NMR metabolomics of symbioses between bacterial vaginosis associated bacteria

Bacterial vaginosis (BV) is a dysbiosis of the vaginal microbiome, characterised by low levels of lacto-bacilli and overgrowth of a diverse group of bacteria, and associated with higher risk of a variety of infections, surgical complications, cancer and spontaneous preterm birth (PTB). Despite the lack of a consistently applicable aetiology, Prevotella spp. are often associated with both BV and PTB and P. bivia has known symbiotic relationships with both Peptostreptococcus anaerobius and Gardnerella vaginalis. Higher risk of PTB can also be predicted by a composite of metabolites linked to bacterial metabolism but their specific bacterial source remains poorly understood. Here we characterise diversity of metabolic strategies among BV associated bacteria and lactobacilli and the symbiotic metabolic relationships between P. bivia and its partners and show how these influence the availability of metabolites associated with BV/PTB and/or pro- or anti-inflammatory immune responses. We confirm a commensal relationship between Pe. anaerobius and P. bivia, refining its mechanism; P. bivia supplies tyrosine, phenylalanine, methionine, uracil and proline, the last of which leads to a substantial increase in overall acetate production. In contrast, our data indicate the relationship between P. bivia and G. vaginalis strains, with sequence variant G2, is mutualistic with outcome dependent on the metabolic strategy of the G. vaginalis strain. Seven G. vaginalis strains could be separated according to whether they performed mixed acid fermentation (MAF) or bifid shunt (BS). In co-culture, P. bivia supplies all G. vaginalis strains with uracil and received substantial amounts of asparagine in return. Acetate production, which is lower in BS strains, then matched that of MAF strains while production of aspartate increased for the latter. Taken together, our data show how knowledge of inter- and intra-species metabolic diversity and the effects of symbiosis may refine our understanding of the mechanism and approach to risk prediction in BV and/or PTB.


51
Although multiple studies have identified Prevotella spp. as being associated with both BV, and preterm 52 birth, 9,12,13,15 their presence has not been found to be predictive of PTB. 15 However, their residence within the 53 vagina correlates with that of a number of other bacteria including Gardnerella vaginalis, 15,16 and P. bivia is known 54 to enjoy symbiotic interactions with both Peptostreptococcus anaerobius and G. vaginalis. 24-26 Two groups have 55 found an association between preterm birth and G. vaginalis, 7,9,16 but its presence alone does not predict PTB.

56
There is though, reason to consider whether the substantial diversity of G. vaginalis affects the ability to establish 57 its functional role(s) in both BV and preterm birth. 27 Studies of microbial communities often sequence and quan-58 tify specific marker genes and cluster such sequences into Operational Taxonomic Units (OTUs). Although such 59 OTUs have been generally shown to have high levels of ecological consistency, 28 and the approach remains pop-60 ular and useful, there remains the possibility that functionally relevant differences in bacterial behaviour are ob-61 scured by this approach. Indeed, in one study that confirmed an association between G. vaginalis and preterm 62 birth, high-resolution statistical bioinformatics was used to detect nine unique G. vaginalis 16S rRNA sequence 63 variants and this revealed that only one of three G. vaginalis clades was responsible for the association of the 64 genus with PTB. 9 Strain level profiling has also helped improve understanding of species co-occurrence profiles. 16

65
In addition, the role of the otherwise dominant lactobacilli may also be critical in defining PTB risk, with Lactoba-66 cillus crispatus dominance frequently associated with term delivery. 9,10,13,15,16 The picture for L. iners is less clear.

67
One study showed an association with PTB, 10 but two subsequent studies found none. 9,15 Instead they found fre-68 quent co-existence of L. iners with G. vaginalis, 9 which contrasts with L. crispatus where an exclusionary relation-69 ship with G. vaginalis is found, 9,16   bivia. In addition, we compare metabolism across a panel of lactobacilli to highlight that variation in metabolic 77 strategy is not limited to BV/sPTB associated bacteria and that the metabolite background will likely vary accord-78 ing to microbiome community state type (CST). 5 The information provided by the present study suggests ways of 79 refining prediction models that include metabolite data and gives insight into how bacterial metabolism and sym-80 biosis influence each other, with implications for functional impact and clinical outcomes.

