Antimicrobial and inflammatory properties of South African clinical Lactobacillus isolates and vaginal probiotics

Bacterial vaginosis (BV) causes genital inflammation and increased HIV acquisition risk. The standard-of-care for BV, antibiotic therapy, is associated with high recurrence rates. Probiotics may improve treatment outcomes, although substantial heterogeneity in efficacy has been observed during clinical trials. To evaluate the potential to improve existing probiotics, we compared the inflammatory and antimicrobial (adhesion, H2O2, D-lactate and L-lactate production) characteristics of 23 vaginal Lactobacillus isolates from South African women, commercial vaginal probiotics (L. casei rhamnosus, L. acidophilus) and 4 reference strains. All lactobacilli induced inflammatory cytokine production by genital epithelial cells and produced D-lactate. Of six isolates assessed, five suppressed inflammatory responses to Gardnerella vaginalis. Although the L. acidophilus probiotic was the most adherent, many clinical isolates produced greater amounts of H2O2, D-lactate and L-lactate than the probiotics. The most L-lactate and H2O2 were produced by L. jensenii (adjusted p = 0.0091) and L. mucosae (adjusted p = 0.0308) species, respectively. According to the characteristics evaluated, the top 10 isolates included 4 L. jensenii, 2 L. crispatus, 1 L. mucosae, 1 L. vaginalis and the L. acidophilus probiotic. There is potential to develop an improved vaginal probiotic using clinical Lactobacillus isolates. Inflammatory profiles are critical to evaluate as some isolates induced substantial cytokine production.

Lactobacillus species, which dominate in a healthy FGT, are thought to protect against BV, HIV and other STIs by a number of mechanisms. Lactic acid produced by lactobacilli hinders the growth of potential pathogens and inactivates HIV 12,13 , partly by maintaining the physiological pH of the vagina below 4.5 14 . Lactic acid exists as both D-and L-isomers; while L-lactic acid has been found to inactivate HIV more effectively than D-lactic acid 15 , D-lactic acid is thought to play a more important role in inhibiting bacterial pathogens, including Chlamydia trachomatis 16,17 . Many lactobacilli also produce hydrogen peroxide (H 2 O 2 ), which has a virucidal effect on HIV by inhibiting viral adhesion and replication 18 . However, the role of H 2 O 2 in protection against BV-associated bacteria is controversial, as some studies have reported that physiological concentrations are not microbicidal and that, at microbicidal concentrations, H 2 O 2 inhibits lactobacilli more effectively than pathogenic bacteria 19 . Competitive exclusion is another important protective mechanism utilized by lactobacilli, whereby adherent lactobacilli prevent the adhesion of pathogens to the vaginal epithelium and thus colonization 20 . Genital inflammation caused by BV and STIs increases risk of HIV acquisition 21 . Suppression of inflammatory responses by lactobacilli and lactic acid is another proposed mechanism for reduced susceptibility to HIV in women with Lactobacillus-dominant microbiota 22 .
As the standard-of-care for BV, antibiotic treatment, is associated with high recurrence rates 14,23 , there is an urgent need to develop better treatment strategies, particularly in regions of high HIV prevalence. Several randomized clinical trials have evaluated Lactobacillus-containing probiotics for BV treatment, alone or as adjunctive therapy with antibiotics 24,25 . Although some trials have demonstrated improved BV outcomes with probiotics, the results have been heterogeneous, with some studies finding no benefit 24,25 . One of the limitations of many of these probiotic formulations is that they include bacterial species that are not adapted for survival in the FGT and are not usually found in women with healthy vaginal microbiota 26 . It is also possible that probiotics containing Lactobacillus isolates from one population may have reduced efficacy when used in another population, as major geographical and ethnic differences have been observed in the vaginal microbiota and host factors that may influence bacterial colonization 27,28 . Therefore, using vagina-specific Lactobacillus species with effective antimicrobial properties that have been isolated from within the population of intended use may improve BV treatment outcomes. The aims of this study were to compare the antimicrobial and inflammatory characteristics of existing vaginal probiotics on the South African market to those of clinical Lactobacillus isolates from the FGTs of South African women.
