To identify potential oligonucleotides for the F. vaginae probe, we selected a set of sequences from 16S and 23S collections, with lengths >1200 bp or >1600 bp, respectively, available at Arb-Silva database (https://www.arb-silva.de/search/). Only sequences with a quality score >90 were considered. Each set contained sequences from F. vaginae strains, as well as species from the same genus and other genera closely related to the bacterium of interest. The sequences were then aligned using the Clustal Omega tool (https://www.ebi.ac.uk/Tools/msa/clustalo/). Regions for potential probes were searched, showing the same sequence in the species of interest and one or more mismatches in the sequences belonging to other species. Theoretical sensitivity and specificity of the PNA probes were determined, as previously described (Almeida et al., 2013). The probes were evaluated using the TestProbe tool (https://www.arb-silva.de/search/testprobe/) with no mismatches allowed. Sequences with the highest theoretical sensitivity and specificity, complementarity with a low number of non-interest sequences, GC content between 40% and 60%, high melting temperature (>50°C) (Almeida et al., 2011), and Gibbs free energy ranging from -13 kcal/mol to -20 kcal/mol (Yilmaz and Noguera, 2004) were selected as the best probes. The selected probe was then synthesized (Eurogentec, Seraing, Belgium) and the oligonucleotide N-terminus was linked to an Alexa Fluor molecule via a double 8-amino-3,6-dioxaoctanoic acid linker (F. vaginae probe: Alexa Fluor 488-OO-CGATGTGCGACTAAA).

F. vaginae ATCC BAA-55, F. vaginae CCUG 42099, F. vaginae CCUG 44116, and 21 other F. vaginae strains, previously isolated from cases of BV (De Backer et al., 2006; De Backer et al., 2010; Santiago et al., 2012), were used to determine F. vaginae probe analytical sensitivity. Forty different bacterial species associated with BV or with the vaginal microbiota, were used to determine F. vaginae probe analytical specificity. The strains were grown in Columbia Blood Agar Base (Oxoid, Basingstoke, UK) supplemented with 5% (v/v) of defibrinated horse blood (Oxoid) for 24 or 48 h, except for Sneathia sanguinegens which was maintained in chocolate agar supplemented with 10% (v/v) inactivated horse serum (Biowest, Nuaillé, France) for 48 h. Actinomyces urogenitalis, Aerococcus christensenii, Bifidobacterium bifidum, Campylobacter ureolyticus, F. vaginae strains, Lactobacillus iners, Megasphaera micronuciformis, Mobiluncus curtisii, M. mulieris, Mycoplasma hominis, Peptostreptococcus anaerobius, Porphyromonas asaccharolytica, Prevotella bivia, Propionibacterium acnes, S. sanguinegens and Veillonella parvula were kept at 37°C under anaerobic conditions (AnaeroGen Atmosphere Generation system, Oxoid). Acinetobacter baumannii was grown at 30°C and the remaining species were maintained at 37°C and 10% CO₂.

For PNA-FISH experiments, a bacterial suspension was prepared in phosphate-buffered saline (PBS), in which we first adjusted it to an optical density (OD) at 620 nm ~ 0.1 and then we performed a 2-fold dilution. Afterward, 30 µL of the suspension was spread on epoxy coated microscope glass slides (Thermo Fisher Scientific, Lenexa, KS) and air-dried. Optimization experiments were performed to identify the optimal hybridization conditions which resulted in the best fluorescence signal. Variations in the hybridization temperature, ranging from 50°C-63°C, and time of 60 min and 90 min were tested using the strain F. vaginae ATCC BAA-55. At the optimized conditions, the cells were fixed with 100% (v/v) methanol (Thermo Fisher Scientific) for 15 min, 4% (w/v) paraformaldehyde (Thermo Fisher Scientific) for 10 min, followed by 50% (v/v) ethanol (Thermo Fisher Scientific) for 15 min, and allowed to dry. After, 20 µL of hybridization solution containing 10% (w/v) dextran sulfate (Sigma, Germany), 10 mM NaCl (Sigma), 30% (v/v) formamide (Thermo Fisher Scientific), 0.1% (w/v) sodium pyrophosphate (Thermo Fisher Scientific), 0.2% (w/v) polyvinylpyrrolidone (Sigma), 0.2% (w/v) Ficoll (Sigma), 5 mM disodium EDTA (Panreac, Spain), 0.1% (v/v) Triton X-100 (Thermo Fisher Scientific), 50 mM Tris–HCl (pH 7.5; Thermo Fisher Scientific), and 200 nM of PNA probe were applied to the slides and covered with a coverslip. The slides were placed in a moist and opaque container and incubated at the selected testing time/temperature. After that, the coverslips were removed, and the slides were immersed in the pre-warmed washing solution containing 5 mM Tris-base (Thermo Fisher Scientific), 15 mM NaCl (Sigma), and 1% (v/v) Triton-X (pH 10; Thermo Fisher Scientific) and incubated for 30 min at the same temperature as hybridization. Hybridization was performed at 56°C for 60 min. For the multiplex experiments in biofilms, a PNA probe specific for Gardnerella (Machado et al., 2013) was added to the hybridization solution at the same concentration of 200 nM.