5
To better understand the contribution of different bacteria to the vaginal metabolome in eubiosis and dysbiosis a 150 panel of lactobacilli and BV associated isolates was assembled. Whole genome sequencing of seven G. vaginalis 151 strains included reference strains from the NCTC and new isolates from vaginal swabs, enables them to be assigned 152 to Clades 9,16,33 or subgroup 34 and identifies genes for sialidase and vaginolysin (

158
All isolates are sensitive to clindamycin and erythromycin.

159
Overview of bacterial metabolism in BHI and identification of major metabolic strategies for BV associated bacte-160 ria. Analysis of BHI spent culture allows comparison of the overall metabolic strategy for each of the BV associated 161 bacteria but also comparison ( Fig. 1) of the relative amounts of key metabolites that define the vaginal chemical 162 environment and/or have been associated with BV and/or PTB ( Fig. 2; S1). The NMR metabolomic approach clearly

169
M. curtisii is known to be capable of using trimethylamine oxide (TMAO) as an electron donor for anaerobic res-170 piration, producing trimethylamine (TMA). 37 In BHI it also conducts anaerobic respiration, but the production of 171 succinate ( Fig. 1; 2C) is suggestive of fumarate acting as the electron donor in place of TMAO which is absent. M.

172
curtisii is known also to consume arginine to produce ornithine, citrulline and ammonia, 38 and both it and A. va-173 ginae do this also in BHI ( Fig. 1A; S1E/T). P. bivia characteristically also produces succinate via anaerobic respiration 174 but also ferments glucose to acetate, 39 and this is observed in BHI alongside avid consumption of asparagine (

178
The result of these differing metabolic strategies is, in every case, an acidification of the spent BHI culture but this 179 is relatively modest for M. curtisii, Pe. anaerobius and A. vaginae compared with that observed for the seven G.   6 Acidification of the spent culture media is likely limited by the relatively low glucose concentration in BHI but the 194 greatest acidification is achieved by L. acidophilus (significantly more than all except L. crispatus 2), which also 195 produces more lactate than any of the other strains (p < 0.05) ( Fig. S2B; S3A).

202
We have shown previously that lower lactate and higher acetate were associated with increased risk of PTB < 37 203 weeks (odds ratios respectively 0.432 and 1.610). 15 As expected, and again despite the relatively low concentration 204 of glucose in BHI, the lactobacilli produce a final lactate concentration of between 30 and 45 mM in spent BHI 205 7 culture (Fig. S3A), which substantially exceed production by BV associated bacteria ( Fig. 2A). Of note however is 206 that A. vaginae spent culture is enriched with around 27 mM lactate and the two BS G. vaginalis produce substan-207 tially more lactate than the five MAF G. vaginalis strains (p < 0.0001) ( Fig. 2A). Except for L. iners, acetate is pro-208 duced by lactobacilli in BHI to achieve final concentrations ranging from 5 mM to 21 mM (Fig. S3). Similar levels of 209 acetate production are achieved by A. vaginae, P. bivia and Pe. anaerobius but this is dwarfed by production by G.

214
Higher aspartate has previously been associated with increased risk of PTB < 37 and < 34 weeks (odds ratios re-215 spectively 1.675 and 1.768). 15 Seven of the nine lactobacilli strains produce this, but this is very modest with spent 216 culture enriched by a maximum of 1.2 mM aspartate (Fig. S3D). In monoculture, none of the G. vaginalis strains 217 produce aspartate but modest amounts are produced by A. vaginae and it is consumed by P. bivia (Fig. 2D). We

226
Higher leucine and betaine and lower tyrosine have also been associated with increased risk of PTB< 37 weeks 227 (odds rations respectively, 3.118, 1.365 and 0.023). 15 None of the lactobacilli or BV associated bacteria in the pre-228 sent study produce leucine when cultured in BHI though it is avidly consumed by P. bivia ( Fig. S1G; S4G). Tyrosine 229 is produced in modest amounts by six of the lactobacilli isolates, most notably by L. acidophilus, L. gasseri 1 and 230 2, and most G. vaginalis strains as well as A. vaginae, P. bivia and M. curtisii (Fig S1C). It is consumed avidly by Pe.