Hydrogen peroxide production. L. mucosae isolates produced the most H 2 O 2 relative to the other strains collectively (p = 0.0044; adjusted p = 0.0308), after a 3 h incubation period under aerobic conditions, followed by L. jensenii isolates (p = 0.0061; adjusted p = 0.0427; data not shown). The other species evaluated produced comparably lower amounts of H 2 O 2 ( Fig. 2A). Only one L. crispatus isolate, none of the L. vaginalis isolates and none of the probiotic Lactobacillus isolates produced significant amounts of H 2 O 2 (Fig. 2B). The rates of H 2 O 2 production were similar for all producing isolates, with production being evident within 2 h (Fig. 2B). The level of H 2 O 2 production was not associated with the growth rates of the lactobacilli (rho = −0.1136, p = 0.5501; data not shown).
D-and L-lactate production and culture acidification. When the lactobacilli were cultured in MRS, it was found that the probiotics lowered the culture pH levels most, followed by the L. crispatus isolates (Fig. 2C,D). In contrast, L. vaginalis and L. mucosae culture pH levels were the highest. Most Lactobacillus isolates produced large amounts of D-lactate, with no significant differences between species after multiple comparisons adjustment (Fig. 2E,F). L-lactate production was however highly varied between isolates (Fig. 2G,H), with L. jensenii isolates producing significantly greater amounts of L-lactate than the other isolates collectively (p = 0.0013, adjusted p = 0.0091; data not shown), followed by L. gasseri and L. vaginalis (Fig. 2G). While all the probiotic isolates produced large amounts of D-lactate, only the probiotic capsule L. acidophilus isolate produced detectable levels of L-lactate (Fig. 2F,H). The concentration of total lactic acid, calculated using the Henderson-Hasselbalch equation 31 , correlated with culture pH (rho = 0.8626, p < 0.0001; data not shown). Additionally, there was a trend towards a significant correlation between D-lactate and L-lactate production (Spearman rho = 0.3272, p = 0.0776; data not shown). Although there was a trend towards an inverse correlation between culture pH and bacterial growth rates (rho = −0.3079, p = 0.0978), neither D-lactate (rho = 0.07787, p = 0.6825) nor L-lactate (rho = 0.2520, p = 0.1791) correlated with growth rates (data not shown).

Inflammatory cytokine responses. Inflammatory cytokines produced by CaSki cells in response to
4.18 × 10 6 colony forming units (CFU) of each Lactobacillus isolate were assessed using Luminex. While several cytokines [including interleukin (IL)-1α, IL-1β, IL-6, IL-8, IFN-γ-inducible protein (IP)-10, macrophage inflammatory protein (MIP)-1α, MIP-1β, MIP-3α, regulated on activation, normal T cell expressed and secreted (RANTES)] were induced, others [including tumor necrosis factor (TNF)-α, interferon (IFN)-γ, IL-17A and IL-10] were not produced (below the minimum levels of detection for all samples; not shown). The relative cytokine production elicited by each of the isolates is shown as a heatmap (Fig. 3A). While most lactobacilli induced very little cytokine production by CaSki cells, some of the isolates were substantially more inflammatory than others, including two L. crispatus isolates, two L. jensenii isolates and the L. acidophilus vaginal tablet isolate which clustered separately from the other isolates assessed (Fig. 3A). Factor analysis was used to group all of the inflammatory cytokines together onto one factor and generate overall inflammatory scores for each isolate (Fig. 3B,C). L. acidophilus was most inflammatory, while L. gasseri isolates were the least inflammatory according to inflammatory factor scores (Fig. 3B), although these differences were not statistically significant. Of the individual isolates assessed, LJ1 was the most inflammatory, while LG1 was the least inflammatory (Fig. 3C).
To evaluate the impact of D-and L-lactate concentrations on inflammatory responses, these metabolites were measured in lactobacilli-CaSki co-cultures following a 24 hour incubation at 37 °C under aerobic conditions (Fig. 4). It was found that neither D-lactate nor L-lactate correlated significantly with inflammatory cytokine production by CaSki cells (data not shown).

Inhibition of G. vaginalis growth.
