Gardnerella Vaginalis Alters Cervicovaginal Epithelial Cell Function Through Epithelial Cell-type Speci c Immune Responses


 Background: The cervicovaginal (CV) microbiome is highly associated with vaginal health and disease in both pregnant and non-pregnant individuals. An overabundance of Gardnerella vaginalis in the CV space is commonly associated with adverse reproductive outcomes including bacterial vaginosis (BV), sexually transmitted diseases and preterm birth while the presence of Lactobacillus spp is often associated with reproductive health. While host-microbial interactions are hypothesized to contribute to CV health and disease, the mechanisms by which these interactions regulate CV epithelial function remain largely unknown. Results: Using an in vitro co-culture model, we assessed the effects of Lactobacillus crispatus and G. vaginalis on the CV epithelial barrier, the immune mediators that could be contributing to decreased barrier integrity and the immune signaling pathways regulating the immune response. G. vaginalis, but not L. crispatus, significantly increased epithelial cell death and decreased epithelial barrier integrity in an epithelial cell-specific manner. A G. vaginalis-mediated epithelial immune response including NFkB activation and proinflammatory cytokine release was initiated partially through TLR2 dependent signaling pathways. Additionally, investigation of the cytokine immune profile in human CV fluid showed distinctive clustering of cytokines by G. vaginalis abundance and birth outcome. Conclusions: The results of this study show both microbe- and epithelial cell-type specific effects on CV epithelial function. Altered epithelial barrier function through cell death and immune mediated mechanisms by G. vaginalis, but not L. crispatus, indicates that host epithelial cells respond to bacteria-associated signals, resulting in altered epithelial function and ultimately CV disease. Additionally, distinct immune signatures associated with G. vaginalis or birth outcome provide further evidence that host-microbial interactions may contribute significantly to the biological mechanisms regulating reproductive outcomes.


Introduction
Host-microbe interactions play a signi cant role in the pathophysiology of human disease. Tissuespeci c microbiomes are fundamental to host physiology and immunology. Interactions range from a mutual symbiosis (commensal) to disease-causing pathogenicity with many levels in between (contextual pathogen/pathobiont). Location-speci c microbiomes have been shown to affect the host through multiple complex mechanisms including direct interactions with host cells, such as immune activation through bacterial cell well components or the release of microbial immune regulators including metabolites or anti-microbial proteins [1]. One of the most prominent and well-studied examples of hostmicrobe interactions occur at mucosal surfaces overlaying epithelial barriers, which can be found in speci c human niches including the gut [2,3]. Crosstalk between microbiota and host epithelial cells can promote or diminish the health of the epithelial barrier causing disease states such as in ammatory bowel disease (IBD) [4][5][6].
proteins [29][30][31][32] , and miRNAs [29,31,[33][34][35]. We propose that the CV microbiome is a common regulator of these molecular pathways. Previous studies published by our laboratory and others have shown that bacteria-free supernatants from G. vaginalis and M. mulieris cause breakdown of the ectocervical and endocervical epithelial barriers through cleavage of adhesion proteins, immune activation, and epigenetic regulation [28,29]. Furthermore, mouse studies have shown that vaginal colonization with G. vaginalis results in sPTB [36]. While the results of these studies provide evidence that alterations in the CV microbiome can signi cantly alter epithelial cell function, they did not investigate the mechanistic drivers of host-microbe epithelial interactions.
In vivo, host-microbial interactions in the CV space are complex and likely driven by microbes, microbial by-products, and their communication with speci c immune and epithelial cell types. As vaginal and cervical epithelial cells originated from different embryological origins [37][38][39], they likely have distinct cellular roles with complex cell speci c functions. Despite their unique origins, very few studies have investigated how host-microbe effects differ between cervical and vaginal cells [40,41]. In vivo, both live bacteria and the molecules they secrete (e.g., proteins, metabolites, extracellular vesicles) may elicit relevant cellular effects, and it is unknown how these effects might be similar or divergent across CV epithelial cells. Therefore, revealing the interactions between bacteria and their by-products with different epithelial cell types within the CV space are necessary to advance our understanding of host-microbial interactions and their contributions to reproductive health and disease.
Therefore, the objectives of this study were 1) to determine the effects of live CV bacteria and their supernatants on the integrity of the cervical and vaginal epithelial barrier; 2) to elucidate the immune pro le resulting from interactions between live bacteria or their supernatants and host CV epithelial cells; 3) to determine if toll-like receptor (TLR) signaling pathways are necessary for the activation of the epithelial cell immune responses from common CV microbes; and 4) to validate in vitro ndings by assessing immune signatures in the CV space of pregnant individuals with a high abundance of G. vaginalis.

Bacterial Cultures and Preparation of Bacteria-Free Supernatants
Bacterial strains, Lactobacillus crispatus (ATCC 33197) or Gardnerella vaginalis (ATCC 14018), were obtained from the American Type Culture Collection (Manassas, VA). G. vaginalis was grown on Human Blood Tween Bilayer agar plates (Fisher Scienti c) and L. crispatus was grown on De Man, Rogosa and Sharpe agar (Fisher Scienti c); both strains were grown in New York City (NYC) III broth. Bacteria were grown at 37 o C in an anaerobic glove box (Coy Labs, Grass Lake, MI).