235
Symbiosis between P. bivia and Pe. anaerobius influences production of key PTB markers. 1 H NMR of the spent 236 culture from P. bivia, Pe. anaerobius and a 1:1 co-culture reveals that combining the two species leads to a sub-237 stantial adjustment in the levels of metabolites that have previously been associated with PTB and/or shown utility 238 in predicting patient outcomes. In the spent BHI media, even though relative abundance could not be enumerated 239 by plating, there is clear evidence from production and consumption of species-specific metabolites that both 240 species proliferate (Fig. 3). In monoculture, only P. bivia consumes asparagine and produces butyrate, fumarate 241 and succinate and this is observed also in co-culture although succinate production is reduced (p < 0.0001) (

255
Previously, a commensal symbiosis between P. bivia and Pe. anaerobius has been demonstrated and ascribed to 256 use of amino acids, by Pe. anaerobius, that were secreted by P. bivia. 24 Here, 1 H NMR identifies enrichment of BHI 257 media with tyrosine, phenylalanine, proline, methionine, alanine, glutamate, glycine, isoleucine, valine and also 258 choline and uracil (Fig. S5/S6). Of these, Pe. anaerobius avidly consumes tyrosine, phenylalanine, proline and ura-259 cil, and modestly consumes methionine and possibly choline (Fig. S5). Levels of alanine, glutamate, glycine, iso-260 leucine and valine are also lower in the co-culture spent media than that of P. bivia but, since these are available 261 in BHI normally and are not consumed in Pe. anaerobius monoculture, it is assumed that this reduction can also 262 be accounted for by a lower overall growth of P. bivia in the combination relative to monoculture (Fig. S6).

9
While the benefits of co-culture to Pe. anaerobius appear manifold, the reverse is not true for P. bivia and 1 H NMR 264 does not detect any metabolites produced by Pe. anaerobius that are consumed by P. bivia. This supports the 265 previous finding of a commensal relationship between the two organisms. 24 There is one possible caveat to this in 266 that, while no effect of Pe. anaerobius conditioned media on P. bivia growth was observed previously, 24 here we 267 find that P. bivia metabolism is likely altered by co-culture with Pe. anaerobius. First, while production of Pe. an-268 aerobius specific metabolites is increased in co-culture relative to monoculture, the same is true for P. bivia only 269 for butyrate (p = 0.0295), with less succinate, alanine, glutamate, glycine and valine than might be expected. Sec-270 ond, the total consumption of formate by P. bivia in monoculture is not observed in co-culture (Fig. 3C) while 271 lactate, produced by both species in monoculture, is no more abundant in the co-culture spent media than in 272 fresh BHI (Fig. 3D) even though acetate production almost doubles (Fig. 3E). Both formate and lactate are potential 273 electron donors for anaerobic respiration and the NMR analysis provides evidence for a switch in electron donor, 274 from formate to lactate, by P. bivia when Pe. anaerobius is present.

275
The spent culture media pH will be affected by the production/consumption of a range of organic and amino acids.

276
Although acidification of spent culture will be limited due to the relatively low levels of glucose interactions be-277 tween these two species will affect the acidity of the environment (Fig. 3B). Despite production of acetate (pKa 278 4.76) and lactate (pKa 3.86), acidification by Pe. anaerobius is relatively modest with a reduction by only 0.65 pH 279 units. In contrast, both the spent P. bivia monoculture and co-culture are reduced by over one pH unit (respectively 280 1.41 and 1.16). In both cases substantial amounts of succinate (pKa 4.2, 5.6) are produced (20 mM vs 9.5 mM for 281 monoculture vs co-culture). More acetate is produced in the co-culture (30.1 vs 15.9 mM) but there is no net 282 lactate production. These effects combine to ensure that the spent co-culture pH is a little higher than that of the 283 P. bivia monoculture but substantially lower than that corresponding to Pe. anaerobius.

290
The Spearman and Pearson r for KC1 are both negative indicating that when G. vaginalis KC1 grew well, P. bivia 291 did not, and vice versa. This is manifested in the metabolomics analysis where levels of some metabolites, known 292 to be produced by P. bivia, notably succinate, fumarate, proline, uracil and alanine are highly variable (Fig. S7D-F,

293
J, K). There is some explanation for this phenomenon in the metabolomics data (Fig.4A). Notably, KC1 may be the 294 only one of the five G. vaginalis strains that is not capable of adequately supplying asparagine to P. bivia (Fig. 4D).

