Host-independent synergism between Lactobacillus crispatus and other vaginal lactobacilli

Summary

The human vagina is a unique microbiome, typically predominated by Lactobacillus species in healthy women. However, we currently lack an understanding of why lactobacilli predominate in this environment and how these bacteria interact, aspects that are crucial for developing microbiome-based therapeutics. In this study, we used cost-efficient synthetic communities (SynComs) to investigate the stability and dynamics of Lactobacillus-predominated vaginal communities from healthy women independent of host influence. Reproducible communities of Lactobacillus crispatus co-occurring with Limosilactobacillus species and Lactobacillus jensenii were established in top-down experiments. Co-occurrence was verified with compositional correlation patterns in metagenome sequencing data and reproduced through a bottom-up approach. This co-occurrence pattern was independent of strain selection, host factors, and inoculation ratio. Genome-scale metabolic models predicted potential cross-feeding involving amino acids (e.g., L-arginine, L-lysine, and γ-aminobutyric acid [GABA]) and vitamins as mechanisms mediating their co-occurrence. This study provides a framework for developing reproducible synthetic vaginal Lactobacillus communities and informs future microbiome-based therapies.

Graphical abstract

Highlights

• •Lactobacilli dominance in the vagina is poorly understood
• •Synthetic communities reveal host-independent lactobacilli dynamics
• •Lactobacillus crispatus co-exist stably with L. jensenii and Limosilactobacillus
• •Cross-feeding of amino acids and vitamins can support stable microbial networks

Vander Donck and Victor et al. use synthetic communities and metabolic modeling to explore how Lactobacillus crispatus dominates the vaginal microbiome independently of host factors. They show that L. crispatus forms stable, host-independent communities with other lactobacilli through nutrient sharing, offering insights for microbiome-based therapies.

Introduction

The vaginal microbiome is frequently cited as a simple ecosystem in the human body, predominantly characterized by low-species diversity and predominated by Lactobacillus species.^1,2 These lactobacilli, especially L. crispatus, are typically associated with vaginal health and protection against pathogens such as Gardnerella vaginalis, which is associated with bacterial vaginosis^3,4; Prevotella species, which is associated with aerobic vaginitis⁵; and Streptococcus agalactiae (GBS), which is associated with preterm birth and neonatal infections.^6,7,8,9,10 The predominance of lactobacilli in the vaginal microbiome is a unique characteristic of humans and cannot be accurately studied in animal models.¹¹ This predominance has traditionally been attributed to specific human habits, such as diet and clothing, and other factors, such as distinctive hormone levels and fluctuations (particularly estrogen) and the presence of glycogen in vaginal secretions.¹² Lactobacilli metabolize glycogen to produce lactic acid, which acidifies the vaginal environment and is thought to protect this environment by inhibiting human viruses or the growth of potentially harmful microorganisms.^13,14,15 However, the factors driving the predominance of specific species such as L. crispatus remain unclear. Given that the metabolic ability to degrade glycogen and its breakdown products (maltose and maltotriose) and its pH tolerance are not unique to this species, other mechanisms likely contribute to its predominance.^16,17,18 Much of the research to date has been epidemiological and observational, focusing on cataloging which taxa are present rather than understanding the interactions and dynamics that occur within the microbial community. This limitation underscores the need for experimental approaches that investigate the ecological relationships among taxa and their functional roles, moving beyond simple taxa identification. This is now increasingly adopted for the gut^19,20,21 but is currently still largely overlooked in research on vaginal communities. We have recently uncovered modules of co-occurring taxa in the vagina, with L. crispatus showing positive co-occurrence patterns with L. jensenii and Limosilactobacillus taxa (referred to as the L. crispatus module) and negative associations with dysbiosis-related Gardnerella- and Prevotella-predominated modules.² However, it is not yet known whether these modules represent microbe-microbe interactions, such as metabolic cross-feeding or physical co-aggregation, or whether they arise from co-selection by shared host factors, such as hormones, low pH, and the immune system. Such knowledge is crucial in designing microbiome-based therapeutics for the vagina. Currently, the development of such therapeutics mainly focuses on the use of either a single strain, such as Lactin-V,²² a combination of L. crispatus strains,²³ or complex vaginal microbiota transplants, which require extensive donor screening.²⁴ Research on the gut microbiome has shown that synthetic communities (SynComs) or cooperating commensals can enhance the persistence and efficacy of live biotherapeutic products (LBPs).^25,26

SynComs are increasingly used in gut microbiome research for controlled studies of microbial interactions, stability, and community assembly, free from host influences.^27,28 They also offer a promising approach to study ecodynamic processes within the vaginal microbiome. Two main approaches are commonly used to construct SynComs: top-down and bottom-up.²⁹ In a top-down approach, microbial communities are cultured directly from biological samples, preserving natural species interactions while eliminating host factors. In contrast, the bottom-up approach involves assembling communities from isolated strains, offering precise control over community composition and allowing the detailed study of pairwise or multi-strain interactions. Both top-down and bottom-up approaches provide valuable and reproducible preclinical ecological models while also serving as practical tools for developing LBP strategies that are accessible to microbiome researchers globally. For example, the US Food and Drug Administration (FDA)-approved Seres³⁰ and Rebiotics³¹ donor-derived products can be considered top-down SynComs. Both approaches start from healthy donor samples to develop microbiome-based therapies to treat and prevent recurrent Clostridioides difficile infection in the gut. However, despite the increasing documentation of the potential of SynComs to better understand microbial ecosystems and develop therapies, neither top-down nor bottom-up approaches have been systematically applied to study the vaginal ecosystem. We have recently implemented a citizen-science approach to facilitate fundamental and applied research on the vaginal microbiome with donations of healthy volunteers.^2,32,33 This platform now provides access to an expanding collection of vaginal isolates and swabs, which can be utilized in top-down and bottom-up SynCom studies.

In this study, we investigated how L. crispatus can establish predominance in the vaginal microbiome independent of host influence and how it interacts with other vaginal taxa under such conditions. We first mined metagenomic data for co-occurrence patterns of L. crispatus with other vaginal taxa and then combined this with top-down and bottom-up SynCom approaches to evaluate the dynamics and stability of L. crispatus-predominated communities when maintained outside the host. To uncover mechanisms supporting community stability, we applied genome-scale metabolic modeling to predict potential cross-feeding interactions. We propose that the integrated approaches presented here can function as valuable preclinical models for the design and selection of LBPs, highlighting the potential benefits of multi-strain formulations for stable engraftment and enhanced antimicrobial properties compared to single-strain products.