For each experiment the following bacterial growth protocol was followed: L. crispatus and G. vaginalis glycerol stocks were streaked on agar plates and grown overnight. Individual colonies were used to inoculate starter cultures and grown overnight. Starter cultures were diluted to an optical density of 0.2 and then used to inoculate working cultures, which were grown for 20 hours prior to use in experiments. Bacterial densities of the working cultures were estimated the day of the experiment based on optical density readings at 600 nm using an Epoch2 plate reader (Biotek, Winooski, VT), and the appropriate volume was centrifuged at 13,000 g for 3 min. The bacterial pellets were resuspended in K-SFM cell culture media without antibiotics and added to cells at 10 4 -10 6 CFUs/well. Precise bacterial densities of the working culture were determined by CFU assays (plating serial dilutions of the working cultures). For all experiments, reported bacterial densities are +/-0.5 log of the noted bacterial density (CFU/well).
To obtain supernatants, the working cultures were centrifuged at 13,000 g for 3 min and the supernatant was ltered through a 0.22 μm lter (Fisher Scienti c) to remove any remaining live bacteria. Bacteria-free supernatants are diluted to 10% v/v in K-SFM cell culture media without antibiotics.
Epithelial Cell/ Bacteria Co-Culture In vitro Model Ectocervical, endocervical and vaginal cells were plated at 2.0 x 10 5 cells/well in twenty-four well plates containing K-SFM without antibiotics. The next day, the cells were treated with either live L. crispatus or G.
vaginalis (1x10 4 -1x10 6 CFU/well) or 10% (v/v) bacteria-free supernatants (1x10 5 -1x10 7 CFU/mL culture density) in K-SFM cell growth media for 24 hours. For cells treated with bacteria-free supernatants from L. crispatus, K-SFM media was supplemented with 50mM HEPES and sodium bicarbonate (3000 mg/L total concentration) to bring the pH of the media up to a physiological level (7.2) as L. crispatus bacteria produces high amounts of lactic acid during growth. Without pH adjustments, even at lower volume per volume percentages, the cells did not survive. In additional experiments, ectocervical, endocervical and vaginal cells were pre-treated with a neutralizing IgA monoclonal antibody to human TLR2 (10ug/mL) (anti-hTLR2-IgA, InvivoGen, San Diego, CA) for one hour prior to exposure to live bacteria or supernatants from L. crispatus or G. vaginalis. For all supernatant experiments, cells were also treated with NYCIII bacterial growth media alone as a negative control to determine any baseline effects of the growth media on the outcome of interest. At the end of each experiment, cell culture media was collected for cell death, ELISA assays and/or the cells were collected in Trizol (Invitrogen, Thermo-Fisher Scienti c) for RNA extraction.
Differential Interference Contrast Imaging Ectocervical, endocervical and vaginal epithelial cells were plated at 2.0 x 10 5 cells/dish on 35 mm high glass bottom µ-dishes (ibidi, Martinsried, Germany) coated with 0.1% gelatin for 24hrs. Live L. crispatus and G. vaginalis were added to the cells and incubated 4-6 hours prior to imaging. Differential Interference Contrast (DIC) images of our epithelial cell / bacteria co-culture with L. crispatus and G. vaginalis were taken using the Zeiss Axio Observer 7 Wide eld microscope using the 100x objective (Zeiss 100x/1.4 NA oil Plan-Apochromat) with ZEN Blue software (version 2.5).
HEK-hTLR2 Treatments and NFkB and IL-8 Detection HEK-hTLR2 cells were plated at 7.5 x 10 4 cells/well in 96 well plates containing DMEM + 10% heatinactivated FBS without antibiotics. The next day, the cells were treated with either live L. crispatus or G. vaginalis (10 4 -10 6 CFU/well) or 10% (v/v) bacteria-free supernatants (10 7 -10 5 CFU/mL culture density) in DMEM cell culture media for 24 hours. In additional experiments, the cells were pre-treated with the TLR2 neutralizing antibody, anti-hTLR2-IgA (InvivoGen) for one hour prior to exposure to live bacteria or supernatants. In these experiments, the TLR2 agonist FSL-1 (10 ng/mL, Sigma-Aldrich, St. Louis, MO) was used as a positive control for antibody e cacy. For detection of a nuclear factor kappa-B (NFkB) response (SEAP reporter), cell culture supernatants were incubated with Quanti-Blue solution (Invivogen) for 1 hour, pictures were taken of the plate, and absorbance was read at 630nm on a SpectraMax i3x plate reader (Molecular Devices). For detection of an IL-8 response (Lucia luciferase reporter), cell culture supernatants from the same experiment were combined with Quanti-Luc solution (Invivogen) and luminescence was read immediately on a SpectraMax i3X plate reader. Additionally, cell culture supernatants were used in cell death assays as described below.

