Microbial and immune based biomarkers and prediction of preeclampsia

Abstract

Methods, systems, and computer-implemented tools are provided for early prediction of preeclampsia from cervicovaginal samples as early as the first trimester. A method includes quantifying local immune factors (for example, comprising least one of Flt-3L, EOF, and IL- IRA) and determining vaginal microbiome features (including taxa such as a Prevotella species, a Bifidobacterium species, and a Gardnerella species), optionally together with maternal characteristics such as BMI, blood pressure, or medical history, and generating a numerical risk score using a trained model. In embodiments, reduced local concentrations of immune factors and BMI-conditioned microbial patterns (e.g., increased P. timonensis or reduced Bifidobacterium abundance) are associated with elevated risk, enabling risk assessment ~6 months prior to conventional clinical diagnosis. The disclosure further provides a system and kit to implement the assays and analytics, and non-transitory computer-readable media storing instructions to compute the risk score and support early clinical management.

Description

MICROBIAL AND IMMUNE BASED BIOMARKERS AND PREDICTIONOF PREECLAMPSIACROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims the benefit of U.S. Provisional Patent Application No. 63 / 764,360, filed February 27, 2025, titled “Microbial and Immune Based Biomarkers and Prediction of Preeclampsia,” and U.S. Provisional Patent Application No. 63 / 717,665, filed November 7, 2024, titled “Microbial and Immune Based Biomarkers and Prediction of Preeclampsia,” the entirety of the disclosure of each is hereby incorporated by this reference.STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH OR DEVELOPMENT

[0002] This invention was made with government support under HD106017 and HD114715 awarded by the National Institutes of Health. The government has certain rights in the invention.TECHNICAL FIELD

[0003] Embodiments of the present disclosure relate to microbial and immune-based biomarkers and predictions of preeclampsia. In some embodiments, the disclosure relates to early prediction and management of preeclampsia (PEC) in pregnancy using local (vaginal / cervicovaginal) immune factor measurements and vaginal microbiome features.BACKGROUND

[0004] Preeclampsia is a multi-system hypertensive disorder of pregnancy typically diagnosed after 20 gestational weeks based on blood pressure and systemic signs (e.g., proteinuria, laboratory abnormalities). Preeclampsia complicates approximately 5% of all pregnancies and contributes to 10-15% of maternal deaths and up to 25% of neonatal deaths. It also increases the risks of additional complications, including intrauterine growth restriction and preterm delivery.

[0005] Despite decades of research into placental vascular remodeling and endothelial dysfunction, preeclampsia is currently diagnosed based on symptoms indicating impending multi-organ damage that manifests late in pregnancy. Routine clinical diagnosis still occurs after 20 gestational weeks, when blood pressure criteria and end-organ findings have already emerged (i.e., elevated transaminase levels). At this stage, the patient may already beexperiencing symptoms such as proteinuria, vision changes, and headaches. This late, symptom-based diagnostic paradigm limits opportunities for preventive interventions and close surveillance earlier in gestation.SUMMARY

[0006] In some embodiments, a method for predicting a risk or likelihood that a pregnant patient will develop preeclampsia includes obtaining a cervicovaginal or vaginal swab from the subject during weeks 6-32 of gestation. The method may also include quantifying, in the sample, concentrations of local immune factors, determining, from the sample, abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, and a Mobiluncus species, and generating, by a prediction model that receives the immune-factor concentrations and microbial abundances as inputs, a numerical risk score indicative of preeclampsia. In some aspects, the concentrations of local immune factors quantified comprise at least one of EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL- 10, IL- 18, IL- IRA, IL- la, IL-ip, IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. In certain implementations, the concentrations of local immune factors quantified comprise at least one of Flt-3L, EGF, IL-IRA, IL-27, MCP-1, and sCD40L. In some aspects, the abundance of at least one of Gardnerella vaginalis, Prevotella timonensis, Prevotella colorans, Prevotella buccalis, Sneathia vaginalis, Lactobacillus crispatus, Megasphaera hutchinsoni, Aerococcus christensenii, Lactobacillus mulieris, Lactobacillus jensenii, Sneathia anginosus, Fenollaria massiliensis, Mobiluncus mulieris, Prevotella melaninogenia, and Lactobacillus iners is determined.

[0007] According to some embodiments, the method further comprises inputting a maternal body mass index (BMI) value, a blood pressure value, type II diabetes status, and / or a hbAlc value, wherein the model is parameterized by a BMI of 25 or greater, a blood pressure level, type II diabetes status, and / or a hbAlc value.

[0008] In some embodiments, the method includes that reduced concentration of at least one of Flt-3L, EGF, and IL-IRA relative to gestational-age-matched controls increases the risk score.

[0009] According to some embodiments, the presence of P. timonensis and / or reduced Bifidobacterium abundance increases the risk score when the maternal BMI is 25 or greater.

