Bacterial vaginosis (BV) is characterised by microbial dysbiosis in the female genital tract and associated with overgrowth of diverse anaerobic pathogens. BV is the most prevalent vaginal infection among women of reproductive age, with high recurrence rates and a significant global economic burden [1]. During BV episodes, beneficial lactobacilli species are displaced by anaerobic pathogens, primarily Gardnerella vaginalis [2]. In previous work, we have developed a polymicrobial biofilm model that represents the microbial community composition of BV, where G. vaginalis acts as the initial biofilm coloniser, providing a scaffold for the subsequent adhesion of secondary pathobionts, including Fannyhessea vaginae, Mobiluncus curtisii, and Prevotella bivia [3]. Whilst these polymicrobial communities have been shown to tolerate elevated concentrations of metronidazole and clindamycin, the interaction between these biofilms and the host remains insufficiently understood. Therefore, the aim of this study was to investigate the development of a sequentially grown polymicrobial biofilm model using an organotypic vaginal system, allowing for the evaluation of biofilm colonisation, susceptibility to antimicrobial treatment, and biofilm-mediated host responses. For these experiments, we used a four-species BV biofilm model previously described [3], comprising G. vaginalis (ATCC 14018) as the biofilm coloniser for an initial 24 h, followed by the addition of the accessory pathogens Fannyhessea vaginae (DSMZ 15829), Prevotella bivia (DSMZ 20514), and Mobiluncus curtisii (CCUG 21018T) for a further 24 h. These isolates were cultured anaerobically on Columbia agar (Merck Millipore, Cork, Ireland), supplemented with 5 % defibrinated horse blood (E&O Laboratories, Bonnybridge, UK), and transferred to NYC III medium as needed. Biofilms were grown at the liquid interface using 1 × 107 cfu/mL of each species as above on five-day-old EpiSkin human vaginal epithelium (HVE) tissue (Skinethic™, Episkin, Lyon, France), which was incubated at 37 °C in 5 % CO2 for biofilm formation. For treatment, co-cultured biofilms were exposed to metronidazole at four times the peak serum concentration (46 μg/mL), a concentration previously shown to be effective against G. vaginalis mono-species but not multi-species BV biofilms [4]. EpiSkin maintenance media served as the vehicle for the BV control, and untreated (UNT) controls consisting of sterile HVE tissue were also included. At the end of the experiments following biofilm growth and treatment at 72 h, tissues were removed from the inserts by gently piercing the outer ring with a sterile 26G needle (Fisher Scientific, Loughborough, UK). For histological analysis and RNA extraction, tissues were immediately placed in 1 mL formalin or 300 μL RNAlater, respectively. For colony counts and DNA extraction, tissues were placed in 1 mL PBS and ultra-sonicated for 10 min to dislodge the biofilms, before biomass was pelleted via centrifugation. Bacteria were treated using PMAxx dye (Biotium, California, USA) for live/dead analysis, prior to extraction using the MasterPure™ complete DNA extraction kit (CAMBIO, Cambridge, UK) as per manufacturer's instructions. The PMAxx staining was performed according to a previously published optimised protocol [5]. Prior to experiments, this protocol was established in-house using heat-killed bacteria as the limit of detection. Colony counts were then determined using the Miles-Misra technique [6].

Biofilm composition was assessed by qPCR, with a standard curve prepared from DNA derived from known quantities of target bacteria (1 × 103–1 × 108 CFU/mL). Gene expression was analysed by RT-qPCR, with results normalized to a housekeeping gene (GAPDH) and the untreated tissue control. Inflammatory gene expression was selected based on previous studies observing protein-level alterations in response to planktonic BV-associated bacteria in a 3D human cervix model [7]. Primers for all PCR based assays are provided in Table 1. For bacterial assays, previously published primers which target 16S rRNA were used [8]. All qPCR reactions were performed on a ViiA 7 real-time PCR system (Applied Biosystems) using 20 μL reactions as previously stated [3]. In brief, each reaction consisted of 10 μL qPCRBIO SyGreen® mastermix (PCR Biosystems, London, UK), 7 μL Hyclone molecular grade water, 1 μL forward primer (10 μM), 1 μL reverse primer (10 μM) and 1 μL of DNA/cDNA where appropriate. The following cycling conditions were utilised; 50 °C for 2 min, 95 °C for 2 min, 40 cycles of 95 °C for 3 s followed by 60 °C for 30 s. Tissue cytotoxicity was measured daily by assessing supernatant lactate dehydrogenase (LDH) levels using the CyQUANT™ LDH Cytotoxicity Assay. Histological analysis was performed at the University of Glasgow Veterinary School. All experiments were carried out with technical duplicates and biological tissue triplicates (n = 3), and statistical analyses are indicated in the figure legends. Biofilm bacterial quantities were determined from qPCR data, expressed as ‘colony-forming equivalents per mL’ (CFE/mL). Gene expression data are presented using the 2-ΔΔCT method to calculate relative expression levels compared to GAPDH and the untreated control.

