WO2022155469A1 - Using the vaginal metabolome for early prediction of spontaneous preterm birth - Google Patents
Using the vaginal metabolome for early prediction of spontaneous preterm birth Download PDFInfo
- Publication number
- WO2022155469A1 WO2022155469A1 PCT/US2022/012512 US2022012512W WO2022155469A1 WO 2022155469 A1 WO2022155469 A1 WO 2022155469A1 US 2022012512 W US2022012512 W US 2022012512W WO 2022155469 A1 WO2022155469 A1 WO 2022155469A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- preterm birth
- woman
- recited
- sptb
- vaginal
- Prior art date
Links
- 208000005107 Premature Birth Diseases 0.000 title claims abstract description 100
- 230000002269 spontaneous effect Effects 0.000 title description 18
- 239000002207 metabolite Substances 0.000 claims abstract description 200
- 244000005700 microbiome Species 0.000 claims abstract description 103
- 238000000034 method Methods 0.000 claims abstract description 44
- 239000000090 biomarker Substances 0.000 claims abstract description 33
- 238000004458 analytical method Methods 0.000 claims abstract description 23
- DZGWFCGJZKJUFP-UHFFFAOYSA-N Tyramine Natural products NCCC1=CC=C(O)C=C1 DZGWFCGJZKJUFP-UHFFFAOYSA-N 0.000 claims description 63
- 229960003732 tyramine Drugs 0.000 claims description 33
- 208000035010 Term birth Diseases 0.000 claims description 24
- ZBCBWPMODOFKDW-UHFFFAOYSA-N diethanolamine Chemical compound OCCNCCO ZBCBWPMODOFKDW-UHFFFAOYSA-N 0.000 claims description 19
- KCXVZYZYPLLWCC-UHFFFAOYSA-N EDTA Chemical compound OC(=O)CN(CC(O)=O)CCN(CC(O)=O)CC(O)=O KCXVZYZYPLLWCC-UHFFFAOYSA-N 0.000 claims description 15
- 238000012360 testing method Methods 0.000 claims description 15
- 230000008774 maternal effect Effects 0.000 claims description 11
- WYUFTYLVLQZQNH-JAJWTYFOSA-N Ethyl beta-D-glucopyranoside Chemical compound CCO[C@@H]1O[C@H](CO)[C@@H](O)[C@H](O)[C@H]1O WYUFTYLVLQZQNH-JAJWTYFOSA-N 0.000 claims description 10
- FEWJPZIEWOKRBE-JCYAYHJZSA-N Dextrotartaric acid Chemical compound OC(=O)[C@H](O)[C@@H](O)C(O)=O FEWJPZIEWOKRBE-JCYAYHJZSA-N 0.000 claims description 9
- 229940095064 tartrate Drugs 0.000 claims description 9
- FBPFZTCFMRRESA-FSIIMWSLSA-N D-Glucitol Natural products OC[C@H](O)[C@H](O)[C@@H](O)[C@H](O)CO FBPFZTCFMRRESA-FSIIMWSLSA-N 0.000 claims description 4
- FBPFZTCFMRRESA-KVTDHHQDSA-N D-Mannitol Chemical compound OC[C@@H](O)[C@@H](O)[C@H](O)[C@H](O)CO FBPFZTCFMRRESA-KVTDHHQDSA-N 0.000 claims description 4
- FBPFZTCFMRRESA-JGWLITMVSA-N D-glucitol Chemical compound OC[C@H](O)[C@@H](O)[C@H](O)[C@H](O)CO FBPFZTCFMRRESA-JGWLITMVSA-N 0.000 claims description 4
- 229930195725 Mannitol Natural products 0.000 claims description 4
- 239000000594 mannitol Substances 0.000 claims description 4
- 235000010355 mannitol Nutrition 0.000 claims description 4
- ZHFMVVUVCALAMY-UHFFFAOYSA-N pipecolate Natural products OC1CNC(C(O)=O)C(O)C1O ZHFMVVUVCALAMY-UHFFFAOYSA-N 0.000 claims description 4
- HXEACLLIILLPRG-UHFFFAOYSA-N pipecolic acid Chemical compound OC(=O)C1CCCCN1 HXEACLLIILLPRG-UHFFFAOYSA-N 0.000 claims description 4
- 239000000600 sorbitol Substances 0.000 claims description 4
- 230000009469 supplementation Effects 0.000 claims description 4
- YCCMTCQQDULIFE-UHFFFAOYSA-N 3-aminopiperidine-2-one Chemical compound NC1CCCNC1=O YCCMTCQQDULIFE-UHFFFAOYSA-N 0.000 claims description 3
- PYMYPHUHKUWMLA-WDCZJNDASA-N arabinose Chemical compound OC[C@@H](O)[C@@H](O)[C@H](O)C=O PYMYPHUHKUWMLA-WDCZJNDASA-N 0.000 claims description 3
- PYMYPHUHKUWMLA-UHFFFAOYSA-N arabinose Natural products OCC(O)C(O)C(O)C=O PYMYPHUHKUWMLA-UHFFFAOYSA-N 0.000 claims description 3
- SRBFZHDQGSBBOR-UHFFFAOYSA-N beta-D-Pyranose-Lyxose Natural products OC1COC(O)C(O)C1O SRBFZHDQGSBBOR-UHFFFAOYSA-N 0.000 claims description 3
- WQZGKKKJIJFFOK-QTVWNMPRSA-N D-mannopyranose Chemical compound OC[C@H]1OC(O)[C@@H](O)[C@@H](O)[C@@H]1O WQZGKKKJIJFFOK-QTVWNMPRSA-N 0.000 claims description 2
- ZGEYCCHDTIDZAE-UHFFFAOYSA-N glutamic acid 5-methyl ester Chemical compound COC(=O)CCC(N)C(O)=O ZGEYCCHDTIDZAE-UHFFFAOYSA-N 0.000 claims description 2
- FKCRAVPPBFWEJD-XVFCMESISA-N orotidine Chemical compound O[C@@H]1[C@H](O)[C@@H](CO)O[C@H]1N1C(=O)NC(=O)C=C1C(O)=O FKCRAVPPBFWEJD-XVFCMESISA-N 0.000 claims description 2
- FKCRAVPPBFWEJD-UHFFFAOYSA-N orotidine Natural products OC1C(O)C(CO)OC1N1C(=O)NC(=O)C=C1C(O)=O FKCRAVPPBFWEJD-UHFFFAOYSA-N 0.000 claims description 2
- DZGWFCGJZKJUFP-UHFFFAOYSA-O tyraminium Chemical compound [NH3+]CCC1=CC=C(O)C=C1 DZGWFCGJZKJUFP-UHFFFAOYSA-O 0.000 claims 2
- 238000002705 metabolomic analysis Methods 0.000 description 51
- 230000001431 metabolomic effect Effects 0.000 description 51
- 238000012384 transportation and delivery Methods 0.000 description 28
- 230000000813 microbial effect Effects 0.000 description 27
- 230000002503 metabolic effect Effects 0.000 description 26
- 230000035935 pregnancy Effects 0.000 description 25
- 238000005259 measurement Methods 0.000 description 21
- 238000009826 distribution Methods 0.000 description 16
- OEYIOHPDSNJKLS-UHFFFAOYSA-N choline Chemical compound C[N+](C)(C)CCO OEYIOHPDSNJKLS-UHFFFAOYSA-N 0.000 description 15
- 229960001231 choline Drugs 0.000 description 15
- 208000034423 Delivery Diseases 0.000 description 14
- 241000894007 species Species 0.000 description 14
- 238000002790 cross-validation Methods 0.000 description 13
- 230000036266 weeks of gestation Effects 0.000 description 13
- 241000736262 Microbiota Species 0.000 description 12
- 230000000875 corresponding effect Effects 0.000 description 12
- 230000000694 effects Effects 0.000 description 9
- 230000003993 interaction Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 9
- 239000002537 cosmetic Substances 0.000 description 8
- 108090000623 proteins and genes Proteins 0.000 description 8
- 239000002676 xenobiotic agent Substances 0.000 description 8
- 230000004060 metabolic process Effects 0.000 description 7
- 230000037361 pathway Effects 0.000 description 7
- 208000004926 Bacterial Vaginosis Diseases 0.000 description 6
- NTYJJOPFIAHURM-UHFFFAOYSA-N Histamine Chemical compound NCCC1=CN=CN1 NTYJJOPFIAHURM-UHFFFAOYSA-N 0.000 description 6
- 206010036590 Premature baby Diseases 0.000 description 6
- 208000037009 Vaginitis bacterial Diseases 0.000 description 6
- 238000004519 manufacturing process Methods 0.000 description 6
- 230000002265 prevention Effects 0.000 description 6
- KIDHWZJUCRJVML-UHFFFAOYSA-N putrescine Chemical compound NCCCCN KIDHWZJUCRJVML-UHFFFAOYSA-N 0.000 description 6
- PFNFFQXMRSDOHW-UHFFFAOYSA-N spermine Chemical compound NCCCNCCCCNCCCN PFNFFQXMRSDOHW-UHFFFAOYSA-N 0.000 description 6
- 239000000126 substance Substances 0.000 description 6
- 108020004465 16S ribosomal RNA Proteins 0.000 description 5
- KWIUHFFTVRNATP-UHFFFAOYSA-N Betaine Natural products C[N+](C)(C)CC([O-])=O KWIUHFFTVRNATP-UHFFFAOYSA-N 0.000 description 5
- 241000186660 Lactobacillus Species 0.000 description 5
- KWIUHFFTVRNATP-UHFFFAOYSA-O N,N,N-trimethylglycinium Chemical compound C[N+](C)(C)CC(O)=O KWIUHFFTVRNATP-UHFFFAOYSA-O 0.000 description 5
- 229960003237 betaine Drugs 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 5
- 150000001875 compounds Chemical class 0.000 description 5
- 229940039696 lactobacillus Drugs 0.000 description 5
- 230000007108 local immune response Effects 0.000 description 5
- 241000207201 Gardnerella vaginalis Species 0.000 description 4
- WHUUTDBJXJRKMK-VKHMYHEASA-N L-glutamic acid Chemical compound OC(=O)[C@@H](N)CCC(O)=O WHUUTDBJXJRKMK-VKHMYHEASA-N 0.000 description 4
- 238000000585 Mann–Whitney U test Methods 0.000 description 4
- 208000006399 Premature Obstetric Labor Diseases 0.000 description 4
- RJKFOVLPORLFTN-LEKSSAKUSA-N Progesterone Chemical compound C1CC2=CC(=O)CC[C@]2(C)[C@@H]2[C@@H]1[C@@H]1CC[C@H](C(=O)C)[C@@]1(C)CC2 RJKFOVLPORLFTN-LEKSSAKUSA-N 0.000 description 4
- 230000002411 adverse Effects 0.000 description 4
- 210000004369 blood Anatomy 0.000 description 4
- 239000008280 blood Substances 0.000 description 4
- 239000002131 composite material Substances 0.000 description 4
- 230000004907 flux Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 229930195712 glutamate Natural products 0.000 description 4
- 239000002243 precursor Substances 0.000 description 4
- 238000012163 sequencing technique Methods 0.000 description 4
- 210000002966 serum Anatomy 0.000 description 4
- 108091093088 Amplicon Proteins 0.000 description 3
- 241000427397 Corynebacterium aurimucosum Species 0.000 description 3
- 239000005700 Putrescine Substances 0.000 description 3
- 239000000654 additive Substances 0.000 description 3
- 230000000996 additive effect Effects 0.000 description 3
- 229940024606 amino acid Drugs 0.000 description 3
- 235000001014 amino acid Nutrition 0.000 description 3
- 150000001413 amino acids Chemical class 0.000 description 3
- 230000015572 biosynthetic process Effects 0.000 description 3
- 230000002596 correlated effect Effects 0.000 description 3
- 239000010432 diamond Substances 0.000 description 3
- 230000007613 environmental effect Effects 0.000 description 3
- 229960001340 histamine Drugs 0.000 description 3
- 230000037356 lipid metabolism Effects 0.000 description 3
- 150000002632 lipids Chemical class 0.000 description 3
- 210000004379 membrane Anatomy 0.000 description 3
- 239000012528 membrane Substances 0.000 description 3
- 239000002858 neurotransmitter agent Substances 0.000 description 3
- 150000003904 phospholipids Chemical class 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- 230000010076 replication Effects 0.000 description 3
- 229940063675 spermine Drugs 0.000 description 3
- 238000013517 stratification Methods 0.000 description 3
- CCEKAJIANROZEO-UHFFFAOYSA-N sulfluramid Chemical group CCNS(=O)(=O)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)C(F)(F)F CCEKAJIANROZEO-UHFFFAOYSA-N 0.000 description 3
- 238000003786 synthesis reaction Methods 0.000 description 3
- 210000002700 urine Anatomy 0.000 description 3
- 238000010200 validation analysis Methods 0.000 description 3
- 230000007279 water homeostasis Effects 0.000 description 3
- 108091032973 (ribonucleotides)n+m Proteins 0.000 description 2
- 241001464890 Anaerococcus prevotii Species 0.000 description 2
- 241001633064 Atopobium vaginae Species 0.000 description 2
- 241000894006 Bacteria Species 0.000 description 2
- 239000002028 Biomass Substances 0.000 description 2
- 241001535083 Dialister Species 0.000 description 2
- 108010016626 Dipeptides Proteins 0.000 description 2
- ULGZDMOVFRHVEP-RWJQBGPGSA-N Erythromycin Chemical compound O([C@@H]1[C@@H](C)C(=O)O[C@@H]([C@@]([C@H](O)[C@@H](C)C(=O)[C@H](C)C[C@@](C)(O)[C@H](O[C@H]2[C@@H]([C@H](C[C@@H](C)O2)N(C)C)O)[C@H]1C)(C)O)CC)[C@H]1C[C@@](C)(OC)[C@@H](O)[C@H](C)O1 ULGZDMOVFRHVEP-RWJQBGPGSA-N 0.000 description 2
- 241000192016 Finegoldia magna Species 0.000 description 2
- 208000022361 Human papillomavirus infectious disease Diseases 0.000 description 2
- 208000022559 Inflammatory bowel disease Diseases 0.000 description 2
- 241000218492 Lactobacillus crispatus Species 0.000 description 2
- 101100478282 Mus musculus Sptb gene Proteins 0.000 description 2
- 241000699670 Mus sp. Species 0.000 description 2
- RWKUXQNLWDTSLO-GWQJGLRPSA-N N-hexadecanoylsphingosine-1-phosphocholine Chemical compound CCCCCCCCCCCCCCCC(=O)N[C@@H](COP([O-])(=O)OCC[N+](C)(C)C)[C@H](O)\C=C\CCCCCCCCCCCCC RWKUXQNLWDTSLO-GWQJGLRPSA-N 0.000 description 2
- 241000566145 Otus Species 0.000 description 2
- 102100033073 Polypyrimidine tract-binding protein 1 Human genes 0.000 description 2
- 238000011529 RT qPCR Methods 0.000 description 2
- 241000270295 Serpentes Species 0.000 description 2
- 241000607762 Shigella flexneri Species 0.000 description 2
- 230000006978 adaptation Effects 0.000 description 2
- 150000001412 amines Chemical class 0.000 description 2
- 210000004381 amniotic fluid Anatomy 0.000 description 2
- 238000013459 approach Methods 0.000 description 2
- 238000012098 association analyses Methods 0.000 description 2
- 210000004027 cell Anatomy 0.000 description 2
- 238000006243 chemical reaction Methods 0.000 description 2
- 238000012937 correction Methods 0.000 description 2
- 238000013399 early diagnosis Methods 0.000 description 2
- 238000010201 enrichment analysis Methods 0.000 description 2
- 230000001605 fetal effect Effects 0.000 description 2
- 210000001320 hippocampus Anatomy 0.000 description 2
- 244000005702 human microbiome Species 0.000 description 2
- 230000036512 infertility Effects 0.000 description 2
- 239000004615 ingredient Substances 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 230000000803 paradoxical effect Effects 0.000 description 2
- 230000009984 peri-natal effect Effects 0.000 description 2
- 229920000768 polyamine Polymers 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 229960003387 progesterone Drugs 0.000 description 2
- 239000000186 progesterone Substances 0.000 description 2
- 235000018102 proteins Nutrition 0.000 description 2
- 102000004169 proteins and genes Human genes 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 230000028327 secretion Effects 0.000 description 2
- ATHGHQPFGPMSJY-UHFFFAOYSA-N spermidine Chemical compound NCCCCNCCCN ATHGHQPFGPMSJY-UHFFFAOYSA-N 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 230000002034 xenobiotic effect Effects 0.000 description 2
- WYUFTYLVLQZQNH-UHFFFAOYSA-N 1-Ethyl-D-galactoside Natural products CCOC1OC(CO)C(O)C(O)C1O WYUFTYLVLQZQNH-UHFFFAOYSA-N 0.000 description 1
- AJKPHHFHSGXPJV-UHFFFAOYSA-N 3,7-dihydroxy-3-[(trimethylazaniumyl)methyl]heptanoate Chemical compound OCCCCC(O)(C[N+](C)(C)C)CC([O-])=O AJKPHHFHSGXPJV-UHFFFAOYSA-N 0.000 description 1
- 241001464988 Actinomyces neuii Species 0.000 description 1
- 239000004475 Arginine Substances 0.000 description 1
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 1
- 108010078791 Carrier Proteins Proteins 0.000 description 1
- 206010009944 Colon cancer Diseases 0.000 description 1
- 208000001333 Colorectal Neoplasms Diseases 0.000 description 1
- 241000186216 Corynebacterium Species 0.000 description 1
- 229920004934 Dacron® Polymers 0.000 description 1
- 208000027244 Dysbiosis Diseases 0.000 description 1
- 241000194032 Enterococcus faecalis Species 0.000 description 1
- 108090000790 Enzymes Proteins 0.000 description 1
- 102000004190 Enzymes Human genes 0.000 description 1
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 1
- 238000000729 Fisher's exact test Methods 0.000 description 1
- 241000207202 Gardnerella Species 0.000 description 1
- 241000193814 Gemella haemolysans Species 0.000 description 1
- 206010019695 Hepatic neoplasm Diseases 0.000 description 1
- 208000008839 Kidney Neoplasms Diseases 0.000 description 1
- ONIBWKKTOPOVIA-BYPYZUCNSA-N L-Proline Chemical compound OC(=O)[C@@H]1CCCN1 ONIBWKKTOPOVIA-BYPYZUCNSA-N 0.000 description 1
- ODKSFYDXXFIFQN-BYPYZUCNSA-P L-argininium(2+) Chemical compound NC(=[NH2+])NCCC[C@H]([NH3+])C(O)=O ODKSFYDXXFIFQN-BYPYZUCNSA-P 0.000 description 1
- 241000186606 Lactobacillus gasseri Species 0.000 description 1
- 241001561398 Lactobacillus jensenii Species 0.000 description 1
- 241000203734 Mobiluncus curtisii Species 0.000 description 1
- 241000699666 Mus <mouse, genus> Species 0.000 description 1
- 241001312780 Mydosama mulleri Species 0.000 description 1