[0010] In some embodiments, the prediction model comprises any machine learning algorithm or any statistical algorithm trained to map input feature data to a predicted risk value. In particular embodiments, the prediction model is a regularized logistic regression trained with nested cross-validation. In some aspect, the regularized logistic regression is evaluated to achieve an area under the receiver operating characteristic curve (AUROC) of at least 0.75 in internal validation and at least 0.78 when combining microbiome and immune inputs.

[0011] According to some embodiments, the method further comprises initiating one or more clinical actions when the risk score exceeds a threshold, wherein the clinical actions are selected from prophylactic therapy, enhanced surveillance, visit cadence adjustment, and additional diagnostic testing.

[0012] According to certain implementations, multiple cervicovaginal or vaginal samples are taken from the pregnant patient, for example, a first sample collection at 6-14 weeks of gestation follow by at least one other sample collection at 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24-28 weeks of gestation, 26-30 weeks of gestation, and 28-32 weeks of gestation. In such embodiments, the prediction model updates the numerical risk score indicative of preeclampsia risk from inputs generated from the subsequent samples.

[0013] In some embodiments, a method for predicting the likelihood of a pregnant patient developing preeclampsia includes providing a biological sample from the pregnant patient, determining the concentration of an immune factor from the biological sample, identifying vaginal microbiome from the biological sample collected from the pregnant patient, wherein the pregnant patient has a body mass index of 25 or more, and determining the presence of Prevotella species in the vaginal microbiome. In such embodiments, reduced concentration of the local immune factor in the pregnant patient compared to a control level and the presence of P. timonensis in the vaginal microbiome indicates the pregnant patient has an increased likelihood of developing preeclampsia. In some aspects, the immune factor is selected from the group consisting of: EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. In certain implementations, the immune factor is selected from least one of Flt-3L, EGF, IL-IRA, IL-27, MCP-1, and sCD40L.

[0014] According to some embodiments, the biological sample is a vaginal swab.

[0015] In some embodiments, the biological sample is collected from the subject before 15 weeks of gestation, before 14 weeks of gestation, between 6 week and 14 weeks of gestation, or between 6 week and 13 weeks 6 days of gestation.

[0016] According to some embodiments, a method for predicting the likelihood of a pregnant patient developing preeclampsia includes providing a biological sample from a pregnant patient and determining the concentration of an immune factor from the biological sample, wherein the immune factor is selected from the group consisting of: Ftl-3L, EGF, and IL- IRA, such that reduced concentration of the local immune factor in the pregnant patient compared to a control level indicates the pregnant patient has an increased likelihood of developing preeclampsia.

[0017] In some embodiments, a method for predicting the likelihood of a pregnant patient developing preeclampsia includes identifying vaginal microbiome from a biological sample collected from the pregnant patient, wherein the pregnant patient has a body mass index of more than 25, and determining the abundance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome, wherein dominance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome indicates the subject has reduced likelihood of developing preeclampsia.

[0018] According to some embodiments, a method for predicting the likelihood of a pregnant patient developing preeclampsia includes identifying vaginal microbiome from a biological sample collected from the pregnant patient, wherein the pregnant patient has a body mass index of 25 or more, and determining the presence of Prevotella species in the vaginal microbiome, wherein the presence of P. timonensis in the vaginal microbiome indicates the subject has increased likelihood of developing preeclampsia.

[0019] In some embodiments, a method for predicting the likelihood of a pregnant patient developing preeclampsia includes obtaining, from a pregnant patient during 6 weeks and 14 weeks of gestation, a cervicovaginal or vaginal swab sample; quantifying in the sample concentrations of a panel of vaginal immune factors comprising at least one of Flt-3L, EGF, and IL- IRA; determining, from the sample, abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, P. timonensis, Sneathia vaginalis, and Gardnerella vaginalis,' and generating, by a prediction model that receives as inputs the immune factor concentrations and the microbial abundances, a numerical risk score for preeclampsia.

[0020] According to some embodiments, the method further comprises measuring a maternal body mass index (BMI) value during the first trimester, wherein the prediction model receives the maternal BMI value as an input to generate the numerical risk score for preeclampsia.

[0021] In some embodiments, the method further comprises initiating a clinical action when the numerical risk score exceeds a threshold, the clinical action selected from the groupconsisting of initiating or intensifying prophylactic therapy, enrolling the subject in enhanced surveillance, adjusting visit cadence, counseling, and scheduling additional diagnostic testing.

[0022] According to some embodiments, the method includes that generating the risk score occurs at least two months prior to a symptom-based clinical diagnosis window for preeclampsia.