Cell Death Assay
Ectocervical, endocervical, vaginal and HEK-hTLR2 cells were grown as described above. The release of lactate dehydrogenase (LDH) from ectocervical, endocervical, and vaginal cells (n=3-9 independent experiments per cell type) was measured using the CytoTox 96 Non-radioactive cytotoxicity assay (Promega, Madison, WI). This assay allows for the quantitative measurement of LDH that is released upon cell lysis using a coupled enzymatic assay that results in the conversion of a tetrazolium salt into a red formazan product. The amount of color formed is proportional to the amount of LDH released as the cells lyse. Color formation was read on a colorimetric plate reader at 490nm and absorbance values were recorded.

Cell Permeability Experiments
Ectocervical, endocervical and vaginal cell permeability was determined using an In Vitro Vascular Permeability Assay (Millipore, Bedford, MA). Brie y, ectocervical, endocervical and vaginal cells were plated at 1.0 x 10 6 cells/mL into 24 well hanging cell culture inserts which contain 1 µm pores with a transparent polyethylene terephthlate (PET) membrane pre-coated with collagen. The cells were treated with either live L. crispatus or G. vaginalis (10 6 -10 4 CFU/well) or 10% (v/v) bacteria-free supernatants from L. crispatus (n=6) and G. vaginalis (n=9) for 24 hours. In additional experiments, ectocervical, endocervical and vaginal cells were pre-treated with the anti-hTLR2 antibody prior to exposure to live bacteria or supernatants from L. crispatus or G. vaginalis (as described above) for 24 hours (n=9). For all supernatant experiments, cells were also treated with NYCIII bacterial growth media alone as a negative control to determine any baseline effects of the growth media on cell permeability. After 24 hours of treatment, the media was removed and phenol red free K-SFM media (ScienCell Laboratories, Carlsbad, CA) containing FITC-Dextran was added to the top of the insert. The movement of FITC-Dextran from the top insert to the bottom was measured after two hours by a uorescent plate reader at 485nm and 535nm, excitation and emission, respectively. ELISA Ectocervical, endocervical and vaginal cells were cultured in 24-well plates and treated with either live bacteria or bacteria-free supernatants as stated above. IL-8 was measured in cell culture media after 24 hours of treatment. The expression of IL-8 was measured by a ligand-speci c commercially available ELISA kit that utilizes a quantitative sandwich enzyme immunoassay technique using regents from R&D Systems (Minneapolis, MN).

Luminex Assay
Ectocervical, endocervical and vaginal cells were cultured and treated with live bacteria or bacteria-free supernatants as stated above. A 41-plex cytokine/chemokine (HCYTMAG-60K-PX41) and a TGFB 3-plex (TGFBMAG-64K-03) human magnetic bead Luminex panel (EMD Millipore, Billerica, MA) was run on 1) ectocervical, endocervical and vaginal cell culture media after 24 hours of treatment with L. crispatus (n=3) or G. vaginalis (n=3) live bacteria or bacteria-free supernatants and 2) cell culture media from ectocervical, endocervical and vaginal cells pre-treated with an anti-TLR2 blocking antibody followed by L. crispatus (n=3) or G. vaginalis (n=3) live bacteria and 3) human cervical vaginal uid (CVF) collected from individuals with or without a sPTB with low or high CV G. vaginalis abundance. Human CVF was collected using Dacron swabs (Starplex, ThermoFisher). Materials on the swabs were eluted in sterile PBS with a protease inhibitor cocktail (Complete Mini) for 5 min to release the soluble proteins. All samples were run in duplicate, per the manufacturer's protocol on the FLEXMAP3D Luminex platform (Luminex, Austin, TX). Absolute quanti cation in pg/mL was obtained using a standard curve generated by a veparameter logistic (5PL) curve t using Xponent 4.2 software (Luminex). Fold change values were calculated between treatment groups and the non-treated control (NTC) for live bacteria or between the treatment groups and the NYC media control for bacteria-free supernatants. For those cells pre-treated with the anti-TLR2 blocking antibody, fold change was calculated between the bacteria treatment alone and the bacteria plus the anti-TLR2 antibody. For human samples, data is expressed as average pg/mL. Heatmaps were created using R.
mRNA Isolation from Epithelial Cells Following treatment with live L. crispatus and G. vaginalis, ectocervical, endocervical and vaginal cells were collected in Trizol (Invitrogen, Thermo-Fisher Scienti c) and underwent phenol-chloroform extraction. The resulting aqueous phase was further column puri ed with the miRNeasy kit (Qiagen, Hilden, Germany) according to the manufacturer's protocol for total RNA isolation. RNA concentration was determined via a NanoDrop 2000 Spectrophotometer (Nanodrop TM Rockland, DE) prior to the generation of cDNA. cDNA generation and qPCR cDNA was generated from 1µg of isolated RNA from ectocervical, endocervical and vaginal cells using the high capacity cDNA reverse transcription kit (Applied Biosystems, Thermo-Fisher Scienti c). qPCR was performed on the 7900HT Real-Time PCR System (Applied Biosystems) using the TaqMan Universal PCR Master Mix (Applied Biosystems) according to the manufacturers' protocols. The standard curve method was used for relative expression quanti cation using the RQ manager software v2.4 (Applied Biosystems). The relative abundance of the target of interest was divided by the relative abundance of 18S in each sample to generate a standardized abundance for the target transcript of interest. All mRNA primers were purchased from Applied Biosystems: TLR2, MYD88, NOD1, NOD2 and 18S.