Results

Vaginal metagenome data reveal co-occurrence of L. crispatus with L. jensenii and Limosilactobacillus species

To uncover co-occurrence patterns of L. crispatus with other vaginal microbiota, we leveraged data from the extensive biorepository obtained in the Isala program.² This continuously growing collection currently comprises over 4,000 vaginal DNA samples from female donors, complemented by a comprehensive phenotype dataset through survey data and almost 3,000 identified vaginal isolates matched to these donors (Figure 1A). From this repository, a subset of samples underwent deep metagenome sequencing (n = 79, min: 0.22 Gbp), and the resulting profiles were analyzed using the SPRING correlation metric³⁴ to assess co-occurrence patterns with L. crispatus. We focused on the 50 most relatively abundant taxa at the species level, revealing an interesting group of positively correlated lactobacilli, in particular L. crispatus with L. jensenii and Limosilactobacillus vaginalis, resembling the structure of the L. crispatus module (Figure 1B). Pearson correlation network analysis confirmed the robust correlation between L. crispatus and Lim. vaginalis (Figures S1 and S2). These two species also exhibited negative correlations with taxa linked to a more diverse vaginal microbiome, such as Prevotella, Dialister, and Anaerococcus (Figure S1). Moreover, while network analyses pointed at a consistent co-occurrence pattern of L. crispatus with Lim. vaginalis, this association was not exclusive. Other Limosilactobacillus species, such as Lim. coleohominis, Lim. fermentum, Lim. oris, Lim. timonensis, and Lim. reuteri, were also detected alongside L. crispatus in Isala samples (Figure 1C). Of note, in most cases where L. crispatus was not detected, L. jensenii and Limosilactobacillus species were also not present above detection limits. For validation, we also analyzed the metagenomic data from the VIRGO database³⁵ and observed similar patterns (Figure S3). Altogether, these findings suggest that L. crispatus does not exist in isolation within the vaginal microbiome but instead forms close interactions with other less abundant Lactobacillaceae, particularly L. jensenii and Lim. vaginalis. This supports the idea of a structured microbial network that plays a crucial role in maintaining vaginal health.

Figure 1.

Lactobacillus crispatus co-occurs with Lactobacillus jensenii and/or different Limosilactobacillus species in vaginal metagenome data

(A) Overview of the Isala citizen-science program providing different data and sample types for vaginal microbial community analysis and functional studies with isolates.

(B) Network of interacting (green) and counteracting (red) vaginal bacterial taxa in vaginal metagenomic data of the Isala cohort (n = 79) as defined by a compositional correlation analysis on the species level using SPRING associations.³⁴ Of note, species labeled here as Bifidobacterium (including B. vaginale, B. piotii, B. swidsinkii, and B. leopoldii) are considered synonymous with Gardnerella in NCBI taxonomy. Our nomenclature presented here follows the Genome Taxonomy Database (GTDB), release R220.³⁶

(C) Robust centered log ratio (CLR)-transformed counts of L. crispatus, L. jensenii, and various Limosilactobacillus species in metagenomic data of the Isala cohort (n = 79).

Figures were created using R (v.4.4.1) and BioRender.

Top-down community assembly shows reproducible and host-independent co-existence of L. crispatus with other lactobacilli

We then investigated whether the in-vivo-observed co-occurrence patterns of L. crispatus with Limosilactobacillus species and L. jensenii in the Isala and VIRGO data reflect functional relationships with biological relevance, are merely coincidental, or are mainly driven by host factors. Here, we explored the dynamics and stability of vaginal communities in the absence of host complexity by employing a top-down SynCom approach. Based on the data from the Isala biorepository, donors with an L. crispatus-predominated vaginal microbiome were selected (Figure 2A). These participants were invited to provide fresh vaginal swabs to initiate the top-down SynCom analysis in a nutrient-rich culture medium that, while different from the vaginal environment, also selectively promotes the growth of lactobacilli, in particular L. crispatus. This top-down approach was monitored over time using Oxford Nanopore Technologies’ full-length 16S rRNA sequencing of community RNA to reflect the metabolically active community and to enable higher-resolution taxonomic profiling than only selected variable regions of the 16S rRNA gene with amplicon sequencing (Figure 2B). Based on these full-length 16S rRNA sequences, closely related and genetically similar species within the Lactobacillus and Limosilactobacillus genera were classified into distinct subgenus groups to ensure consistent and reliable classification across samples (see STAR Methods). We distinguished two main groups: the L. jensenii group (including L. jensenii, L. mulieris, and L. psittaci, with 5 single-nucleotide polymorphisms [SNPs] between their respective 16S rRNA genes) and the Lim. vaginalis group (including Lim. vaginalis and Lim. urinaemulieris, with 4 SNPs between respective 16S rRNA genes). Additionally, L. paragasseri and L. gasseri only differ in 2 SNPs in their respective 16S rRNA genes. The other taxa could be identified up to the species level.

Figure 2.

Vaginal top-down selection, setup, and validation, showing robust and reproducible SynComs with sustained L. crispatus predominance

(A) Donor selection and experimental setup of top-down SynCom approach.

(B) Evolution of relative microbial abundances in top-down synthetic communities shown for 5 different donors with initial (day 0) L. crispatus group-predominated vaginal microbiome over 24 days with medium renewal every 3 days. One of three biological replicates is shown.

(C) Evolution of relative microbial abundances to validate the reproducibility of top-down synthetic communities originating from new fresh swabs of donors 1 and 2 over 10 days with medium renewal at time points indicated with an asterisk (^∗).

(D) A t-distributed stochastic neighbor embedding (tSNE) analysis of Aitchison distances between vaginal top-down synthetic communities established from 5 different donors for all time points and replicates (n = 3) shows that all donor samples cluster together.

Figures were created using R (v.4.4.1) and BioRender.

The relative abundance of L. crispatus in the original samples was confirmed to be between 35.3% and 90.4%, indicating a long-lasting L. crispatus predominance within these donor women. During the 24-day SynCom experiment, Lactobacillaceae taxa remained predominant in all the samples, with predominance defined as the most relatively abundant taxon that constituted at least 30% of the profile. At the final time point, all top-down SynComs exhibited predominance of L. crispatus supplemented with at least one other taxon from the L. crispatus module. In most SynComs tested, a co-occurrence with the two main members of the L. crispatus module, namely the L. jensenii group and the Lim. vaginalis group, became apparent. The intermediate time points of some donors were also predominated by other taxa of the module (L. jensenii and Limosilactobacillus), but they also shifted to L. crispatus predominance at the later time points.

Stable co-existence of L. crispatus with L. jensenii and Limosilactobacillus within phylogenetically distinct bottom-up communities

Building on the robust co-occurrence of L. crispatus, L. jensenii, and Limosilactobacillus groups observed in vivo and their co-existence in top-down SynComs independently of host factors, we next investigated this dynamic by constructing two bottom-up assembled SynComs in nutrient-rich conditions with single-strain representatives of the three members of the L. crispatus module (Figure 3A). Lim. reuteri was chosen as a representative species of the Limosilactobacillus genus. Both tricultures were manually assembled in two inoculation ratios; the first (1:1:1) contained equal proportions of all three species, while the second (100:10:1) mimicked the ratios of L. crispatus-predominant vaginal communities observed in the Isala participants (Figure 3B), with low Limosilactobacillus abundance, a 10-fold increase in L. jensenii, and L. crispatus as the predominant species.²

Figure 3.

Selection of vaginal isolates, experimental setup, and stability of bottom-up tri- and six-cultures over time

(A) Selection of phylogenetically distant vaginal isolates based on whole genomes.

(B) Experimental setup of bottom-up-designed tricultures. Tricultures were assembled in two inoculation ratios (1:1:1 and 100:10:1) in triplicate and cultivated over 12 days.

(C) Dotplot of the absolute abundances (colony-forming unit [CFU]/mL) of 2 triculture synthetic vaginal communities containing L. crispatus, L. jensenii, and Lim. reuteri grown over 12 days, with inoculation of the tricultures in 1:1:1 and 100:10:1 ratios. Bars represent the standard deviation of the average CFU/mL.

(D) Experimental setup of bottom-up six-cultures assembled by combining triculture 1 and 2 at day 12 and grown together until day 18.

(E) Dotplot of the ratio of the absolute abundances (CFU/mL) of 2 strains of the same species in six-culture synthetic vaginal communities containing 2 L. crispatus strains, 2 L. jensenii strains, and 2 Lim. reuteri strains grown over 6 days, a six-culture created by adding tricultures 1 and 2 originating from composition 1 at day 12, and a six-culture created by adding tricultures 1 and 2 originating from composition 2 at day 12. Each condition in both tricultures and six-cultures is grown in triplicate. Bars represent the standard deviation of the mean ratio.