- QHZIGNLCLJPLCU-QPPIDDCLSA-N N-hexadecanoylsphinganine-1-phosphocholine Chemical compound CCCCCCCCCCCCCCC[C@@H](O)[C@H](COP([O-])(=O)OCC[N+](C)(C)C)NC(=O)CCCCCCCCCCCCCCC QHZIGNLCLJPLCU-QPPIDDCLSA-N 0.000 description 1
- 208000006816 Neonatal Sepsis Diseases 0.000 description 1
- 108700026244 Open Reading Frames Proteins 0.000 description 1
- 208000009608 Papillomavirus Infections Diseases 0.000 description 1
- 208000001300 Perinatal Death Diseases 0.000 description 1
- 206010036600 Premature labour Diseases 0.000 description 1
- 206010036603 Premature rupture of membranes Diseases 0.000 description 1
- 241000605861 Prevotella Species 0.000 description 1
- ONIBWKKTOPOVIA-UHFFFAOYSA-N Proline Natural products OC(=O)C1CCCN1 ONIBWKKTOPOVIA-UHFFFAOYSA-N 0.000 description 1
- 241000700159 Rattus Species 0.000 description 1
- 208000007313 Reproductive Tract Infections Diseases 0.000 description 1
- 206010040047 Sepsis Diseases 0.000 description 1
- 241000123706 Sneathia sanguinegens Species 0.000 description 1
- 241000191963 Staphylococcus epidermidis Species 0.000 description 1
- 101150067147 TDC gene Proteins 0.000 description 1
- 101500027794 Thermococcus kodakarensis (strain ATCC BAA-918 / JCM 12380 / KOD1) Homing endonuclease PI-PkoII Proteins 0.000 description 1
- 108010035075 Tyrosine decarboxylase Proteins 0.000 description 1
- 241000202921 Ureaplasma urealyticum Species 0.000 description 1
- 208000036029 Uterine contractions during pregnancy Diseases 0.000 description 1
- 206010046914 Vaginal infection Diseases 0.000 description 1
- 201000008100 Vaginitis Diseases 0.000 description 1
- -1 alkyl glucoside Chemical class 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000006907 apoptotic process Effects 0.000 description 1
- ODKSFYDXXFIFQN-UHFFFAOYSA-N arginine Natural products OC(=O)C(N)CCCNC(N)=N ODKSFYDXXFIFQN-UHFFFAOYSA-N 0.000 description 1
- 238000003556 assay Methods 0.000 description 1
- 230000000712 assembly Effects 0.000 description 1
- 238000000429 assembly Methods 0.000 description 1
- 230000001580 bacterial effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000035 biogenic effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 150000001720 carbohydrates Chemical class 0.000 description 1
- 235000014633 carbohydrates Nutrition 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 108091092259 cell-free RNA Proteins 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 239000002738 chelating agent Substances 0.000 description 1
- 230000035606 childbirth Effects 0.000 description 1
- 229940018560 citraconate Drugs 0.000 description 1
- HNEGQIOMVPPMNR-IHWYPQMZSA-N citraconic acid Chemical compound OC(=O)C(/C)=C\C(O)=O HNEGQIOMVPPMNR-IHWYPQMZSA-N 0.000 description 1
- KRKNYBCHXYNGOX-UHFFFAOYSA-N citric acid Chemical class OC(=O)CC(O)(C(O)=O)CC(O)=O KRKNYBCHXYNGOX-UHFFFAOYSA-N 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 231100000433 cytotoxic Toxicity 0.000 description 1
- 230000001472 cytotoxic effect Effects 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 206010012601 diabetes mellitus Diseases 0.000 description 1
- 235000005911 diet Nutrition 0.000 description 1
- 230000000378 dietary effect Effects 0.000 description 1
- 235000014113 dietary fatty acids Nutrition 0.000 description 1
- 229940043237 diethanolamine Drugs 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 231100000673 dose–response relationship Toxicity 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 230000007140 dysbiosis Effects 0.000 description 1
- 230000002500 effect on skin Effects 0.000 description 1
- 239000000598 endocrine disruptor Substances 0.000 description 1
- 210000000750 endocrine system Anatomy 0.000 description 1
- 229940032049 enterococcus faecalis Drugs 0.000 description 1
- 210000002919 epithelial cell Anatomy 0.000 description 1
- 229960003276 erythromycin Drugs 0.000 description 1
- 235000020774 essential nutrients Nutrition 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 210000002219 extraembryonic membrane Anatomy 0.000 description 1
- 229930195729 fatty acid Natural products 0.000 description 1
- 239000000194 fatty acid Substances 0.000 description 1
- 150000004665 fatty acids Chemical class 0.000 description 1
- 210000004700 fetal blood Anatomy 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 239000012530 fluid Substances 0.000 description 1
- 235000013305 food Nutrition 0.000 description 1
- 235000013373 food additive Nutrition 0.000 description 1
- 239000002778 food additive Substances 0.000 description 1
- 238000010230 functional analysis Methods 0.000 description 1
- 125000000524 functional group Chemical group 0.000 description 1
- 229960003692 gamma aminobutyric acid Drugs 0.000 description 1
- BTCSSZJGUNDROE-UHFFFAOYSA-N gamma-aminobutyric acid Chemical compound NCCCC(O)=O BTCSSZJGUNDROE-UHFFFAOYSA-N 0.000 description 1
- 230000002496 gastric effect Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 229930182478 glucoside Natural products 0.000 description 1
- 150000002306 glutamic acid derivatives Chemical class 0.000 description 1
- ZWZWYGMENQVNFU-AKGZTFGVSA-N glycerol 1-phosphoserine Chemical compound OC(=O)[C@@H](N)COP(O)(=O)OCC(O)CO ZWZWYGMENQVNFU-AKGZTFGVSA-N 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 208000002566 gonadal dysgenesis Diseases 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000002440 hepatic effect Effects 0.000 description 1
- 230000002519 immonomodulatory effect Effects 0.000 description 1
- 230000028709 inflammatory response Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000004895 liquid chromatography mass spectrometry Methods 0.000 description 1
- 210000004185 liver Anatomy 0.000 description 1
- 208000014018 liver neoplasm Diseases 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 208000018773 low birth weight Diseases 0.000 description 1
- 231100000533 low birth weight Toxicity 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004949 mass spectrometry Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000005555 metalworking Methods 0.000 description 1
- 150000004702 methyl esters Chemical class 0.000 description 1
- 230000007269 microbial metabolism Effects 0.000 description 1
- 210000003470 mitochondria Anatomy 0.000 description 1
- 230000011278 mitosis Effects 0.000 description 1
- 210000005155 neural progenitor cell Anatomy 0.000 description 1
- 230000000926 neurological effect Effects 0.000 description 1
- 239000002773 nucleotide Substances 0.000 description 1
- 125000003729 nucleotide group Chemical group 0.000 description 1
- 210000000056 organ Anatomy 0.000 description 1
- 230000007170 pathology Effects 0.000 description 1
- 230000003169 placental effect Effects 0.000 description 1
- 210000005059 placental tissue Anatomy 0.000 description 1
- 239000005020 polyethylene terephthalate Substances 0.000 description 1
- 201000011461 pre-eclampsia Diseases 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000002028 premature Effects 0.000 description 1
- 208000026440 premature labor Diseases 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000001850 reproductive effect Effects 0.000 description 1
- 210000005000 reproductive tract Anatomy 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 201000005608 severe pre-eclampsia Diseases 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 150000003384 small molecules Chemical class 0.000 description 1
- 229940063673 spermidine Drugs 0.000 description 1
- 239000004094 surface-active agent Substances 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 210000002504 synaptic vesicle Anatomy 0.000 description 1
- 230000009885 systemic effect Effects 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 231100000378 teratogenic Toxicity 0.000 description 1
- 230000003390 teratogenic effect Effects 0.000 description 1
- 210000001519 tissue Anatomy 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 230000004102 tricarboxylic acid cycle Effects 0.000 description 1
- 210000001215 vagina Anatomy 0.000 description 1
- 208000010603 vasculitis due to ADA2 deficiency Diseases 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
- G01N33/50—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing
- G01N33/68—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving proteins, peptides or amino acids
- G01N33/689—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing involving proteins, peptides or amino acids related to pregnancy or the gonads
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/48—Biological material, e.g. blood, urine; Haemocytometers
- G01N33/50—Chemical analysis of biological material, e.g. blood, urine; Testing involving biospecific ligand binding methods; Immunological testing
- G01N33/53—Immunoassay; Biospecific binding assay; Materials therefor
- G01N33/5308—Immunoassay; Biospecific binding assay; Materials therefor for analytes not provided for elsewhere, e.g. nucleic acids, uric acid, worms, mites
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2800/00—Detection or diagnosis of diseases
- G01N2800/36—Gynecology or obstetrics
- G01N2800/368—Pregnancy complicated by disease or abnormalities of pregnancy, e.g. preeclampsia, preterm labour
Definitions
- the disclosure of the present patent application relates to identifying and using vaginal microbiome metabolites as early biomarkers for identifying potential risk factors and predicting spontaneous preterm birth.
- PTB Preterm birth
- PTB also reflects a significant racial disparity, manifesting in a substantially higher risk for PTB in black women compared to non-Hispanic white women.
- the human microbiome is a strong biomarker of many complex diseases, often predicting host phenotypes even better than host genetics.
- the vaginal microbiome specifically, is a promising area of research for early diagnosis of sPTB: studies in multiple cohorts, clinical settings, and populations have found various associations with sPTB and other adverse pregnancy outcomes. See, for example, Brown, R. G. et al., Vaginal dysbiosis increases risk of preterm fetal membrane rupture, neonatal sepsis and is exacerbated by erythromycin, BMC Med. 16, 9 (2016); Callahan, B. J.
- Metabolomics the study of the metabolites present in an organism, cell, or tissue - enables the measurement of thousands of small molecules present in a particular ecosystem, and has revealed intriguing new insights on host-microbiome and other interactions in a variety of contexts, from colorectal cancer to diabetes.
- Spontaneous preterm birth is a leading cause of maternal and neonatal morbidity and mortality. Nevertheless, available means for both its prevention and early risk stratification are limited.
- the vaginal microbiome has been associated with PTB risk, possibly via metabolic or other interactions with its host.
- vaginal metabolome partially corresponds to CSTs, and reveal novel associations between metabolites measured in the middle of pregnancy and subsequent sPTB. Exploring the potential origins of these metabolites, we propose that some are of an exogenous source, suggesting a novel risk factor and potential targets for novel prevention strategies. Finally, using the vaginal metabolome, measured an average of 3 months before delivery, we devise machine learning algorithms that accurately predict subsequent preterm birth, weeks to months in advance, with a high degree of accuracy. This predictor, more accurate than predictors based on microbiome or clinical data, is validated on two external cohorts. Our results demonstrate a promising new approach for studying potential causes of prematurity as well as for early risk stratification, and highlight the need to study environmental exposures as a potential factor in sPTB.
- CST I indicates domination by Lactobacillus crispatus;
- CST II is dominated by Lactobacillus gasseri;
- CST III is dominated by Lactobacillus triers',
- CSTs IV- A and IV -B contains diverse anaerobes;
- CST-V is dominated by Lactobacillus jensenii.
- the metabolite sub-pathway most enriched among metabolites associated with each MC were polyamine metabolism, dipeptides, dicarboxylated fatty acids, glutamate metabolism, TCA cycle, and dipeptides for MC A-F, respectively (hypergeometric p ⁇ 2.4x10-4, 1.1x10-4, 0.025, 0.02, 0.015 and 3.6x10-4, respectively).
- Fig. ID is a chart correlating the distribution of CSTs for each MC. The number above the horizontal lines is Fisher’s exact p, FDR (False Discovery Rate) ⁇ 0.1.
- Fig. IE is a chart correlating the relative MC distribution for black women and for white women.
- Fig. IF is a chart showing frequency of pregnancy outcome (full term birth (TB) vs. sPTB) broken out by CST in white women and in black women.
- This Fig. I F is a replication of an analysis done in Elovitz, M. A. el al., Cervicovaginal microbiota and local immune response modulate the risk of spontaneous preterm delivery, Nat. Commun. 10, 1305 (2019). The figure shows that CST IV- A is enriched for sPTB among white women.
- Fig. 1G is a chart showing frequency of pregnancy outcome (full term birth (TB) vs. sPTB) broken out by MC in white women and in black women. The number abo ve the horizontal lines is Fisher’s exact. p, FDR ⁇ 0.1.
- Fig. 2 A is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes, for all women, and broken out for black women and for white women.
- Fig. 2B shows box and swarm plots of three metabolites with significant, associations with sPTB - diethanolamine, choline, and betaine.
- Fig. 2C is an illustration depicting common wisdom that diethanolamine (DEA) (here found to be associated with sPTB) inhibits choline uptake, while choline and betaine (both found here to be associated with TB) are important for membrane lipid synthesis and osmoregulation.
- DEA diethanolamine
- Fig. 2D is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes for black women, broken out by gestational age at birth (GAB).
- Fig. 3 A depicts a network of microbial correlations with metabolites associated with sPTB. Ellipses depict microbial species; diamonds depict metabolites enriched in TB and sPTB, respectively.
- Fig. 3B shows box and swarm plots of tyramine levels as measured, comparing preterm deliveries (blue boxes) and term deliveries (red boxes) in self-identified white and black women.
- Fig. 3C shows box and swarm plots of tyramine levels as predicted using metabolic models, comparing preterm deliveries (red boxes) and term deliveries (blue boxes) in selfidentified white and black women.
- Fig. 3D is a plot of tyramine production derived from microbiome metabolic models against measured tyramine levels, colored by race, and by birth outcome for white women.
- Fig. 4A is a graph showing receiver operating characteristic (ROC) curves comparing sPTB prediction accuracy based on clinical, microbiome, and metabolomics data and plotting true positive rate vs. false positive rate.
- ROC receiver operating characteristic
- Fig. 4B is a graph showing precision-recall (PR) curves comparing sPTB prediction accuracy based on clinical, microbiome, and metabolomics data.
- Fig. 4C is a graph showing ROC curves evaluating the performance of the metabolomics-based predictor on two external cohorts, comparing true positive rate to false positive rate.
- PR precision-recall
- Fig. 4D depicts the effect on total prediction for the 10 most predictive metabolites in the metabolomics-based predictor, sorted with descending importance.