To investigate if the cytokine signature in pregnant individuals with high and low G. vaginalis abundance was different, we utilized CVF samples from a prospective cohort study of 2000 individuals with singleton pregnancies. Details of this study are available in the primary report [10]. All participants provided written informed consent and the study was approved by the Institutional Review Board at the University of Pennsylvania (IRB #818914) and the University of Maryland School of Medicine (HP-00045398). Cervicovaginal specimens were self-collected by the participant or collected by a research coordinator between 16 and 20 weeks. A set of cervicovaginal swabs was obtained including an ESwab (COPAN) stored in 1 mL of Amies Transport Medium and a Dacron swab stored without buffer. All samples were immediately frozen at -80 °C until processing. All delivery outcomes were recorded. Cases of PTB were adjudicated by the PI (Elovitz) to determine if they were spontaneous as previously described [10]. To assess the CV microbial communities, 16s rRNA gene ampli cation and sequence was performed as described previously [10].
For this study, a nested case-control selection of subjects was performed (Table 1). Inclusion criteria were individuals identi ed as being in CST IV at 16-20 weeks by microbial sequencing (n=131). From this cohort of individuals with CST IV, four groups were identi ed by birth outcome and G. vaginalis abundance. The four groups were 1) Term with low G. vaginalis (GV abundance lower than 0.2), 2) sPTB with low G. vaginalis, 3) Term with high G. vaginalis (GV abundance of 0.2 or greater) or 4) sPTB with high G. vaginalis. G. vaginalis abundance cutoff of 0.2 was determined based on the calculated median from the whole cohort. Analysis of Luminex cytokine data was performed using a one-way ANOVA followed by different post hoc tests comparing treatment groups to 1) a non-treated control (NTC) was done using a Dunnett's multiple comparison test or 2) between select pairs of treatment groups was done using Sidak's multiple comparisons test. Human CVF Luminex data were analyzed by two tailed t-test with term delivery compared to sPTB in both the low and high G. vaginalis abundance groups. If the variances between groups were signi cantly different, then a t-test with Welch's correction was used.

G. vaginalis co-localizes with cervicovaginal epithelial cells resulting in increased cell death in a host-microbial coculture model
In creating our in vitro host-microbial co-culture model, characterization of bacterial interactions with host epithelial cells was necessary before moving forward with subsequent experiments. Differential Interference Contrast (DIC) images of L. crispatus and G. vaginalis co-cultured with ectocervical ( Figure  1A p<0.001, vaginal: p<0.0001), with no signi cant effects seen with live L. crispatus at the same doses ( Figure 2A, C, E). CV cells co-cultured with live G. vaginalis had signi cantly higher cell permeability when compared to those co-cultured with L. crispatus at both the 10 6 (ecto: p<0.0001, endo: p=0.0072, vaginal: p<0.0001) and 10 5 (ecto: p<0.0001, endo: p=0.0233, vaginal: p<0.0001) doses of bacteria but not at 10 4 (CFU/well). Exposure of ectocervical, endocervical, and vaginal cells to bacteria-free supernatants from G. vaginalis had no effect on cell permeability at any of the doses tested, while L. crispatus supernatants signi cantly decreased cell permeability, when compared to the NYC control, similarly at all three doses tested (ecto: p<0.0001, endo: NS, vaginal: p<0.0001) ( Figure 2B, D, F). There was no signi cant effect of the bacterial growth media alone (NYC control) on cell permeability of ectocervical, endocervical, or vaginal cells.
G. vaginalis and L. crispatus bacteria and their bacterial-free supernatants induce distinct immune pro les in a CV epithelial cell-type speci c manner We have previously shown that G. vaginalis bacteria-free supernatants can increase the release of a varied group of in ammatory mediators from ectocervical cells [29]; however, the contribution of live bacteria (including bacterial cell wall and its associated proteins) to activating a host-epithelial cell immune response was not assessed. Therefore, we sought to examine the effects of both live bacteria and bacteria-free supernatants on the activation of the immune response from all three CV epithelial cell lines. Of the 44 cytokines/chemokines included in the discovery-based Luminex arrays (41-plex immune and 3-plex TGFB), 28 were detectable in at least one of the three cell lines after exposure to live bacteria ( Figure 3A), while 27 were detectable after exposure to bacteria-free supernatants ( Figure 3B). A full list of all cytokines/chemokines included in the Luminex assays and their average values (pg/mL) in each cell type can be found in Supplemental Table 1A (live bacteria) and 1B (bacteria-free supernatants). Most of the detectable cytokines were signi cantly increased after exposure to live G. vaginalis versus exposure to L. crispatus (when each are compared to NTC) (Supplemental Table 2). PDGF-AA was the only cytokine decreased after live G. vaginalis exposure in all three cell lines. Very few analytes were altered by live L. crispatus with two, three, and four cytokines changed in ectocervical, endocervical, and vaginal cells, respectively (Supplemental Table 2A). In cells exposed to bacteria-free supernatants, there were signi cant numbers of cytokines altered in both the L. crispatus and G. vaginalis groups. In cells exposed to bacteria-free supernatants from L. crispatus, the majority of altered cytokines were decreased, while in cells exposed to G. vaginalis, the majority of altered cytokines were increased (Supplemental Table 2B).