[0023] In some embodiments, a system for early prediction of preeclampsia includes a sample-processing module configured to receive a first-trimester cervicovaginal or vaginal swab and (A) quantify levels of at least one of Flt-3L, EGF, and IL- IRA, and (B) determine abundances of vaginal microbes comprising at least one of a Bifidobacterium species, P. limonensis. S. vaginalis, and G. vaginalis,' a memory storing a trained prediction model that outputs a risk score for preeclampsia based on immune factor levels and microbial abundances and that is optionally stratified or parameterized by a maternal BMI threshold; and a processor operatively coupled to the sample-processing module and the memory and configured to execute the trained prediction model to generate the risk score and provide, to a user interface, an indication of the risk score and a corresponding clinical action recommendation.

[0024] According to some embodiments, a kit for first-trimester assessment of preeclampsia risk includes (A) assay reagents for detecting in a cervicovaginal or vaginal swab at least the immune factors Flt-3L, EGF, and IL- IRA; (B) nucleic-acid reagents sufficient to detect or quantify, in the sample, at least one microbial taxon selected from a Bifidobacterium species, P. timonensis, S. vaginalis, and G. vaginalis by sequencing, target enrichment, or amplification; and (C) instructions specifying that the sample is collected during weeks 6-14 of gestation and that measured immune factor levels and microbial abundances are combined, optionally with maternal BMI information, to compute a risk score indicative of preeclampsia or preeclampsia with severe features.

[0025] In some embodiments, a non-transitory computer-readable medium stores instructions that, when executed by a processor, cause a computer system to perform a method comprising receiving, as input, data representing immune factor concentrations and microbial biomarker abundances from a vaginal sample of a pregnant patient; inputting the data into a machine learning model trained to predict preeclampsia risk; generating, using the machine learning model, an output indicative of the likelihood of preeclampsia; and providing the output to a user or system.

[0026] In some embodiments, an in vitro system for predicting a risk that a pregnant patient will develop preeclampsia comprises a sample-processing module configured to receive a cervicovaginal or vaginal swab obtained from the subject during weeks 6-14 of gestation,generate sequencing reads from the sample, and quantify concentrations of local immune factors comprising at least one of Flt-3L, EGF, and IL-IRA. The system further comprises one or more processors and a non-transitory memory storing instructions that, when executed, cause the system to estimate and correct laboratory processing bias across taxa using a model that imposes similarity constraints based on phylogenetic relatedness; optionally detect and remove contaminant microbial signal from the sequencing reads by identifying clonal strain profiles recurring across samples and inconsistent with biological co-occurrence and / or negative-control profiles; determine abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, P. limonensis. S. vaginalis, and G. vaginalis,' optionally receive a maternal body-mass-index (BMI) value and parameterize or stratify subsequent modeling by a BMI threshold; train, validate, and / or apply a prediction model that receives as inputs the bias-corrected microbial abundances, the immune-factor concentrations, and optionally the BMI value, wherein training employs a rebalanced leave-one-out cross-validation scheme with intra-training hyperparameter selection to improve generalization across collection sites and independent cohorts; generate a numerical risk score indicative of preeclampsia; and provide, to a user interface, the risk score and, when the risk score exceeds a threshold, one or more clinical action recommendations selected from prophylactic therapy, enhanced surveillance, visit-cadence adjustment, counseling, and additional diagnostic testing.

[0027] According to some embodiments, the method includes that the inputs comprise a vector of immune-factor concentrations and a vector of centered log-ratio (CLR)-transformed microbial abundances.

[0028] The foregoing and other aspects, features, and advantages will be apparent from the DESCRIPTION and DRAWINGS, and from the CLAIMS if any are included.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The patent or application file contains at least one drawing executed in color. Copies of this patent or patent application publication with color drawing(s) will be provided by the Office upon request and payment of the necessary fee.

[0030] Implementations will hereinafter be described in conjunction with the appended and / or included DRAWINGS.

[0031] FIG. 1 illustrates an overview of the study design and analytical workflow and method for early prediction of preeclampsia using first-trimester vaginal samples.

[0032] FIGs. 2A-2G illustrate vaginal immune factors are associated with preeclampsia with severe features (sPEC). FIG. 2A shows a PCA of vaginal immune factor profiles colored byfuture sPEC status. FIG. 2B shows a box and swarm plot of immune factor (first principal component, PCI) by outcome. FIG. 2C shows a box and swarm plot comparing the sum concentration of 20 immune factors between sPEC and non-sPEC patients. FIG. 2D shows the significance (y-axis; two-sided t-test) and effect size (x-axis; Cohen’s d) for associations between individual immune factors and sPEC. FIGs. 2E-2G show the box and swarm plots for Fit 3L (FIG. 2E), EGF (FIG. 2F), and IL- IRA (FIG. 2G).

[0033] FIG. 2H depicts an exemplary barplot of the coefficients of immune factors PCI indicating that the association of immune factor PCI with sPEC is driven by multiple immune factors. FIG. 21 and 2J show stratified analysis of associations of host immune factors with sPEC in individuals with a BMI <25 (FIG. 21) and in individuals with a BMI >25 (FIG. 2J), which show vaginal immune factors are more strongly associated with sPEC in individuals with BMI >25.

