic analysis of proteinogenic amino acids, L. crispatus (strain 233) and L. iners (strain    Table 11). In 5a,e significance is depicted as * p £ 0.05, ** p Relative concentrations of (a) GSH and (b) Cys-Gly by BV status in CVL fluid from the 53 women in (Fig. 3a).

(c & d)
Relative concentrations of (c) GSH and (d) Cys-Gly by cervicotype in CVL fluid from the 142 women depicted in (Fig. 3b,c). In (a-d) the red dotted line represents the