295
As noted above, P. bivia avidly consumes asparagine since this can be used to produce aspartate and, in turn, 296 fumarate which is an important electron acceptor anaerobic respiration. Asparagine is produced in substantial 297 amounts by all G. vaginalis isolates (p < 0.0001), with the exception of KC1 (p = 0.0333). The two BS strains increase 298 the availability of asparagine by 42% (10915) and 45% (10287). This is modest when compared with MAF strains 299 11292 and KC3 which respectively increase the availability of asparagine by 63% and 70%, such that approximately 300 double the amount of asparagine that is consumed by P. bivia in monoculture is available in co-culture. In contrast, 301 KC1 only increases the amount available by 18.5%.

11
While supply of asparagine from G. vaginalis to P. bivia is observed for both MAF and BS strains, a further means 313 by which BS strains, but not MAF strains, may supply P. bivia is also apparent. Unlike the BS G. vaginalis strains, P. 314 bivia and all three MAF G. vaginalis strains consume pyruvate from BHI (Fig. S7C). With two species growing to-315 gether the metabolite data for co-culture has greater variance but considering just the data from monocultures 316 (as above) indicates that some pyruvate is likely secreted from 10287 (p = 0.0035) and 10915 (p = 0.0046). As such 317 the BS strains differ from the MAF strains in that they avoid competition with P. bivia for pyruvate and, likely, may 318 supply it in co-culture.

319
As noted above, in monoculture the MAF strains 11292 and KC3 (p < 0.0001) and KC1 (p < 0.05) produce more 320 acetate than the BS strains 10287 and 10915 but less lactate. In co-culture however acetate produced by P. 321 bivia/10287 and P. bivia/10915 increases by 42-45% over that produced by G. vaginalis alone while the corre-322 sponding figure for the MAF strains is between 2 and 11%. Lactate production is largely unchanged in co-culture 323 for any of the strains. Co-culture with P. bivia therefore has the potential to substantially increase overall acetate 324 levels and change the acetate/lactate ratio when BS strains are present but not MAF strains. Further, while P. bivia 325 was confirmed to consume formate, ethanol and aspartate by spiking experiments (Fig. S8) there is insufficient 326 evidence here that production of these metabolites by MAF G. vaginalis provides substantial benefit for P. bivia 327 with no apparent consumption of these metabolites in the respective co-cultures (Fig. S7G, H; Fig. 4F). Indeed, 328 while both 11292 (p = 0.015) and KC3 (p = 0.008) produce aspartate in monoculture, the amount found in the 329 spent co-culture media is increased respectively 2-and 3-fold (Fig. 4F). As previous work has indicated P. bivia 330 supplies ammonia to G. vaginalis, 25 this suggests that MAF G. vaginalis might be performing a detoxification role 331 by consuming both ammonia and fumarate (Fig. S7E), secreted by P. bivia, to produce aspartate. 44 332 While the symbiotic relationship between P. bivia and Pe. anaerobius is commensal in BHI, we suggest here that 333 the relationship between P. bivia and G. vaginalis is mutualistic since, as well as the presumed supply of ammonia 334 and fumarate, we show P. bivia also likely supplies G. vaginalis with uracil ( Fig. 5; S7J). As above, in monoculture

339
The levels of other metabolites vary little between the spent monoculture and co-cultures though choline, pro-340 duced by P. bivia but not G. vaginalis in monoculture, is further increased in three out of the five spent co-cultures 341 10915, 11292, KC3; Fig S7N).

343
The present study describes the metabolic strategies, and quantifies the relative metabolites produced and con-   Pe. anaerobius. These all stimulate glucose uptake by Pe. anaerobius and increased proline availability also causes a switch 357 from ethanol to acetate production, accounting for a 56% increase in acetate secretion. The relationship between G. vaginalis 358 and P. bivia is mutualistic with the former supplying asparagine and the latter again supplying uracil. However, the relationship 359 between MAF or BS G. vaginalis strains and P. bivia will differ with MAF strains competing with P. bivia for pyruvate but 360 potentially supplying formate as an electron donor for anaerobic respiration. The origin of the increased aspartate found in 361 MAF G. vaginalis and P. bivia co-culture is as yet unclear. FDH -formate dehydrogenase; SDH -succinate dehydrogenase; FRD 362 -fumarate reductase. Image created with BioRender.com.