- Fig. 5A is a graph showing the distribution of metabolite super pathways among assayed metabolites.
- Fig. 5B is a graph showing the distribution of metabolite prevalence across samples.
- Fig. 6A is a graph plotting within cluster sum of squared distances for k-medoids clustering using Canberra distances for k from 1 to 15.
- Fig. 6B is a graph plotting gap statistic for k-medoids clustering using Canberra distances for k from 1 to 15.
- Fig. 6C is a heatmap showing metabolite levels for each subject, sorted by assigned metabolites cluster (MC) with metabolites clustered hierarchically using Canberra distance and Ward linkage.
- Fig. 6D is a Principal Component Analy sis (PCA) plot of metabolites data colored by MCs A to F.
- PCA Principal Component Analy sis
- Fig. 6E is a Canberra distance-based Principal Coordinate Analysis (PCoA) plot of metabolites data colored by MCs A to F.
- Fig. 6F is a iSNE (t-disiriaded Stochastic Neighbor Embedding) plot of metabolites data colored by MCs A to F.
- PCoA Principal Coordinate Analysis
- Fig. 6G is a histogram of consistency of MC assignment.
- Fig. 7 A is a chart showing CST distribution within each MC, for all women, white women, and black women.
- Fig. 7B is a chart correlating the distribution of CSTs for each MC, separately for black women and for white women.
- Fig. 7C is a microbiome UMAP colored by race.
- Fig. 7D is a metabolomics UMAP plot colored by race.
- Fig. 7E is a chart showing frequency of pregnancy outcome (TB vs. sPTB) broken out by various CSTs for all women combined.
- Fig. 7F is a chart showing frequency of pregnancy outcome (TB vs. sPTB) broken out by MC for all women combined.
- Fig. 7G is a chart showing frequency of pregnancy outcome (early sPTB) broken out by MC in black women.
- Fig. 8A is a set of box and swarm plots of the levels of metabolites associated with sPTB, comparing preterm and term deliveries, and stratified by self-identified race.
- Fig. 8B is a kernel density estimation showing the distribution across all samples of four xenobiotics associated with sPTB or early sPTB.
- Fig. 8C is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes, for women not treated with progesterone, and also broken out for black women and for white women.
- Fig. 8D is a heatmap showing metabolite sets altered in sPTB in various subsets.
- Fig. 9 A depicts a network of microbial correlations with metabolites associated with sPTB. Ellipses depict microbial species; diamonds depict metabolites enriched in TB and sPTB, respectively. This is similar to Fig. 3A, but without the grouped nodes - each microbial taxa is instead represented as an individual node.
- Fig. 9B is a volcano plot with every point representing a comparison of a specific microbe-metabolite correlation between black and white women.
- Fig. 9C depicts a network of microbial correlations with metabolites associated with sPTB, with individual nodes, for black women.
- Fig. 9D depicts a network of microbial correlations with metabolites associated with sPTB, with individual nodes, for white women.
- Fig. 9E depicts a network of microbial correlations with metabolites associated with early sPTB, with individual nodes, for black women with early sPTB.
- Fig. 9F is a volcano plot with every point representing a comparison of a specific microbe-metabolite correlation between black women with early sPTB and all other black women.
- Fig. 10A is a plot comparing prediction of microbiome putrescine from microbiome metabolic models (net maximal production capacity, or NMPC) against measured metabolite levels.
- Fig. 10B is a plot comparing prediction of microbiome histamine from microbiome metabolic models (NMPC) against measured metabolite levels.
- Fig. IOC is a plot comparing prediction of microbiome tyramine from microbiome metabolic models (NMPC) against measured metabolite levels.
- Fig. 10D is a set of plots showing the percent of total sample abundance represented by metabolic models, separately for black women, white women with sPTB, and white women with TB.
- Fig. 10E shows the correlation between NMPCs and metabolite measurements for models that contain a maximum of N most abundant species across the cohort.
- Fig. 11 A is an ROC curve comparing performance of different sPTB prediction algorithms on metabolomics data.
- Fig. 11B is an ROC curve comparing performance of models stratified for race against a model based on all samples (and not stratified for race).
- Fig. 11C is an ROC curve evaluated in nested cross-validation, comparing sPTB prediction accuracy for models based on metabolomics data alone and on metabolomics data combined with microbiome and clinical data.
- Fig. 11 D is a graph showing precision-recall (PR) curve evaluated in nested cross- validation, comparing sPTB prediction accuracy for models based on metabolomics data alone and on metabolomics data combined with microbiome and clinical data.
- PR precision-recall
- Fig. 1 IE is a SHAP-based (SHapley Additive exPlanations) effect plot for the top 10 features in the combination models, sorted with descending importance.
- Fig. 1 IF is a graph showing ROC curves comparing sPTB prediction accuracy based on metabolomics data and plotting true positive rate vs. false positive rate, for prediction evaluated for extremely PTB ( ⁇ 28 weeks of gestation) or very PTB ( ⁇ 32 weeks of gestation). Extremely and very PTB, in which sPTB happens even sooner during pregnancy, are considered more severe forms of sPTB and are associated with worse outcomes.
- Fig. 11G is a graph showing ROC curve comparing sPTB prediction accuracy based on microbiome data, for prediction evaluated for extremely PTB or very PTB.
- Fig. UH is a graph showing PR curve evaluating the performance of the metabolomics-based predictor on one of two external cohorts.
- Fig. HI is a graph showing PR curve evaluating the performance of the metabolomics-based predictor on the second of the two external cohorts.
- Fig. 1 11 is a SHAP-based effect plot for the top 10 features in the microbiome model, sorted with descending importance.
- the levels of 748 metabolites were measured from vaginal samples collected in the second trimester of 232 pregnant women, for whom the composition of the microbiota was previously characterized using 16S rRNA gene amplicon sequencing.
- the vaginal metabolome partially corresponds to CSTs.
- vaginal metabolome measured early in pregnancy, can be used to accurately predict subsequent preterm birth, more accurately than predictions based on microbiome or clinical data alone.
- Our results demonstrate a promising new approach for further analyzing potential causes of premature birth, as well as for early risk stratification, and highlight the need to study environmental exposures as a potential factor in sPTB.
- biological marker means at least one metabolite of a woman’s vaginal metabolome, and/or at least one biological component of the woman’s vaginal microbiome.
- a method of predicting preterm birth is provided, based on obtaining a vaginal ecosystem sample from a pregnant woman, testing the sample to determine the presence of at least one biological marker, or of a combination of biological markers, associated with preterm birth, and predicting a preterm birth if the at least one biological marker or the combination of biological markers is detected at level(s) above a pre-determined threshold.
- the sample may be taken from the pregnant woman at any appropriate time as determined by the practitioner, such as, for example, at about 20-24 weeks of gestation.
- the at least one biological marker, or the combination of biological markers may include at least one metabolite of the pregnant woman’s vaginal metabolome, and or at least one biological component of the pregnant woman’s vaginal microbiome.
- the biological marker being considered could be, for example, one or more of tartrate, ethyl glucoside, mannose, arabinose, mannitol, sorbitol, EDTA, tyramine, pipecolate, orotidine, (R)-hydroxybutyl carnitine, glutamate, gamma-methyl ester, 3--amino-2-piperidone, and diethanolamine.
- Another embodiment provides a method of reducing the risk of preterm birth in a pregnant woman by reducing the pregnant woman’s exposure to one or more metabolites associated with preterm birth, exogenous or otherwise.
- the one or more metabolites could include, for example, ethyl glucoside, tartrate, EDTA, or diethanolamine.
- the method could include first determining that preterm birth is more likely for the pregnant woman than for an average pregnant woman based on analysis of a vaginal ecosystem sample obtained from the pregnant woman.
- the method could also include determining whether the pregnant woman’s race, such as by self-identification, affects the analysis of the woman’s vaginal ecosystem.
- the method could also include administering to the woman a supplementation of one or more metabolites associated with term birth.
- Another embodiment provides a method of reducing the risk of preterm birth in a woman, comprising determining the woman’s race, predicting a preterm birth based on the woman’s race, and administering to the woman a supplementation of one or more metabolites associated with term birth, or reducing the woman’s exposure to metabolites - exogenous or otherwise - associated with preterm birth, or both.
- the method could also include obtaining a vaginal ecosystem sample from the woman, and testing the sample for presence of at least one biological marker, or a combination of biological markers, associated with preterm birth.
- the method could be applied to a woman known to be pregnant, or to a woman not pregnant or not known to be pregnant.
- vaginal samples collected between 20-24 weeks of gestation from women with singleton pregnancies were profiled using mass spectrometry.
- the vaginal metabolome partially preserves CST structure
- vaginal microbiome clusters to well-defined community state types (CSTs). See, Ravel, J. el al., Vaginal microbiome of reproductive- age women, Proc. Natl. Acad. Set. 108, 4680-4687 (2011). We demonstrated the same for this cohort. See Fig. 1A (PERMANOVA p ⁇ 0.001 for separation between CSTs). See also Fig. 7C (stratified by race).
- Figure 6C is a heatmap showing metabolite levels for each subject (rows) and metabolite (columns), with the subjects sorted by their assigned metabolites cluster (MC) and metabolites clustered hierarchically using Canberra distance and Ward linkage. The color above each column reflects metabolite annotations.
- Figure 6D, 6E, and 6F show PCA (Fig. 6D), Canberra distance-based PCoA (Fig. 6E), and t-SNE (Fig. 6F) metabolome plots colored by MCs A to F.
- Figure 6G depicts a histogram of consistency of MC assignment, defined as the fraction of samples assigned to the same MC (x-axis) in 100 iterations in which we randomly selected 90% (209 women) of the cohort, and generated 6 metabolite clusters de novo. The analysis shows that many of the iterations (29 iterations, 29%) had over 95% consistency, with an overall mean consistency of 84%.
- Figure 7B is a chart similar to that in Fig. ID, correlating the distribution of CSTs for each MC but stratified by race.
- Figures 7C and 7D are similar to Figs. 1A and IB, showing a microbiome UMAP (70 or a metabolome UMAP (7D) colored by race.
- Figures 7E and 7F are similar to Figs.
- Figure 7G is similar to Fig. 1G, showing frequency of pregnancy outcome (TB vs. sPTB) for black women with early sPTB (gestational age at birth ⁇ 32 weeks).
- Xenobiotics are overrepresented among metabolites associated with MC-C (Fisher's exact p - 0.0014, FDR ⁇ 0.1).
- MC-F is composed entirely of CST IV
- MC-E is evenly split (50 CST IV). See Figs. ID, 7A. Reciprocally, we found various enrichmentsof CSTs in
- metabolites are significantly associated with sPTB (Mann-Whitney U test/? ⁇ 0.05, False Discovery Rate (FDR) ⁇ 0.1; see Fig. 2 A (showing only metabolites with at least one association with FDR ⁇ 0.1 ; metabolites are sorted by their average signed (direction of fold change) logp - value). See also Fig. 8 A. Three of these metabolites, all higher in women who delivered preterm (p ⁇ 10 3 , FDR ⁇ 0.1 for all; see Fig.
- ethyl glucoside ethyl beta-glucopyranoside
- diethanolamine DEA; p ⁇ 10 i0 ), commonly used in personal care and cosmetic products, and which was shown, in mice, to cause liver and kidney tumors, as well as induce apoptosis in fetal hippocampus cells, decrease neural progenitor cell mitosis in the hippocampus, and reduce littersize in a dose dependent manner.
- Figure 8A depicts a series of box and swarm plots (line, median; box, IQR (Interquartile Range); whiskers, 1.5*IQR) of the levels of metabolites associated with sPTB, comparing preterm and term deliveries, and stratifying by maternal self-identified race (p -- Mann Whitney).
- Figure 2B shows box and swarm plots (line, median; box, IQR; whiskers, 1.5*IQR) of three metabolites with significant associations with sPTB (p - Mann- Whitney U).
- Gioline is an essential nutrient required for membrane phospholipids and neurotransmitter synthesis, and lower choline levels were previously found in cord blood from premature infants.
- DEA diethanolamine
- Metabolite associations with sPTB interact with race and sPTB timing
- Phospholipids, however, and specifically palmitoyl sphingomyelin were recently found to be increased in placental tissue of sPTB deliveries.
- Citraconate a derivative of citric acid, was likewise negatively associated with extremely PTB (p - 0.0014), and was previously found to be present in significantly lower concentrations in placental mitochondria of women with severe preeclampsia.
- EDTA is another metabolite that can be found in cosmetic products, where it acts as a chelating agent. Potentially at different concentrations that those measured here, EDTA has also been shown to be cytotoxic in vaginal epithelial cells, in which it provokes an inflammatory response, and is teratogenic, causing fetal gonadal dysgenesis in rats at non-matemotoxic doses. We note that while EDTA is also present in the buffer in which our samples were collected, this is unlikely to explain the observed association. Overall, we find that the associations between metabolites and sPTB interact with both race and timing of sPTB. Functional metabolite sets are enriched with sPTB -associated metabolites
- Figure 8D is a heatmap showing metabolite sets altered in sPTB in various subsets, broken out for all women, black women, white women, black women with very PTB, and black women with extreme PTB (colors correspond to p- value of metabolite set enrichment analysis; only associations with FDR ⁇ 0.1 are shown).
- Figures 3A address microbe-metabolite correlations with sPTB.
- Figure 3A shows a network of microbial correlations with metabolites associated with sPTB (ellipses, microbial species; blue and red diamonds, metabolites enriched in TB and sPTB, respectively; blue and red edges, negative and positive Spearman correlations with FDR ⁇ 0.1, !pl > 0.25, respectively; edge width, median p). (See Fig. 9A for the same network without grouped nodes).
- Figure 9A to 9F generally show networks of microbial correlations with PTB- associated metabolites.
- Figures 9A all women
- 9C black women
- 9D white women
- 9E black women with extremely or very FI B
- Figure 9B is a volcano plot where every point represents a microbe-metabolite association.
- the x-axis displays the difference between spearman p’s calculated separately among black and white women.
- the y-axis displays the significance of the difference, using Fisher’s R-to-z transform.
- the gold points indicate associations where there is a difference in sign between the correlations among black and white women.
- Figure 9F is similar to Fig. 9B, showing differences in associations in black women with extremely or very preterm births, and the rest, of the black women subjects.
- vaginal microbiome may be more strongly associated with earlier sPTB
- vaginal microbes were correlated with metabolites associated with these earlier sPTBs in samples from black women (Fig. 2D).
- Fig. 2D We identified significant associations between only two microbes, Anaerococcus prevotii and Shigella flexneri, and two metabolites, EDTA and the unnamed metabolite X-l 1381, which were elevated in early PTB and TB, respectively (Fig. 2D). Both A. prevotii and 5.
- S. flexneri has been previously associated with both vaginitis and preterm labor, and its correlation with early-PTB- associated metabolites may reflect the increased incidence of reproductive tract infections in earlier PTBs.
- Fig. 9F we investigated whether the vaginal microbiome of women who subsequently had early sPTB has different associations with sPTB-associated metabolites.
- Microbisme metabolic models support microbial production of tyramine
- community-level metabolic models to predict the metabolic output of each microbiome sample (community net maximal production capacity [NMPC]).
- NMPC community net maximal production capacity
- These models combine genetic and biochemical knowledge to generate predictions of metabolite output, using only microbial relative abundances.
- Figures 4A to 4D relate to prediction of subsequent sPTB using metabolomics, microbiome, or clinical data.
- Figure 4D shows the effect on total prediction (SHAP- (SHapley Additive exPlanations-) based) for the 10 most predictive metabolites in the metabolomics-based predictor, sorted with descending importance. Each dot represents a specific sample, with the color corresponding to the relative level of the metabolite in the sample compared to all other samples.
- HTP- SHapley Additive exPlanations-
- Figures 11A to 11J relate to performance and features of prediction models for sPTB.
- Figure 11A shows a receiver operating characteristic (ROC) curve comparing the performance of different sPTB prediction algorithms on metabolornics data.
- Figure 11 B shows a ROC curve comparing the performance of a composite model for all women, compared to results stratified for race (black women and non-black women).
- the model trained on samples from all women achieves the same accuracy as the model trained only on samples from black women, when evaluated in 10-fold cross-validation on sPTB prediction for black women (auROC of 0.83 for both).
- a different model is learned on each subgroup, models trained separately on each subgroup do not generalize well to the other subgroup (auROC of 0.63 and 0.66). This demonstrated the utility of training a predictor stratified for race.
- Figure 1 IE shows a SHAP- based (SHapley Additive exPlanations-based) effect on total prediction (x-axis) for the top 10 features used in the combination models, sorted with descending importance.