Figures were created using R (v.4.4.1) and BioRender.

Within the four SynComs, all members of the L. crispatus module co-existed with each other after 12 days, irrespective of the phylogenetic differences between strains in the tricultures. No strain was outcompeted (Figure 3D). Moreover, the inoculation ratios (1:1:1 and 100:10:1) resulted in comparable community dynamics within each triculture. For instance, in triculture 1, all strains were present at representative absolute abundances, with a gradual decrease in L. jensenii abundance, while L. crispatus and Lim. reuteri levels remained relatively stable over 12 days. In triculture 2, all strains showed enhanced growth by day 3, followed by stabilization over the subsequent 9 days, with Lim. reuteri reaching higher levels than both L. crispatus and L. jensenii by day 12.

To assess whether this stable co-existence also persisted in a more complex bottom-up community comprising multiple strains of each species, where we expected more strain-strain competition, we assembled two six-cultures, incorporating two strains from each species, based on the members from both tricultures 1 and 2 (Figure 3C). To assess the ability of two strains of the same species to stably co-exist in a mixed culture, we mapped the ratio of their average absolute abundances. In both six-cultures (Figure 3E), initial imbalances at the time of merging (day 12) stabilized by days 15 and 18. Notably, both strains of L. crispatus reached nearly equal absolute abundances in both communities. As observed in the triculture experiments, this demonstrates that community assembly is not heavily impacted by starting inoculum. Moreover, in both six-cultures, the ratios rapidly adjusted to a stable structure where all six strains persisted at notable levels, achieving a balanced distribution between strains of the same species. Overall, these findings confirm that the members of the L. crispatus module can be stably combined in mixed cultures without competitive exclusion, encompassing multiple strains and species, regardless of initial inoculum composition or host influence. This is suggestive of a form of non-competitive synergism between L. crispatus, L. jensenii, and Limosilactobacillus strains. The stability of this module in such a nutrient-rich, non-host-associated context suggests that the vaginal ecosystem may function more as a permissive environment, allowing this community to emerge, rather than acting as the primary driver of its assembly.

Genome-scale metabolic modeling reveals putative cross-feeding within the L. crispatus module

To investigate whether the exchange of nutrients or metabolites among community members could play a crucial role in the establishment and stability of such non-competitive synergism between taxa in the L. crispatus module, we next used genome-scale metabolic models (GEMs) to predict potential cross-feeding interactions (Figure 4). These models were constructed using a top-down reconstruction approach with CarveMe³⁷ at the species level, utilizing publicly available genomes for each module member. We started with the extraction of core genes with the recently developed tool SCARAP,³⁸ which enabled us to construct high-quality GEMs based on robust core metabolic pathways present in the different strains of the species. Triculture community interaction was modeled with GEMs for each species of the L. crispatus module (i.e., L. crispatus, L. jensenii, and two Limosilactobacillus species: Lim. vaginalis [Figure 4A] and Lim. reuteri [Figure 4B]) in a minimal medium environment. Based on these models, Limosilactobacillus species and L. crispatus appeared to primarily act as metabolite donors, whereas L. jensenii mainly seemed to act as metabolite acceptors. Potential cross-feeding of amino acids was identified between Limosilactobacillus, L. crispatus, and L. jensenii, in particular for L-lysine, L-glutamic acid, L-arginine, and L-leucine. Specifically, we consistently observed predicted L-lysine production by the Limosilactobacillus species (Lim. vaginalis and Lim. reuteri) and L. crispatus and its uptake by L. jensenii, along with predicted L-arginine production by Limosilactobacillus and uptake by L. crispatus and L. jensenii, in models involving Lim. vaginalis and Lim. reuteri. Notably, Lim. fermentum did not exhibit similar predicted amino acid cross-feeding patterns, in line with its reduced prevalence in the vaginal ecosystem compared to Lim. vaginalis (Figure S8A). Besides amino acids, Lim. reuteri was predicted to function as a donor of 4-aminobenzoic acid (pABA), a precursor for folate (vitamin B9) production³⁹ to L. crispatus and L. jensenii. Additionally, cross-feeding of the non-proteinogenic amino acid and neurotransmitter γ-aminobutyric acid (GABA)⁴⁰ was predicted to occur between the members of our SynCom, with L. jensenii appearing to be the main donor of this product to Lim. vaginalis and L. crispatus.

Figure 4.

Exploration of predicted cross-feeding patterns within the L. crispatus module in a minimal medium environment using GEMs

Genome-scale metabolic model interactions were obtained by constructing GEMs from core genes of selected species, predicting cross-feeding patterns among L. crispatus, L. jensenii, and different Limosilactobacillus species.

(A) Genome-scale metabolic model interactions including Lim. vaginalis.

(B) Genome-scale metabolic model interactions including Lim. reuteri. The quality of the metabolic models was assessed using the MEMOTE server⁴¹ with a minimum score of 70. Metabolic interactions were predicted for the synthetic triculture using iterative SMETANA⁴² (n = 600 per community) in detailed mode. Metabolites that were cross-fed were filtered with an SMETANA score of >0.5.

Figures were created using R (v.4.4.1).

These interaction models were also run using strain-specific single GEMs from the members of the two tricultures that were also used in the bottom-up approach, again in a minimal medium environment (Figures S8B and S8C). While similar patterns emerged, we also observed interesting strain-specific patterns, with the predicted number of exchanged metabolites varying depending on the specific strains used in the models. Triculture 2 exhibited a higher number of predicted exchanged metabolites than triculture 1. Although Lim. reuteri was not a predicted acceptor in the model based on the core genome, Lim. reuteri 1 and 2 could accept multiple metabolites in the strain-specific models of the tricultures. These findings highlight the complexity of metabolic interactions within the L. crispatus module, implying a potential role for species- and strain-specific cross-feeding in community stability and functioning. Furthermore, as medium complexity and composition could influence cross-feeding patterns, we examined the metabolic interactions within (modified) MRS (De Man, Rogosa, Sharpe) medium (Figure S9). In this nutrient-rich medium standard for lactobacilli, only less pronounced metabolic interactions were observed, such as some predicted cross-feeding of D-alanine. We detected no potential L-arginine cross-feeding when simulated in rich MRS medium containing L-arginine. Yet, when L-arginine was removed from the simulation medium, the predicted cross-feeding patterns re-emerged, indicating the importance of the environmental conditions in shaping metabolic interactions.

Discussion

The human vagina is an often-oversimplified ecosystem, characterized by low species diversity and predominance of Lactobacillaceae.^43,44 Within this ecosystem, L. crispatus is considered to play a crucial role in supporting vaginal health.⁴⁵ Its high abundance is associated with reduced risk of bacterial vaginosis,^44,46 preterm birth,⁴⁷ and improved outcomes in assisted reproductive technologies.⁴⁸ The multifaceted protective nature of L. crispatus stems from its capacity to strongly acidify its environment via lactic acid production,⁴⁹ its immunomodulation capacity,^50,51 and its capacity to produce other antimicrobials, such as bacteriocins.^52,53 This plethora of functions, together with the recently uncovered strain-level variation of L. crispatus in the vagina, shows the underestimated complexity of the vaginal ecosystem.⁵⁴ However, understanding the importance of L. crispatus in the vagina requires a broader ecological perspective, considering its interactions with other microbial inhabitants. In this study, we integrated in vivo, in vitro, and in silico approaches to investigate its role beyond species predominance within the vaginal ecosystem. Network analysis using SPRING associations³⁴ on a subset of metagenome-sequenced vaginal samples from the Isala platform revealed robust species-level co-occurrence patterns between L. crispatus, L. jensenii, and members of the genus Limosilactobacillus. These patterns were further validated using the VIRGO metagenomic database.³⁵ Interestingly, these co-occurrence patterns indicate that taxa typically classified into distinct community state types (CSTs)^1,55 may frequently co-occur and share strong ecological associations. For instance, our data show that L. jensenii (associated with CST V) was rarely observed in the absence of L. crispatus (associated with CST I). This finding also highlights a limitation of the CST classification system, which, while valuable for organizing vaginal microbiota into discrete types, may not fully capture the complexity of microbial interactions within these communities.

