Advances in mass spectrometry technologies to characterize cervicovaginal microbiome functions that impact spontaneous preterm birth

Preterm birth (PTB) is a leading cause of morbidity and mortality in young children. Infection is a major cause of this adverse outcome, particularly in PTBs characterised by spontaneous rupture of membranes, referred to as spontaneous (s)PTB. However, the aetiology of sPTB is not well defined and specific bacteria associated with sPTB differ between studies and at the individual level. This may be due to many factors including a lack of understanding of strain‐level differences in bacteria that influence how they function and interact with each other and the host. Metaproteomics and metabolomics are mass spectrometry‐based methods that enable the collection of detailed microbial and host functional information. Technological advances in this field have dramatically increased the resolution of these approaches, enabling the simultaneous detection of thousands of proteins or metabolites. These data can be used for taxonomic analysis of vaginal bacteria and other microbes, to understand microbiome‐host interactions, and identify diagnostic biomarkers or therapeutic targets. Although these methods have been used to assess host proteins and metabolites, few have characterized the microbial compartment in the context of pregnancy. The utilisation of metaproteomic and metabolomic‐based approaches has the potential to vastly improve our understanding of the mechanisms leading to sPTB.

with recent estimates suggesting individual care costs between $USD 76-604,000, inversely related to infant gestational age at birth. 9 Infection is a major cause of this adverse outcome, especially in PTBs where the onset of labour occurs spontaneously, accounting for up to 40%−50% of such cases 10,11 and associated with the highest morbidity and mortality. 12 Current clinical protocols used to diagnose spontaneous (s)PTB risk are largely centred around previous clinical and family history of the condition, combined with cervical measurement in mid-gestation and use of progesterone and cervical cerclage in cases where cervical insufficiency is identified. 1 Quantitative foetal fibronectin (FFN) is the only validated biomarker in widespread clinical use to predict risk of imminent PTB, however, it has a poor positive predictive value and there are no associated PTB prevention measures beyond increased obstetric management. 13 Continued research utilising state-of-the-art technologies is thus critical to develop better strategies to diagnose and prevent sPTB. This review focuses on how recent advances in the sensitivity and coverage of protein and metabolite detection using metaproteomics and metabolomics can be utilised for this purpose.

1.1
The causes of sPTB sPTB has long been thought to primarily occur when bacteria ascend into the foetal membranes and/or uterus, triggering an inflammatory response that results in the release of prostaglandins, followed by uterine contractions, labour and cervical ripening. 14 Advances in highthroughput sequencing studies (e.g., 16S rRNA gene or internal transcribed spacer region sequencing, shotgun metagenomics) and vaginal inflammatory marker profiling studies have significantly improved our understanding of the aetiology of sPTB, identifying single microbes and combinations of microbes associated with this outcome, as well as localised inflammatory responses in sPTB cases. 15,16 In general, Lactobacillus spp. dominance in the vaginal microbiome is associated with optimal vaginal health, and women are more likely to have an optimal stable cervicovaginal microbiome during pregnancy. 15,[17][18][19] However when the vaginal microbiome becomes non-optimal (Lactobacillusdepleted), women may be at greater risk of sPTB. Bacterial vaginosis (BV), a condition that occurs when the relative abundance of diverse non-optimal bacteria increases with concurrent decreases in optimal Lactobacillus spp., is associated with sPTB risk. 20