[0034] FIGs. 3A-3G indicate that the abundance of early pregnancy vaginal microbes is associated with sPEC development. FIG. 3 A shows PCA of microbiome profiles by outcome. FIG. 3B shows RPC1 distributions by BMI stratum and outcome. FIGs. 3C-3D show taxonlevel associations in the full cohort in individuals with BMI >25. FIGs. 3E-3G provide box and swarm plots from three vaginal taxa with the most significant associations with sPEC in individuals with BMI >25, stratified by BMI group and sPEC. Box, IQR; line, median; whiskers, nearest point to 1.5*IQR; p, Mann-Whitney U test.

[0035] FIGs. 3H-3K show a weak association between vaginal microbes and sPEC in dominance analysis and in individuals with BMI <25. FIGs. 3H and 31 are bar plots showing the faction of individuals with and without sPEC, separate by vaginal microbiome dominance. FIG. 3H presents results for the full cohort, while FIG. 31 presents results for individuals with BMI >25. Dominance was defined as >30% relative abundance. These graphs show no microbiome with >30% abundance; p, Fisher’s exact. However, Gardnerella dominance is weakly associated with non-sPEC among BMI >25 (FIG. 31). FIG. 3J show taxon-level associations in the full cohort in individuals with BMI <25. FIG. 3K is a bar plot showing taxon-level associations in the full cohort in individuals with BMI <25 during the first trimester.

[0036] FIGs. 4A-4E depict that complex interactions among microbiome, immune factors, and sPEC vary with BMI. FIG. 4A is scatter plot showing the first immune factors principal components (y-axis) and the first microbiome robust principal component (x-axis) for each individual in the cohort (N=110; Pearson R=0.39, p=2.4xl0'5). FIGs. 4B and 4C show DIABLO loadings that of the associations identified between immune factors and SPEC in a multi-omic analysis, within the group with BMI <25 (FIG. 4B) and the group with BMI >25(FIG. 4B). FIGs. 4D and 4E show network of clinical, microbiome, and immune factor features identified via a DIABLO multi-omic analysis, restricted to a correlations cutoff of 0.5. FIG. 4D show the network for BMI <25, and FIG. 4E show the network for with BMI >25.

[0037] FIGs. 5A and 5B depict the associations between vaginal microbiome and immune factors in the context of sPEC. FIG. 5 A depicts Pearson correlations of the first three microbiome robust principal components and the first three immune factor principal components. FIG. 5B shows a network of clinical, microbiome, and immune factor features identified via a DIABLO multi-omic analysis, restricted to a correlations cutoff of 0.5, for the full cohort.

[0038] FIGs. 5C-5H show the associations of microbes, immune factors, and clinical covariants with sPEC in a sparse multivariate model. Coefficients of fitted DIABLO model’s associations with the first principal components of each data modality (microbes, FIG. 5C; immune factors, FIG. 5D; clinical data, FIG. 5E) when trained on individuals with BMI <25 are shown in FIGs. 5C-5E. Coefficients of fitted DIABLO model’s associations with the first principal components of each data modality (microbes, FIG. 5F; immune factors, FIG. 5G; clinical data, FIG. 5H) when trained on individuals with BMI >25 are shown in FIGs. 5F-5H. The colors of the bars indicate the sPEC status association for each variable.

[0039] FIG. 6 shows, in accordance with certain embodiments, a schematic of a nested cross- validation pipeline.

[0040] FIGs. 7A and 7B show early pregnancy vaginal microbiome and host immunity are predictive of sPEC ~6 months in advance. They show results for modeling considering only the microbiome, immune factor, and clinical data, respectively along with combinations of the microbiome and Luminex predictions. FIG. 7A depicts the receiver operating characteristics, while FIG. 7B depicts the precision-recall curves of leave-one-out nested cross-validation predictions of linear models. Dashed horizontal line in FIG. 7B indicated class balance.

[0041] FIGs. 8A and 8B show model predictions of sPEC approximately 6 months in advance using early pregnancy vaginal microbiome and host immunity. They show results from models trained on data from the nuMoM2b cohort and tested on data from the Multi-Omic Microbiome Study-Pregnancy Initiative (MOMS-PI).

[0042] FIGs. 9A and 9B depict exemplary linear regression coefficients for the model based on immune factors (FIG. 9A) and the model based on the microbiome (FIG. 9B).DETAILED DESCRIPTION

[0043] The following detailed description provides numerous specific details. Those skilled in the relevant arts understand that embodiments of the disclosure may be practiced without these specific details. The disclosure may also be practiced in different and alternative configurations.

