We searched for data on mammalian vaginal lactobacilli and pH using appropriate search terms in Web of Science, Google Scholar, PubMed, and in the references of other publications. For information on lactobacilli prevalence and relative abundance, we focused on culture-independent studies (i.e., lactobacilli identification based on Illumina or 454 sequencing) because cultivation-based methods can under-represent or overlook some species (Zhou et al., 2004; Lamont et al., 2011). For data on vaginal pH, all pH measurements were collected using standard methods with pH meters or paper. Of particular note, human pH measurements from O'Hanlon et al. (2013) were not included because they collected data under hypoxic conditions and therefore were not comparable to measurements on non-human mammals. In addition, because the human vaginal microbiome changes in response to puberty and menopause and between reproductive states (e.g., Cauci et al., 2002; Thoma et al., 2011a; MacIntyre et al., 2015), we confined our data set to pH values collected during the ovarian cycle (i.e., follicular and luteal phases, anestrus) of sexually mature subjects. Furthermore, humans or non-human animals who had recently mated were not included as semen is alkaline and can temporarily neutralize vaginal tract pH (Tevi-Bénissan et al., 1997). Finally, because testing the reproductive phase hypothesis required data on how natural fluctuations in estrogen impact vaginal pH, we excluded subjects who were sterilized or supplemented with exogenous hormones. However, we did include studies of humans where women were taking hormonal birth control because past work has found no relationship between vaginal pH and birth control (Drake et al., 1980; Wagner and Ottesen, 1982). In total, we found 50 studies with data on vaginal pH and/or lactobacilli relative abundance across 26 mammalian species. Specifically, we found information on vaginal pH for 21 different species of non-human mammals from 44 studies (Figure 1; Table S1) and data on relative abundance of lactobacilli for 14 mammals, 8 of which also had information on vaginal pH (Table 2).

For each mammalian species, we calculated the average vaginal pH across all available studies (Table S1). Because some studies measured vaginal pH at only a few cycle phases, the average pH of a species reflects the available data and not necessarily all cycle phases. If a study reported a pH range instead of a single value, we calculated the midrange (i.e., added the minimum and maximum values and divided by 2). For studies that represented their data graphically, but did not give numerical values, we used WebPlotDigitizer (Rohatgi, 2015) to extract pH values from the relevant plots. To test the reproductive phase hypothesis, we used vaginal pH values from the phases of the ovarian cycle when estrogen was highest and lowest (Table S1). For the majority of mammalian species, peak estrogen occurs during late proestrus or estrus, and the lowest level of estrogen occurs during the luteal phase or menses (Figure S1). However, for several species, including cows and dogs, peak estrogen occurs at both the end of proestrus and beginning of estrus. In these instances, we used the pH value from estrus to represent the high estrogen time point because proestrus can span multiple days and can encompass considerable variation in estrogen levels (Figure S1).

To test the disease risk and obstetric protection hypotheses, we required information on interspecies risk associated with STDs and reproduction. We calculated proxies of both STD and obstetric risk using species-specific trait data compiled from primary literature, edited books, and the life history database, PanTHERIA (Jones et al., 2009; Table S2). Whenever possible, we used data from wild, feral, or free-ranging populations over data from captive or domesticated individuals. In 13 instances, we could not find a particular life history trait for a species. In these cases, we used data from the closest related species with available data. For example, there was no information on the age of sexual maturity or life expectancy for the red-bellied tamarin (Sanginus labiatus), so we used data from the closely related cotton-top tamarin (S. oedipus). All proxies and their descriptions are listed in Table 1. Specifically, to test the disease risk hypothesis, we used five proxies: testes mass relative to body size, baseline white blood cell (WBC) count, annual sexual receptivity, total lifetime reproductive events, and intromission pattern (Table 1). For the obstetric protection hypothesis, we used gestation length, relative neonatal mass, and relative maternal pelvic area (Table 1).