The varying aetiology of sPTB confounds the development of effective prevention tools
The composition of the vaginal microbiota of pregnant women is racially-and geographically-variable and this variability influences pregnancy outcomes. 30 For example, Caucasian and Asian women are more likely to have a L. crispatus-dominant vaginal microbiome than African and Hispanic women. 51 Correlates of sPTB in one race may not necessarily be applicable in different racial groups. 29,36,37 Callahan et al. (2017) found that greater depth was needed to find concordant associations across races. For example, low Lactobacillus spp. with high G. vaginalis and Ureaplasma spp. relative abundance was associated with sPTB in Caucasian but not African-American women. 29 It was further shown that species-level identification of Lactobacillus spp.
was critical for sPTB prediction across cohorts and certain strains of G. vaginalis with differences in functional potential were more likely to be associated with sPTB than others. 29 Differences between ethnicities may also be due to ethnicity-specific host responses to F I G U R E 1 Approaches for the development of diagnostic strategies to predict spontaneous preterm birth (sPTB). Clinical samples including amniotic fluid, cervicovaginal fluid, vaginal swabs, cervical cytobrushes, urine, serum and plasma have been used to identify potential biomarkers of sPTB using methods such as microbial culture, sequencing, metaproteomics and metabolomics coupled with untargeted and machine learning analytical approaches. Other diagnostic strategies include microscopy, antigen point-of-care tests and probe-based methods. Create with BioRender.com particular pathogens. Therefore, the accuracy of sPTB prediction may be further increased by integrating markers of host inflammatory response with taxonomic analysis, since induction of the inflammatory cascade is a key mechanism that underlies initiation of preterm labour (PTL). 14,30,52-54 Kindinger et al. (2016) first demonstrated that a persistent shift towards a non-optimal microbiome in women with braided suture cerclage was associated with increased inflammatory cytokine and interstitial collagenase production, as well as premature cervical remodelling. 52 Elovitz et al. (2019) recently found that while seven cervicovaginal bacterial taxa were significantly associated with increased sPTB risk, higher vaginal levels of β-defensin-2 lowered the risk of sPTB associated with cervicovaginal microbiota in an ethnicity-dependent manner. 30 Pro-inflammatory cytokine concentrations, mannose binding lectin, complement proteins and antibody concentrations have been found to be associated with non-optimal microbiota and increased risk of sPTB. 54 On the other hand, chemotactic cytokines were negatively correlated with non-optimal taxa and these associations were stronger in women who delivered preterm. 53 Additionally, in vitro models using foetal tissues have shown that inflammatory responses differ by bacterial combination and load. 55,56 Thus 16S rRNA gene sequencing and analysis of individual taxa in isolation may not necessarily resolve these key factors. Instead, combinations of 16S rRNA gene sequencing, qPCR and bacterial culture or shotgun metagenomics together with the analysis of host responses are likely to yield higher quality data to answer these questions. Different strains of bacterial species have also been found to have diverse phenotypic and functional characteristics that likely depend on both their environment and gene content. 57,58 Non-iners lactobacilli from BV-negative women were reported to be less inflammatory in vitro than isolates from BV-positive women. 57 L. iners and G. vaginalis were less likely to express virulence factors in the presence of optimal lactobacilli, and G. vaginalis has several subtypes with very different functions. 58,59 A recent study utilised shotgun metagenomics to characterise both structural and predicted functional differences in microbial composition between women at low and high risk of sPTB, where women with higher risk of sPTB exhibited higher levels of BV-associated bacteria. 60 Bacterial genes involved in metabolic functions and biological processes including methionine and folate metabolism functions were also significantly associated with sPTB. 60 Taken together, these studies suggest that understanding bacterial functional activities and their interactions with the host are more important than only understanding bacterial presence or absence. 61 In addition to hampering the development of sPTB prediction tools, the diverse aetiology of the condition confounds the development of effective treatment strategies. While a Cochrane systematic review found that antibiotic use does not appear to prevent BV-related sPTB, 62 this review included heterogenous studies with differences in the type and timing of antibiotic use during gestation. 63 In contrast, a review that specifically focussed on clindamycin treatment before 22 weeks' gestation in women with BV showed a 40% reduction in sPTB and an 80% reduction in late miscarriage. 63 The vaginal microbiome has been found to be more predictive of earlier PTB (<32 weeks, <34 weeks) than late PTB (34-37 weeks) across different studies, 40 therefore, the timing of intervention is likely to be key. It is possible that a better understanding of the causes and mechanisms underlying sPTB may help to further improve treatment outcomes. From here, either data-dependent acquisition (DDA) or data independent acquisition (DIA) can be used. During DDA, peptides are ionized and these 'precursor' ions are identified by measuring the mass to charge ratio before selection of the highest intensity precursors and fragmentation to produce 'product' ions for detection. On the other hand, during DIA, all peptides are fragmented, increasing the depth and reproducibility. As DIA generates much larger amounts of data, requiring greater computational capacity, and data analysis is highly complex, DDA is currently utilised more frequently. The metabolic products of enzymatic reactions that occur within these organisms (termed metabolomics) can also be measured using LC-MS/MS, with the majority of published studies using this approach. However, due to the complexity and structural diversity of the metabolome, no single method has been established that covers the entire rage of metabolites and other methods such as gas chromatography-MS and nuclear magnetic resonance spectroscopy (NMR) are often employed in tandem. 