[0044] Unless specifically noted, it is intended that the words and phrases in the specification and the claims be given their plain, ordinary, and accustomed meaning to those of ordinary skill in the applicable arts. The singular forms “a,” “an,” and “the” include plural referents unless the context clearly dictates otherwise. Thus, for example, a reference to “a step” includes a reference to one or more of such steps. The words “exemplary,” “example,” “embodiment,” or various forms thereof are used herein to mean serving as an example, instance, or illustration. Any aspect or feature described herein as “exemplary” or as an “example” is not necessarily to be construed as preferred or advantageous over other aspects or designs. The examples are provided solely for purposes of clarity and understanding and do not limit or restrict the disclosure. It is to be appreciated that a myriad of additional or alternate examples of varying scope could have been presented but have been omitted for purposes of brevity.

[0045] Throughout the description and claims of this specification, the words “comprise” and “contain” and variations of the words, for example “comprising” and “comprises”, mean “including but not limited to”, and are not intended to (and do not) exclude other components.

[0046] When a range of values is expressed, another embodiment includes from the one particular value and / or to the other particular value. Similarly, when values are expressed as approximations, by use of the antecedent “about,” it will be understood that the particular value forms another embodiment. All ranges are inclusive and combinable. As used herein, the term “about” when used in the context of numeric values denotes an interval of accuracy that is familiar and acceptable to a person skilled in the art.

[0047] The present disclosure may be understood more readily by reference to the following detailed description taken in connection with the accompanying figures and examples, which form a part of this disclosure. It is to be understood that this disclosure is not limited to the specific materials, devices, methods, applications, conditions, or parameters described and / or shown herein, and that the terminology used herein is for the purpose of describing particular embodiments by way of example only and is not intended to be limiting of the claimed inventions. The term “plurality”, as used herein, means more than one.

[0048] As used herein, the terms “preeclampsia with severe features” (sPEC) and “preeclampsia without severe features” is based on a definition recognized and accepted by obstetricians and gynecologists, for example the definition from the American College ofObstetricians and Gynecologists’ Task Force on Hypertension in Pregnancy’s 2013 guidelines (AGOC 2013). New onset gestational hypertension was defined as systolic blood pressure of 140 mm Hg or greater or diastolic blood pressure of 90 mm Hg or greater on two occasions six hours apart or on one occasion that then prompted antihypertensive treatment. sPEC was defined as new onset gestational hypertension plus any of the following symptoms: thrombocytopenia (platelet count < 100,000 / pl), pulmonary edema, serum creatinine >1.1 mg / dL, severe headache, scotoma, serum aspartate aminotransferase (AST) >100 IU / L, epigastric pain, or severe hypertension (systolic blood pressure >160 mm Hg or diastolic blood pressure >110 mm Hg on 2 occasions >6 hours apart or on 1 occasion requiring antihypertensive therapy and excluding blood pressure readings during the second stage of labor).

[0049] The prevailing clinical model posits that PEC pathogenesis begins early, with inadequate placental vascular remodeling and subsequent placental hypoperfusion, followed by the release of soluble factors that promote systemic endothelial dysfunction later in pregnancy. These processes are intertwined with maternal immune dysregulation, reflected in altered cytokine profiles and immune-cell composition locally (e.g., in the reproductive tract and placenta) and systemically. However, most clinical assays and investigational biomarkers have focused on blood-borne signals measured after the first trimester, creating a temporal gap between underlying disease processes and the point at which risk can be assessed in routine care.

[0050] Several molecular approaches have been explored to narrow this gap. For example, serum biomarkers based on the ratio of soluble fms-like tyrosine kinase and placental growth factor (sFIt-l / PIGF ratio) have demonstrated good performance for predicting short-term evolution to preeclampsia in already high-risk patients. For example, the area under the receiver operating curve (ROC) curve (auROC) of the sFIt-l / PIGF ratio is approximately 0.82 for prediction within four weeks. Yet, these tests are typically employed later in gestation and are not optimized for first-trimester deployment. Similarly, cell-free RNA (cfRNA) signatures measured in second-trimester maternal serum have shown comparable auROC (-0.82) in research settings, again leaving a window of several months during which early biology is active but not captured by standard-of-care biomarkers. In addition, multi-protein plasma signatures developed in single cohorts have suffered from limited cross-cohort generalizability, highlighting the need for biomarkers that are robust to differences in population characteristics and laboratory workflows.

[0051] As shown in the Examples herein, maternal microbiome — particularly in the genital tract — plays a role in modulating local inflammation, epithelial integrity, and pregnancy outcomes. The vaginal microbiome can shape the cervicovaginal immune milieu, influencing epithelial barrier function and mucosal cytokine levels, and has been associated with outcomes such as preterm birth. Significant associations were found between both vaginal microbes and immune factors and the development of sPEC, which differ with maternal BMI, and involve known preeclampsia risk factors such as blood pressure and anemia. Integrating immune factor, microbiome and clinical data allows for prediction of sPEC with comparable accuracy to previous molecular tests but much earlier in pregnancy.