Overall, similar cytokines/chemokines were detected after co-culture with either live bacteria or bacteriafree supernatants; however, the levels of those cytokines varied widely between the two exposures with 13 cytokines being higher after exposure to live bacteria when compared to supernatants. Nine were lower after live bacteria exposure compared to supernatants, and ve were similar. A representative subset of cytokines ( Figure 3C, D) demonstrates the differential cell-speci c responses of ectocervical, endocervical, and vaginal cells to either live bacteria ( Figure 3C) or bacteria-free supernatants ( Figure  3D).

L. crispatus and G. vaginalis bind to TLR2 activating NFkB
To identify the intracellular immune pathways being activated by G. vaginalis we utilized HEK TLR2 reporter cells. Live G. vaginalis and L. crispatus induced NFkB activation in a dose-dependent manner with G. vaginalis-exposed cells sustaining a higher level (than L. crispatus) of NFkB across all three bacterial doses tested ( Figure 4A, B). IL-8 activation was signi cantly elevated in a dose-dependent manner to live G. vaginalis but not L. crispatus (except at the 10 6 dose) ( Figure 4C). A similar response was seen after exposure to bacteria-free supernatants from G. vaginalis and L. crispatus (Figure 4 E-G). HEK TLR2 cell death was similar to that seen in the CV epithelial cells with a dose-dependent increase in LDH in live G. vaginalis co-cultures and no signi cant difference found after exposure to live L. crispatus nor any of the bacteria-free supernatants ( Figure 4D, H).
G. vaginalis-mediates increased cytokine release partially through TLR2 activated signaling pathways While this study has shown that NFkB and a multitude of cytokines are activated/increased in response to live G. vaginalis, at least partially, through activation of the TLR2 receptor, we have not investigated if the TLR2 receptor is a necessary activation signal in the CV immune response. TLR2, MYD88, NOD1, and NOD2 are all expressed in ectocervical, endocervical, and vaginal epithelial cell lines ( Figure 5A). activation is essential for the epithelial immune response and consequent cytokine increase, we blocked the TLR2 receptor and examined cytokine output after exposure to L. crispatus and G. vaginalis in both HEK TLR2 cells and all three CV cell lines. In HEK TLR2 cells, FSL (a potent TLR2 agonist, p<0.0001), L. crispatus (p<0.0001) and G. vaginalis (p<0.0001)-mediated increases in NFkB were signi cantly reduced with TLR2 blockade ( Figure 5B). Similarly, pretreatment with the anti-TLR2 antibody also reduced IL-8 activation after exposure to FSL (p<0.0001) and G. vaginalis (p=0.0007) but not L. crispatus ( Figure 5C). Preliminary investigation into the G. vaginalis-mediated increase in cytokine expression in CV cell lines showed that blocking TLR2 also signi cantly reduced IL-8 expression in a cell-type speci c manner: 17% reduction in ectocervical cells (p=0.0067, Figure 5D), 66% reduction in endocervical cells (p<0.0001, Figure 5E), and 37% reduction in vaginal cells (p<0.0001, Figure 5F).
To further investigate the role of TLR2 activation in G. vaginalis-mediated cytokine production in all the CV cell lines, we assessed the immune response after TLR2 blockade using the same Luminex panels as described above ( Figure 6A, B, C). As we observed previously, there was a varied immune response between the three cell lines with ectocervical cells exhibiting the greatest overall number of cytokines altered after TLR2 blockade in both L. crispatus and G. vaginalis exposed cells followed by endocervical and then vaginal cells (Supplemental Table 3). TLR2 blockade inhibited the G. vaginalis-induced increase in cytokine expression across all cell lines, except for IL-1RA which was increased. In contrast, TLR2 blockade only mitigated the L. crispatus-induced cytokine expression in ectocervical but not endocervical or vaginal cells ( Figure 6). Interestingly, TLR2 blockade also increased seven L. crispatus-induced cytokines in ectocervical cells and one cytokine in vaginal cells.
Activation of TLR2 is su cient, but not essential, for the G. Pregnant individuals with a high CV G. vaginalis abundance who deliver preterm have a distinct CV immune signature from those with a term birth To investigate if G. vaginalis induces a distinct immune signature in those who ultimately have a sPTB versus term birth, we performed Luminex on CVF from 131 pregnant individuals categorized into CST-IV by 16S rRNA gene sequencing [10]. Four groups comprised of individuals with high or low G. vaginalis CVF abundance and term or sPTB delivery outcomes were analyzed. Secondarily, the CVF immune pro le of individuals with high G. vaginalis abundance and a sPTB was compared against the immune pro le our in vitro co-culture model (CV epithelial cells exposed to live G. vaginalis). All 44 cytokines studied were detectable in human CVF. Of those 44, seven were signi cantly (p<0.05) increased in individuals with high CVF G. vaginalis abundance who underwent a preterm delivery compared to those that delivered at term ( Figure 8A). Six additional cytokines (EGF, G-CSF, IL-13, MIP-1a, TNF-a, sCD40L) were non-signi cantly increased (p<0.10) in those who had a SPTB and high G. vaginalis abundance. Four cytokines were signi cantly decreased (p<0.05) in CVF from individuals with a low G. vaginalis abundance who had a sPTB compared to those with a term birth ( Figure 8A). Of the 20 cytokines that were signi cantly increased in the G. vaginalis epithelial cell co-culture model ( Figure 3A), six cytokines overlapped between human CVF from individuals with high G. vaginalis and a sPTB and CV epithelial cells exposed to live G. vaginalis ( Figure 8B and 8C).