- FIGS. 11F and 11G show ROC curves for the same metabolomics-based (1. I F) and microbiome-based (11G) models as in Figs. 4A and 4B, for extremely ( ⁇ 28 weeks of gestation) and very ( ⁇ 32 weeks) PTB.
- Figures 11H and 1 11 show precision-recall curves for sPTB prediction based on two external cohorts, using the metabolomics-based predictor without retraining or adaptation.
- Figure 11 J is similar to Fig. HE, for the microbiome-based model.
- vaginal metabolites were measured in a cohort of 232 pregnant women.
- vaginal metabolome largely separates by microbial community state types, but that de novo clustering of the metabolome reveals clusters enriched for sPTB among black women.
- Our results highlight several exogenous metabolites with strong associations with sPTB, which we suggest constitute novel risk factors.
- microbe- metabolite associations we uncover interesting interactions between metabolites associated with term birth and potentially suboptimal microbes, and propose a difference between the metabolism of tyramine in the vaginal microbiome of white women who delivered preterm. Finally, we demonstrate that supervised learning models trained on metabolomics data can accurately predict subsequent sPTB, potentially paving the way for new diagnostic panels.
- the cohort analyzed here includes a majority of black women, offering an opportunity to study preterm birth in women who are disproportionately burdened by it and other adverse pregnancy outcomes, while at the same time are often represented in small numbers in many cohort studies.
- the enrichment of sPTB associations among the xenobiotic metabolite set in black women may potentially reflect disparities in environmental and exogenous exposures, consistent with reports that black women may have greater exposures to endocrine disrupting chemicals through personal care products and consistent with other studies that have identified exogenous chemicals as possible drivers of PTB.
- Metabolomic exposure patterns could differ between cohorts, and could contribute to the association between racial disparities in prematurity rates and racial differences in the vaginal microbiome. Further study is warranted to identify the sources of these metabolites and disentangle their effects on the host, microbiome, and pregnancy outcomes.
- M&M Motherhood & Microbiome
- vaginal microbiota of 503 women was characterized via 16S rRNA gene amplicon sequencing (V3-V4 region) of vaginal swabs collected between 20 to 24 weeks of gestation, and total bacterial load was assessed using the TaqMan® BactQuant assay.
- V3-V4 region the vaginal microbiota of vaginal swabs collected between 20 to 24 weeks of gestation.
- total bacterial load was assessed using the TaqMan® BactQuant assay.
- Metabolite levels were measured from vaginal swabs by Metabolon Inc. (Durham, NC, USA), using an untargeted LC/MS platform. Metabolite measurements were volume normalized, followed by robust standardization of the log (base 10) transformed values (subtracting the median and dividing by the standard deviation calculated while clipping the top and bottom 5% of outliers).
- PERMANOVA analysis was performed using Bray-Curtis distance for microbiome data, and the Canberra dissimilarity metric for metabolites data. De novo clustering of metabolite vectors was done using K-medoids algorithm, also with Canberra dissimilarity. This metric is robust to outliers and sensitive to differences in common features; used with metabolomics data, it has previously produced robust results under bootstrapping and generated compact dusters corresponding with prior knowledge. We determined the optimal number of clusters by comparing the within cluster sum of square error and the gap statistic for clustering solutions with K between 1 and 15 (see Figs. 6A and 6B).
- UMAP Uniform Manifold Approximation and Projection
- Microbiome metabolic modeling was done using Microbiome Modeling Toolbox (COBRA toolbox commit: 71 cl 17305231 I77a0292856e292b95ab32040711), using models from AGORA2. All computations were performed in MATLAB version 2019a (Mathworks, Inc.), using the IBM CPLEX (IBM, Inc.) solver.
- NMPCs Net Maximal Production Capacities
- Metabolic modeling requires environmental conditions such as media and carbon source availability.
- This vaginal media was applied to each microbiome model input compartment in the form of constraints on metabolite uptake reactions, constraining uptake of compounds not present in the environment to zero.
- Samples were split into training and test sets using 10-fold cross validation (“outer folds”), block-stratified for deciles of gestational age at birth (GAB), and for microbiome, metabolomics, and combined models, also stratified for race. To account for stochasticity in the division to 10 folds, we repeated this process 5 times. Train-test sterility was strictly maintained. To tune the optimal set of hyperparameters (including parameters for feature engineering and selection), and to obtain a robust estimate of the generalization error, we used nested cross-validation. In this extension of the training-test-validation framework, the training set was further split to 5 folds (“inner folds”), on which we used 1,000 iterations of a random set of hyperparameters.
- hyperparameter set As the model with the top average R 2 score out of the top 10 most accurate models based on average auROC for sPTB classification, based on performance on the inner folds. We then used these hyperparameters to train a model on the entire training data for the outer fold, and evaluated it on the held- out test data.
- hyperparameters are selected using strictly the training data of each outer 10-fold cross-validation fold, and are evaluated just once on the test set.
- Our prediction pipeline included standardization and imputation (for metabolomics data), optional PC A transformation, and feature selection using sparsity, SNAP feature importance, information gain and/or Spearman correlation, followed by prediction using LightGBM, with ail steps performed strictly using training data.
- vaginal metabolome for early prediction of spontaneous premature labor is not limited to the specific embodiments described above, but encompasses any and all embodiments within the scope of the generic language of the following claims enabled by the embodiments described herein, or otherwise shown in the drawings or described above in terms sufficient to enable one of ordinary skill in the art to make and use the claimed subject matter.
Abstract
The method of predicting human preterm birth is based on analysis of the vaginal microbiome and metabolome of the mother. A sample is obtained from a woman and tested to determine the presence and quantity of at least one biological marker, or a combination of biological markers, associated with preterm birth. Preterm birth may then be predicted through a combination of the levels of various biomarkers. The sample taken from the woman may be a vaginal ecosystem sample, and the at least one biological marker may be at least one metabolite of the woman's vaginal metabolome and/or at least one biological component of the woman's vaginal microbiome.
Description
USING THE VAGINAL METABOLOME
FOR EARLY PREDICTION OF SPONTANEOUS PRETERM BIRTH
CROSS-REFERENCE TO RELATED APPLICATIONS
This application claims the benefit of U.S. Provisional Patent Application No. 63/210,231, filed on June 14, 2021; U.S. Provisional Patent Application No. 63/178,666, filed on April 23, 2021 ; and U.S. Provisional Patent Application No. 63/137,956, filed on January 15, 2021.
GOVERNMENT LICENSE RIGHTS
This invention was made with government support under grant no. R01NR014784, awarded by the National Institutes of Health. The government has certain rights in the invention.
TECHNICAL FIELD
The disclosure of the present patent application relates to identifying and using vaginal microbiome metabolites as early biomarkers for identifying potential risk factors and predicting spontaneous preterm birth.
BACKGROUND ART
Preterm birth (PTB), defined as childbirth before 37 weeks of gestation have passed, is the leading cause of neonatal death, and may lead to a variety of gastrointestinal, neurological, and other lifelong morbidities. Despite extensive medical and research efforts to prevent PTB and ameliorate its consequences, its prevalence remains high, both globally and specifically in the United States.
PTB also reflects a significant racial disparity, manifesting in a substantially higher risk for PTB in black women compared to non-Hispanic white women. Spontaneous preterm birth (sPTB) - preterm birth not arising due to medical indication - accounts for two thirds of all PTBs. Nevertheless, despite extensive study, a great need remains for consistent and reliable methods for early prediction, prevention, or treatment of sPTB.
The human microbiome is a strong biomarker of many complex diseases, often
predicting host phenotypes even better than host genetics. The vaginal microbiome, specifically, is a promising area of research for early diagnosis of sPTB: studies in multiple cohorts, clinical settings, and populations have found various associations with sPTB and other adverse pregnancy outcomes. See, for example, Brown, R. G. et al., Vaginal dysbiosis increases risk of preterm fetal membrane rupture, neonatal sepsis and is exacerbated by erythromycin, BMC Med. 16, 9 (2018); Callahan, B. J. etal., Replication and refinement of a vaginal microbial signature of preterm birth in two racially distinct cohorts of US women, Proc. Natl. Acad. Sci. 114, 9966-9971 (2017); Elovitz, M. A. et al., Cervicovaginal microbiota and local immune response modulate the risk of spontaneous preterm delivery, Nat. Commun. 10, 1305 (2019); Fettweis, J. M. et al., The vaginal microbiome and preterm birth, Nat. Med. 25, 1012-1021 (2019); DiGiulio, D. B. et al., Temporal and spatial variation of the human microbiota during pregnancy. Proc. Natl. Acad. Sci. U. S. A. 112, 11060— 11065 (2015); Romero, R. et al., The vaginal microbiota of pregnant women who subsequently have spontaneous preterm labor and delivery and those with a normal delivery at term, Microbiome 2, 18 (2014).
However, while multiple associations have been shown, a clear consensus on the relationship between the vaginal microbiome and sPTB has yet to emerge. See, Bayar, E., Bennett, P. R_, Chan, D., Sykes, L. & MacIntyre, D. A., The pregnancy microbiome and preterm birth, Semin. Immunopathol. 42, 487-499 (2020). Several studies of the vaginal microbiome have reported associations between sPTB and microbial diversity or microbiome community state types (CSTs), but many of these associations have not generalized across cohorts. Some of these discrepancies may be linked to the underlying structure of the population studied. For example, Callahan et al. found lower Lactobacillus and higher Gardnerella abundances to be associated with sPTB risk in a low risk, predominately white cohort, but not in a high risk, predominantly African American cohort.
Callahan, B. J . et al., Replication and refinement of a vaginal microbial signature of preterm birth intwo racially distinct cohorts of US women, Proc. Natl. Acad. Sci. 114, 9966-71 (2017). In addition, our knowledge of specific mechanisms underlying potential hostmicrobiome interactions in sPTB still lacks much detail, providing little real guidance. In general, metabolites produced or modified by the microbiome have emerged as a prominent factor with other potential local and systemic effects on the host. Metabolomics - the study of the metabolites present in an organism, cell, or tissue - enables the measurement of thousands of small molecules present in a particular ecosystem, and has
revealed intriguing new insights on host-microbiome and other interactions in a variety of contexts, from colorectal cancer to diabetes.
A few studies of the vaginal metaboiome have demonstrated association with sPTB. For example, see Ghartey, J., Bastek, J. A., Brown, A. G., Anglim, L. & Elovitz, M. A., Women with preterm birth have a distinct cervicovaginal metaboiome, Am. J. Obslet. Gynecol. 212, 776<el-12 (2015); Ghartey, J., Anglim, L., Romero, J., Brown, A. & Elovitz, M. A. Women with Symptomatic Preterm Birth Have a Distinct Cervicovaginal Metaboiome, Am. J. Perinatal. 34, 1078-1083 (2017). However, these studies had limited sample sizes, and were not paired with measurements of the microbiome. Such paired microbiome-metabolome studies have yielded potential mechanistic insights in other pathologies, including inflammatory bowel disease and HPV infection. See, Lloyd-Price, J. et al., Multi-omics of the gut microbial ecosystem in inflammatory bowel diseases, Nature 569, 655-62 (2019); Borgogna, J. C. el al., The vaginal metaboiome and microbiota of cervical HPV-positive and HPV-negative women: a cross- sectional analysis, BJOG bit. J. Obstet. Gynaecol. 127, 182-192 (2020).
The few paired studies of the vaginal metaboiome and microbiome performed to date, however, suffer from a low representation of the demographic groups at highest risk for sPTB, identified only a limited set of metabolites, and did not generate robust prediction models for sPTB. Even so, the associations reported between both the vaginal microbiome, metaboiome and sPTB highlight the need for paired and predictive analysis in diverse cohorts in order to advance our understanding of the role of this ecosystem in pregnancy outcomes.
DISCLOSURE
Spontaneous preterm birth (sPTB) is a leading cause of maternal and neonatal morbidity and mortality. Nevertheless, available means for both its prevention and early risk stratification are limited. The vaginal microbiome has been associated with PTB risk, possibly via metabolic or other interactions with its host.
Here, we analyzed the levels of 748 metabolites which were measured from vaginal samples collected in the second trimester of 232 pregnant women for whom the composition of the microbiota was previously characterized using 16S rRNA gene amplicon sequencing. Elovitz, M. A. et al., Cervicovaginal microbiota and local immune response modulate the risk of spontaneous preterm delivery, Nat. Commitn. 10, 1305 (2019).
Samples were collected at zO-24 weeks of gestation from women with singleton pregnancies, of which 80 delivered spontaneously before 37 weeks of gestation. The vaginal metabolome was found to correlate with the microbiome, while it separates into six clusters, three of which are associated with spontaneous preterm birth (sPTB) in black women. Furthermore, while we identify five metabolites that associate with sPTB, another five metabolites associate with sPTB only when stratified by race. We identify multiple microbial correlations with metabolites associated with sPTB, including intriguing correlations between vaginal bacteria that are considered sub-optimal and metabolites that were enriched in women who delivered at term.
Accordingly, we show that the vaginal metabolome partially corresponds to CSTs, and reveal novel associations between metabolites measured in the middle of pregnancy and subsequent sPTB. Exploring the potential origins of these metabolites, we propose that some are of an exogenous source, suggesting a novel risk factor and potential targets for novel prevention strategies. Finally, using the vaginal metabolome, measured an average of 3 months before delivery, we devise machine learning algorithms that accurately predict subsequent preterm birth, weeks to months in advance, with a high degree of accuracy. This predictor, more accurate than predictors based on microbiome or clinical data, is validated on two external cohorts. Our results demonstrate a promising new approach for studying potential causes of prematurity as well as for early risk stratification, and highlight the need to study environmental exposures as a potential factor in sPTB.
These and other features of the present subject matter will become readily apparent upon further review of the following specification.
BRIEF DESCRIPTION OF THE DRAWINGS
Fig. 1A is a microbiome Uniform Manifold Approximation and Projection (UMAP) colored by community state types (CSTs) (N = 503). CST I indicates domination by Lactobacillus crispatus; CST II is dominated by Lactobacillus gasseri; CST III is dominated by Lactobacillus triers', CSTs IV- A and IV -B contains diverse anaerobes; and CST-V is dominated by Lactobacillus jensenii.
Fig. IB is a metabolomics UMAP colored by CSTs (N = 232).
Fig. 1C is a metabolomics UMAP plot colored by metabolites clusters (MCs) A to F (N = 232). The metabolite sub-pathway most enriched among metabolites associated with each MC were polyamine metabolism, dipeptides, dicarboxylated fatty acids, glutamate metabolism, TCA cycle, and dipeptides for MC A-F, respectively
(hypergeometric p ~ 2.4x10-4, 1.1x10-4, 0.025, 0.02, 0.015 and 3.6x10-4, respectively). Furthermore, amino-acid-related ffletabolit.es are overrepresented among metabolites significantly associated with MC-A, MC-B, and MC-D (Mann-Whitney U p < 0.05) compared to other MCs (Fisher’s exact p = 4.3x10-8, 0.0011 and 1.8x10-8, respectively; FDR < 0.1 for all) and xenobiotics are overrepresented among metabolites associated with MC-C (Fisher’s exact p = 0.0014, FDR < 0.1).
Fig. ID is a chart correlating the distribution of CSTs for each MC. The number above the horizontal lines is Fisher’s exact p, FDR (False Discovery Rate) < 0.1.
Fig. IE is a chart correlating the relative MC distribution for black women and for white women.
Fig. IF is a chart showing frequency of pregnancy outcome (full term birth (TB) vs. sPTB) broken out by CST in white women and in black women. This Fig. I F is a replication of an analysis done in Elovitz, M. A. el al., Cervicovaginal microbiota and local immune response modulate the risk of spontaneous preterm delivery, Nat. Commun. 10, 1305 (2019). The figure shows that CST IV- A is enriched for sPTB among white women.
Fig. 1G is a chart showing frequency of pregnancy outcome (full term birth (TB) vs. sPTB) broken out by MC in white women and in black women. The number abo ve the horizontal lines is Fisher’s exact. p, FDR < 0.1.
Fig. 2 A is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes, for all women, and broken out for black women and for white women.
Fig. 2B shows box and swarm plots of three metabolites with significant, associations with sPTB - diethanolamine, choline, and betaine.
Fig. 2C is an illustration depicting common wisdom that diethanolamine (DEA) (here found to be associated with sPTB) inhibits choline uptake, while choline and betaine (both found here to be associated with TB) are important for membrane lipid synthesis and osmoregulation.
Fig. 2D is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes for black women, broken out by gestational age at birth (GAB).
Fig. 3 A depicts a network of microbial correlations with metabolites associated with sPTB. Ellipses depict microbial species; diamonds depict metabolites enriched in TB and sPTB, respectively.