The six Lactobacillus isolates above were selected for G. vaginalis inhibition assays, including 2 L. jensenii (LJ2 and 5), 1 L. crispatus (LC2), 1 L. mucosae (LM2), 1 L. vaginalis (LV6) and 1 L. gasseri (LG1) strains. It was found that 4/6 isolates significantly inhibited G. vaginalis growth (Fig. 6A) and viability (Fig. 6B). The culture pH levels of the isolates that inhibited G. vaginalis growth and viability, LJ2, LJ5, LM2 and LG1, were significantly lower than those of the isolates that were not inhibitory, LC2 and LV6 [mean: 4.551 (range: 4.398-4.826) versus 5.478 (range: 5.283-5.673), respectively; p = 0.0082]. D-lactate, L-lactate and lactic acid levels did not differ significantly between the groups (p = 0.0703, p = 0.2711 and p = 0.0690, respectively), although the sample size was small and the differences in D-lactate and lactic acid concentrations approached significance. Furthermore, reduction of H 2 O 2 and degradation of bacteriocins using proteolytic enzymes did not influence the inhibitory activities of LJ2, LM2 and LG1 (Fig. 6C,D), suggesting that culture medium acidification was the primary mechanism for G. vaginalis inhibition in these assays. overall probiotic-relevant performance of isolates. A scoring system was devised to compare and rank the isolates based on all of the characteristics investigated. Each isolate was given scores out of a maximum score of three per characteristic, according to (1) growth rates; (2) culture pH acidification; the levels of (3) H 2 O 2 , (4) L-and (5) D-lactate produced, (6) ability to adhere to epithelial cells and (7) induction of inflammatory responses by CaSki cells (Table 2). Relative scores per category were assigned as follows: <25 th percentile (score = 0); 25 th −50 th percentile (score = 1); 50 th -75 th percentile score = 2); ≥75 th (score = 3). Lower pH levels and lower levels of inflammation were considered advantageous, so isolates were scored as follows: <25 th percentile (score = 3); 25 th -50 th percentile (score = 2); 50 th -75 th percentile score = 1); ≥75 th (score = 0). The L. jensenii strain (LJ5) ranked highest with a score of 18/21, higher than the probiotic isolates (L. acidophilus and L. casei rhamnosus). Interestingly, the top 10 isolates included 4 L. jensenii, 2 L. crispatus, 1 L. mucosae, and 1 L. vaginalis strain and the L. acidophilus probiotic isolates. The commercial vaginal probiotics were ranked 6 th (LA capsule; 13/21), 10 th (LA tablet; 12/21) and 20 th (LCR; 9/21). The performance of the LA strains was inconsistent between tablets and capsules, while both grew rapidly and lowered the culture pH levels effectively, both induced inflammatory responses and the capsule LA isolate was poorly adherent to epithelial cells, while the tablet LA isolate produced very little H 2 O 2 and L-lactate. The LCR probiotic strain fared worse with poor adhesion to epithelial cells and low H 2 O 2 , L-and D-lactate production in comparison to clinical isolates. Similar rankings were obtained if H 2 O 2 was omitted, with the top 10 containing 3 L. jensenii, 2 L. crispatus, 1 L. mucosae, 2 L. vaginalis, and the 2 L. acidophilus probiotic isolates.

Discussion
BV, an imbalance in vaginal microbiota that predisposes women to HIV infection and reproductive complications, remains prevalent, in part because standard-of-care antibiotic therapy is largely ineffective 14,32 . Adjunctive probiotic treatment for BV may promote vaginal recolonization with healthy lactobacilli, however BV treatment outcomes have been heterogeneous in probiotic clinical trials. Here, we show that the performance of the only vaginal probiotics available on the South African market were highly varied based on the criteria evaluated in this study, providing a possible explanation for the heterogeneous efficacy of probiotic treatment in BV. Neither L. acidophilus nor L. casei rhamnosus species predominate in the FGTs of women with healthy microbiota 27 and were not isolated from the cohort of women included in this study, suggesting that these species might not be ideal for treatment of BV in South African women. In this study, we evaluated Lactobacillus species that have been associated with vaginal health, including L. crispatus, L. gasseri, L. jensenii, L. mucosae and L. vaginalis. L. iners was not included as this species is associated with increased risk of conversion from a healthy to a dysbiotic Mann-Whitney test was used for comparisons between species and p-values < 0.05 after adjustment for multiple comparisons were considered statistically significant. LA: Lactobacillus acidophilus probiotic (n = 2); LC: Lactobacillus crispatus (n = 9); LCR: Lactobacillus casei rhamnosus probiotic (n = 1); LG: Lactobacillus gasseri (n = 2); LM: Lactobacillus mucosae (n = 4); LV: Lactobacillus vaginalis (n = 6); LJ: Lactobacillus jensenii (n = 6). *ATCC reference strains.   