To explore whether the observed in vivo co-occurrence patterns were mediated by host factors or microbe-microbe interactions, we established SynComs to eliminate the multitude of host-related factors that could influence the vaginal ecosystem and impact microbial interactions and community stability.⁵⁶ We further evaluated whether these co-occurrence patterns of L. crispatus with L. jensenii and Limosilactobacillus species reflected functional relationships with biological relevance or were merely coincidental. SynComs, which are comprehensive systems of reduced complexity,⁵⁷ have been increasingly adopted to study plant-soil^58,59,60 and gut ecosystems.^61,62 They support diverse research applications, from ecosystem functioning⁶³ to designing microbial consortia for health-related purposes. There exist different approaches in developing SynComs,⁶⁴ yet we pursued a cost-effective approach so that it can be easily adopted in vaginal microbiome research across the world, including in resource-limited laboratory settings. This is especially relevant given the increasing global interest in elucidating microbe-microbe interactions to inform the development of vaginal microbiome-based therapies, as demonstrated by initiatives such as the Vaginal Microbiome Research Consortium (VMRC)⁶⁵ and the Isala sisterhood.^32,33 Using a top-down, abundance-based assembly, we established L. crispatus-dominated SynComs, consistently detecting notable abundances of the Lim. vaginalis group and L. jensenii group across experiments and replicates. This suggests that vaginal SynComs dominated by L. crispatus tend to exhibit single-stable states, which aligns with findings from a recent bottom-up soil SynCom⁵⁸ but contrasts with findings in other human niches, such as the upper respiratory tract microbiome⁶⁶ and gut,⁶⁷ where identical community replicates often resulted in multiple stable states. Our bottom-up SynComs, composed of strains representing the L. crispatus module, further validated the robustness and stability of this microbial network outside the host. A key factor in this stability could be L. crispatus’ ability to cohabitate with other module members rather than promoting antagonism or competitive exclusion based on nutrient availability.⁶⁸ Unlike many other microbial communities where resource competition leads to dynamic shifts, the persistence of this module suggests a functionally interdependent or synergistic relationship.

One possible explanation for this stability has been shown to be cross-feeding, wherein metabolic byproducts from one species serve as nutrients for others.⁶⁹ Our SynCom design introduced periodic nutrient depletion due to medium renewal every 3 days, leading to the accumulation of bacterial metabolites, which may act as a selective pressure shaping synergistic community interactions.⁷⁰ Cross-feeding has been demonstrated in other microbial ecosystems, where, for example, formate, a short-chain fatty acid (SCFA), facilitates cooperative interactions within bacterial consortia derived from the gut.⁷¹ Similarly, our in silico GEM analysis predictions suggest that amino acid cross-feeding could be a key mechanism facilitating the co-occurrence and synergism between L. crispatus, L. jensenii, and various Limosilactobacillus species in the vaginal microbiome. Lactic acid bacteria from dairy products, an environment also dominated by lactobacilli, similar to the vagina, are found to be more auxotrophic for amino acids.⁷² This inability possibly stems from niche modification and aligns with the Black Queen hypothesis,⁷³ which posits that gene loss can confer a selective advantage by conserving an organism’s limited resources if the lost function is compensated by others in the community or, in this case, a network. Consequently, the energetic burden of amino acid biosynthesis, which is metabolically costly, is likely distributed among members of the L. crispatus module through cross-feeding.⁷⁴ In particular, Limosilactobacillus species such as Lim. vaginalis and Lim. reuteri, which show a more flexible habitat preference than L. crispatus, appeared to donate the most metabolites, whereas L. crispatus and L. jensenii appear to benefit from this and mainly act as metabolite receivers, which could be due to their host-adapted nature.^75,76 Limosilactobacillus species are often present in low abundance or may go undetected in vaginal samples due to technical limitations or even misclassification.⁷⁷ However, our data presented here highlight that we should not ignore their function. Their association with health-promoting effects in other body sites, especially the gut,⁷⁸ along with the potential cross-feeding interactions predicted in our data between Limosilactobacillus species (specifically Lim. vaginalis and L. reuteri) and L. crispatus and L. jensenii (acting as metabolite acceptors), suggests that they may play a more regulatory role in shaping the vaginal ecosystem than their abundance alone would indicate. Furthermore, the predicted cross-fed metabolites also include compounds with potential host-interactive functions, such as the neurotransmitter GABA and precursors of essential vitamins (e.g., pABA and niacinamide). This raises the possibility that cross-feeding not only stabilizes community structure but may also influence host-microbe interactions.^79,80 Future studies are needed to explore how these metabolic exchanges impact host physiology and contribute to the beneficial effects associated with L. crispatus-predominated vaginal microbiota.

Our discovery of an L. crispatus guild or module is consistent with the previously identified L. crispatus module based on 16S amplicon-based analysis of 3,195 vaginal microbiome data from the Isala platform.² Since our findings in the first Isala cohort, more recent studies on smaller cohorts have also explored co-occurrence networks among vaginal microbiota.^81,82,83 For example, Dong et al. found a similar co-occurrence of L. crispatus and L. jensenii within a module using SparCC based on publicly available vaginal microbiome data from 1,941 reproductive-aged women across different geographical regions (315 samples from Africa, 566 from Asia, 396 from Europe, 197 from North America, and 467 from Oceania), including healthy and women with vaginal infections.⁸² While these studies provide valuable insights, they lack experimental validation and operate at the (sub)genus level, limiting their ability to resolve species- or strain-level networks. In contrast, we present here a refined analysis of networks underlying the health-associated abundance of L. crispatus.

Taking these results together, in this study, we mechanistically demonstrated that L. crispatus may not function in isolation within the vagina using metagenomic data, SynComs, and genome-scale metabolic modeling. We present evidence that cross-feeding likely plays a key role in driving non-competitive synergism between L. crispatus, species of the Limosilactobacillus genus, and L. jensenii, where vaginal host factors act more as a permissive environment than as a primary driver of community assembly.