Fig. 3B shows box and swarm plots of tyramine levels as measured, comparing preterm deliveries (blue boxes) and term deliveries (red boxes) in self-identified white and black women.
Fig. 3C shows box and swarm plots of tyramine levels as predicted using metabolic models, comparing preterm deliveries (red boxes) and term deliveries (blue boxes) in selfidentified white and black women.
Fig. 3D is a plot of tyramine production derived from microbiome metabolic models against measured tyramine levels, colored by race, and by birth outcome for white women. Fig. 4A is a graph showing receiver operating characteristic (ROC) curves comparing sPTB prediction accuracy based on clinical, microbiome, and metabolomics data and plotting true positive rate vs. false positive rate.
Fig. 4B is a graph showing precision-recall (PR) curves comparing sPTB prediction accuracy based on clinical, microbiome, and metabolomics data. Fig. 4C is a graph showing ROC curves evaluating the performance of the metabolomics-based predictor on two external cohorts, comparing true positive rate to false positive rate.
Fig. 4D depicts the effect on total prediction for the 10 most predictive metabolites in the metabolomics-based predictor, sorted with descending importance. Fig. 5A is a graph showing the distribution of metabolite super pathways among assayed metabolites.
Fig. 5B is a graph showing the distribution of metabolite prevalence across samples.
Fig. 6A is a graph plotting within cluster sum of squared distances for k-medoids clustering using Canberra distances for k from 1 to 15. Fig. 6B is a graph plotting gap statistic for k-medoids clustering using Canberra distances for k from 1 to 15.
Fig. 6C is a heatmap showing metabolite levels for each subject, sorted by assigned metabolites cluster (MC) with metabolites clustered hierarchically using Canberra distance and Ward linkage. Fig. 6D is a Principal Component Analy sis (PCA) plot of metabolites data colored by MCs A to F.
Fig. 6E is a Canberra distance-based Principal Coordinate Analysis (PCoA) plot of metabolites data colored by MCs A to F.
Fig. 6F is a iSNE (t-disiributed Stochastic Neighbor Embedding) plot of metabolites data colored by MCs A to F.
Fig. 6G is a histogram of consistency of MC assignment.
Fig. 7 A is a chart showing CST distribution within each MC, for all women, white women, and black women.
Fig. 7B is a chart correlating the distribution of CSTs for each MC, separately for black women and for white women.
Fig. 7C is a microbiome UMAP colored by race.
Fig. 7D is a metabolomics UMAP plot colored by race. Fig. 7E is a chart showing frequency of pregnancy outcome (TB vs. sPTB) broken out by various CSTs for all women combined.
Fig. 7F is a chart showing frequency of pregnancy outcome (TB vs. sPTB) broken out by MC for all women combined.
Fig. 7G is a chart showing frequency of pregnancy outcome (early sPTB) broken out by MC in black women.
Fig. 8A is a set of box and swarm plots of the levels of metabolites associated with sPTB, comparing preterm and term deliveries, and stratified by self-identified race.
Fig. 8B is a kernel density estimation showing the distribution across all samples of four xenobiotics associated with sPTB or early sPTB. Fig. 8C is a heatmap showing statistically significant associations between specific metabolite measurements and birth outcomes, for women not treated with progesterone, and also broken out for black women and for white women.
Fig. 8D is a heatmap showing metabolite sets altered in sPTB in various subsets.
Fig. 9 A depicts a network of microbial correlations with metabolites associated with sPTB. Ellipses depict microbial species; diamonds depict metabolites enriched in TB and sPTB, respectively. This is similar to Fig. 3A, but without the grouped nodes - each microbial taxa is instead represented as an individual node.
Fig. 9B is a volcano plot with every point representing a comparison of a specific microbe-metabolite correlation between black and white women. Fig. 9C depicts a network of microbial correlations with metabolites associated with sPTB, with individual nodes, for black women.
Fig. 9D depicts a network of microbial correlations with metabolites associated with sPTB, with individual nodes, for white women.
Fig. 9E depicts a network of microbial correlations with metabolites associated with early sPTB, with individual nodes, for black women with early sPTB.
Fig. 9F is a volcano plot with every point representing a comparison of a specific microbe-metabolite correlation between black women with early sPTB and all other black women.
Fig. 10A is a plot comparing prediction of microbiome putrescine from microbiome metabolic models (net maximal production capacity, or NMPC) against measured metabolite levels.
Fig. 10B is a plot comparing prediction of microbiome histamine from microbiome metabolic models (NMPC) against measured metabolite levels.
Fig. IOC is a plot comparing prediction of microbiome tyramine from microbiome metabolic models (NMPC) against measured metabolite levels.
Fig. 10D is a set of plots showing the percent of total sample abundance represented by metabolic models, separately for black women, white women with sPTB, and white women with TB.
Fig. 10E shows the correlation between NMPCs and metabolite measurements for models that contain a maximum of N most abundant species across the cohort.
Fig. 11 A is an ROC curve comparing performance of different sPTB prediction algorithms on metabolomics data. Fig. 11B is an ROC curve comparing performance of models stratified for race against a model based on all samples (and not stratified for race).
Fig. 11C is an ROC curve evaluated in nested cross-validation, comparing sPTB prediction accuracy for models based on metabolomics data alone and on metabolomics data combined with microbiome and clinical data. Fig. 11 D is a graph showing precision-recall (PR) curve evaluated in nested cross- validation, comparing sPTB prediction accuracy for models based on metabolomics data alone and on metabolomics data combined with microbiome and clinical data.
Fig. 1 IE is a SHAP-based (SHapley Additive exPlanations) effect plot for the top 10 features in the combination models, sorted with descending importance. Fig. 1 IF is a graph showing ROC curves comparing sPTB prediction accuracy based on metabolomics data and plotting true positive rate vs. false positive rate, for prediction evaluated for extremely PTB (<28 weeks of gestation) or very PTB (<32 weeks of
gestation). Extremely and very PTB, in which sPTB happens even sooner during pregnancy, are considered more severe forms of sPTB and are associated with worse outcomes.
Fig. 11G is a graph showing ROC curve comparing sPTB prediction accuracy based on microbiome data, for prediction evaluated for extremely PTB or very PTB. Fig. UH is a graph showing PR curve evaluating the performance of the metabolomics-based predictor on one of two external cohorts.
Fig. HI is a graph showing PR curve evaluating the performance of the metabolomics-based predictor on the second of the two external cohorts.
Fig. 1 11 is a SHAP-based effect plot for the top 10 features in the microbiome model, sorted with descending importance.
Similar reference characters denote corresponding features consistently throughout the attached drawings.
DESCRIPTION OF EMBODIMENTS The levels of 748 metabolites were measured from vaginal samples collected in the second trimester of 232 pregnant women, for whom the composition of the microbiota was previously characterized using 16S rRNA gene amplicon sequencing. The vaginal metabolome partially corresponds to CSTs.
There are novel associations between metabolites measured early in pregnancy and subsequent sPTB . Exploring the potential origins of some of these metabolites, some appear to be of an exogenous source, suggesting a novel risk factor and potential targets for novel prevention strategies. On this basis, the vaginal metabolome, measured early in pregnancy, can be used to accurately predict subsequent preterm birth, more accurately than predictions based on microbiome or clinical data alone. Our results demonstrate a promising new approach for further analyzing potential causes of premature birth, as well as for early risk stratification, and highlight the need to study environmental exposures as a potential factor in sPTB.
As used herein, the term “biological marker’ means at least one metabolite of a woman’s vaginal metabolome, and/or at least one biological component of the woman’s vaginal microbiome.
In one embodiment, a method of predicting preterm birth is provided, based on obtaining a vaginal ecosystem sample from a pregnant woman, testing the sample to determine the presence of at least one biological marker, or of a combination of biological
markers, associated with preterm birth, and predicting a preterm birth if the at least one biological marker or the combination of biological markers is detected at level(s) above a pre-determined threshold. The sample may be taken from the pregnant woman at any appropriate time as determined by the practitioner, such as, for example, at about 20-24 weeks of gestation. The at least one biological marker, or the combination of biological markers, may include at least one metabolite of the pregnant woman’s vaginal metabolome, and or at least one biological component of the pregnant woman’s vaginal microbiome. The biological marker being considered could be, for example, one or more of tartrate, ethyl glucoside, mannose, arabinose, mannitol, sorbitol, EDTA, tyramine, pipecolate, orotidine, (R)-hydroxybutyl carnitine, glutamate, gamma-methyl ester, 3--amino-2-piperidone, and diethanolamine.
Another embodiment provides a method of reducing the risk of preterm birth in a pregnant woman by reducing the pregnant woman’s exposure to one or more metabolites associated with preterm birth, exogenous or otherwise. The one or more metabolites could include, for example, ethyl glucoside, tartrate, EDTA, or diethanolamine. The method could include first determining that preterm birth is more likely for the pregnant woman than for an average pregnant woman based on analysis of a vaginal ecosystem sample obtained from the pregnant woman. The method could also include determining whether the pregnant woman’s race, such as by self-identification, affects the analysis of the woman’s vaginal ecosystem. The method could also include administering to the woman a supplementation of one or more metabolites associated with term birth.
Another embodiment provides a method of reducing the risk of preterm birth in a woman, comprising determining the woman’s race, predicting a preterm birth based on the woman’s race, and administering to the woman a supplementation of one or more metabolites associated with term birth, or reducing the woman’s exposure to metabolites - exogenous or otherwise - associated with preterm birth, or both. The method could also include obtaining a vaginal ecosystem sample from the woman, and testing the sample for presence of at least one biological marker, or a combination of biological markers, associated with preterm birth. The method could be applied to a woman known to be pregnant, or to a woman not pregnant or not known to be pregnant.
METHODS AND ANALYSIS
Vaginal microbiota and metabalome
232 vaginal samples collected between 20-24 weeks of gestation from women with singleton pregnancies were profiled using mass spectrometry. The microbiota of these women was previously characterized from the same double shaft swab used in this study. See, Elovitz, M. A. el al., Cervico vaginal microbiota and local immune response modulate the risk of spontaneous preterm delivery, Nat. Commun. 10, 1305 (2019). Included were all women with available samples who had a subsequent spontaneous preterm delivery in the parent cohort (sPTB; N = 80), as well as similar controls who delivered at term (TB; bi ~ 152). See Table 1 (BMI = body mass index; GA = gestational age; p = Fisher's Exact or Mann- Whitney U test.)
As expected, a higher fraction of women who delivered preterm had a history of PTB compared to those who delivered at term (42.5% vs. 19.2%, respectively, Fisher’s Exact p = 3x10-4).
748 unique metabolites were quantified, of which 637 could be named. Metabolites belonged to diverse biochemical classes, including amino acids, lipids, nucleotides, carbohydrates, and xenobiotics. Most metabolites (549) were measured in over 50% of the cohort, and 108 metabolites were present in all samples. See Figs. 5A, 5B (super pathways and prevalence of assayed metabolites). Fig. 5 A depicts the distribution of metabolite super pathways among assayed metabolites. Metabolite super pathway assignments were
provided by Metabolon Inc. (Durham, NC, USA). Fig. depicts the distribution of metabolite prevalences across samples, distinguishing prevalence of all metabolites (N = 748) vs. prevalence of named metabolites (N ~ 637). The dashed lines distinguish metabolites prevalent in more than 80%' (N = 352) and more than 20% of the samples (N - 694).
We previously showed that similar metabolite measurements are in excellent agreement with measurements performed by an independent certified medical laboratory. Bar, N. el al., A reference map of potential determinants for the human serum metabolome, Nature 588, 135-140 (2020).
The vaginal metabolome partially preserves CST structure
The vaginal microbiome clusters to well-defined community state types (CSTs). See, Ravel, J. el al., Vaginal microbiome of reproductive- age women, Proc. Natl. Acad. Set. 108, 4680-4687 (2011). We demonstrated the same for this cohort. See Fig. 1A (PERMANOVA p < 0.001 for separation between CSTs). See also Fig. 7C (stratified by race).
We investigated whether the vaginal metabolome space recapitulates this structure. While the metabolome is separated by CSTs (p < 0.001; See Fig. IB) and is generally associated with the microbiome (Mantel p < 0.001), specific CSTs are not as well separated. See also Fig. 7D (stratified by race). While women with microbiomes characterized as CST I, dominated by Lactobacillus crispatus, and CST IV (IV-A and IV-B combined), characterized by diverse anaerobes, are well separated from the rest of the cohort in their metabolite measurements (PERMANOVA p < 0.001 for both), the metabolite measurements from women with CSTs IV-A and IV-B, nor CSTs II and III, were not as well separated from one another (p - 0.169 and p - 0.171, respectively). Overall, these results demonstrate a strong but imperfect correspondence between the vaginal microbiome and metabolome spaces.
Metabolome-dusters associate with sPTB
While the metabolome is significantly clustered by CSTs, this clustering is not perfectly corresponding. We therefore performed de novo clustering of the metabolome using k-medoids clustering, revealing six "metabolite-clusters" we labeled as MCs A-F. See Figs. 1C, 6A-6G, and 7A-7G.
Figures 6A and 6B show within cluster sum of squares distances (Fig. 6A) and gap statistic (Fig. 6B) for k-rnedoids clustering using Canberra distances with k from 1 to 15. A shoulder (Fig. 6A) and peak (Fig. 6B) are visible for k - 6. Figure 6C is a heatmap showing metabolite levels for each subject (rows) and metabolite (columns), with the subjects sorted by their assigned metabolites cluster (MC) and metabolites clustered hierarchically using Canberra distance and Ward linkage. The color above each column reflects metabolite annotations. Figure 6D, 6E, and 6F show PCA (Fig. 6D), Canberra distance-based PCoA (Fig. 6E), and t-SNE (Fig. 6F) metabolome plots colored by MCs A to F. Figure 6G depicts a histogram of consistency of MC assignment, defined as the fraction of samples assigned to the same MC (x-axis) in 100 iterations in which we randomly selected 90% (209 women) of the cohort, and generated 6 metabolite clusters de novo. The analysis shows that many of the iterations (29 iterations, 29%) had over 95% consistency, with an overall mean consistency of 84%.
Figure 7 A shows the distribution of CSTs within each metabolite cluster, for all women (N = 232), and separated for black women (N = 173) and for white women (N = 51). Each group of bars corresponds to a single metabolite cluster and bars within a group sum to 100%'. Figure 7B is a chart similar to that in Fig. ID, correlating the distribution of CSTs for each MC but stratified by race. Figures 7C and 7D are similar to Figs. 1A and IB, showing a microbiome UMAP (70 or a metabolome UMAP (7D) colored by race. Figures 7E and 7F are similar to Figs. IF and 1G, showing frequency of pregnancy outcome by various CSTs (7E) or by MC (7F) for all women combined. Figure 7G is similar to Fig. 1G, showing frequency of pregnancy outcome (TB vs. sPTB) for black women with early sPTB (gestational age at birth < 32 weeks).
Amino- acid-related metabolites are overrepresented among metabolites significantly associated with MC-A, MC-B, and MC-D (Mann- Whitney Up < 0.05) compared to other MCs (Fisher's exact p = 4.3x10% p - 0.0011, p = 1.8x10 -8, respectively; FDR < 0.1 for all). Xenobiotics are overrepresented among metabolites associated with MC-C (Fisher's exact p - 0.0014, FDR < 0.1). While four MCs are mostly paired with Lactobacillus dominated CSTs (54%-93% for MC A-D), MC-F is composed entirely of CST IV, and MC-E is evenly split (50 CST IV). See Figs. ID, 7A. Reciprocally, we found various enrichmentsof CSTs in
MCs (see Figs. ID, 7B) as well as enrichments for white women in MC-B and black women in MC-F (p = 0.049 and p = 0.044, respectively; see Fig. IE). These results demonstrate that the correspondence between the vaginal metabolome and microbiome is imperfect,
and somewhat corresponds to Lactobacillus dominance. See also Fig. IE (stratified by race).
A previous analysis of the same cohort showed that CST IV- A is associated with subsequent sPTB only among white women (Fisher's exact p = 0.047 ; see Figs. IF, 7E). We therefore performed the same stratified test using MCs. Interestingly, while CSTs are associated with sPTB only in white women, we find that MCs are associated with sPTB in black women, with a significant association with MC-A, MC-B, and MC-D (p = 0.047, p = 0.025, p = 0.006, respectively. See Figs. 1G, 7F.
Taken together, our results demonstrate that the general metabolome structure in our cohort better captures associations with prematurity in black women than the general microbiome structure.
Multiple metabolites associate with sPTB
Having demonstrated that the global metabolome structure associates with sPTB, we next investigated its associations with the levels of specific metabolites.