vaginal microbiome 33 , and acquisition of STIs 34 . We found that several of the clinical isolates performed better than the probiotic strains, demonstrating the potential to improve existing probiotic formulations and treatment outcomes using novel clinical isolates. We also found that, although some Lactobacillus isolates induced inflammatory responses when cultured with ectocervical epithelial cells in isolation, when we pre-incubated the cells with a subset of six isolates prior to addition of G. vaginalis, 5/6 of the lactobacilli significantly reduced pro-inflammatory cytokine responses to G. vaginalis. Several antimicrobial and other characteristics that are thought to be important for effective probiotic activity were evaluated for each Lactobacillus isolate 16,18,20 . Adhesion, thought to be mediated by specialized pili 35 , is an essential probiotic property as it enables the bacteria to persist in the FGT 36 . In this study, there were no significant differences in the level of adhesion between species. Larger bacterial size has been linked to superior adhesion capabilities 30 , however the relative length of bacteria in this study did not correlate significantly with adhesion. Moreover, although it has been reported that rapid bacterial multiplication leads to a high adhesion capacity 30 , in the present study, bacterial growth rates were not significantly associated with adhesion. H 2 O 2 production differed between species, with L. mucosae isolates producing the most H 2 O 2 , followed by L. jensenii isolates, while the others produced minimal H 2 O 2 . The importance of H 2 O 2 in protection against pathogens in the FGT is controversial. Earlier clinical studies found that women with H 2 O 2 -producing lactobacilli were at reduced risk of dysbiosis 8,18,37 . However, it has since been found that, under hypoxic conditions like those found in the vagina, lactobacilli produce very little H 2 O 2 38 and H 2 O 2 may not be microbicidal at physiological concentrations 19,39 . Furthermore, at microbicidal concentrations H 2 O 2 inactivated lactobacilli more effectively than BV-associated bacteria in vitro 19,39 . It is important to note that, using an in vitro model to evaluate the characteristics of Lactobacillus isolates has limitations as this environment does not perfectly mimic in vivo conditions.
In contrast, lactic acid at physiological concentrations was shown to be microbicidal against pathogenic BV-associated bacteria (including G. vaginalis, P. bivia and P. corporis), but not vaginal lactobacilli 16,17,19,39 . Physiological concentrations of lactic acid have also been shown to have broad-spectrum virucidal activity against HIV that is dramatically more rapid and potent than media acidified to the same pH with HCl or acetic acid 15,40 . Interestingly, all isolates in this study produced D-lactate with no significant differences between species. However, L-lactate production was highly varied, with L. jensenii isolates producing the most L-lactate in both MRS under anaerobic conditions and in DMEM under aerobic conditions overall. The relative importance of L-lactic acid versus D-lactic acid is also controversial. L-lactic acid was found to be 17-fold more potent than D-lactic acid in inactivating HIV BaL in vitro 15,40 . However, D-lactic acid was found to have a greater inhibitory effect on C. trachomatis infectivity than L-lactic acid, which was partly 41 or entirely 39 pH-dependant.
Similar to the finding that lactic acid concentrations correlate inversely with pH in the FGTs of women with Lactobacillus-dominated microbiota 31 , we found that lactic acid concentrations correlated significantly with culture pH. As found in the present study, a previous study showed that the vaginal acidity achieved by L. crispatus was the highest compared to L. jensenii and L. gasseri 27 . Interestingly, in the present study, the L. casei rhamnosus probiotic isolate was associated with the lowest culture pH, followed by L. acidophilus probiotic.