64,65 Since the use of 2D electrophoresis techniques for protein characterisation in the 1970′s to the rapid advancement of high throughput MS workflows decades later, the proteomics and metabolomics fields have had a simultaneous increase in data depth and decrease in sample run time. 66 The human cervicovaginal proteome was first investigated in 2007 where 150 proteins were detected, including largely metabolic (32%) and immune response-related (22%) proteins. 67,68 Since then, the depth, accuracy and dynamic range of MS platforms has greatly increased and one recent study was able to detect over four thousand (4370)  Protein biomarkers can also be utilised to develop accessible, lowcost, equipment-free and user-friendly diagnostics. A recent cost effectiveness analysis showed that the use of a hypothetical proteomics prognostic test to identify women at risk of sPTB combined with appropriate treatment is cost-effective and may reduce total costs, while preventing sPTBs and their consequences. 74 These F I G U R E 2 Metaproteomics and metabolomics allow for the exploration of actual microbial and host function and predicted microbial taxonomy. Metataxonomic and metagenomics methods are gold standards for taxonomic analysis; however, these methods can only generate predicted functional profiles. Metatranscriptomics can be used to predict both taxonomic profiles and function. Created with BioRender.com characteristics enable the use of these tests in remote and resourcelimited areas, administration by community health workers and even self-testing. fetoprotein. [75][76][77][78][79][80][81][82][83] Metabolites that have been found to be associated with sPTB include amniotic fluid glutamate, which is involved in energy metabolism, and prostaglandins, which are known to induce uterine contractions. [84][85][86] Other vaginal metabolites, including ethyl β-glucopyranoside, tartrate and diethanolamine, predicted to be of exogenous origin, were elevated in women who delivered preterm. 87 In the same study, vaginal levels of choline, previously found to be lower in cord blood from premature infants, were inversely associated with sPTB. 87,88 Models to predict sPTB utilising this data achieved greater accuracy than previous models including only microbiome or maternal covariates data and the biomarkers included were influenced by ethnicity. IL-6 and other inflammatory mediators are induced as part of the response to infection. 75,[77][78][79] Measurement of amniotic fluid IL-6 is useful in sPTB prediction in both women with and without clinical symptoms of infection. 89,90 Similarly, neutrophil defensins-1 and -2 and calgranulins A and C are antimicrobial peptides that are expressed by neutrophils in response to an infection. 91,92 Calgranulin C acts as a ligand for the immunoglobulin super family receptor for advanced glycation endproducts (RAGE) that is crucial for inflammation. 93,94 However, many of these biomarkers are only predictive of sPTB if measured in amniotic fluid collected through an invasive process known as amniocentesis. 95 The process itself carries some risks to the mother and/or foetus, and since the advent of non-invasive prenatal testing using cell-free foetal DNA sequencing, the procedure is no longer commonly performed. 96 It is thus important to consider non-invasive alternatives such as CVF, plasma or urine for novel proteomic or metabolomic biomarkers. For example, maternal serum CRP has been associated with amniotic fluid infection and increased risk of sPTB. [97][98][99] CRP is a sensitive inflammatory marker that is primarily produced by hepatocytes in response to increased levels of cytokines such as IL-6 and high levels of CRP in the first and second trimester have previously been found to be indicative of sPTB. 97,98,100 FFN is another biomarker commonly detected in the CVF that is frequently utilised to determine whether an individual is at risk of imminent sPTB. FFN is an extracellular matrix glycoprotein that acts as a 'glue' between the placenta and the maternal decidua, and the presence of FFN in the CVF can be attributed to detachment of the foetal chorion from the maternal decidua as a result of inflammation or mechanically induced damage to the extracellular matrix. [101][102][103][104] However, there is a lack of consensus on when FFN testing should be performed and, even with quantitative assessment, the positive predictive value is low, with false positive results also arising due to various reasons such as unprotected sexual intercourse, further limiting the test's effectiveness as a screening tool for sPTB. [105][106][107] There is recent evidence that human proteases, a broad group of enzymes that can modify protein products and function by breaking down polypeptides, 108 may increase the inflammatory response elicited by less-optimal microbiota (such as L. iners), which in turn could increase the risk of sPTB. 109 Proteases also have the potential to weaken cervical function and other host defences against sPTB. 110,111 Several host protease inhibitors (such as cystatin A, inter-alpha-trypsin inhibitor and serine protease inhibitor Kazal-type 5) were found to be associated with optimal Lactobacillus dominant microbiomes in a metaproteomics study. 112 One of the proteins identified in this study, cystatin A, was found to decrease towards the onset of labour in another study by Heng et al. 113 Furthermore, secretory leukocyte protease inhibitor, which has a protective and anti-inflammatory role in the genital tract, 114 is decreased in women with STIs or BV 115 and has been found to be associated with lower risk of sPTB when present at high levels. 116 Another anti-protease, SerpinA1, was found to be downregulated in placentas from women who delivered preterm. 117 Together these findings suggest that anti-protease activity may be protective against BV-related adverse reproductive health outcomes including sPTB.