[0052] Embodiments of the present disclosure are directed to methods, systems, and devices for predicting the likelihood of a pregnant patient developing preeclampsia by analyzing local immune factor concentrations and / or vaginal microbiome features from biological samples collected in early pregnancy. These approaches enable risk assessment months before the typical symptom-based diagnosis window, such as during the first trimester or at least two months earlier, facilitating earlier clinical management. The clinical management includes prophylactic therapy, enhanced surveillance, visit cadence adjustment, and / or additional diagnostic testing. In some implementations, the biological sample is collected before 17 weeks of gestation, before 15 weeks of gestation, between 6 week and 17 weeks of gestation, between 6 weeks and 15 weeks of gestation, between 6 weeks and 14 weeks of gestation, during the first trimester, or between 6 week 0 days and 13 weeks 6 days of gestation. In some implementations, multiple cervicovaginal or vaginal samples are taken from the pregnant patient, for example, a first sample collection at 6-14 weeks of gestation follow by at least one other sample collection at 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24-28 weeks of gestation, 26-30 weeks of gestation, and 28-32 weeks of gestation. Changes in the local immune factor concentrations and / or vaginal microbiome features across the samples are then used to update the prediction for the pregnant patient’s likelihood of developing preeclampsia. In certain embodiments, this data is combined — optionally with maternal BMI, a blood pressure, type II diabetes status, and / or a hbAlc level — in a prediction model to output a numerical risk score that can inform clinical action months before the typical symptom-based diagnostic window.

[0053] In some aspects, the model utilizes logistic regression, random forest, or any appropriate statistical model that can be trained to predict risk scores using tabular data. In other aspects, the model utilizes machine learning model trained to predict risk scores using tabular data.

[0054] In some embodiments, the method for predicting a risk that a pregnant patient will develop preeclampsia comprises providing a biological sample from a pregnant patient and determining the concentration of an immune factor from the biological sample, wherein the immune factor is selected from the group consisting of: EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. In some implementations, the concentration of Flt-3L, EGF, and / or IL- 1RA is determined, and reduced concentration of these immune factor in the pregnant patient compared to a control level indicates the pregnant patient has an increased likelihood of developing preeclampsia. In some aspects, the pregnant patient has BMI of 25 or more.

[0055] In other embodiments, the method of predicting the likelihood of a pregnant patient developing preeclampsia comprises identifying vaginal microbiome from a biological sample collected from the pregnant patient and determining the abundance of at least one of Bifidobacterium species, Prevotella species, Gardnerella species, Lactobacillus species, Bergeylla species, Megasphaera species, Aerococcus species, Fenollaria species, Sneathia species, and Mobiluncus species or the presence of at least one of G. vaginalis, P. timonensis, P. colorans, P. buccalis, S. vaginalis, L. crispatus, M. hutchinsoni, A. christensenii, L. mulieris, L. jensenii, S. anginosus, F. massiliensis, M. mulieris, P. melaninogenia, and L. inersin the vaginal microbiome. In some aspects, dominance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome indicates the subject has reduced likelihood of developing preeclampsia. In some aspect, the presence of P. timonensis in the vaginal microbiome indicates the subject has increased likelihood of developing preeclampsia. In some aspects, the pregnant patient has BMI of 25 or more.

[0056] In certain implementation, the method of predicting the likelihood of a pregnant patient developing preeclampsia comprises providing a biological sample from the pregnant patient; determining the concentration of an immune factor from the biological sample, wherein the immune factor is selected from the group consisting of: Flt-3L, EGF, and IL-IRA; and determining the abundance of Gardnerella species and / or Bifidobacterium species or the presence of Prevotella species in the vaginal microbiome. Reduced concentration of the immune factor in the pregnant patient compared to a control level indicates the pregnant patient has an increased likelihood of developing preeclampsia. Dominance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome indicates the subject has reduced likelihood of developing preeclampsia. The presence of P. timonensis in the vaginal microbiome indicates the subject has increased likelihood of developing preeclampsia. In some aspects, the pregnant patient has BMI of 25 or more.

[0057] In some aspects, the method of predicting the likelihood of a pregnant patient developing preeclampsia comprises providing a biological sample from the pregnant patient; quantifying the concentration of an immune factor from the biological sample; characterizing microbiome from the biological sample to determine abundances of microbial taxa; and generating, by a prediction model that receives as inputs the immune factor concentrations and abundances of microbial abundances, a numerical risk score for preeclampsia. In some embodiments, the method further comprises measuring a maternal BMI value during the first trimester, wherein the prediction model receives the maternal BMI value as an input to generate the numerical risk score for preeclampsia. In some implementations, the method further comprises initiating a clinical action when the numerical risk score exceeds a threshold, the clinical action selected from the group consisting of: initiating or intensifying prophylactic therapy, enrolling the subject in enhanced surveillance, adjusting visit cadence, counseling, and scheduling additional diagnostic testing.