Microscopic visualization was performed using an Olympus BX51 epifluorescence microscope (Olympus, Lisbon, Portugal) equipped with a FITC filter (BP 470-490, FT500, LP 516 sensitive to the Alexa Fluor 488 molecule). Filters that do not detect the probe fluorescence were used as controls to confirm if the cells did not have auto-fluorescence. For every experiment, a negative control was performed with hybridization solution without a probe. The experiments were performed with at least two independent assays.

Inoculums of G. vaginalis ATCC 14018 and F. vaginae ATCC BAA-55 were prepared in New York City III (NYCIII) broth (Rosca et al., 2020a) supplemented with 10% (v/v) inactivated horse serum and incubated for 24 h at 37°C under anaerobic conditions. After 24 h, the bacterial concentration was adjusted to 1 × 10⁷ CFU/mL in NYCIII broth and biofilms were formed on eight-well chamber slides (Thermo Fisher Scientific™ Nunc™ Lab-Tek™, Rochester, NY, USA) by inoculating each of the respective species for single biofilms and both species for dual-species biofilms, for a final volume of 400 µL. Biofilms were incubated at 37°C in anaerobic conditions. After 24 h, the medium was removed, and the biofilm was washed once with NaCl and air-dried. Subsequently, fixation was performed (as described above) and PNA-FISH was conducted at 60°C for 90 min using the F. vaginae and G. vaginalis (Machado et al., 2013) probes. CLSM images were acquired using an Olympus™ Fluo-View FV1000 (Olympus) confocal laser scanning microscope. The experiments were performed in duplicate.

The bacterial population within dual-species biofilms of G. vaginalis and F. vaginae was differentiated by PNA-FISH, as described previously (Castro et al., 2020). Briefly, non-adherent cells were removed by one gentle wash with PBS and, afterward, biofilms were scraped vigorously from the well. Then, 30 μL of each resuspended biofilm was spread on epoxy-coated microscope glass slides (Thermo Fisher Scientific) and the FISH procedure was performed as described above. Microscopic visualization was performed using filters capable of detecting the PNA Gard162 probe (BP 530-550, FT 570, LP 591 sensitive to the Alexa Fluor 594 molecule) and the PNA FvagPNA651probe (BP 470-490, FT500, LP 516 sensitive to the Alexa Fluor 488 molecule). Twenty fields were randomly acquired in each sample. The number of bacteria was counted using ImageJ software (Rasband, 1997), applying automated counts and specific thresholds as indicated in a previous protocol (Labno). Biofilm assays were repeated three times on separate days.

Genomic DNA (gDNA) was extracted from the dual-species biofilms, as well as from pure cultures of the 2 species under study (for the qPCR calibration curves), using the DNeasy Ultraclean microbial kit (Qiagen), following the manufacturer instructions, with minor adaptations. In brief, bacterial samples were centrifuged at 14000 rpm for 5 min, the supernatant carefully removed, and the pellet frozen at -20°C, overnight. This step increased the DNA yield up to 2-fold. The cells were lysed in a BeadBug 6 Microtube Homogenizer (Benchmark Scientific, NJ, USA) using 2×3 cycles of 30 s at 4350 rpm, and samples were kept on ice between the 2 cycles. To assess the efficiency and variability of gDNA extraction between samples, 10 µL of luciferase cDNA, obtained as described before (Magalhães et al., 2019), were added to each sample, before transferring the lysate to the spin column. gDNA was eluted in 50 µL of DNase-free water. To determine the bacterial load, a calibration curve was generated with gDNA isolated from pure bacterial cultures with concentrations ranging from 1 × 10⁹ CFU/mL to 5 × 10⁶ CFU/mL. The primers used to quantify G. vaginalis (Fw CCTCATGCAAAATGTGATGC; Rv CCAAAACAGAAGCACGGAAT; amplifying the locus GAVG_1017, obtained from GenBank: AP012332.1) or F. vaginae (Fw CCTCATGCAAAATGTGATGC; Rv CCAAAACAGAAGCACGGAAT; amplifying the locus I6G91_00565, obtained from GenBank: CP065631.1) were designed with CLC genomics workbench version 21 (QIAGEN). Primer specificity was specifically designed to differentiate these two species in this controlled in vitro study and was first confirmed using Primer-BLAST and then experimentally determined by qPCR. All samples, including the standard curves, were diluted 10× in DNase-free water and then 2 µL of these solutions were mixed with 8 µL of reaction buffer containing 5 µL of Xpert Fast SYBR (Grisp, Porto, Portugal), 1 µL of primers mixture (at 10 µM) and 2 µL of water. All samples were analyzed in triplicates. Non-template controls were performed to evaluate reagent contamination. To assess the efficiency of gDNA extraction, and to calibrate data between qPCR runs, a control was used by adding 2 µL of cDNA luciferase to each qPCR plate. qPCR runs were performed in a CFX96™ (Bio-Rad, CA, USA) with the following cycle parameters: 95°C for 3 min, and 40 cycles of 95°C for 5 s and 60°C for 20 s. Melt analysis was performed to ensure the absence of unspecific products and primer-dimers. PCR amplification efficiency was determined from the slope of a standard curve and efficiencies of 82% for G. vaginalis primers and 79% for F. vaginae were obtained. Bacterial load in each sample was interpolated from the averaged standard curves. This experiment was repeated three times on separate days.