All statistical tests were implemented in the R statistical environment (R Core Team, 2015). Before testing our primary hypotheses, we first tested for a phylogenetic signature on vaginal pH such that mammals with more similar evolutionary histories had more similar vaginal pH values. We obtained divergence times between the species in our data set from Bininda-Emonds et al. (2007), with an updated primate phylogeny by Perelman et al. (2011) and Papio spp. values from Purvis (1995). We constructed a distance matrix and phylogenetic tree using the R package, ape (Paradis et al., 2004), along with a distance matrix of vaginal pH values between species. We correlated phylogenetic distance with differences in vaginal pH using a Mantel test, which calculates a P-value based on comparison of the observed Pearson correlation coefficient to the distribution of coefficients from 10,000 permutations.

To test the reproductive phase hypothesis, we compared the vaginal pH during high- and low-estrogen phases of the female ovarian cycle using a paired t-test in R. To test the disease risk and obstetric protection hypotheses, we correlated vaginal pH and lactobacilli relative abundance with each proxy of STD or obstetric risk using linear regressions and ANOVAs. To calculate the minimum sample size required to detect a significant relationship, we used the R package, pwr (Champely, 2015), with a power value of 0.80. We further constructed multivariate linear regression models to test whether any combination of proxies predicted vaginal pH. Predictor variables in each model included relative testes mass, baseline WBC count, annual sexual receptivity, total lifetime reproductive events, intromission pattern, gestation length, and relative neonatal mass. Model selection was performed using stepwise backward regression with the stepAIC function from the R package, MASS (Venables and Ripley, 2002) and a 5% significance level was used as a threshold for inclusion in the final model.

Of the 26 mammalian species for which we found data on vaginal pH and/or lactobacilli relative abundance, most were captive primates used for medical studies, domesticated ungulates (e.g., horse, cow, pig), and common laboratory rodents (e.g., mouse, rat, and guinea pig). Of note, only one study measured vaginal pH in a wild population (yellow baboons; Miller et al., in review).

Together, our results confirm that humans are distinct from other mammals in dominance of lactobacilli and the acidity of their vaginal tract. For instance, across 10 studies of healthy human women, median vaginal pH was 4.5 (range = 4.0–4.9), while median vaginal pH across non-human mammals was 6.8, with no species falling in the range of healthy human pH, and only two species with a pH below 6.0 (Figure 1). Furthermore, while 13 of 14 mammals had detectable levels of Lactobacillus spp., the average relative abundance of lactobacilli was only 1.1% (±0.39% SEM) in non-human mammals compared to 69.6% in human women (±0.046% SEM) (Figure 2). This disparity is unlikely to be due to differences in the species of Lactobacillus in non-human mammals as compared to humans as the same four species that dominate the vaginal tract of women (L. cripsatus, L. gasseri, L. iners, and L. jensenii), were also frequently found in other mammals, albeit in low relative abundance. However, non-human mammals also harbored other lactobacilli generally not found in humans, including L. animalis, L. fornicalis, L. amylovorus, and L. johnsonii. Other members of the phylum Firmicutes were also common in non-human mammals, particularly species of two lactic acid-producing genera Aerococcus and Facklamia. Among primates, there was also high relative abundance of multiple genera linked to bacterial vaginosis (BV) in women, including Gardnerella, Sneathia, and Prevotella (Onderdonk et al., 2016).

Importantly, as has been reported previously, human women with BV had significantly higher vaginal pH than healthy women (N = 6 studies; two-sample t-test: t_(8.14) = 3.65, P = 0.0063; Figure 1). Women with BV also exhibited compositional similarities to healthy baboons and macaques, including moderately high relative abundances of Gardnerella, Mobiluncus, Sneathia, and Prevotella (e.g., Miller et al., in review; Spear et al., 2010, 2012; Uchihashi et al., 2015; Onderdonk et al., 2016). However, of the five macaque and baboon species in our data set, all but the olive baboon had significantly higher vaginal pH than BV women (ANOVA: F_{(5, 14)} = 12.84, P = 8.12e-05; Figure 1), which suggests that similar microbial composition may not always translate into similar ecological functions and consequences (e.g., Mirmonsef et al., 2012).