As outlined in Fig. 1, there was clear colonisation of EpiSkin vaginal tissues through histological, molecular and conventional microbiological approaches. As evidenced through hematoxylin and eosin and Gram-staining (Fig. 1A), microcolonies and aggregates can be seen adhering to the epithelium surface. All organisms were able to successfully colonise and remain viable at varying quantities within the biofilm following at least 48 h of growth, despite less favourable atmospheric conditions (Fig. 1B–F). These findings agree with a previous study who demonstrated tissue colonisation with each of these pathogens with a six-species consortia [4]. Importantly, we have been able to show that the full biofilm infection model can be performed in 5 % CO2 conditions, without the requirement for potentially cytotoxic anaerobic incubation of the tissue as per the aforementioned study. We hypothesise that the initial biofilm scaffold formed by G. vaginalis, helps promote growth and survival of the more fastidious pathobionts, P. bivia and M. curtisii. In agreement with these findings, previous in vitro studies have identified that G. vaginalis can increase cellular aggregation and promote survival of these pathogens in CO2 when grown as dual-species biofilms with F. vaginae, P. bivia and M. curtisii [9]. Importantly, our model was able tolerate an elevated treatment with metronidazole and only reducing the bioburden by approximately 1.5 log10 cfu/mL (Fig. 1G). When evaluated using live/dead qPCR, unsurprisingly there were no antimicrobial effects on the metronidazole resistant species F. vaginae and M. curtisii, with significant reductions in total and live G. vaginalis and total P. bivia following treatment (p < 0.01). Despite optimally working in an anaerobic environment, metronidazole has been previously shown to demonstrate significant antibiofilm activity against mono-species G. vaginalis when grown in CO2 enriched conditions [10].

One finding of interest is the change in model composition in comparison to when grown anaerobically using simple biofilm substrates, whereby G. vaginalis dominates the biofilm community [3]. In contrast, when using this complex host-pathogen system grown in 5 % CO2, F. vaginae becomes compositionally dominant and the most abundant pathogen, despite colonising the biofilm for 24 h less than G. vaginalis. Previous studies have identified that F. vaginae exerts the most significant proinflammatory properties across a range of vaginal pathobionts including G. vaginalis [7,11]. This is in agreement with our cytotoxicity data assessed by measuring LDH (Fig. 2A), whereby only a small increase in cytotoxicity was observed following the addition of G. vaginalis at day 1, with a more pronounced cytotoxic response following the addition of the accessory pathogens including F. vaginae at day 2. Following metronidazole treatment, there is a subtle but non-significant (p = 0.07) decrease in cytotoxicity in comparison to the untreated BV biofilm. Despite this small decrease, there were no statistically significant changes in gene expression of proinflammatory cytokines and chemokine. When comparing untreated and metronidazole treated tissue, levels of IL-1β were similar, with small decreases in median log2 fold change of CXCL-8 (0.87–0.13) and IL-6 (0.13 to −0.49), all non-significant. Future work evaluating immune protein-release or screening a larger panel of pro/anti-inflammatory mediators may provide further insight into the host-BV biofilm interaction.

In recent years, there has been an advancement in the design and implementation of organotypic models for the pre-clinical study of vaginal dysbiosis [4,12]. The pioneering findings of microbiome modelling on the human vagina-on-a-chip have provided an excellent platform for understanding microbial interactions with the host [12]. It does however have the disadvantage as not being as accessible as the commercially available reconstituted vaginal epithelium used in our study, with additional complexities associated with using a microfluidic device and the requirement for primary vaginal epithelial cells.

In conclusion, we have developed a sequentially grown polymicrobial biofilm model using reconstituted vaginal epithelium that allows for host-pathogen studies in BV. Our model can efficiently colonise the epithelium for up to 72 h when grown in a CO2 enriched atmosphere and can tolerate an elevated concentration of metronidazole. Tissue cytotoxicity increased following the sequential addition of the accessory pathogens following G. vaginalis colonisation and this was only partially reduced following metronidazole treatment, with minor changes in proinflammatory gene expression. Overall, this model can act as a testbed for both host-pathogen and novel antimicrobial discovery in BV.