[0058] In some aspects, the biological sample from the pregnant patient is cervicovaginal sample or vaginal swab sample. For example, the method of predicting a risk that a pregnant patient will develop preeclampsia comprises obtaining a cervicovaginal or vaginal swab from the subject during weeks 6-32 of gestation; quantifying, in the cervicovaginal or vaginal sample, concentrations of local immune factors comprising at least one of Flt-3L, EGF, and IL- IRA; and determining, from the sample, abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, P. limonensis. and a Gardnerella species; and generating, by a prediction model that receives the immune-factor concentrations and microbial abundances as inputs, a numerical risk score indicative of preeclampsia. In some implementations, the cervicovaginal sample or vaginal swab sample is obtained during 6-14 weeks of gestation, 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24-28 weeks of gestation, 26-30 weeks of gestation, or 28- 32 weeks of gestation. In certain implementation, the first sample collection takes place at 6- 14 weeks of gestation follow by at least one other sample collection at 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24-28 weeks of gestation, 26-30 weeks of gestation, and 28-32 weeks of gestation. In such embodiments, the prediction model updates the numerical risk score indicative of preeclampsia risk from inputs generated from the subsequent samples. The model is a regularized logistic regression trained with nested cross-validation. In some aspects, the model is evaluated to achieve AUROC >0.75 in internal validation and >0.78 when combining microbiome and immune inputs. In other aspects, the model utilizes random forest or any appropriate statistical model that can be trainedto predict risk scores using tabular data. In other aspects, the model utilizes machine learning model trained to predict risk scores using tabular data.

[0059] In another implementation, the method of predicting the likelihood of a pregnant patient developing preeclampsia comprising obtaining, from a pregnant patient between 6 weeks 0 days and 13 weeks 6 days of gestation, a cervicovaginal or vaginal swab sample; quantifying in the sample concentrations of a panel of vaginal immune factors comprising at least one of Flt-3L, EGF, and IL- IRA; and determining, from the sample, abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, P. limonensis, S. vaginalis, and G. vaginalis. The method next comprises generating, by a prediction model that receives as inputs the immune factor concentrations and the microbial abundances, a numerical risk score for preeclampsia. In some embodiments, the panel of vaginal immune factors comprises EGF, Flt- 3L, G-CSF, IFNa2, IFN-y, IL- 10, IL- 18, IL- IRA, IL- la, IL-ip, IL-27, IL-4, IL-6, IL-8, MCP- 1, RANTES, TGFa, TNFa, VEGF, and sCD40L. In some embodiments, the step of determining abundances of vaginal microbial taxa evaluates the abundance or presence of a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, and Mobiluncus species. In some aspects, the abundance or presence of G. vaginalis, P. timonensis, P. colorans, P. buccalis, S. vaginalis, L. crispatus, M. hutchinsoni, A. christensenii, L. mulieris, L. jensenii, S. anginosus, F. massiliensis, M. mulieris, P. melaninogenia, and L. iners.

[0060] Embodiments disclosed herein are supported by data from a multi-center observational cohort of nulliparous pregnant individuals with singleton gestations (nuMoM2b). nuMoM2b refers to a study that collected extensive clinical data and medical history, including first- trimester maternal BMI, age, and hypertension status.

[0061] In some embodiments, clinical data associated with the patient is collected, including maternal characteristics such as BMI, blood pressure, and medical history (such as type II diabetes status and hbAlc level). The patient data may be collected from an electronic health database. In some embodiments, the method or system may include statistical analyses to identify immune and microbial features associated with sPEC. The sample results and medical data may be added to the machine-learning models to predict risk.

[0062] In some embodiments, FIG. 1 also highlights the multi-omic nature of the approach, showing how immune factor data, microbiome profiles, and clinical variables are combined to generate a numerical risk score. This integrated design enables early identification of high-riskpregnancies and supports clinical decision-making well before conventional diagnostic criteria are met.

[0063] The multi-omic analysis described herein also revealed complex interactions among vaginal microbes, immune factors, and clinical features such as blood pressure and anemia. These findings suggest that the vaginal ecosystem reflects broader systemic processes relevant to preeclampsia risk, and that integrating data across these domains can improve early prediction. Predictive models developed in this study achieved robust performance, with combined microbiome and immune factor models reaching auROC values of 0.78 in internal validation and 0.80 in an independent external cohort. This level of accuracy is comparable to other molecular tests for preeclampsia, but with the advantage of much earlier sampling in pregnancy.

[0064] In some embodiments, reduced concentration of one or more of Flt-3L, EGF, or IL- 1RA in the biological sample, relative to a control level, indicates an increased likelihood of developing preeclampsia. Control levels may be established using gestational-age-matched reference distributions or internal standards.