Discussion
This study provides evidence that CV bacteria have signi cant microbe-and cell speci c effects on host epithelial cell function indicating that complex host-microbe interactions contribute to vaginal health. We have shown that G. vaginalis, a vaginal microbe commonly associated with adverse reproductive health, results in both functional and immune-based host responses as evidenced by increased CV epithelial cell death, epithelial barrier breakdown and immune activation partially through TLR2/NFkB induced expression of chemokines/cytokines. However, L. crispatus, a microbe associated with optimal vaginal health, does not induce changes in epithelial cell death nor epithelial barrier function. Unique to this work, we reveal that vaginal and cervical epithelial cells possess different immune responses to similar microbe challenges and thus emphasize the need to understand the complexity of the CV space for optimizing reproductive health. Importantly, these studies demonstrate that activation of TLR2 is a shared pathway across epithelial cell types in response to G. vaginalis exposure. Results from our human study con rm our in vitro ndings that the host-microbe mediated CV immune response is associated with the abundance of select microbes and notably that an increased cytokine response is associated with birth outcome.
In both pregnant and non-pregnant individuals, the primary function of the cervical and vaginal epithelial cells is to create a barrier against harmful microbes. To accomplish this, multiple diverse mechanisms are used including the creation of a physical barrier made up of the epithelial cells themselves, the expression of adhesion and tight junction proteins responsible for holding the cells tightly together, and the production of a mucous layer that protects the epithelial cells by entrapping or killing pathogens. Pathogenic microbes have historically developed ways to circumvent this barrier, as is the case with G. vaginalis,which expresses a number of factors that enhance its virulence potential including adherence to CV epithelial cells, bio lm forming capabilities, mucous degradation (salidase) [42], and host cell cytotoxicity [43]. G. vaginalis has been shown to produce vaginolysin [44,45], a cholesterol dependent cytolysin that acts as cytotoxic pore forming toxin and, consequently, kills host epithelial cells. Similar to previous studies investigating the cytotoxic effects of G. vaginalis[46], we observed CV cell death in our in vitro host-microbial co-culture model providing evidence that G. vaginalis functional outcomes were conserved in this model. Interestingly, the cytotoxic effect of G. vaginalis was similar across all three epithelial cell types with a slightly more potent effect in ectocervical versus endocervical and vaginal cells. L. crispatus, as expected, had no effect on CV epithelial cell death despite being exposed to a large dose (10 6 CFU/well) of bacteria.
The integrity of the CV barrier is essential not only to vaginal health (vaginal epithelial barrier), but also to a successful pregnancy (cervical epithelial barrier). The breakdown of the cervical epithelial barrier is thought to be a critical step in the cervical remodeling and dilation process that occurs prior to the onset of labor [47]. Consequently, early disruption of the epithelial barrier is hypothesized to contribute to preterm delivery through premature cervical remodeling or microbe-associated in ammatory mechanisms. A previous study from our laboratory supports this hypothesis as it demonstrated that intravaginal G. vaginalis colonization in a pregnant mouse model results in preterm birth with an activated immune response, breakdown of the epithelial barrier, induced cervical remodeling, and altered cervical biomechanics [32]. Consistent with our previous ndings in an animal model, we show that live G. vaginalis decreases epithelial barrier integrity in vitro in all three CV cell types. We have previously demonstrated that G. vaginalis bacteria-free supernatants (factors secreted by bacteria) are also able to breakdown the cervical epithelial barrier [29]. In contrast to this prior report, in our current study, we found that bacteria-free supernatants did not have the same effect on epithelial cell permeability. One notable difference is the exposure time used in this study (24 hours) versus our previous study (48 hours). While exposure time may be a cause of the difference in ndings with bacteria-free supernatants, it is also worth considering the more potent effects of live bacteria on epithelial cell function (predominantly increased cell death) as this likely occurs in vivo. While there was a signi cant difference in both ectocervical and vaginal (but not endocervical) cell permeability between G. vaginalis and L. crispatus bacteria-free supernatant exposed cells, the effect of G. vaginalis supernatants was diminished compared to live G. vaginalis. These results do provide evidence that live bacteria (and their cell wall) may be a more effective or signi cant contributor to epithelial barrier function than bacterial secreted factors alone. However, a key nding of these experiments is that microbes and their supernatants elicit different biological effects on CV epithelial cells. In vivo, both the live bacteria and their secreted factors likely contribute to the CV host response. Therefore, to fully understand the biological pathways that govern vaginal/reproductive health, it will be essential to reveal the harmful and bene cial effects of both live bacteria and their supernatants. As such, further investigation into the composition (e.g., proteins, metabolites, etc.) of bacteria-free supernatants would be warranted to fully understand their role in regulating CV epithelial function.