363
Commensal supply of proline by P. bivia increases acetogenesis by Pe. anaerobius -commensal symbiosis of P. 364 bivia and Pe. anaerobius is known to depend on provision of amino acids from the former to the latter. 24 Here we 365 show that, in addition, uracil supply is substantial and that these amino acids are limited to methionine, tyrosine, 366 phenylalanine and proline with the last three consumed avidly by Pe. anaerobius. All four of these amino acids 367 have been shown to stimulate glucose uptake, with leucine and tyrosine having the greatest effect. 45 The increased 368 13 availability of tyrosine and phenylalanine is associated with, respectably, a 23% increase in desaminotyrosine and 369 a 33% increase in 3-phenylpropionate production in co-culture compared with Pe. anaerobius conditioned media.

370
In contrast, proline availability increases by 470% in P. bivia conditioned media, and this leads to a 243% increase 371 in 5-aminopentanoate production in co-culture. Proline has been shown to not only be capable of initiating glu-372 cose uptake, but also high proline levels are associated with a switch from ethanol to acetate production, a process 373 which generates additional ATP. 45 Here, acetate in co-culture increases by 56% over the Pe. anaerobius monocul-374 ture while the corresponding increase for ethanol is only 17%. As such then, co-existence of P. bivia with Pe. an-375 aerobius and/or greater availability of proline from other sources can be expected to substantially increase pro-376 duction of acetate (Fig. 5).

377
Diversity in G. vaginalis metabolism influences symbiosis with P. bivia. When originally described, G. vaginalis was 378 proposed to be the sole aetiological agent of BV since it was found in 127 out of 138 cases but in none of 78 379 healthy women examined. 27,46 Since then more doubt has been expressed that G. vaginalis alone is the causative 380 agent of BV as its distribution is more widespread and is frequently found colonising the vagina of healthy or non-381 symptomatic women. At the same time, there is recognition that there is considerable diversity in the G. vaginalis 382 genus with both different species and clades or sub-groups being proposed. 33,34,47 The functional relevance of 383 diversity in G. vaginalis has been highlighted by the finding in one study that an association, between G. vaginalis 384 and PTB, was driven exclusively by sequence variants G2 with an association absent for other variants and the 385 association for the genus lost when G2 variants were excluded. 9 The implication from this is that associations and 386 mechanistic links between G. vaginalis and PTB, if they exist, will be obscured if diversity is not considered.

387
It has been shown recently that G. vaginalis enhances the invasive potential of P. bivia, 26 aiding its ascension into 388 the uterus. A commensal metabolic symbiotic relationship between these two species was proposed over 20 years 389 ago. 25 Here we use NMR metabolomics to characterise the symbiosis between P. bivia and G. vaginalis. Of the five 390 G. vaginalis isolates tested here, those four that are identified as sequence variants G2, 9 benefit from a relation-391 ship that this is mutualistic rather than commensal and further show that the outcome is specific to the metabolic 392 strategy of specific G. vaginalis isolates. As such we show that diversity in G. vaginalis metabolism is manifested 393 both in monoculture and co-culture and has potential to alter the vaginal chemical environment. Lower lactate 394 and higher acetate levels in the vagina are considered hallmarks of BV and are associated with sPTB. 15,17,22 Such 395 conditions would be consistent with a depletion of lactobacilli and increase in G. vaginalis, but this relative differ-396 ence also describes the relationship between BS and MAF strains of G. vaginalis albeit not to the same magnitude.