Five metabolites are significantly associated with sPTB (Mann-Whitney U test/? < 0.05, False Discovery Rate (FDR) < 0.1; see Fig. 2 A (showing only metabolites with at least one association with FDR < 0.1 ; metabolites are sorted by their average signed (direction of fold change) logp - value). See also Fig. 8 A. Three of these metabolites, all higher in women who delivered preterm (p < 103, FDR < 0.1 for all; see Fig. 2A), appear to be of exogenous source: ethyl glucoside (ethyl beta-glucopyranoside; p = 1.9x10-4), an alkyl glucoside used as a surfactant in cosmetic products; tartrate (p = 4.8x10-4), used, along with its products, in the food, pharmaceutical, and cosmetics industries; and diethanolamine (DEA; p < 10 i0), commonly used in personal care and cosmetic products, and which was shown, in mice, to cause liver and kidney tumors, as well as induce apoptosis in fetal hippocampus cells, decrease neural progenitor cell mitosis in the hippocampus, and reduce littersize in a dose dependent manner.
Figure 8A depicts a series of box and swarm plots (line, median; box, IQR (Interquartile Range); whiskers, 1.5*IQR) of the levels of metabolites associated with sPTB, comparing preterm and term deliveries, and stratifying by maternal self-identified race (p -- Mann Whitney). Figure 8B shows the distribution (kernel density estimation) of four xenobiotics associated with sPTB or early sPTB across this cohort; samples with no metabolite detected are excluded.
We further find lower levels of the amine choline in women with subsequent sPTB (p = 5.5x10 FDR < 0.1; see Figs. 2A, 2B). Figure 2B shows box and swarm plots (line, median; box, IQR; whiskers, 1.5*IQR) of three metabolites with significant associations with sPTB (p - Mann- Whitney U). Gioline is an essential nutrient required for membrane phospholipids and neurotransmitter synthesis, and lower choline levels were previously found in cord blood from premature infants. Choline is also a precursor of betaine, a metabolite mainly involved in osmoregulation which was also negatively associated with sPTB (p = 0.007, FDR = 0.14; Fig. 2B). DEA (diethanolamine) is known to disrupt several enzymes and transporters involved in choline metabolism, and its dermal administration in mice was reported to deplete hepatic stores of choline. High levels of the exogenous chemical DEA, which was higher in samples from women who delivered preterm, may be linked to the lower choline and betaine levels that we found. See Figs. 2B, 2C. Figure 2C depicts the understandings that DEA (here shown to be associated with sPTB) inhibits choline uptake, and that, choline and betaine, both associated with TB, are important for membrane lipid synthesis and osmoregulation.
Figure 8C is a heatmap showing statistically significant associations between metabolite measurements and birth outcome, for women not treated with progesterone, for all women (N = 219), and broken out for black women (N = 162) and for white women (N - 50).
Taken together, these results highlight the potential role of exogenous metabolites in premature birth, potentially from environmental exposures to hygienic and cosmetic products.
Metabolite associations with sPTB interact with race and sPTB timing
As the global metabolome structure shows differences between black and white women, we performed the same association analysis while stratifying by race. We detected five additional metabolites negatively associated with sPTB (Mann-Whitney U p < 0.05; FDR < 0.1; see Fig. 2A). For black women, these additional metabolites include glycerophosphoserine (p = 3x10-5), previously reported to be altered in preeclampsia; spermine (p ~ 2.4x104), which has immuno-modulatory roles in the gut and was increased in the blood of preterm infants; hydroxybutyl carnitine (p = 2.6x10-4), a ketocamitine
involved in lipid metabolism and which has been shown to be depleted in the blood of low birth weight full-term neonates; and glutamate gamma-methyl ester (p = 4.9x104), a derivative of glutamate and a precursor of the inhibitory neurotransmitter GABA. Tyramine, a biogenic amine with neuromodulator activity, was significantly lower in samples from white women who delivered preterm (p = 2.8x104; see Figs. 2A and 8A). Tyramine was shown to colocalize with synaptic vesicles in the mouse uterine plexus, highlighting a possible role in uterine contractions. Altogether, these results highlight the potential connection between vaginal metabolites, metabolite levels in various reproductive organs, and preterm delivery. Earlier preterm deliveries are associated with worse maternal and neonatal outcomes.
Therefore, we investigated associations between vaginal metabolites and subsequent very and extremely preterm deliveries (gestational age at birth <32 and <28 weeks, respectively). Due to the high proportion of black women among pregnancies delivering prior to 32 and 28 weeks (21 of 26 and 14 of 15, respectively), we performed this analysis only among black women, and identified 12 metabolites associated only with these earlier sPTBs. See Fig. 2D, similar to Fig. 2A but focused on very and extremely preterm deliveries for black women.
The phospholipids palmitoyl sphingomyelin and palmitoyl dihydro sphingomyelin were both negatively associated with extremely PTB (p = 0.00087 and p = 0.0011, respectively). Phospholipids, however, and specifically palmitoyl sphingomyelin, were recently found to be increased in placental tissue of sPTB deliveries. Citraconate, a derivative of citric acid, was likewise negatively associated with extremely PTB (p - 0.0014), and was previously found to be present in significantly lower concentrations in placental mitochondria of women with severe preeclampsia. Ethylenediaminetetraacetic acid (EDTA) was one of two metabolites increased in extremely PTB and very PTB (p = 0.0008 and p - 1.5x104, respectively). EDTA is another metabolite that can be found in cosmetic products, where it acts as a chelating agent. Potentially at different concentrations that those measured here, EDTA has also been shown to be cytotoxic in vaginal epithelial cells, in which it provokes an inflammatory response, and is teratogenic, causing fetal gonadal dysgenesis in rats at non-matemotoxic doses. We note that while EDTA is also present in the buffer in which our samples were collected, this is unlikely to explain the observed association. Overall, we find that the associations between metabolites and sPTB interact with both race and timing of sPTB.
Functional metabolite sets are enriched with sPTB -associated metabolites
In addition to identifying specific metabolites with strong associations with sPTB, we next checked whether certain functional groups of metabolites (e.g. KEGG pathways) are enriched for associations with sPTB, compared to all other metabolites, even if changes to any specific metabolite are small in scale. See Fig. 8D. We find significant sPTB- associated deviations in metabolites related to proline and arginine metabolism (p = 0.0018, FDR < 0.1).
Figure 8D is a heatmap showing metabolite sets altered in sPTB in various subsets, broken out for all women, black women, white women, black women with very PTB, and black women with extreme PTB (colors correspond to p- value of metabolite set enrichment analysis; only associations with FDR < 0.1 are shown).
Consistent with the association found between tyramine levels and term births among white women, see Fig. 2A, we find a global deviation of metabolites related to the endocrine system among white women (p = 0.0046, FDR < 0.1). See Fig. 8D. We further identify lipid-metabolism-related metabolites to be enriched for associations with early sPTB among black women (p - 0.0021 and p - 0.0048 for very and extremely PTB, respectively, FDR < 0.1), potentially related to other alterations in lipid metabolism reported in women who delivered preterm. Notably, we identify a global enrichment of xenobiotic metabolites associated with sPTB among black women (p = 0.006, FDR < 0.1). This is consistent with our finding of exogenous metabolites associated with sPTB in this population, and could potentially be related to the higher burden of exogenous and environmental exposures in black communities, which have been identified as potential drivers of preterm birth. In all, our analyses highlight multiple metabolites associated with sPTB, and that these associations interact with both race and sPTB severity.
A network of microbe-metabolite associations in sPTB
We next investigated associations between the absolute abundances of various microbial species and metabolites associated with sPTB. Our results replicate several known associations, such as thosebetween Dialister species or Enterococcus faecalis and tyramine (Spearman p > 0.54, p < 10-5 for all; see Figs. 3A, 9 A), as well as evidence for choline metabolism genes in Gardnerella vaginalis and Corynebacterium aurimucosum (p = 0.34, p < 10-6 and p = 0.40, p = 0.0004, respectively). Additionally, higher concentrations of tyramine were previously found in bacterial vaginosis (BV), supporting many of the
associationswe find with bacteria that are also associated with BV. See Fig. 3A. Finally, we observe a positive correlationbeiween C. aurimucosum and spermine (p = 0.27, p = 0.02), and it has been shown that spermine and its precursor spermidine are the key polyamines in several Corynebacterium species. The strongest and most numerous microbe-metabolite correlations were for tyramine
(35 associations, Spearman 0.27 < p < 0.73; see Fig. 3A) which was higher in term deliveries among white women (see Fig. 2A).
Figures 3A address microbe-metabolite correlations with sPTB. Figure 3A shows a network of microbial correlations with metabolites associated with sPTB (ellipses, microbial species; blue and red diamonds, metabolites enriched in TB and sPTB, respectively; blue and red edges, negative and positive Spearman correlations with FDR <0.1, !pl > 0.25, respectively; edge width, median p). (See Fig. 9A for the same network without grouped nodes).
Eight of the 35 tyrarnine-correlated microbes are also correlated with choline, which was enriched in term deliveries across all women. See Fig. 2 A.
Many of the species positively correlated with metabolites associated with term delivery - including Atopobium vaginae, G. vaginalis, Sneathia sanguinegens, C. aurimucosum, Mobiluncus curtisii, Actinomyces neuii, Ureaplasma urealyticum, Gemella haemolysans, several Prevotella species, Candidates Lachnocurva vaginae (BVAB1), BVAB2 and BVAB3 - were previously reported to be associated with negative outcomes, such as BV, preterm birth, and other adverse pregnancy and neonatal outcomes. We find a similarly paradoxical negative correlation between Staphylococcus epidermidis, previously shown to be associated with B V and late-onset sepsis in preterm neonates, and both tartrate and ethyl glucoside (p - -0.28, p ~ 0.00069; p = -0.26, p ~ 0.0015, respectively; see Fig. 3A), which were positively associated with sPTB. These results suggest the existence of a potential beneficial metabolic effect for some microbial populations that were previously considered dysbiotic in nature.
As many of the associations between metabolites and sPTB were modulated by race, we investigated whether the associations in our network are likewise influenced. We find that nine of the 75 microbe-metabolite associations we detected were significantly different
(Fisher's R-to-z p < 0.05; Methods) between black and white women, although a different direction of association was detected in only four of these. See Fig. 9B.
Specifically, G. vaginalis, A. vaginae, and three other species that were positively associated with tyramine, had significantly stronger associations in black women (p < 0.02 for all). Recent studies show that biofilm interactions between G. vaginalis and other microbes, including
may contribute to BV, suggesting that the differences in tyramine associations between black and white women may be related to differences in community structure and microbial interactions. Taken together, however, we find a relatively small effect of race on microbe-metabolite correlations.
Figure 9A to 9F generally show networks of microbial correlations with PTB- associated metabolites. Figures 9A (all women), 9C (black women), 9D (white women), and 9E (black women with extremely or very FI B ) are similar to Fig. 3 A, but each microbial taxa is represented as an individual node. Figure 9B is a volcano plot where every point represents a microbe-metabolite association. The x-axis displays the difference between spearman p’s calculated separately among black and white women. The y-axis displays the significance of the difference, using Fisher’s R-to-z transform. The horizontal maroon line designates p = 0.05. The gold points indicate associations where there is a difference in sign between the correlations among black and white women. Figure 9F is similar to Fig. 9B, showing differences in associations in black women with extremely or very preterm births, and the rest, of the black women subjects.
Since there is evidence that the vaginal microbiome may be more strongly associated with earlier sPTB, we investigated whether vaginal microbes were correlated with metabolites associated with these earlier sPTBs in samples from black women (Fig. 2D). We identified significant associations between only two microbes, Anaerococcus prevotii and Shigella flexneri, and two metabolites, EDTA and the unnamed metabolite X-l 1381, which were elevated in early PTB and TB, respectively (Fig. 2D). Both A. prevotii and 5. flexneri were positively associated with X-l 1381 (p = 0.36, p = 0.0004 and p = 0.36, p - 0.0002, respectively, FDR < 0.1) and negatively associated with EDTA (p = -0.37, p - 5.1x10’5 and p = -0.37, p = 0.0002, respectively, Fig. 9E). S. flexneri has been previously associated with both vaginitis and preterm labor, and its correlation with early-PTB- associated metabolites may reflect the increased incidence of reproductive tract infections in earlier PTBs. Next, we investigated whether the vaginal microbiome of women who subsequently had early sPTB has different associations with sPTB-associated metabolites (Fig. 9F). We find only one association, the negative correlation between Ca. L. vaginae (BVAB1) and glutamate gamma methyl-ester, which was significantly stronger in subjects
who subsequently had early PTB. Overall, we find a relatively small effect of sPTB timing on microbe-metabolite correlations.
Microbisme metabolic models support microbial production of tyramine To gain some mechanistic insight into the correlations we found, we next used community-level metabolic models to predict the metabolic output of each microbiome sample (community net maximal production capacity [NMPC]). These models combine genetic and biochemical knowledge to generate predictions of metabolite output, using only microbial relative abundances. Our models showed accurate predictions for several metabolites that are known to be produced by the vaginal microbiome, such as putrescine and histamine (Spearman p 0.64 between NMPCs and metabolomic measurements, N = respectively; Figs. 10A, lOB).
Two sPTB-associated metabolites were represented in our models - tyramine and choline. As our models predicted that choline was not significantly affected by the vaginal microbiome (NMPCs of 0 predicted for all women), we focused on tyramine, which previous studies suggest is produced by vaginal microbes. Following genomic curation of our metabolic models, the predictions of our models were highly accurate (Spearman p = 10C). When examining samples from white women, we find
that while the measured levels of tyramine were enriched in TB (Mann-Whitney U p = 0.00028; Fig. 3B), its predicted output by the microbiome was not, and was even somewhat higher in sPTB (p - 0.26; Fig. 3C). This stems from a lower accuracy in tyramine predictions in white women who delivered preterm (Spearman p - 0.19 versus p - 0.65, p - 0.02 for difference in p’s; Fig. 3D).
To check whether this difference in accuracy stems from lower representation of microbes in our metabolic models, we compared the coverage of the models between the different subgroups. We found that black women had somewhat lower model coverage (Mann Whitney p ~ 0.05, Fig. 10D), which could be expected, as the vaginal microbiome in this population tends to be more diverse. As the metabolic models account for the abundance of each microbe, and the vaginal microbiome has a skewed distribution, the models are robust to lack of representation of low- abundance microbes.
These analyses show that the lower tyramine prediction accuracy in white women who delivered preterm is not the result of lower model coverage or of ill-defined metabolic constraints. Our results therefore suggest a difference between the subgroups with respect
to tyramine levels, which could potentially stem from a difference in strains, functional capacity, or a non-microbial effect, with either potentially explaining the aforementioned paradoxical microbial associations with tyramine (Fig. 3 A). lire possibility of either a microbial difference or a non-microbial effect explaining the discrepancy in the accuracy of tyramine predictions is supported by AMON, which predicts that tyramine may be either microbial or host derived, Shaffer, M. et al., AMON: annotation of metabolite origins via networks to integrate 617 microbiome and metabolome data, BMC Bioinformatics 20, 614 (2019). Overall, our results demonstrate the utility of metabolic models in microbiome- metabolome interactions, and raise intriguing hypotheses for further investigation. Early prediction of sPTB risk using the vaginal metabolome
Early diagnosis of pregnancies with high risk for prematurity is crucial for the development of prevention and intervention strategies, yet is largely still lacking. We therefore explored whether we can use clinical covariates, microbiome data or metabolome data, collected at weeks 20-24 of gestation, to predict sPTB. Of note, prediction was of deliveries occurring 14.5+4.2 weeks (mean±std; i.e., ~3 months) after the samples were taken. We trained predictive models using gradient boosted decision trees (LightGBM), as they were superior to alternative models (Fig. HA). For models using microbiome and metabolome data, we trained composite predictors, such that a separate model was used for white and black women. Despite the smaller effective sample size for each model, this resulted in better performance (Fig. 1 IB), allowing the model to account for the differences between the two subgroups. We evaluated all models on held-out samples using nested cross-validation with strict train-test sterility.
Our models using clinical (age, BMI, race, parity status, history of sPTB and nulliparity) and the model using microbial abundances data obtained limited accuracy (area under receiver operating characteristic [auROC] = 0.58, area under precision recall curve [auPR] = 0.44 for clinical data; auROC - 0.56, auPR = 0.40 for microbiome data; p ~ 0.47 for difference between the models; Figs. 4A, 4B). Notably, using metabolomics data, we were able to generate a model with superior accuracy (auROC = 0.77, auPR = 0.61: p < 10' 10 and p < 7.1X10-8 for comparison of auROCs with clinical and microbiome models, respectively; Figs. 4A, 4B).