Limitations of the study

Understanding the ecological dynamics that support L. crispatus predominance in the vaginal microbiome is crucial for developing microbiome-based strategies that promote women’s health. While our study provides key insights, several limitations remain. Our use of a nutrient-rich medium and microaerophilic atmospheric conditions (5% CO₂), ideal for L. crispatus and other lactobacilli growth but not growth of L. iners, has introduced inevitable culture bias and is not highly representative of the native environment but was considered necessary here to focus on L. crispatus as the predominant member. Future studies should include more defined media that better reflect the nutritional composition of the vaginal environment, including key host-derived compounds such as mucins, antimicrobial peptides, and hormonal factors, and are suitable for a larger variety of vaginal taxa. Additionally, we focused here on the health-associated L. crispatus module but highlight that the exploration of microbial modules linked to suboptimal health, such as Prevotella- and Gardnerella-centered modules, is equally interesting to better understand vaginal microbiome interactions and design new therapeutics and diagnostics. Their study would require further optimization of culture conditions. Expanding SynCom replication in Isala sisterhood and related vaginal microbiome projects across the world in diverse cohorts will be crucial for assessing their relevance across diverse populations. Although our top-down and bottom-up approaches are scalable, reproducible, and feasible with minimal equipment and resources, they remain labor intensive and currently limit our ability to process large numbers of vaginal swabs or strains in parallel. More high-throughput systems like mini-bioreactor arrays could improve scalability and reduce contamination risks.⁸⁴ An RNA-based sequencing strategy was used to focus on metabolically active bacterial cells of both bottom-up and top-down SynComs while minimizing host read contamination. For the top-down approach in particular, this method has proven effective in capturing the active microbial dynamics in other Lactobacillaceae-enriched systems⁸⁵ and simple synthetic cultures.⁸⁶ However, the use of specific primers targeting unique genes relies on the assumption that gene expression levels remain relatively stable across different conditions, which may not always hold true. As a complementary strategy, a DNA-based approach that incorporates propidium monoazide (PMA) staining to distinguish live from dead cells could offer a more direct assessment of viable community members in SynCom cultures.⁸⁷ Another important limitation we acknowledge is related to the genome-scale metabolic modeling we used. Interactions were simulated in a minimal medium environment to generate hypotheses about potential cross-feeding patterns. To assess the effect of medium complexity and composition, simulations were also run in nutrient-rich conditions that were used for the in vitro SynCom experiments. Predicted cross-feeding interactions were shown to be influenced by the simulation environment, such as weaker predicted cross-feeding patterns in nutrient-rich environments before nutrient depletion. Importantly, both minimal medium and nutrient-rich MRS are not fully reflective of vaginal fluid; thus, future dedicated experimental metabolomics analyses, including of vaginal samples, will be essential to validate the GEM-predicted cross-feeding mechanisms for specific strains and uncover additional (more strain-specific) ecological drivers, such as the production of bacteriocins that target closely related competing strains. We expect that, besides the general mechanism of synergy uncovered here, more strain-specific mechanisms of inhibition will also become relevant. Overcoming these challenges will be key to refining microbiome-based interventions that promote L. crispatus predominance and long-term vaginal health.

Resource availability

Lead contact

Requests for additional information on resources used in the study can be directed to the lead contact, Sarah Lebeer (sarah.lebeer@uantwerpen.be).

Materials availability

Commercially unavailable bacterial strains are available from the lead contact upon reasonable request through a material transfer agreement with the University of Antwerp.

Data and code availability

• •This paper does not report original code.
• •All 16S amplicon (V4) sequencing data from the Isala cohort (n = 3,345) and deep shotgun metagenome sequencing data of a subset of samples (n = 79) are available at the European Nucleotide Archive (ENA) under BioProject: PRJEB50407. Sequencing data from the top-down SynCom experiments (16S amplicon V4 and full-length 16S V1–V9) are available at the ENA under BioProject: PRJEB87999.
• •GEMs are available at the EBI Biomodels repository under accession numbers MODEL2507070001–2507070005
• •Any additional information required to reanalyze the data reported in this study is available upon request from the lead contact.

Acknowledgments

We first want to thank all Isala donors for their samples, data, and project idea donations. We also thank the entire Isala team (now across the world in a sisterhood) for all their previous, current, and future work on the vaginal microbiome. We want to sincerely thank Prof. J. Ravel and Dr. M. France for providing us with the metagenomes used for the creation of the VIRGO database. We thank Dr. T. Eilers for his help with the design and development of the strain-specific primers needed for bottom-up SynCom analyses. We also thank D. Bandalac for her help as a bachelor’s student with SynCom analyses. The authors acknowledge the following funding agencies and institutions: the Research Foundation—Flanders (FWO), which currently supports M.V. and L.V.D. as PhD students with an aspirant strategic basic research grant (1SC2725N and 1SD0622N, respectively) and S.W. and W.V.B. as postdoctoral fellows with PhD Fellowship Fundamental Research (12AZ624N and 1224923N, respectively). The lab of Prof. Lebeer acknowledges the following research grants for research on the vaginal microbiome: FWO SBO project DeVEnIR (S006424N), FWO research projects G049022N and G031222N; the Inter-University Special Research Fund of Flanders (iBOF) for the POSSIBL project; the industrial research fund UAntwerpen for IOF POC project CRUCIAL; BOF funding for Isala; and the European Research Council (ERC) for starting grant Lacto-Be (H2020) (grant ID 852600) and proof-of-concept VALERIE (Horizon) (grant ID 101213306).

Author contributions

L.V.D., M.V., and S.L. designed the study and worked on the conceptualization of the research project. L.V.D. and M.V. carried out the experimental and logistical work. M.V., L.V.D., and W.V.B. processed the sequencing data and performed the analyses. M.V., L.V.D., T.V.R., W.V.B., C.N.A., and S.L. worked on the visualizations. L.V.D., M.V., J.D., S.W., C.N.A., W.V.B., T.V.R., S.A., and S.L. contributed to the interpretation of the results. M.V., L.V.D., C.N.A., and S.L. wrote the original manuscript. All authors contributed to the review and editing of the final manuscript.