Before testing our main hypotheses, we first investigated whether shared evolutionary history among mammals explained variation in vaginal pH among mammal species. Figure 1 indicates that primates exhibited the widest variation in vaginal pH, with species-specific averages in non-human primates ranging from 5.4 to 7.8, and chimpanzees and olive baboons exhibiting pH values most similar to humans. Indeed, more variation in vaginal pH existed among primate species than among all other mammal lineages, which, despite encompassing many more years (and units of branch length) of evolutionary history, had vaginal pH values that ranged from just 6.5–7.4 (Figure 1). However, evolutionary distance between mammalian species did not predict similarity in vaginal pH, either among humans and non-human primates (Mantel test: N = 10, Mantel statistic r = −0.01, P = 0.47), or among all mammals (N = 22, Mantel statistic r = −0.2, P = 0.93). Hence, shared evolutionary history is unlikely to play a dominant role in shaping interspecies variation in vaginal pH.

The reproductive phase hypothesis proposes that differences between human and non-human primate vaginal microbiota are due to between-species variation in reproductive physiology—especially the continuous nature of human ovarian cycling. This hypothesis predicts that non-human mammals will exhibit lactobacilli dominance and a vaginal pH similar to humans during the period in their ovarian cycle when they experience highest estrogen levels. To test this idea, we compared vaginal pH during high- and low-estrogen phases of the female ovarian cycle in 10 non-human mammals where we had enough information on pH throughout the ovarian cycle. Notably, some of the mammal species we considered exhibit continuous, year-round reproductive cycling (i.e., spider monkey, cow, rat, pig-tailed macaque, common brushtail possum, and olive and yellow baboons), allowing us to make a strong comparison with humans. We found that, similar to humans, non-human mammals exhibit the lowest vaginal pH during high-estrogen phases (paired t-test: t_(9.00) = −3.16, P = 0.012; Figure 3). However, even during peak estrogen, the pH of non-human mammals never declined to a level comparable to humans (Figure 3). Data on the relative abundance of Lactobacillus spp. as a function of ovarian cycle phase are rare in non-human mammals, but one study on wild baboons found that the order Lactobacillales, which comprises all lactic acid-producing bacteria, including lactobacilli, was more abundant in individuals experiencing ovulation compared to other ovarian cycle phases (Miller et al., in review). However, the mean relative abundance of Lactobacillus spp. in these ovulating baboons was only 0.0058%, which is considerably lower than the level typical in healthy human women (Miller et al., in review).

Together, these results suggest that estrogen plays a similar role in shaping vaginal microbial composition and pH in humans and in other mammals. Specifically, rising estrogen levels increase available glycogen in the vaginal epithelium, which in turn, provides an energy source for lactobacilli to produce lactic acid. Indeed, like humans, many non-human mammals exhibit a thickening of the vaginal epithelium and increasing glycogen content in response to rising estrogen levels during the ovarian cycle (e.g., Gregoire and Guinness, 1968; Gregoire and Parakkal, 1972; Williams et al., 1992; Nyachieo et al., 2009). However, direct correlations between estrogen, glycogen, lactic acid, lactobacilli, and vaginal pH have yet to be explored in non-human mammals to the extent they have been in humans (e.g., Boskey et al., 1999, 2001; Mirmonsef et al., 2014; but see Mirmonsef et al., 2012). In summary, in terms of the reproductive phase hypothesis, we find that differences in reproductive cycling between humans and other mammals are not sufficient to explain why the human vaginal pH and lactobacilli abundance are such outliers relative to other mammals. Hence the uniqueness of the human vaginal microbiome is unlikely to be a consequence of sampling non-human mammals during the incorrect reproductive state.