Figures 4A to 4D relate to prediction of subsequent sPTB using metabolomics, microbiome, or clinical data. Figures 4A and 4B are ROC and PR curves, respectively, comparing sPTB prediction accuracy for models based on clinical (auROC = 0.58, auPR =
0.44), microbiome (auROC = 0.56, auPR - 0.40) and metabolomics (auROC = 0.77, auPR = 0.61) data, evaluated in nested cross-validation. (N = 232 for all). Shaded lines show results from five independent outer 10-fold cross-validation draws. Figure 4C is a ROC curve evaluating the performance of our metabolomics-based predictor on two external cohorts, obtaining relatively accurate predictions without retraining (auROC = 0.65, auROC = 0.69, for the Ghartey 2017 [N=50] and Ghartley 2015 [N— 20] cohorts, respectively. Figure 4D shows the effect on total prediction (SHAP- (SHapley Additive exPlanations-) based) for the 10 most predictive metabolites in the metabolomics-based predictor, sorted with descending importance. Each dot represents a specific sample, with the color corresponding to the relative level of the metabolite in the sample compared to all other samples.
“Ghartley 2015” was a case-control study of 20 women, mostly (75%) white, at high-risk for PTB, from whom samples were collected between 24 to 28 weeks of gestation, and of which 10 delivered preterm. Ghartey, J., Bastek, J. A., Brown, A. G., Anglim, L. & Elovitz, M. A. Women with preterm birth have a distinct cervicovaginal metabolome, Am.
J. Obstet. Gynecol. 212, 776.el-12 (2015). “Ghartley 2017” was a case-control study of 50 women, mostly (88%) black, who had no prior history of PTB but presented with symptoms of preterm labor, and of whom 20 indeed delivered preterm. Samples in this cohort were collected between 22-34 weeks of gestation. Ghartey, J., Anglim, L., Romero, J., Brown, A. & Elovitz, M. A. Women with Symptomatic Preterm Birth Have a Distinct Cervicovaginal
Metabolome. Am. J. Perinatal 34, 1078-1083 (2017).
Lastly, a model combining clinical, microbiome, and metabolomics data obtains similar accuracy to the model which used only metabolomics data (auROC = 0.79, auPR = 0.65; p ~ 0.49 for comparison of auROC with metabolomics model; Fig. 11C, 1 ID), with metabolomics-based features as the most prominent contributors to the model (10 most predictive features; Fig. HE). This suggests that metabolite measurements are a sufficient representation of information contained in these three data types. Our metabolomics-based model is superior or similar in accuracy to several previously-published models, such as those using amniotic fluid metabolomics (auROC = 0.65-0.70, N = 24), maternal serum metabolome and clinical data (auROC - 0.73, N = 164). maternal urine and plasma metabolomics (auROC - 0.69-0.79, N = 146), cell-free blood RNA measurements (auROC = 0.81, N = 38), or vaginal protein biomarkers (auROC = 0.86, N = 150, sPTB N - 11), many of which have small sample sizes, lack demographic diversity, or focus on high-risk
cohorts. Overall, our results demonstrate the promising utility of vaginal metabolites as early and accurate biomarkers of spontaneous preterm birth.
Figures 11A to 11J relate to performance and features of prediction models for sPTB. Figure 11A shows a receiver operating characteristic (ROC) curve comparing the performance of different sPTB prediction algorithms on metabolornics data. LightGBM (auROC - 0.82) outperforms support vector regression (auROC - 0.79, p = 0.066 for auROC comparison) and logistic regression (auROC = 0.73, p ~ 9.6x10"4 for auROC comparison against LightGBM). Figure 11 B shows a ROC curve comparing the performance of a composite model for all women, compared to results stratified for race (black women and non-black women). The model trained on samples from all women achieves the same accuracy as the model trained only on samples from black women, when evaluated in 10-fold cross-validation on sPTB prediction for black women (auROC of 0.83 for both). However, the model trained on samples from all women significantly underperforms a model trained only on samples from women who do not identify as black when evaluated in 10-fold cross-validation on the same subgroup (auROC of 0.64 vs. 0.82, p = 0.0017 for auROC comparison). A different model is learned on each subgroup, models trained separately on each subgroup do not generalize well to the other subgroup (auROC of 0.63 and 0.66). This demonstrated the utility of training a predictor stratified for race.
Figures 11C and 1 ID show ROC characteristic ( 11C) and precision-recall (PR, 1 ID) curves, evaluated in nested cross-validation, comparing sPTB prediction accuracy for models based on metabolornics data alone (auROC - 0.77, auPR - 0.61), and on metabolornics data combined with microbiome and clinical data (“combination”; auROC - 0.79, auPR = 0.65; p ~ 0.49 for comparison between auROCs). Figure 1 IE shows a SHAP- based (SHapley Additive exPlanations-based) effect on total prediction (x-axis) for the top 10 features used in the combination models, sorted with descending importance. Each dot represents a specific sample, with the color corresponding to the relative level of the metabolite in the sample compared to all other samples. Figures 11F and 11G show ROC curves for the same metabolomics-based (1. I F) and microbiome-based (11G) models as in Figs. 4A and 4B, for extremely (<28 weeks of gestation) and very (<32 weeks) PTB. The microbiome-based models show increasing accuracy tor predicting extremely and very PTB (auROC of 0.66 and 0.61, respectively, compared to auROC of 0.57 for all sPTB, p = 0.005 and p ~ 0.034, respectively).
Figures 11H and 1 11 show precision-recall curves for sPTB prediction based on two external cohorts, using the metabolomics-based predictor without retraining or adaptation. Figure 11 J is similar to Fig. HE, for the microbiome-based model.
We next checked the performance of the microbiome- and metabolomics-based predictors at predicting early sPTB. We evaluated the same models, without retraining, for predicting extremely (<28 weeks) and very (<32 weeks) PTB in the same held-out data (i.e., only the ground-truth classification of outcome changed), limiting our evaluation to samples from black women. Interestingly, while the metabolites-based model shows a slight decrease in accuracy across different sPTB timings (auROC of 0.69 and 0.73 for extremely and very PTB, respectively, compared to auROC of 0.77 for all sPTB; p ~ 6.2xlO"4 and p ~ 0.019, respectively; Fig. 1 IF), our microbiome-based model shows increasing accuracy for predicting extremely and very' PTB (auROC of 0.66 and 0.61, respectively, compared to auROC of 0.57 for all sPTB, p = 0.005 and/? = 0.034, respectively; Fig. 11G).
These results suggest that metabolite measurements are a sufficient representation of information contained in these three data types. Our metabolomics-based model is superior to or similar in accuracy to several previously-published models, such as those using amniotic fluid metabolomics (auROC = 0.65-0.70, N = 24), maternal serum metabolome and clinical data (auROC = 0.73, N = 164), maternal urine and plasma metabolomics (auROC = 0.69-0.79, N = 146), cell-free blood RNA measurements (auROC = 0.81 , N = 38), or vaginal protein biomarkers (auROC = 0.86, N = 150, sPTB N = 11), many of which have small sample sizes, lack demographic diversity, or focus on high-risk cohorts. Overall, our results demonstrate the promising utility of vaginal metabolites as early and accurate biomarkers of spontaneous preterm birth. Interpretation of predictive models reveals novel contributing features
To interpret our predictive models and obtain insights into the features they use, we performed feature attribution analysis (using SHAP) allowing us to infer the contribution of each feature towards the final prediction for each sample. As expected, seven of the ten most predictive metabolites, namely diethanolamine, tyramine, arabinose, X- 11381, (R)- hydroxybutyl carnitine, Glutamate gamma-methyl ester, and ethyl glucoside, were also identified in our association analysis, with a similar direction of association (Figs. 2, 4D). In addition to these previously identified associations, our metabolomics-based model also indicates that high levels of the metabolites pipecolate and 3-amino-2-piperidone and low
levels of orotidme contribute to sPTB predictions. Of these, pipecolate was previously shown to be elevated in women with BV.
A similar analysis, performed on our microbiome-based predictor, has also captured previously-detected associations between various vaginal microbes and sPTB, including those of M. mulleris, and of Lactobacillus and Dialister species (Fig. 11J). We further identify a new association with Finegoldia magna, an anaerobic gram-positive species whose levels contributed to prediction of sPTB (Fig. 11 J). Tills replicates the results from a previous study, which showed that F. magna was more prevalent in women who delivered preterm. These results highlight the interpretability of our models and their reliance on complex, non-linear interactions of metabolites with both sPTB and other features, enabling us to expose associations not detected by univariate analyses.
DISCUSSION AND ANALYSIS
We measured the levels of 748 vaginal metabolites in a cohort of 232 pregnant women. We show that the vaginal metabolome largely separates by microbial community state types, but that de novo clustering of the metabolome reveals clusters enriched for sPTB among black women. We further identify multiple metabolites that are associated with sPTB, separately for black and white women, and for early preterm births. Our results highlight several exogenous metabolites with strong associations with sPTB, which we suggest constitute novel risk factors.
Using microbe- metabolite associations, we uncover intriguing interactions between metabolites associated with term birth and potentially suboptimal microbes, and propose a difference between the metabolism of tyramine in the vaginal microbiome of white women who delivered preterm. Finally, we demonstrate that supervised learning models trained on metabolomics data can accurately predict subsequent sPTB, potentially paving the way for new diagnostic panels.
We detected a group of xenobiotics, namely diethanolamine, ethyl glucoside, tartrate, and EDTA, which were elevated in women who subsequently gave birth preterm. These metabolites were largely independent from the microbiome, and both prior literature and a functional analysis suggest they are of exogenous source. Diethanolamine, a chemical with no known natural source commonly used in drilling and metalworking fluids, and ethyl glucoside, which is present in alcohol-containing products, are both ingredients or
precursors for ingredients in hygienic and cosmetic products. Tartrate and EDTA are used ubiquitously as food additives and are also common in hygienic and cosmetic products.
While we cannot identify the exact sources of these metabolites in this cohort, the fact that all are used in hygienic and cosmetic products raises concern that the use of products containing these chemicals may increase the risk for spontaneous preterm birth. Our results coincide with recent studies which raise concern regarding environmental exposures in pregnant women, and demarcate the presence of these chemicals in the reproductive tract. Of note, the United States Department of Environmental Protection has estimated that diethanolamine is in the 98th percentile of chemicals (by volume) to which reproductive- aged women are exposed.
The cohort analyzed here includes a majority of black women, offering an opportunity to study preterm birth in women who are disproportionately burdened by it and other adverse pregnancy outcomes, while at the same time are often represented in small numbers in many cohort studies. We note that the enrichment of sPTB associations among the xenobiotic metabolite set in black women may potentially reflect disparities in environmental and exogenous exposures, consistent with reports that black women may have greater exposures to endocrine disrupting chemicals through personal care products and consistent with other studies that have identified exogenous chemicals as possible drivers of PTB. Metabolomic exposure patterns could differ between cohorts, and could contribute to the association between racial disparities in prematurity rates and racial differences in the vaginal microbiome. Further study is warranted to identify the sources of these metabolites and disentangle their effects on the host, microbiome, and pregnancy outcomes.
To obtain insights into microbial metabolism of tyramine by the microbiome, we used community-scale metabolic models. Our models provided accurate predictions of several metabolites, and offered insights regarding potential sources of tyramine in the context of its association with preterm delivery.
As this cohort was focused on sPTB, we are unable to assess if our models are specific to sPTB or are detecting a general risk for adverse pregnancy outcomes. Further, while our models use samples from the middle of pregnancy to predict delivery occurring up to 19 weeks later, there is additional unexplored potential in using even earlier samples for prediction. A larger sample size, and combination with other sources of data, such as maternal urine or serum metabolomics, vaginal metagenomics, or cell free RNA
measurements, could further improve prediction accuracy. Nevertheless, our results demonstrate the potential of vaginal metabolites to serve as early biomarkers of preterm delivery and to potentially enable die personalization of treatment and prevention strategies.
METHODS
Study design and cohort description
We analyzed banked samples from the previously collected and described Motherhood & Microbiome (M&M) cohort. This cohort was approved by the Institutional Review Board at the University of Pennsylvania (IRB #818914) and the University of Maryland School of Medicine (HP-00045398), and all participants provided written informed consent. The M&M cohort recruited 2,000 women with a singleton pregnancy prior to 20 weeks of gestation. Women were followed to delivery, and spontaneous preterm birth was defined as delivery before 37 weeks of gestation with a presentation of cervical dilationand/or premature rupture of membranes. Of these, the vaginal microbiota of 503 women was characterized via 16S rRNA gene amplicon sequencing (V3-V4 region) of vaginal swabs collected between 20 to 24 weeks of gestation, and total bacterial load was assessed using the TaqMan® BactQuant assay. For this study, out of women with available microbiome data, all available samples were selected from women who delivered preterm (N = 80), in addition to samples from 152 controls who delivered at term. The selected samples were replicates of those used for 16S rRNA gene sequencing, collected using a double shaft Dacron swab.
Metabolomics profiling and preprocessing
Metabolite levels were measured from vaginal swabs by Metabolon Inc. (Durham, NC, USA), using an untargeted LC/MS platform. Metabolite measurements were volume normalized, followed by robust standardization of the log (base 10) transformed values (subtracting the median and dividing by the standard deviation calculated while clipping the top and bottom 5% of outliers).
Microbiome data processing
Ail microbiome-based analyses were done using data previously processed with DADA2 and SpeciatelT software, available from Supplementary Data 2 of Elovitz, M, A. et al., Cervicovaginal microbiota and local immune response modulate the risk of
spontaneous preterm delivery, Nat. Commun. 10, lj(b (2019). The exception to this are predictive models, which were trained on 97%-clustered OTUs using the USEARCH pipeline. We obtained raw sequences from the database of Genotypes and Phenotypes (dbGaP) under study accession: phs001739.vl.pl. Primers were aligned to reads and then trimmed, followed by end merging and quality filtering (-fastq_maxee 1.0). The filtered reads were then pooled together, dereplicated, clustered with a 97% threshold, and chimera filtered with the UPARSE algorithm to produce the OTU count matrix.
Global microbiome and metabolome structure
PERMANOVA analysis was performed using Bray-Curtis distance for microbiome data, and the Canberra dissimilarity metric for metabolites data. De novo clustering of metabolite vectors was done using K-medoids algorithm, also with Canberra dissimilarity. This metric is robust to outliers and sensitive to differences in common features; used with metabolomics data, it has previously produced robust results under bootstrapping and generated compact dusters corresponding with prior knowledge. We determined the optimal number of clusters by comparing the within cluster sum of square error and the gap statistic for clustering solutions with K between 1 and 15 (see Figs. 6A and 6B). Uniform Manifold Approximation and Projection (UMAP) was performed using the Python umap-leam package, with n_neighbors=15 and min_dist=0.05 for microbiome data and n_neighbors=15 and min. dist=0.25 for metabolomics data.
Differential abundance testisig and metabolite set enrichment analysis
Differential abundance tests between metabolite levels were done using the Mann- Whitney U test for metabolites which were present in at least half of the cases. To identify functional sets of metabolites that were perturbed between sPTB and TB, we compared, for each set, the Mann- Whitney p values for differential abundance between PTB and sPTB for metabolites within the set to the same p values for metabolites outside the sets, using an additional Mann- Whitney U test. We calculated significance by comparing the p value of the latter test to 10,000 similar p values calculated on random pennutations of sPTB and TB labels. For functional sets, we used definitions of super and sub pathways provided by Metabolon, as well as KEGG pathways. FDR correction was performed separately for each metabolite set type.
Microbe-metabolite correlations
To identify associations between microbes and metabolites, we estimated microbial absolute abundance by multiplying the relative abundances of each taxa by the total 16S rRNA copy number for the sample, obtained using the TaqMan qPCR-based panel, and calculated Spearman correlations with the levels of metabolites we found to be associated with sPTB. Across all correlation network analyses (Figs. 3 A, 9 A, 9C, 9D, 9E) we included correlations with at least 22% of paired measurements, corresponding to 50 samples of 232 for Fig. 3A. All correlation measurements used available data without imputation, and correction for multiple testing was performed via the Benjamini-Hochberg FDR method. To determine whether edges in our network were influenced by race (Fig. 9B) or by the severity of sPTB (Fig. 9F), we used a two-sided Fisher R-to-z transform to compare these correlations in black women to the same correlations in white women as well as compare these correlations in black women who delivered prior to 32 weeks to the same correlations in all other black women.
Creating and interrogating vaginal microbiome models
Microbiome metabolic modeling was done using Microbiome Modeling Toolbox (COBRA toolbox commit: 71 cl 17305231 I77a0292856e292b95ab32040711), using models from AGORA2. All computations were performed in MATLAB version 2019a (Mathworks, Inc.), using the IBM CPLEX (IBM, Inc.) solver.