A breakdown in the CV barrier likely has multiple overlapping causes. Therefore, while epithelial cell death de nitively plays a signi cant role in decreased barrier integrity, there are other mechanisms that have been shown to contribute to a breakdown in the cervical and vaginal epithelial barriers including in ammation [30]. As part of the epithelial barrier, one of the main functions of epithelial cells is to initiate an innate immune response to ght invading pathogens. Several previous studies from our laboratory and others have shown that G. vaginalis (live and supernatants) induces a robust but widely varied and complex in ammatory response in CV epithelial cells [29,48,49]. Similarly, in this study, the presence of live bacteria revealed a microbe-and cell-speci c immune pro le. G. vaginalis exposure resulted in the activation of an immune response across all three CV epithelial cells lines, however, the level of cytokine production across the three cell types was often diverse and varied by each cytokine analyzed. At a macroscopic level, vaginal epithelial cells seemed to have a more robust innate response to live G. vaginalis when compared to the cervical epithelial cells. This nding suggests that vaginal epithelial cells may act as a rst line of defense against pathogens entering the vaginal canal and may help to protect the cervix from infection, especially during pregnancy. The immune response evoked by G. vaginalis bacteria-free supernatants was also robust in activating cytokines, and like our previous published study [29], L. crispatus also signi cantly altered cytokine production albeit mostly in a downward direction indicative of its commensal, and perhaps protective, functions. These results indicate that CV epithelial cells can recognize multiple different bacterial components, either from bacterial cell wall or secreted factors, and mount an essential immune response. We hypothesized that this type of redundancy makes it more di cult for harmful bacteria to evade host immune recognition.
Epithelial cell recognition of bacterial components has been shown to activate the innate immune response through toll like receptor (TLR) signaling. Of the TLRs, TLR2 is known to be activated bypathogen-associated molecular patterns (PAMPs)includinglipoproteins which are ubiquitous to all bacteria and highly expressed in the outer membrane of gram-positive bacteria.Once activated, TLR2 initiates intracellular signalingpathways which induce nuclear translocation of NFkB to modulate gene transcription and consequent in ammatory cytokine productionand immune cell in ltration [50,51].Previous studies have shown that CV lavage samples from clinical cases of BV upregulate cytokines through TLR2-mediated mechanisms [52,53]. There is a paucity of data regarding the role of TLR signaling in the setting of CV host-microbe interactions. Using a TLR2 reporter cell line, we found that exposure to live and bacteria-free supernatants of G. vaginalis signi cantly induced NFkB and IL-8. Interestingly, L. crispatus also activated NFkB but that activation did not result in increased IL-8, except at the highest bacterial inoculum. These results indicate that high doses of bacteria, independent of bacteria species, activate a TLR2-mediated immune response. The non-speci c activation of TLR2 by assumed healthy and non-healthy microbes may be a regulatory mechanism by the CV epithelium to prevent bacterial overgrowth. Alternatively, it is possible that a secondary signal or that a threshold of NFkB activation may be needed to activate cytokine production in response to L. crispatus or other CV microbes. The ability of G. vaginalis to cause host immune activation at a dose where L. crispatus does not suggest that host cells are able to recognize bacteria speci c PAMPs and mount an immune response to more pathogenic bacteria. Additionally, like our cell death and permeability results, live bacteria exposure, compared to bacterial supernatants, results in a more robust IL-8 response, again indicating that whole bacteria have more substantial effects on these cellular functions than bacterial secreted factors alone.
It is well known that immune system activation is a redundant process with the ability of many different signals/receptors to be activated resulting in the same outcome of upregulated cytokines which also have overlapping functions (cytokine pleiotropy) [54]. Therefore, identifying which factors within the immune response are essential for altering CV epithelial cell function become necessary to target these pathways for future therapeutic or preventative strategies against conditions associated with non-optimal microbial communities in the CV space such as BV, STI acquisition, or sPTB. Since we have shown in this study that CV epithelial cells express TLR2 and the adaptor protein, MYD88, needed for NFkB activation [51,55,56],we investigated if TLR2 activation is an essential mechanism for bacteria-induced host immune activation. After blocking the TLR2 receptor, the reduction in NFkB activation seen with both live L. crispatus and G. vaginalis exposure also resulted in a mitigated IL-8 response. The complete mitigation of L. crispatus-induced NFkB and IL-8, but not G. vaginalis, indicates that TLR2 may not be the only immune pathway activated by G. vaginalis. Interestingly, in CV epithelial cells, blocking TLR2 resulted in varying levels of IL-8 reduction depending on the cell type. The biggest reduction in IL-8 (66% after blocking TLR2) was seen in endocervical cells. This result agrees with the results of our Luminex assay, where the greatest number of G. vaginalis-induced cytokines were reduced after blocking TLR2 in endocervical cells perhaps indicating that TLR2 signaling is a bigger contributor to the immune response in endocervical cells than the other cell types. Of note, all three CV cell types express NOD1 and NOD2 which recognize intracellular PAMPs that can enter the cell through phagocytosis or membrane pores (such as those created by vaginolysin). NODs are part of the NLRP3 in ammasome which can activate NFkB signaling independently of membrane bound TLRs. G. vaginalis has been shown to activate the NLRP3 in ammasome in macrophages and monocytes [57,58]. Therefore, it is possible that NOD signaling may be a signi cant regulator of G. vaginalis-mediated in ammation in CV cells, especially in ectocervical and vaginal cells where TLR2 is likely not the only mediator of host-microbial mediated in ammation.