The disease risk hypothesis proposes that high risk of STD exposure in humans, compared to other mammals, has selected for a protective, lactobacilli-dominated community. This hypothesis predicts that, in general, mammals, such as humans, with greater STD risk will have higher abundance of lactobacilli and lower vaginal pH compared to species with minimal STD risk. To test this hypothesis, we correlated the vaginal pH and lactobacilli relative abundance of mammals with species-specific proxies of STD risk (Table 1). Overall, we found no significant relationships between vaginal pH and any of the five STD risk proxies, including relative testes mass (linear regression: N = 21, F_{(1, 19)} = 0.16, P = 0.70; Figure 4A), baseline WBC count (N = 21, F_{(1, 19)} = 0.90, P = 0.36; Figure 4B), annual sexual receptivity (N = 21, F_{(1, 19)} = 0.00030, P = 0.99; Figure 4C), number of lifetime reproductive events (N = 21, F_{(1, 19)} = 2.56, P = 0.13; Figure 4D), and intromission pattern (ANOVA: N = 21, F_{(2, 18)} = 1.07, P = 0.36; Figure 4E). Additionally, there was no significant relationship between lactobacilli relative abundance and any of the same STD risk proxies (N = 14; relative testes mass: F_{(1, 12)} = 0.0050, P = 0.95; baseline WBC count: F_{(1, 12)}= 0.072, P = 0.79; annual sexual receptivity: F_{(1, 12)} = 0.79, P = 0.39; number of lifetime reproductive events: F_{(1, 12)} = 2.66, P = 0.13; copulation pattern: F_{(2, 11)} = 2.52 P = 0.13).

Parallel to the disease risk hypothesis, the obstetric protection hypothesis proposes that the unique nature of the human vaginal microbiome is due to risk of infection during human pregnancy and birth. By extension, across mammals, species experiencing higher risk during gestation and parturition should have higher abundance of lactobacilli and lower vaginal pH. We correlated vaginal pH and lactobacilli relative abundance with three common proxies of obstetric risk across mammals (Table 1). Again, there was no relationship between vaginal pH and any obstetric risk proxy, including gestation length (linear regression: N = 21, F_{(1, 19)} = 0.18, P = 0.67; Figure 4F), relative neonatal mass (N = 21, F_{(1, 19)} = 0.54, P = 0.47; Figure 4G), and relative maternal pelvic area (N = 11, F_{(1, 6)} = 0.42, P = 0.54; Figure 4H). There was also no relationship between lactobacilli relative abundance and any proxy (gestation length: N = 14, F_{(1, 12)} = 0.69, P = 0.42; relative neonatal mass: N = 14, F_{(1, 12)} = 0.20, P = 0.66; relative maternal pelvic area: N = 5, F_{(1, 3)} = 0.0054, P = 0.95).

Finally, because disease and obstetric risk could be acting as simultaneous selection pressures on the vaginal microbiome, we also tested both hypotheses together using combinations of STD and obstetric risk proxies as factors in multivariate linear regressions predicting vaginal pH and lactobacilli relative abundance. Consistent with univariate results, no combination of proxies was significantly correlated with either vaginal pH or lactobacilli relative abundance (P-value range = 0.89–0.13). While the sample sizes for these correlations were small, a power analysis suggests that we still would need a sample size at least 2 times our current sample size and often more than 10 times the current sample size to detect significant effects. Hence, STD and obstetric risk are, at best, minor forces shaping mammalian vaginal pH and the relative abundance of lactobacilli.

Our results suggest that, contrary to the assumptions underlying the disease risk and obstetric protection hypotheses, humans do not have considerably higher disease or obstetric risk compared to other mammals (see asterisks on Figure 4). For example, many mammals have STDs and may experience high STD risk due to sexually promiscuous behavior (Smith and Dobson, 1992; Lockhart et al., 1996; Altizer et al., 2003). Additionally, a number of mammals have longer gestation duration than humans, and may experience extreme complications associated with parturition (Aksel and Abee, 1983; Frank and Glickman, 1994; Rosenberg and Trevathan, 2002). For instance, squirrel monkeys give birth to exceptionally large neonates for the diameter of the maternal pelvis, which can lead to an almost 50% perinatal mortality in some captive populations, yet vaginal pH in squirrel monkeys is around 6.75 (Aksel and Abee, 1983). Thus, the human STD and obstetric risk selective pressures do not appear to be particularly strong compared to selective pressures experienced by other mammals. Together, these results do not support the hypotheses that STD or obstetric risks have shaped the mammalian vaginal microbiome.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.