Declaration of interests

S.L. declares to be a voluntary academic board member of the International Scientific Association on Probiotics and Prebiotics (ISAPP, www.isappscience.org), cofounder of YUN, and scientific advisor for Freya Biosciences. The team of S.L. declares research funding or consumables for research from YUN, BioOrg, Puratos, DSM-Firmenich, Drylocks, and Lesaffre/Gnosis. None of these organizations or companies were involved in the design or data analysis of this study, which was fully funded by university, governmental, and European funding. S.A. declares to be a voluntary member of the student and fellows’ association of ISAPP. S.A., S.W., J.D., C.N.A., and S.L. are inventors on patent applications of specific strains, including patent application EP20210606.8 (owned by the University of Antwerp with S.A., S.W., and S.L. as inventors) for a strain used in this work.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used ChatGPT by OpenAI exclusively to assist with paraphrasing and to gather inspiration for scientific rephrasing. After using this tool/service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
Lactobacillus crispatus 1	Isolated from Isala participant (Lebeer et al.2)	AMBV-815
Lactobacillus crispatus 2	Isolated from Isala participant (Lebeer et al.2)	AMBV-961
Lactobacillus jensenii 1	Isolated from Isala participant (Lebeer et al.2)	AMBV-494
Lactobacillus jensenii 2	Isolated from Isala participant (Lebeer et al.2)	AMBV-805
Limosilactobacillus reuteri 1	Isolated from Isala participant (Lebeer et al.2)	AMBV-339
Limosilactobacillus reuteri 2	Isolated from Isala participant (Lebeer et al.2)	AMBV-532
Biological samples
Vaginal samples	Isala project, University of Antwerp	ClinicalTrials.gov: NCT04319536
Chemicals, peptides, and recombinant proteins
ReadyScript cDNA Synthesis Mix	Sigma Aldrich	UNSPSC: 12352200
Power Up SYBR Green	Thermo Fisher Scientific	Cat no. A25743
Critical commercial assays
Illumina DNA Prep (M) Tagmentation kit	Illumina	https://www.illumina.com/products/by-type/sequencing-kits/library-prep-kits/illumina-dna-prep.html; "RRID:SCR_010233"
NovaSeq 6000 S4 Reagent Kit v1.5	Illumina	https://www.illumina.com/products/by-type/sequencing-kits/cluster-gen-sequencing-reagents/novaseq-reagent-kits.html; "RRID:SCR_010233"
NexteraXT DNA Sample Preparation kit	Illumina	https://www.illumina.com/products/by-type/sequencing-kits/library-prep-kits/nextera-xt-dna.html; "RRID:SCR_010233"
RNeasy Mini Kit	Qiagen	Cat no. 74106
16S Barcoding Kit 24 V14	Oxford Nanopore Technologies	SQK-16S114.24; "RRID:SCR_003756"
Flongle flow cell (R10.4.1)	Oxford Nanopore Technologies	FLO-MIN114; "RRID:SCR_003756"
Deposited data
16S rRNA amplicon sequencing data, Isala project	Lebeer et al.,2 European Nucleotide Archive	PRJEB50407
Metagenomic sequencing data, VIRGO	Ma et al.88	http://virgo.igs.umaryland.edu.
Metagenomic sequencing data, Isala project	This paper, European Nucleotide Archive	PRJEB87999
Top-down SynCom sequencing data	This paper, European Nucleotide Archive	PRJEB87999
Genome-scale metabolic models	This paper, European Bioinformatics Institute BioModels repository41	MODEL250707000, https://github.com/LebeerLab/2025_GEMS_syncom_vagina
Oligonucleotides
27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′- CTACGGCTACCTTGTTACGA-3′) universal 16S primers	This paper	N/A
Strain-specific primers for bottom-up SynComs, see Table S2	This paper	N/A
Software and algorithms
Fastp (version 0.23.2)	Chen89	https://github.com/OpenGene/fastp
Hostile (version 0.2.0)	Constantinides et al.90	https://github.com/bede/hostile
MultiQC	Ewels et al.91	https://github.com/MultiQC/MultiQC; "RRID:SCR_014982"
Sylph (version 0.8.02)	Shaw and Yu92	https://github.com/bluenote-1577/sylph
R (version 4.4.1)	Online	https://cran.r-project.org/bin/windows/base/old/
Tidytacos (version 1.0.6)	Wittouck et al.93	https://github.com/LebeerLab/tidytacos
Genome Taxonomy Database R220	Parks et al.36	https://gtdb.ecogenomic.org/
Shovill	Seemann94	https://github.com/tseemann/shovill
CheckM	Parks et al.95	https://github.com/Ecogenomics/CheckM
GUNC (version 1.0.13)	Orakov and Fullam et al.96	https://github.com/grp-bork/gunc
NetCoMi package (version 1.1.0)	Peschel et al.97	https://github.com/stefpeschel/NetCoMi
Emu	Curry et al.98	https://github.com/treangenlab/emu
Uniortho	Eilers and Delanghe et al.88	https://github.com/TomEile/uniortho
SCARAP	Wittouck et al.38	https://github.com/SWittouck/SCARAP/blob/master/pyproject.toml
CarveMe	Machado et al.37	https://github.com/cdanielmachado/carveme
MEMOTE	Lieven et al.41	https://github.com/opencobra/memote
SMETANA	Zelezniak et al.42	GitHub - cdanielmachado/smetana
Other
PrimerQuest™ Tool	Online, Integrated DNA Technologies	https://www.idtdna.com/pages/tools/primerquest; "RRID:SCR_001363"
Isala clinical trial identifyer	This paper, Lebeer et al.2	ClinicalTrials.gov: NCT04319536

Experimental model and study participant details

Vaginal samples used in this study were obtained from the Isala project, which received approval from the Ethical Committee of the Antwerp University Hospital/University of Antwerp (B300201942076) and is registered on ClinicalTrials.gov (NCT04319536). A total of 3,345 healthy, non-pregnant women participated by donating vaginal samples for microbiome profiling, culturomics, and metabolomics analyses, and by completing an extensive questionnaire covering lifestyle-related factors. A subset of these participants was selected and asked to provide additional vaginal samples based on their predetermined microbiome composition, for use in in vitro top-down experiments. Further methodological details are available in our previous publication.²

Method details

Generation of a large biorepository of vaginal microbiomes and isolates

Sampling, transport, and processing

In the Isala project, approved by the Ethical Committee of the Antwerp University Hospital/University of Antwerp (B300201942076) and registered on ClinicalTrials.gov (NCT04319536), 3,345 healthy, non-pregnant women donated two vaginal swabs: an eNAT (Copan, Brescia, Italy), intended for microbiome profiling, and an ESwab (Copan, Brescia, Italy), intended for culturomics. Briefly, all swabs were stored at home in the fridge until transported at room temperature with prepaid services by the national parcel service (Bpost). The potential effects of transport duration and sampling season on sequencing outcomes were assessed, but no significant impact was observed.² Upon arrival, the eNAT swabs were immediately stored at −20°C until further processing in the lab, while the ESwabs were stored at 4°C. After a maximal storage time of 6 h, the ESwabs were vortexed for approximately 15 s, after which 500 μL was combined with 500 μL of 50% glycerol in a 96-tube Micronic plate and stored at −80°C for culturomics. The remaining fluid was stored for metabolomic analysis in a 96-tube Micronic plate at −80°C until processing.

16S rRNA amplicon sequencing of vaginal samples

The vaginal microbial community composition of the 3,345 Isala samples was previously determined from the eNAT swabs using 16S rRNA amplicon sequencing, as previously described.² Sequencing data is available at the European Nucleotide Archive (ENA) under bioproject PRJEB50407.

Metagenomic sequencing of a subset of vaginal samples

A subset of 79 vaginal samples from the Isala biorepository were used for metagenomic sequencing. DNA was extracted from the eNAT swabs, as previously described,² following the previously optimized in-house protocol.⁹⁹ DNA extracts were stored at −20°C until further processing. Library preparation and metagenomic sequencing were outsourced to a subcontractor and performed using the Illumina DNA Prep (M) Tagmentation kit (Illumina) following the manufacturer’s protocol. The dual-index-paired-end sequencing was performed on the NovaSeq 6000 DX platform using the NovaSeq 6000 S4 Reagent Kit v1.5 at a target sequencing depth of ∼45 Gbp per sample following the manufacturer’s guidelines. Adapter and quality trimming of raw metagenomic reads were performed with fastp⁸⁹ version 0.23.2. Subsequently, human reads were removed using Hostile⁹⁰ version 0.2.0, and quality was assessed with MultiQC⁹¹ version 1.13. Samples were taxonomically profiled using Sylph⁹² version 0.8.02, using default GTDB R220 species representative genomes. The taxonomic composition of each metagenome at species-level was visualized using R (version 4.4.1) and the tidytacos package⁹³ version 1.0.6. Sequencing data is available at ENA under bioproject PRJEB87999.