While the TLR2 blockade inhibited G.vaginalis-mediated epithelial immune activation, it did not prevent G. vaginalis-induced changes in epithelial cell permeability. Our data suggest that while TLR2 receptor blockade prevents some immune response, it is insu cient to limit the diverse downstream molecular effects from exposure to G.vaginalis. Consistent with the redundancy of the host immune response, it is plausible that multiple pathways need to be blocked to signi cantly mitigate the G. vaginalis-mediated breakdown of the CV epithelial barrier. In addition to in ammation, other biological mechanisms are known to regulate the epithelial barrier including cell death and tight junctions, therefore, it is quite likely that a combination of interconnected factors are necessary for a functional change to the barrier. Future research targeting a combination of biological mechanisms would be warranted to fully assess the essential mechanisms regulating CV epithelial barrier function.
We acknowledge that an in vitro model of host-microbial interactions has its limitations mostly due to the nature of a single cell type acting alone without the effects of surrounding cells including stromal cells and immune effector cells (neutrophils, T-cells, NK cells, macrophages). Therefore, we chose to compare our in vitro cytokine results with those seen in a well characterized human cohort of CVF samples collected from pregnant individuals characterized as having CST-IV, with high or low G. vaginalis abundance, who ultimately had a term delivery or sPTB [10]. Although human data inherently has wide variability, the immune signature revealed a distinctive clustering of cytokines within each group analyzed. Interestingly, there seems to be a larger number of elevated cytokines in CVF from pregnant individuals with low compared to high G. vaginalis abundance early in pregnancy. Since all individuals in this study have a CST-IV microbial state, these results would indicate that the interaction of select CST-IV In this cohort, since we did not categorize patients by BV status, but by GV abundance and delivery outcome instead, it is di cult to determine if the G. vaginalis strain subtypes present in these individuals fall into the more virulent category. However, it is interesting to note that when comparing individuals with high G. vaginalis abundance who had a term versus sPTB, there is a cluster of elevated cytokines (EGF, G-CSF, GM-CSF, IL-2, PDGF, Eotaxin and IL-9) that may give some insight into identifying those individuals who are at higher risk for sPTB. A more indepth analysis of this data including correlations with patient metadata, other bacterial species abundance and immune regulators is planned in future studies.
In comparing the immune signature from our human dataset to that of our in vitro host-microbial coculture model, we found both similarities and differences in the baseline presence and elevation of the cytokines. Not surprisingly, there were more detectable cytokines in human cervicovaginal samples when compared to our in vitro cultures (44 vs 28 cytokines). Additionally, there was some overlap in elevated cytokines between the high G. vaginalis abundance with sPTB group and our CV cells exposed to live G. vaginalis (six cytokines) indicating that our in vitro co-culture partially mimics the activated immune response seen in human samples. Of note, IL-8 was not increased in the high G. vaginalis abundance group in our human cohort. While this was surprising given the strong correlation observed with IL-8 and G. vaginalis exposure in our in vitro cultures, we acknowledge that the in ammatory pro les between a single cell culture and a true physiological response will be different. However, the primary outcome of a G. vaginalis-mediated in ammatory response is similar in both sample subsets. Given this result, the in vitro co-culture model remains a useful tool in studying the mechanisms regulating host-microbial interactions with the knowledge that ndings using this model should be veri ed in human samples as well.

Conclusion
Overall, the results of this study have provided critical insight into the host-microbial interactions between CV epithelial cells and the microbiota that inhabit the CV space. This study has shown that G. vaginalis alters cervical and vaginal epithelial cell function by decreasing epithelial barrier integrity, increasing cell death, and activating an in ammatory response that is partially regulated by TLR2/NFkB-based signaling pathways. Importantly, this study shows distinct epithelial cell-speci c immune responses to microbespeci c signals that reveal complex interactions within the CV space. Understanding differences in the epithelial immune response in the cervix versus vagina will also provide insights into cervical or vaginal speci c immune-based therapies/treatments for known G. vaginalis-associated states such as BV, sexually transmitted diseases (e.g., HIV), or adverse pregnancy outcomes including preterm birth. Additionally, differential immune pro ling of CST-IV dominant pregnant individuals with altered levels of G. vaginalis abundance and delivery outcomes provides a distinct immune signature for risk strati cation of individuals who may ultimately have a sPTB. The results from this study begin to ascribe a biological mechanism as to how host epithelial cells and CV microbes interact to regulate cellular function in both commensal and dysbiotic states.

Consent for Publication
Not Applicable Availability of data and materials All data generated or analyzed during this study are included in this published article and its supplementary information les.   Live G. vaginalis activates the host-epithelial immune response while bacteria-free supernatants alter immune activation in a non-bacteria speci c manner. Immune cytokines/chemokines released from