The alignment of the two sets of rRNA sequences gathered for the 16S and 23S rRNA evaluation revealed that the large subunit sequences presented conserved regions of potential interest (i.e. consistent among the F. vaginae sequences and with mismatches in the non-F. vaginae sequences) ( Supplementary Figure 1 ). The candidate target region for probe design was then selected based on the number of target strains, the position of the mismatches in the close related strains used for the alignment, % of GC, melting temperature, and free energy. The probe was named FvagPNA651 considering the target starting position in the 23S rRNA (E. coli numbering).

In silico evaluation of F. vaginae probe performance was done using the TestProbe tool that searches for targets of the probe in an online database of rRNA sequences. A total of 157859 sequences, from the REF sequence collection, large subunit, 23S database (Arb-Silva) were analyzed. From those sequences, only six sequences corresponded to F. vaginae strains. The determination of sensitivity and specificity was done using the equations previously described by Almeida et al. (Almeida et al., 2013). The analysis of the probe resulted in a theoretical value of sensitivity of 100% and specificity of 99.9%; but it should be considered that the sensitivity value was estimated based on the very low available number of F. vaginae target sequences. The values of melting temperature and Gibbs free energy were also calculated in silico, resulting in 61.16°C and -17.71 kcal/mol, respectively.

PNA-FISH procedure can be affected by several factors that will influence the fluorescence signal of the probe. Factors such as pH, dextran sulfate, probe concentration (Rocha et al., 2016), fixation and permeabilization steps (Rocha et al., 2018), as well as the time and temperature of hybridization, are important in the outcome of the FISH method. As such, we first performed pilot experiments testing different times (60 and 90 min) and temperatures (50 to 63°C) of hybridization to obtain the best signal of the probe. The optimization assays ( Supplementary Table 1 ) resulted in an optimal signal-to-noise ratio at a temperature of 56°C and 60 min, which was the selected temperature for the determination of the probe analytical sensitivity and specificity.

F. vaginae probe analytical sensitivity was determined using 24 different isolates of F. vaginae. As described in Table 1 , the results of hybridization were qualitatively classified into four levels: absence (-), poor (+), moderate (++), and good (+++) hybridization. All tested F. vaginae isolates showed hybridization with the probe, although with different efficiencies, resulting in a value of analytical sensitivity of 100%. For the determination of analytical specificity, 40 different bacterial species associated with BV or with vaginal microbiota were used. Under the tested conditions, no hybridization was detected, which result in an analytical specificity of 100% ( Table 2 ). Figure 1 presents examples of the hybridization results for F. vaginae strains with either good or poor hybridization, as well as for some of the most common BV-associated bacteria, as well as L. crispatus which is typically associated with optimal vaginal microbiota. Results from the other tested species are available in Supplementary Figure 2 .

Since BV development has been strongly associated with Gardnerella and F. vaginae biofilms (Swidsinski et al., 2005; Hardy et al., 2016), we also tested if the probe could be used in a multiplex assay, to discriminate species within a biofilm. A mixed biofilm of G. vaginalis and F. vaginae was grown and the presence of both species was assessed using our novel probe and a previously developed Gardnerella probe (Machado et al., 2013). As shown in Figure 2 , both probes were able to detect the respective species in mono- and dual-species biofilms. We further assessed if this method could be used to estimate the abundance of F. vaginae in this complex biofilm structure. PNA-FISH image quantification resulted in F. vaginae abundance of 50.0 ± 7.8% but when the same biofilms were quantified by qPCR, F. vaginae abundance was slightly lower, around 39.1 ± 3.8%.