Isolation and identification of vaginal bacteria

From each participant of the large Isala cohort, 10 μL of the ESwab glycerol stock was inoculated on a small Petri dish (10 mL) with growth media and grown for 24-48h at 37°C and 5% CO₂. In a small pilot study, several growth media were tested (MRS, MRS at pH 4.6, MRS + cysteine, MRS + vancomycin, MRS + vit K + hem, YPD, BHI, TH, TSB/TSA, or CB), as at the start of the isolation campaign the main goal was to recover members of the Lactobacillaceae family, MRS, MRS + vancomycin and CB were selected as three main growth media. In later smaller isolation campaigns, which added to the total amount of currently more than 3,000 isolates, other growth media listed above were used. The isolates used in this paper are part of the initial isolation campaign. After 24h, the plates were checked for colonies, and if present, one colony per plate was selected at random (max three isolates/woman). Part of this colony was inoculated in 10 mL MRS and grown overnight (ON) at 37°C and 5% CO2. If no growth was observed on the initial media, CB broth was used as an alternative. However, due to the scale and labor-intensive nature of the isolation campaign, no further attempts were made for samples that consistently failed to grow or could not be reliably identified. Of ON growth, 800 μL was mixed with 800 μL 50% glycerol in labeled cryovials (Greiner Bio-one Cryo) and stored at −80°C for future purposes. The rest of the colony was used for colony PCR for taxonomic identification with 16S rRNA gene Sanger sequencing, using universal primers 27F (5′-AGAGTTTGATCCTGGCTCAG-3′) and 1492R (5′- CTACGGCTACCTTGTTACGA-3′). The success of the colony PCR was checked with gel electrophoresis. Low-quality bases (with a mean Q < 20) of forward and reverse reads were trimmed using fastp¹⁰⁰ with a sliding window of 4 bp. Both reads were merged if enough overlap was possible, and otherwise the longest read was used. The resulting sequences were then blasted against the full-release 220 GTDB 16S rRNA database.³⁶ The hit with the highest bit-score was used as the result of classification.

In-house pipeline DNA extraction and quantification of vaginal bacteria

Promising vaginal isolates were selected for further characterization (out of the scope of this paper) and DNA was extracted in preparation for whole genome sequencing. The selection of promising isolates encompasses those utilized in the bottom-up SynCom approach. These isolates were grown overnight with 10 mL MRS at 37°C and 5% CO₂. The overnight culture was inactivated with ampicillin (100 mg/mL). This step was performed twice. The resulting solution was incubated for one hour at 37°C, afterward centrifuged, the supernatant was removed, and the pellet was washed three times with NaCl-EDTA. Next, the two pellets were combined and resuspended in NaCl-EDTA to ensure enough product. The pH was adjusted to 8. Then, lysozyme (10 mg/mL) and RNase (20 mg/mL) were added, and the mixture was incubated for one hour at 37°C while shaking periodically. Next, NaCl-EDTA, 10% SDS, and Proteinase K (20 mg/mL) were added to the solution, which was vortexed and incubated for one hour at 55°C. After incubation, a cold protein precipitation solution was added, vortexed, and the mixture was shortly put on ice. Then, the solution was centrifuged, and the supernatant was transferred to a clean tube and centrifuged again. This supernatant, which contained the DNA, was transferred again to a clean tube and the DNA was precipitated with ice-cold isopropanol. This tube was centrifuged, the supernatant was discarded, and the pellet was resuspended with fresh 70% EtOH. This solution was centrifuged again and while the supernatant was discarded, the pellet was left to air-dry until all the ethanol was evaporated. Finally, the pellet was dissolved in water (molecular grade). DNA of the samples was quantified using the Qubit 3.0 Fluorometer (Life Technologies) using the manufacturer’s instructions and standard kit. The DNA concentration should be between 25 and 50 ng/μL in a minimum 20 μL volume.

Whole genome sequencing of vaginal bacteria

Vaginal isolates were grown overnight in MRS at 37°C, 5% CO₂, and their DNA was extracted using methods described previously.¹⁰¹ Illumina-based genomes were sequenced using the NexteraXT DNA Sample Preparation kit (Illumina, United States of America) and the MiSeq platform (Illumina, United States of America) using 2 × 250 cycles at the Laboratory of Medical Microbiology (University of Antwerp, Belgium) as previously described. The genomes sequenced with DNBseq were sent for sequencing to BGI (Hong Kong). Genome sequencing data were assembled using Shovill.⁹⁴ Genome completeness and contamination were assessed using checkM,⁹⁵ and GUNC⁹⁶ (version 1.0.13) using GTDBTk version 2.4.0.¹⁰²

Network analyses of vaginal metagenomes

Co-occurrence networks

To uncover co-occurrence patterns of vaginal micro-organisms within 79 vaginal metagenomes obtained from the Isala biorepository, SPRING and Pearson correlation coefficients between the pairwise taxa were calculated using the NetCoMi package⁹⁷ version 1.1.0, which is a pipeline for running the different correlation methods. The following settings were used for determining any correlations using the different methods: abundances were transformed using a modified centrum log ratio (mCLR) transformation,¹⁰³ to further sparsify the network, the 50 taxa with the highest frequencies were included in the networks.

Correlation analyses

To assess the robustness of co-occurrence patterns involving Lactobacillus crispatus and other vaginal lactobacilli, correlation matrices between pairs of three vaginal bacterial taxa associated with the L. crispatus module (L. crispatus, L. jensenii, and Lim. vaginalis) in the Isala metagenome dataset (n = 79) were calculated, using all 10 different correlation metrics, including gconda, ccrepe, cclasso, spring, sparcc, spieceasi, pearson, propr, spearman, and bicor.⁹⁷ Furthermore, the presence of L. crispatus, L. jensenii, and various Limosilatobacillus taxa in metagenomic datasets from both the Isala cohort (n = 79) and the VIRGO database³⁵ (n = 264) was visualized in Figures 1C and S2, respectively. For this, centered log-ratio (CLR) transformed abundance counts were used to account for compositionality of the vaginal samples.

Establishment of vaginal synthetic communities

Top-down approach

Five participants with L. crispatus-predominated vaginal microbiomes were selected from the large Isala biorepository. On site, each provided one fresh vaginal eNAT for microbiome profiling and one fresh vaginal ESwab to conduct a top-down synthetic community experiment. The eNAT was immediately stored at −20°C until further processing in the laboratory, while each ESwab was vortexed and divided into three 300 μL aliquots, creating replicates. This rapid processing time minimizes potential microbial loss caused by temperature fluctuations or repeated freeze-thaw cycles. These were inoculated in 10 mL MRS and incubated at 37°C with 5% CO₂ for 24 h. Then, 500 μL from each culture was transferred to 30 mL fresh MRS for 24 days of incubation under the same conditions. Every 3 days, all cultures were vortexed for 60 s, and 600 μL of the community was sampled. The samples were centrifuged for 10 min at 6,000g, the supernatant was discarded, and 600 μL of RLT lysis buffer from the RNeasy Mini Kit (Qiagen) was added to the cell pellet. The samples were snap-frozen in liquid nitrogen and stored at −80°C until RNA extraction. Following sampling, 15 mL of medium was replaced with fresh MRS for continued incubation.

Validation of top-down approach

Two participants (donors 1 and 2) were selected to validate the top-down approach. At a different timepoint, they provided a fresh eNAT swab, processed as before, and a fresh ESwab, which was immediately vortexed for 15 s, and 100 μL inoculated into 50 mL MRS. Cultures were incubated at 37°C with 5% CO₂ for 10 days, with samples collected on days 1, 2, 3, 4, 7, and 10, following the procedure described earlier. Optical density at 600 nm using a spectrophotometer (Genesys 20, Thermo Scientific) was measured at each timepoint to monitor biomass (Figure S6). After sampling on days 3 and 7, 25 mL of medium was replaced with fresh MRS.