397
Further, since co-culture of MAF strains, but not BS strains, with P. bivia leads to an increase in aspartate produc-  microbiota, respectively at pH 3.9 and pH 7, revealed that the mixture chosen to mimic BV increased basal and 406 toll-like receptor (TLR) induced production of pro-inflammatory cytokines including tumor necrosis factor-α (TNF-407 α) but decreased basal production of CCL5 and IP-10 chemokines. 20 When tested alone, 100 mM acetate at pH 7 408 largely recapitulated the effects of the BV mixture. Since the pKa of acetic acid is 4.75 and those of succinic acid 409 are 4.2 and 5.6, these will exist respectively as the carboxylate or dicarboxylate anions at such an extreme as pH 410 7. As both the relative concentrations and the ionization state of the organic acids are changing under these ex-411 perimental conditions, it is yet unclear as to the relative importance of these two factors and the impact of acetic 412 14 acid/acetate may depend also on the vaginal pH, driven by relative concentrations of, primarily lactic acid. The 413 absolute and relative proportions of these two organic acids may therefore have substantial impact on the vaginal 414 inflammatory state and need to be considered.

415
The description of metabolism, in pairings of P. bivia with diverse G. vaginalis isolates, reveals symbiosis has the 416 potential to substantially increase the amount of acetate excreted by BS but not MAF strains. Similarly, co-culture 417 between P. bivia and Pe. anaerobius modulates pH, eliminates net lactate production and increases acetate pro-418 duction. Together, these observations raise the prospect that co-existence of P. bivia with either of the two species 419 might affect their physiological impact.

420
Further, while this study is predominantly focused on the metabolism of PTB and/or BV associated bacteria, it is 421 also important to consider the metabolism of lactobacilli that often dominate the vaginal microbiome, and hence 422 contribute to the metabolite background, and their known relationships with BV/PTB associated bacteria. Patterns 423 of cooccurrence between L. crispatus and G. vaginalis have been shown to be highly exclusive. 9 In contrast, L. iners 424 has been shown to co-exist with G. vaginalis at high frequency and its dominance has been found to be associated where other lactobacilli co-exist e.g. L. rhamnosus, and the relative change will be greater. Similarly, although less 432 abundant, succinate is produced by nine out of eleven lactobacilli strains tested here, with none detected for L. 433 iners and L. rhamnosus 1. Again, detection of succinate produced by PTB and/or BV associated P. bivia and BV 434 associated M. curtisii will be easier to detect in the L. iners CST background than in others.

435
A comparison between representative isolates of L. iners and L. crispatus dominated microbiomes is therefore 436 warranted but beyond the scope of the present study. Notably, substantial variation in metabolism was observed 437 in the two L. crispatus isolates, notably for asparagine consumption and aspartate and acetate production, and 438 there is a need to establish the extent to which metabolism varies across a larger panel of isolates to appreciate 439 its possible impact.

440
Finally, we assess whether the current study sheds any light on the protection against PTB suggested to be pro-441 vided by L. acidophilus. 15 Of note L. acidophilus does make the highest amount of lactate of all the lactobacilli 442 isolates grown here in BHI (p < 0.0001 for all but L. rhamnosus p < 0.05 and L. jensenii 2 p = 0.0047) and it produces 443 the spent culture with the lowest pH. Lactate production is correlated with H2O2, which would inhibit anaerobes, 444 and bacteriocins lose activity and hydrogen peroxide becomes unstable as the pH increases. Peroxide is however 445 only produced in presence of oxygen and L. gasseri may make more H2O2 while cervicovaginal fluid has been 446 shown to attenuate its microbicidal activity. 48,49 As such, the extent to which higher lactate production and greater 447 ability to acidify the environment, from certain less-dominant lactobacilli, is protective against BV or PTB should 448 be explored further, especially if able to co-exist within more diverse communities.

450
The diversity of intraspecies BV/PTB associated bacteria, and interspecies lactobacilli, metabolism as well as the 451 commensal and mutualistic symbiotic relationships of P. bivia have the potential to alter pro-inflammatory ace-452 tate, and other metabolites in the vaginal metabolome and consequently alter risk of bacterial vaginosis and/or 453 spontaneous preterm birth.
19 Table 2. P. bivia vs G. vaginalis co-culture correlation. Relationship between CFU counts for each species as a function of G. 579 vaginalis isolate as determined by parametric Pearson or non-parametric Spearman correlation coefficients. Only for KC1 is a 580 negative correlation between the two species found while positive correlations exist for the remaining four isolates.