We measured inflammatory responses to lactobacilli, as genital inflammation has been found to increase risk of HIV acquisition in women 21 and it is thus essential that probiotics induce little or no inflammation. Most of the isolates in isolation induced low levels of cytokine production by CaSki cells following a 24 h incubation period. However we also found that most of the lactobacilli evaluated were able to significantly suppress inflammatory cytokine production by CaSki cells in response to the BV-associated pathogen, G. vaginalis. In vivo, in the same cohort from whom the lactobacilli were isolated, we found that endogenous L. reuteri, L. gasseri, L. crispatus and L. jensenii were all significantly associated with low inflammation in the FGT 42 . Other studies have reported that lactobacilli and lactic acid suppress inflammatory responses to pathogens and pattern recognition receptor ligands in vitro 22,26 . It is interesting that, in this study, some of the lactobacilli in isolation were able to induce increases in pro-inflammatory cytokine production and 1/6 of the lactobacilli included in the G. vaginalis co-culture experiment had an additive effect on cytokine induction. Additionally, anti-inflammatory IL-10 production was not detected in these cultures and IL-1RA production was suppressed by lactobacilli. Another study showed that, while L. rhamnosus and L. reuteri suppressed the expression of certain inflammatory mediators by vaginal epithelial cells in response to Candida albicans [including nuclear factor-kappa B inhibitor kinase alpha, toll-like receptor (TLR)-2, TLR-6, IL-8, and TNF-α], the lactobacilli also induced expression of pro-inflammatory cytokines IL-1α and IL-1β 43 . The authors suggested that the lactobacilli may suppress inflammatory responses induced by the NF-κB signal transduction pathway, but induce other inflammatory responses via an alternate signal pathway, such as the mitogen activated protein kinase and activator protein-1 (MAPK/AP-1) signal transduction pathway 43 . Similarly Rose et al. (2012) found that lactobacilli induced non-significant increases in the production of some inflammatory cytokines by vaginal epithelial cells cultured in transwell culture systems 44 . Proteins present in the peptidoglycan layer of the Lactobacillus cell wall and lipoteichoic acids present in the cell membrane may have inflammatory properties and the inflammatory nature of particular Lactobacillus strains has been found to be partly dependent on the structure of the peptidoglycan and the presence of exopolysaccharides which prevent the interaction of these TLR agonists with pattern recognition receptors 45 . Therefore it is not surprising that some isolates are capable of inducing inflammatory responses. Lactic acid itself was found to have direct pro-inflammatory effects on immune and vaginal epithelial cells in some studies 46,47 , but anti-inflammatory effects in others 22 . In this study, we found that neither D-lactate nor L-lactate production correlated with cytokine LA: Lactobacillus acidophilus probiotic (n = 2); LC: Lactobacillus crispatus (n = 9); LCR: Lactobacillus casei rhamnosus probiotic (n = 1); LG: Lactobacillus gasseri (n = 2); LM: Lactobacillus mucosae (n = 4); LV: Lactobacillus vaginalis (n = 6); LJ: Lactobacillus jensenii (n = 6). *ATCC reference strains.   properties of these cells may differ from healthy cells and it will thus be important to confirm that probiotic candidates do not induce inflammatory responses using other models. Interestingly, the characteristics of the Lactobacillus isolates from South African women investigated in this study differed in some respects to the characteristics of isolates from other geographical regions and women of different ethnicities 27,28 . While studies in other regions, including America and Spain 37,48 , have found that L. crispatus strains produce high levels of H 2 O 2 and lactic acid, this study found that these South African L. crispatus isolates generally produced relatively little of these antimicrobials. Similarly, major geographical and ethnic differences in the vaginal microbiome have been noted. African women were found to have a low abundance of Lactobacillus that was not associated with sexual behavior, contraceptive usage, or demographic characteristics 4 .   According to the characteristics that were assessed in this study that may influence Lactobacillus antimicrobial activity (including growth rates, culture pH, adhesion, H 2 O 2 , lactic acid and inflammatory responses), several South African clinical isolates performed better than the commercial probiotic Lactobacillus isolates evaluated. Of the top ten Lactobacillus isolates, L. jensenii and L. crispatus were well represented, while L. mucosae and L. vaginalis strains were mostly absent. Five clinical isolates performed better than the probiotic L. acidophilus strain, suggesting that there is indeed potential to improve existing probiotics that may, in turn, improve BV treatment outcomes.   