Bottom-up approach

From the culturable Isala biorepository of almost 3,000 identified vaginal isolates, two strains from each species of L. crispatus (AMBV-815 and AMBV-961), L. jensenii (AMBV-494 and AMBV-805), and Lim. reuteri (AMBV-339 and AMBV-532) were selected to create two genetically distinct combinations (ANI <99.5%, Table S1) of bottom-up SynComs representing the L. crispatus module. First, all six isolates were inoculated overnight in 10 mL MRS at 37°C with 5% CO₂. Subsequently, two combinations (triculture 1 and triculture 2) were established by assembling the strains for overnight growth in 10 mL of MRS using two different starting compositions, composition 1 with equal ratios (1:1:1) and composition 2 in a ratio of 100:10:1 for L. crispatus, L. jensenii, and Lim. reuteri, respectively. The inoculation volumes were determined using the optical density at 600nm to achieve the desired ratio between the species. 100μL of the overnight culture was then inoculated in triplicate in 30mL MRS. The bottom-up SynComs were incubated for 12 days, during which sampling and medium renewal were carried out as described above in the top-down approach. On day 12, tricultures were combined to establish sixcultures, which were then incubated for an additional 6 days. Sixculture 1 was assembled by combining both tricultures inoculated at a 1:1:1 ratio, while sixculture 2 was formed using tricultures established with a 100:10:1 inoculation ratio (Figure 3C). Additionally, a control community containing all six isolates in equal abundances was constructed (Figure S7).

RNA-based 16S rRNA sequencing of top-down SynComs

An RNA-based approach was employed to process all SynCom samples, as it has been demonstrated to reliably reflect the dynamics and metabolically active fraction of closed microbial systems, predominated by lactobacilli, such as those observed in fermentation processes.⁸⁵ RNA was extracted from all top-down samples using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s protocol. The quantity of isolated RNA was measured using the NanoDrop One spectrophotometer (Thermo Fisher Scientific), and 500 ng of RNA was used for cDNA synthesis with the ReadyScript cDNA Synthesis Mix (Sigma Aldrich). A microbial mock community (ZymoBIOMICS Microbial Community Standard, Zymo Research) was included as a control to evaluate potential bias during RNA extraction, and microbial mock community DNA (ZymoBIOMICS Microbial Community DNA Standard) was used to assess bias introduced during sequencing library preparation and subsequent processes (Figure S5). Full-length 16S rRNA gene (V1-V9 region) sequencing of the cDNA from the top-down samples was performed using the MinION Mk1C device (Oxford Nanopore). The DNA library was prepared using the 16S Barcoding Kit 24 V14 (SQK-16S114.24, Oxford Nanopore), according to the manufacturer’s instructions, but 75 fmol of pooled cDNA instead of 50 fmol was loaded onto the Flongle flow cell with R10.4.1 chemistry (FLO-MIN114, Oxford Nanopore). DNA reads were classified on species level using emu¹⁰² with GTDB R220 with representative 16S rRNA sequences. Species were grouped together into subgenus groups when emu alignment probability scores were <99%, due to high sequence similarity of their respective 16S rRNA genes. Visualizations were done in R version 4.4.1 using tidytacos⁹³ version 1.0.6. The samples for validating the top-down approach were sequenced using Illumina MiSeq 16S rRNA gene amplicon sequencing, following the method outlined by Lebeer et al. (2023).² The sequencing data is available at ENA under bioproject PRJEB87999.

Determining the composition of bottom-up SynComs using qPCR

Strain-specific primers for all bacterial strains (Table S2) were designed based on the unique orthogroups using the Uniortho Tool (github.com/TomEile/uniortho)⁸⁸ and using the Integrated DNA Technologies (IDT) PrimerQuest Tool and OligoAnalyzer Tool on a selection of the unique genes. Primer efficiency and strain specificity were tested, and the best primers were selected. Strain specificity was tested in silico by mapping the developed primers to publicly available genomes and in vitro by testing the primer efficiency against the other isolate we used of the same species. However, due to the genomic nature of L. crispatus and L. jensenii, the primers developed for L. crispatus 2 (AMBV-961) and L. jensenii 1 (AMBV-494) mismatched to L. crispatus 1 (AMBV-815) and L. jensenii 2 (AMBV-805), respectively. In tri-cultures, these mismatches were not an issue because only one strain of each species was present, allowing each to be individually quantified using its corresponding primer. However, in six-member communities containing two strains of the same species, primer cross-reactivity became a limitation. To address this, we used a subtraction-based approach: the combined signal from both strains (e.g., L. crispatus 1 + L. crispatus 2) was measured using the L. crispatus 2 primer set, which also amplified L. crispatus 1. The signal specific to L. crispatus 1, obtained using its strain-specific primer, was then subtracted to estimate the abundance of L. crispatus 2. This method allowed for approximate quantification of each strain despite the lack of complete primer specificity. Samples of the bottom-up SynComs were stored, and RNA extraction was analogous to the top-down SynComs. The generated cDNA was used in RT-qPCR with SYBR Green (Thermo Fisher Scientific) as an intercalating dye and the designed primers to determine the absolute bacterial abundances.

Metabolic analyses of vaginal synthetic communities

Metabolic modeling

Metabolic interactions (cross-feeding) were predicted using genome-scale metabolic models (GEMs). Online available genomes were downloaded as protein sequence fasta file (.faa) for L. crispatus (n = 454), L. jensenii (n = 78), Lim. reuteri (n = 157), Lim. vaginalis (n = 44), and Lim. fermentum (n = 196). To construct high-quality models, species-level models were established using the core genome of these species, which was determined using SCARAP³⁸ with default settings. From each core orthogroup, a representative was chosen to construct the metabolic model using CarveMe,³⁷ with the Gram-positive model as backbone. The quality of the models was assessed using the MEMOTE server⁴¹ with a minimum score of 70. Metabolic interactions were predicted for the synthetic triculture using iterative SMETANA⁴² (n = 600 per community) in detailed mode. Metabolites that were cross-fed were filtered with an SMETANA score >0.5, indicating a high metabolite production and uptake rate of either the donor or receiver model. Inorganic molecules (e.g., zinc, iron) were removed. To assess the effect of medium composition and complexity on cross-feeding simulations, SMETANA simulations were run using MRS and modified MRS (exclusion of L-arginine) environments using settings described above. Additionally, to assess the importance of cross-feeding in the L. crispatus modules used in our in vitro work strain strain-specific GEMS were constructed and SMETANA analysis was performed as described above. Genome-scale metabolic models were deposited at the European Bioinformatics Institute BioModels repository¹⁰⁴ and assigned the identifiers: MODEL2507070001-2507070005, and are available on github (https://github.com/LebeerLab/2025_GEMS_syncom_vagina).

Quantification and statistical analysis

Absolute abundance data represents the average CFU/mL ± SD, the ratio represents the mean ratio ±SD. Details are provided in the figure legends. The co-occurrence patterns were determined using the SPRING correlation metric³⁴ and the Pearson correlation metric, using the NetCoMi package⁹⁷ version 1.1.0. The co-occurrence pattern between L. crispatus, L. jensenii and Lim. vaginalis was validated using all 10 different correlation metrics, including gconda, ccrepe, cclasso, spring, sparcc, spieceasi, pearson, propr, spearman, and bicor.⁹⁷ To account for compositionality in the samples, centered log-ratio (CLR) transformed abundance counts were used. The quality of the metabolic models was assessed using the MEMOTE server⁴¹ with a minimum score of 70. Metabolic interactions were predicted for the synthetic triculture using iterative SMETANA (n = 600 per community) in detailed mode. Metabolites that were cross-fed were filtered with an SMETANA score >0.5. Details of the statistical tests and methodology are provided in the respective STAR Methods sections.

Additional resources

The vaginal swabs and isolates were part of the clinical trial called Isala, with identifier NCT04319536 (clinicalTrials.gov).

Published: August 19, 2025