For each sample, tailored microbiome models were created through the compartmentalization technique: metabolic reconstructions of species present in the sample are merged into a shared compartment, and input and output compartments are added, lire shared compartment enables microbes to share metabolites while input and output compartments are present to enable compounds intake and secretion. Coupling constraints are added to ensure a dependency between relative abundances and each species network fluxes. Finally, sample-specific microbiome biomass objective functions, composed by the sum of each microbial biomass multiplied by the corresponding relative abundance value, are added to each microbiome model. To interrogate the secretion potential of each sample-specific microbiome model, we computed Net Maximal Production Capacities (NMPCs) using the pipeline mgPipe.m of the Microbiome Modeling Toolbox. NMPC calculation accounts for maximal microbiome compound production and uptake rates and aims at predicting the overall
contribution of microbiomes to the metabolism of specific compounds. To later assess prediction accuracy, we computed Spearman correlations between NMPCs and the corresponding metabolite measurements without imputation.
To support and improve the accuracy of our tyramine predictions, we validated the presence of the TDC gene, coding for tyrosine decarboxylase. For each species represented in our metabolic models (N = 95), we used Prodigal to predict open reading frames in up to 200 randomly selected Refseq assemblies, and searched them for evidence of TDC using the hmmsearch function of Hmmer3.3.2 and a profile hmm for TDC (NCBI HMM accession: T1GR0381 1.1). We then curated our metabolic models, making sure that the corresponding reaction exists in models for which at least one assembly contained the corresponding gene.
To compile the metabolic models, we matched between the species detected in the microbiome samples and those present in AGORA2. To increase the representativeness of our models, we added three representatives for abundant vaginal species without a corresponding AGORA2 model that were present with >5% relative abundance in at least 20 samples. The only species that passed this threshold which was not included in our models was Ca. L. vaginae (BVAB 1), for which no suitable AGORA model was available.
To generate species level models, we combined metabolic models from available strains using the function createPanModels.m of the Microbiome Modeling Toolbox. Altogether, our microbiome metabolic models included 95 different species, with an average of 20 species in each sample. As the vaginal microbiome has a very skewed distribution, this resulted in a median [IQR] of 96.7% [88.4-98.8%] of the total abundance across samples represented by our models (Fig. 10D).
As a test of the sensitivity of our models to the lack of representation of low abundance microbes, we performed simulations where we iteratively removed the 10 least abundant species from consideration by our models, and evaluated the accuracy of our models in predicting the well-modeled metabolites tyramine, putrescine, and histamine. As expected, as our models account for the abundance of each microbe and as the vaginal microbiome has a skewed distribution, our models were not sensitive to the representation of low abundance microbes (Fig. 10E), even when removing 70 out of 95 models.
Metabolic modeling requires environmental conditions such as media and carbon source availability. We therefore formulated a “general vaginal media”, as the union of all metabolites present in at least 50 samples to which a corresponding metabolite was
identified in AGORA, assuming them to be present m an unlimited (i.e., very high) concentration. This vaginal media was applied to each microbiome model input compartment in the form of constraints on metabolite uptake reactions, constraining uptake of compounds not present in the environment to zero. Uptake of specific gut-related dietary compounds, automatically performed in mgPipe, was disabled acknowledging the different metabolic environment in the vagina, and essential metabolites required for achieving microbiome growth, together with their respective flux value, were detected and added to the vaginal media using the fasiFVA and findMIIS functions of the COBRA Toolbox. A comparison of the “general” media to subgroup-specific media, defined as metabolites present in 75% of samples from black and white women separately, with uptake fluxes constrained to the mean value across the subgroup; and to a person- specific media, in which uptake fluxes were constrained for each sample separately, showed similar accuracy with respect to tyramine predictions.
Training and testing of sPTB classifiers
We constructed predictive models separately using the clinical (age, race, paritystatus, history of sPTB, and BMI), microbiome, and metabolomics data, as well as a combination model consisting of all of these data types combined. As race had very strong interactions with microbiome and metabolomics data, we trained a composite predictor for microbiome, metabolomics, and combination models, whereas a separate model was trained for black women. Despite the smaller sample size for each model, this empirically improved prediction performance (Fig. 11B). Microbiome-based models used absolute abundances, calculated from USEARCH-processed OTUs as described above. In cases were qPCR based total load was not available (N = 14), it was imputed to the mean total load using only training samples.
Samples were split into training and test sets using 10-fold cross validation (“outer folds”), block-stratified for deciles of gestational age at birth (GAB), and for microbiome, metabolomics, and combined models, also stratified for race. To account for stochasticity in the division to 10 folds, we repeated this process 5 times. Train-test sterility was strictly maintained. To tune the optimal set of hyperparameters (including parameters for feature engineering and selection), and to obtain a robust estimate of the generalization error, we used nested cross-validation. In this extension of the training-test-validation framework, the training set was further split to 5 folds (“inner folds”), on which we used 1,000 iterations of
a random set of hyperparameters. Once more, to account for stochasticity, we repeated this process 5 times. We selected the best hyperparameter set as the model with the top average R2 score out of the top 10 most accurate models based on average auROC for sPTB classification, based on performance on the inner folds. We then used these hyperparameters to train a model on the entire training data for the outer fold, and evaluated it on the held- out test data. Of note, in this framework, hyperparameters are selected using strictly the training data of each outer 10-fold cross-validation fold, and are evaluated just once on the test set. Our prediction pipeline included standardization and imputation (for metabolomics data), optional PC A transformation, and feature selection using sparsity, SNAP feature importance, information gain and/or Spearman correlation, followed by prediction using LightGBM, with ail steps performed strictly using training data.
The selected models were then evaluated, without retraining, on classification of extremely (GAB < 28 weeks) or very (GAB < 32 weeks) PTB on the outer fold. Benchmark analyses (Figs. 11 A, 11B) were done using 10-fold cross-validation. We assessed the significance of the difference in auROC between two models by computing z-scores of the normal distributions of auROC on 30 10-fold cross-validation splits.
To obtain a final model for interpretation and validation, we trained new composite models on the entire cohort (N-232), using the hyperparameters selected for each of the outer folds (50 models), and picked the model with the best auROC on the same cohort (training fit). For validation on external vaginal metabolome datasets, we note that information on maternal race at the subject level was not available to us. We therefore applied the metabolomics model used for non-black women, without retraining or adaptation, to metabolomics data from the Ghartey 2015 cohort, as this cohort contained mostly white women; and similarly applied the metabolomics model used for black women to metabolomics data from the Ghartey 2017 cohort.
It is to be understood that the use of vaginal metabolome for early prediction of spontaneous premature labor is not limited to the specific embodiments described above, but encompasses any and all embodiments within the scope of the generic language of the following claims enabled by the embodiments described herein, or otherwise shown in the drawings or described above in terms sufficient to enable one of ordinary skill in the art to make and use the claimed subject matter.
Claims
1. A method of predicting preterm birth, the method comprising the steps of: obtaining a vaginal ecosystem sample from a pregnant woman; testing the sample to determine the presence of at. least one biological marker, or the presence of a combination of biological markers, associated with preterm birth; and predicting a preterm birth if the at least one biological marker is detected at a level above a pre-determined threshold, or if the combination of biological markers are detected at one or more levels above pre-determined thresholds.
2. The method of predicting preterm birth as recited in claim 1, wherein the step of obtaining the sample comprises obtaining a vaginal ecosystem sample from the pregnant woman.
3. The method of predicting preterm birth as recited in claim 1, wherein the at least one biological marker or the combination of biological markers includes at least one metabolite of a vaginal metabolome of the pregnant woman.
4. The method of predicting preterm birth as recited in claim 1, wherein the at least one biological marker or the combination of biological markers includes at least one biological component of a vaginal microbiome of the pregnant woman.
5. The method of predicting preterm birth as recited in claim 3, wherein the biomarkers and thresholds used for predicting preterm birth are based on maternal selfidentified race.
6. The method of predicting preterm birth as recited in claim 4, wherein the biomarkers and thresholds used for predicting preterm birth are based on maternal selfidentified race.
7. The method of predicting preterm birth as recited in claim 3, wherein the at least one metabolite includes at least one metabolite selected from the group consisting of
tartrate, ethyl glucoside, mannose, arabinose, mannitol, sorbitol, EDTA, tyramine, pipecolate, orotidine, (R)-hydroxybutyl carnitine, glutamate, gamma-methyl ester, 3- amino-2-piperidone, and diethanolamine.
8. The method of predicting preterm birth as recited in claim 3, wherein the at least one metabolite includes at least one metabolite selected from the group consisting of mannitol, sorbitol, EDTA, tyramine, and diethanolamine.
9. The method of predicting preterm birth as recited in claim 7, wherein the at least one metabolite includes at least one metabolite selected from the group consisting of ethyl glucoside, tartrate, and diethanolamine.
10. The method of predicting preterm birth as recited in claim 7, wherein the at least one metabolite includes at least one metabolite selected from the group consisting of mannitol, sorbitol, and EDTA.
11. The method of predicting preterm birth as recited in claim 8, wherein the at least one metabolite includes diethanolamine.
12. A method of reducing the risk of preterm birth in a pregnant woman, the method comprising reducing the pregnant woman's exposure to one or more metabolites associated with preterm birth.
13. The method of reducing the ri sk of preterm birth in a pregnant woman as recited in claim 12, wherein the one or more metabolites associated with preterm birth includes at least one metabolite selected from the group consisting of ethyl glucoside, tartrate, EDTA and diethanolamine.
14. The method of reducing the risk of preterm birth in a pregnant woman as recited in claim 12, the method further comprising first determining that preterm birth is more likely for the pregnant woman than for an average pregnant woman based on analysis of a vaginal ecosystem sample obtained from the pregnant woman.
15. The method of reducing the risk of preterm birth in a pregnant woman as recited in claim 14, the method further comprising determining whether the pregnant woman’s selfidentified race affects the analysis of the vaginal ecosystem sample obtained from the pregnant woman.
16. The method of reducing the risk of preterm birth in a pregnant woman as recited in claim 12, the method further comprising administering to the woman a supplementation of one or more metabolites associated with term birth.
17. A method of reducing the risk of preterm birth in a woman, the method comprising the steps of: determining the woman’s race, predicting a preterm birth based on the woman’s race, and administering to the woman a supplementation of one or more metabolites associated with term birth, or reducing the woman’s exposure to one or more metabolites associated with preterm birth, or both.
18. The method of reducing the risk of preterm birth in a woman as recited in claim 17, the method further comprising obtaining a vaginal ecosystem sample from the woman, and testing the sample to determine the presence of at least one biological marker, or a combination of biological markers, associated with preterm birth.
19. The method of reducing the risk of preterm birth in a woman as recited in claim 17, wherein the woman is known to be pregnant.
20. The method of reducing the risk of preterm birth in a woman as recited in claim 17, wherein the woman is not pregnant or not known to be pregnant.
Applications Claiming Priority (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US202163137956P | 2021-01-15 | 2021-01-15 | |
US63/137,956 | 2021-01-15 | ||
US202163178666P | 2021-04-23 | 2021-04-23 | |
US63/178,666 | 2021-04-23 | ||
US202163210231P | 2021-06-14 | 2021-06-14 | |
US63/210,231 | 2021-06-14 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2022155469A1 true WO2022155469A1 (en) | 2022-07-21 |
Family
ID=82448694
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2022/012512 WO2022155469A1 (en) | 2021-01-15 | 2022-01-14 | Using the vaginal metabolome for early prediction of spontaneous preterm birth |
Country Status (1)
Country | Link |
---|---|
WO (1) | WO2022155469A1 (en) |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080090759A1 (en) * | 2004-08-30 | 2008-04-17 | Robert Kokenyesi | Methods and kits for predicting risk for preterm labor |
WO2019045246A1 (en) * | 2017-09-04 | 2019-03-07 | 이화여자대학교 산학협력단 | Preterm birth risk prediction using change in microbial community in sample |
WO2020168118A1 (en) * | 2019-02-14 | 2020-08-20 | Mirvie, Inc. | Methods and systems for determining a pregnancy-related state of a subject |
-
2022
- 2022-01-14 WO PCT/US2022/012512 patent/WO2022155469A1/en active Application Filing
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20080090759A1 (en) * | 2004-08-30 | 2008-04-17 | Robert Kokenyesi | Methods and kits for predicting risk for preterm labor |
WO2019045246A1 (en) * | 2017-09-04 | 2019-03-07 | 이화여자대학교 산학협력단 | Preterm birth risk prediction using change in microbial community in sample |
WO2020168118A1 (en) * | 2019-02-14 | 2020-08-20 | Mirvie, Inc. | Methods and systems for determining a pregnancy-related state of a subject |
Non-Patent Citations (3)
Title |
---|
BORKOWSKI KAMIL, NEWMAN JOHN W., AGHAEEPOUR NIMA, MAYO JONATHAN A., BLAZENOVIĆ IVANA, FIEHN OLIVER, STEVENSON DAVID K., SHAW GARY : "Mid-gestation serum lipidomic profile associations with spontaneous preterm birth are influenced by body mass index", PLOS ONE, vol. 15, no. 11, 17 November 2020 (2020-11-17), pages e0239115, XP055958949, DOI: 10.1371/journal.pone.0239115 * |
CARTER ET AL.: "Metabolomics to reveal biomarkers and pathways of preterm birth: A systematic review and epidemiologic perspective", METABOLOMICS, vol. 15, no. 9, 10 September 2019 (2019-09-10), pages 1 - 34, XP036887366, DOI: 10.1007/s11306-019-1587-1 * |
GHARTEY JENY, BASTEK JAMIE A., BROWN AMY G., ANGLIM LAURA, ELOVITZ MICHAL A.: "Women with preterm birth have a distinct cervicovaginal metabolome", AMERICAN JOURNAL OF OBSTETRICS & GYNECOLOGY, vol. 212, no. 6, 1 June 2015 (2015-06-01), US , pages 776.e1 - 776.e12, XP055958951, ISSN: 0002-9378, DOI: 10.1016/j.ajog.2015.03.052 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Serrano et al. | Racioethnic diversity in the dynamics of the vaginal microbiome during pregnancy | |
Yousefi et al. | DNA methylation-based predictors of health: applications and statistical considerations | |
Yassour et al. | Sub-clinical detection of gut microbial biomarkers of obesity and type 2 diabetes | |
Kho et al. | Transcriptomic analysis of human lung development | |
WO2020168118A1 (en) | Methods and systems for determining a pregnancy-related state of a subject | |
US20190172555A1 (en) | Method and system for microbiome-derived diagnostics and therapeutics for oral health | |
Flaviani et al. | Cervicovaginal microbiota and metabolome predict preterm birth risk in an ethnically diverse cohort | |
WO2021022225A1 (en) | Methods and systems for detecting microsatellite instability of a cancer in a liquid biopsy assay | |
Abbas et al. | Machine learning based refined differential gene expression analysis of pediatric sepsis | |
Kindschuh et al. | Preterm birth is associated with xenobiotics and predicted by the vaginal metabolome | |
Liu et al. | Multi-kingdom gut microbiota analyses define COVID-19 severity and post-acute COVID-19 syndrome | |
Oliver et al. | Cervicovaginal microbiome composition is associated with metabolic profiles in healthy pregnancy | |
Baud et al. | Microbial diversity in the vaginal microbiota and its link to pregnancy outcomes | |
WO2021081253A1 (en) | Systems and methods for predicting therapeutic sensitivity | |
Zhang et al. | Variation of the vaginal microbiome during and after pregnancy in Chinese women | |
Witt et al. | Detection of microbial cell-free DNA in maternal and umbilical cord plasma in patients with chorioamnionitis using next generation sequencing | |
Laghi et al. | Vaginal metabolic profiles during pregnancy: Changes between first and second trimester | |
Villani et al. | Identifying predictive bacterial markers from cervical swab microbiota on pregnancy outcome in woman undergoing assisted reproductive technologies | |
Cheng et al. | Microbiome resilience and health implications for people in half-year travel | |
Correia et al. | Moving beyond DNA: towards functional analysis of the vaginal microbiome by non-sequencing-based methods | |
Liu et al. | Characterization of vaginal microbiota in third trimester premature rupture of membranes patients through 16S rDNA sequencing | |
WO2022155469A1 (en) | Using the vaginal metabolome for early prediction of spontaneous preterm birth | |
Li et al. | TimiGP: Inferring cell-cell interactions and prognostic associations in the tumor immune microenvironment through gene pairs | |
WO2022192904A1 (en) | Systems and methods for identifying microbial biosynthetic genetic clusters | |
EP4347887A1 (en) | Biomarker panel for increased likelihood of blastocyst implantation success |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 22740145 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 22740145 Country of ref document: EP Kind code of ref document: A1 |