dilutions were plated on MRS agar and colonies were counted following anaerobic incubation at 37 °C for 48 h. CFU/ml was calculated and expressed as a percentage of the baseline CFU/ml added to each monolayer. This experiment was repeated three times. For visual confirmation of bacterial adhesion to CaSki cells, bacteria were added to cell monolayers in Nunc TM Lab-Tek TM II Chamber Slides (Thermo Fisher Scientific Inc., USA) and incubated for 2 h at 37 °C, 5% CO 2 . Cell culture medium was removed and wells were washed 3 times with PBS. Chambers were removed before each slide was heat-fixed and Gram-stained. Representative images were captured (Leica ICC50 HD, Leica Microsystems, Germany). Hydrogen peroxide production. H 2 O 2 production was evaluated in aerobic cultures in order to assess the maximal H 2 O 2 production capacity of the isolates. Cultures were incubated aerobically with agitation at 170 rpm for 3 h at 37 °C. Aliquots were collected hourly for 3 h, centrifuged and the supernatants stored at −80 °C. pH levels of the cultures were measured using a pH meter (Jenway Bench pH Meter 2510, Bibby Scientific, UK). Measurement of Lactobacillus H 2 O 2 concentrations was performed using a tetramethylbenzidine assay in duplicate as described previously 51 . Inflammatory cytokine responses to lactobacilli. CaSki monolayers were grown to near confluence in 24 well plates. Culture medium was removed and 4.18 × 10 6 CFU of each bacterial species or OD 600 of 0.1 ± 0.01 adjusted bacteria were added to the cells in antibiotic-free cell culture medium (DMEM with L-Glutamine, 10% fetal calf serum) and incubated aerobically at 37 °C for 24 h under 5% CO 2 . Culture supernatants were removed, centrifuged at 6000xg and stored at −80 °C. CaSki cell viability in these co-cultures was confirmed using an MTT assay (Roche Diagnostics, Germany). In order to account for the impact of cleavage of the yellow tetrazolium salt MTT to purple formazan crystals by metabolically active lactobacilli, lactobacilli-only cultures incubated in antibiotic-free cell culture medium under aerobic conditions were included as controls. In all cases, the metabolic activity of the CaSki cells in co-culture with the lactobacilli was either comparable to or higher than the CaSki  Supernatants were collected and processed as above and a Magnetic Luminex Screening Assay kit (R&D Systems, Minneapolis, USA) was then used to measure IL-1α, IL-6, IL-8, IP-10 and IL-1RA concentrations. Data was collected using a Bio-Plex TM Suspension Array Reader and a 5 parameter logistic regression was used to calculate cytokine concentrations from the standard curves using BIO-plex manager software (version 4; Bio-Rad Laboratories Inc ® , USA). Cytokine concentrations below the detectable limit were assigned the value of half the lowest recorded concentration of that cytokine.

Study participants.
D-and L-lactic acid production. Following a 24 h incubation period at 37 °C under anaerobic conditions, 4.18 × 10 6 CFU of each isolate was added to 15 ml MRS broth and incubated for an additional 24 h at 37 °C under anaerobic conditions. Supernatants were collected and D-and L-lactate concentrations were measured in duplicate using D-Lactate Colorimetric and Lactate Assay kits according to the manufacturer's instructions (Sigma-Aldrich, USA). Both D-and L-lactate concentrations were also measured in lactobacilli-CaSki cell co-cultures following aerobic incubation at 37 °C for 24 h under 5% CO 2 using D-Lactate Colorimetric and Lactate Assay kits according to the manufacturer's instructions (Sigma-Aldrich, USA) . Statistical Analysis. GraphPad Prism 5 ® (GraphPad Software, USA), STATA version 11.0 (StataCorp, USA), and R were used for statistical analyses. Mann Whitney U and Student's t-tests were used for unpaired comparisons of non-parametric and parametric data, respectively; Spearman Rank test was used for non-parametric correlations. A false-discovery rate (FDR) step-down procedure was used to adjust p-values for multiple comparisons and adjusted p-values < 0.05 were considered statistically significant. Unsupervised hierarchical clustering was used to evaluate the relationship between lactobacilli and cytokine profiles. To compare overall inflammatory cytokine responses, confirmatory factor analysis was used to generate inflammatory factor scores, which are linear combinations of the concentrations of each inflammatory cytokine (IL-1α, IL-1β, IL-6, IL-8, IP-10, MIP-1α, MIP-1β, MIP-3α, RANTES) in the factor, weighted according to their factor loadings.

G. vaginalis inhibition assay. Six
ethics approval and consent to participate. The parent study was approved by the University of Cape Town (UCT) human research ethics committee (UCT HREC: 267/2013). The microbiological sub-study was approved by the UCT human research ethics committee (UCT HREC: 551/2016). Women older than 18 years provided written informed consent, while those who were 16-17 years old provided assent and written informed consent was obtained from their parents or legal guardians.

Data Availability
The datasets used and/or analysed during this study are available from the corresponding author on reasonable request.