Human protein and metabolite biomarkers of sPTB
Although many of these studies have identified promising biomarkers, many have included small sample sizes, have not characterised the participants in detail, have not separated PTB into indicated and spontaneous, and have not validated the candidate biomarkers in inde-pendent cohorts of women. 73 Accordingly, very few biomarkers are used in clinical practice. Additionally, many of these biomarkers have been detected during the later weeks of pregnancy or in women with imminent PTL. However, greater emphasis should be placed on the detection of these biomarkers in early pregnancy to allow for timely interventions to reduce sPTB risk utilising existing and novel approaches to modify the vaginal microbiome. 118

1.5
Moving beyond microbial identification to understanding microbial functions and interactions with the host that impact sPTB samples to detect >1790 human and >1609 microbial proteins, identifying candidate protein biomarkers of sPTB. 122 Metaproteomics in non-pregnant women has previously been utilised to show that CVF bacterial functional profiles may more accurately predict inflammation than corresponding 16S rRNA gene sequencing analyses alone. 123 This finding suggests that functional profiles may also be more predictive of the inflammatory responses that lead to the initiation of sPTB. This study was able to detect differences in microbial metabolic pathways, biofilm formation, stress responses and lactic acid production between women with low versus high inflammation. 123 Lactic acid is a key metabolite produced by lactobacilli that protects against BV-associated bacteria and also regulates inflammatory responses 124 and promotes cervicovaginal epithelial barrier integrity. 125 Bacterial virulence factors are also detectable using metaproteomics, including vaginolysin and other cytolysins produced by G. vaginalis and L.
iners, 123 both of which may be relevant to sPTB. This study also compared 16S rRNA gene sequencing to metaproteomics for taxonomy, finding that while genus-level taxonomic profiles were similar, specieslevel differences were noted. 123 These differences are likely due to database limitations, where some taxa have not yet been included, as well as possible misidentification based on short peptides and peptides of housekeeping proteins that may be conserved between species. Significant differences in the relative abundance of taxa were also evident and is likely a reflection of differences in the metabolic activities of taxa. A recent study similarly showed that metabolomics can also be used to predict bacterial community state types and even host inflammatory status. 75 This study further found that microbiome instability and inflammatory responses predicted sPTB. Stafford et al. (2017) showed that elevated vaginal lactate and succinate were associated with L. crispatus and L. gasseri dominant microbiomes over L. jensenii dominance in women who delivered at term compared with those who delivered preterm. 126 Another recent study found that the combination of metabolites including leucine, tyrosine, aspartate, lactate, betaine, acetate, and Ca2+ predicted risk of sPTB < 37 weeks, while glucose, aspartate, Ca2+, L. crispatus, and L. acidophilus relative abundance together identified women at risk of early sPTB < 34 weeks. 127 Importantly, stratification by ethnicity significantly improved the accuracy of the latter model. 127 Furthermore, in addition to measuring host and bacterial proteins, metaproteomics is also able to detect vaginal fungal taxa (mycobiome) and a limited number of archaeal and viral proteins (the 'virome'). 123 These data are relatively sparse compared to bacterial research, even though there is evidence that these other microorganisms may also have a role in adverse pregnancy outcomes. 128

CONCLUSION
Overall, through the simultaneous characterisation of host immune responses, predicted microbial taxonomy and microbial function, the utilisation of metaproteomic and metabolomic-based approaches has the potential to improve our understanding of the mechanisms leading to sPTB. Proteins and metabolites associated with sPTB could be used to develop low-cost rapid point-of-care tests to identify women at risk and identify diagnostic or screening biomarkers.

AUTHOR CONTRIBUTIONS
LM, JW, SAH, GT and MP co-wrote the manuscript.

DATA AVAILABILITY STATEMENT
None.