[0065] In some embodiments, a system for early prediction of preeclampsia comprises a sample-processing module configured to receive a first-trimester vaginal swab and quantify levels of Flt-3L, EGF, and IL- IRA, as well as determine abundances of key vaginal microbes. In some aspects, the level of EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL- la, IL-ip, IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L are quantified. The key vaginal microbes are selected from Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, and a Mobiluncus species. In another aspect, the key vaginal microbes are selected from G. vaginalis, P. timonensis, P. colorans, P. buccalis, S. vaginalis, L. crispatus,M. hutchinsoni,A. christensenii, L. mulieris, L. jensenii, S. anginosus, F. massiliensis, M. mulieris, P. melaninogenia, and L. iners. The system includes a memory storing a trained prediction model and a processor configured to execute the model and provide risk scores and clinical recommendations via a user interface.

[0066] In some aspects, a kit for first-trimester assessment of preeclampsia risk includes assay reagents for detecting Flt-3L, EGF, and / or IL-IRA, nucleic-acid reagents for detecting or quantifying Bifidobacterium species, P. timonensis, S. vaginalis, and G. vaginalis, and instructions specifying sample collection timing and risk score computation. In some aspects, the assay reagent detect EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. In particular embodiments, the kit comprises nucleic-acid reagents for detecting or quantifying a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, a Mobiluncus species, or any combination thereof. In certain embodiments, the kit comprises nucleic-acid reagents for detecting or quantifying G. vaginalis, P. timonensis, P. colorans, P. buccalis, S. vaginalis, L. crispatus, M. hutchinsoni, A. christensenii, L. mulieris, L. jensenii, S. anginosus, F. massiliensis, M. mulieris, P. melaninogenia, and L. iners.

[0067] According to some embodiments, a system for early prediction of preeclampsia is provided. The system comprises a sample-processing module configured to receive a cervicovaginal or vaginal swab and to quantify levels of immune factors in the sample. The sample-processing module may further be configured to determine abundances of vaginal microbes. The system may also comprise a memory storing a trained prediction model that outputs a risk score for preeclampsia based on immune factor levels and microbial abundances, and that is optionally stratified or parameterized by a maternal BMI threshold. A processor operatively coupled to the sample-processing module and the memory may be configured to execute the trained prediction model to generate the risk score and provide, to a user interface, an indication of the risk score and a corresponding clinical action recommendation. In some embodiments, the user interface may display the risk score in real time and provide suggested next steps for clinical management based on the predicted risk. In certain implementations, the sample-processing module is configured to receive a cervicovaginal or vaginal swab and to quantify levels of at least one of EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL- la, IL- 10, IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, sCD40L in the sample and / or configured to determine abundances of vaginal microbes comprising at least one of a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, a. Mobiluncus species. In some aspects, multiple cervicovaginal or vaginal samples are taken from the pregnant patient, for example, a first sample collection at 6-14 weeks of gestation follow by at least one other sample collection at 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24- 28 weeks of gestation, 26-30 weeks of gestation, and 28-32 weeks of gestation. In such embodiments, the trained prediction model updates the numerical risk score indicative of preeclampsia risk from inputs generated from the subsequent samples.

[0068] In some embodiments, a kit for first-trimester assessment of preeclampsia risk is provided. The kit comprises assay reagents for detecting in a cervicovaginal or vaginal swab at least one immune factor. The kit may further comprise nucleic-acid reagents sufficient to detect or quantify, in the sample, at least one microbial taxon by sequencing, target enrichment, or amplification. The kit may also include instructions specifying that the sample is collected during weeks 6-32 of gestation (for example, at 6-14 weeks of gestation, 14-18 weeks of gestation, 16-20 weeks of gestation, 18-22 weeks of gestation, 22-26 weeks of gestation, 24- 28 weeks of gestation, 26-30 weeks of gestation, or 28-32 weeks of gestation) and that measured immune factor levels and microbial abundances are combined, optionally with maternal BMI information, a blood pressure value, type II diabetes status, and / or a hbAlc value, to compute a risk score indicative of preeclampsia or preeclampsia with severe features. According to some embodiments, the kit may further comprise positive and negative controls for assay calibration, as well as software access or a web portal for uploading assay results and receiving a computed risk score. In some implementations, the kit comprises assay reagents for detecting EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL-27, IL- 4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and / or sCD40L. In some implementations, the nucleic-acid reagents detect or quantify at least one microbial taxon selected from a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, and a Mobiluncus species. In particular embodiments, the nucleic-acid reagents detect or quantify at least one microbial taxon selected from G. vaginalis, P. timonensis, P. co I or an s, P. buccalis, S. vaginalis, L. crispatus, M. hutchinsoni, A. christensenii, L. mulieris, L. jensenii, S. anginosus, F. massiliensis, M. mulieris, P. melaninogenia, and L. iners.

[0069] According to some embodiments, a non-transitory computer-readable medium is provided, storing instructions that, when executed by a processor, cause a computer system to perform a method for early prediction of preeclampsia. The method may comprise receiving, as input, data representing immune factor concentrations and microbial biomarker abundances from a vaginal sample of a pregnant patient. The instructions may further cause the system to input the data into a machine learning model trained to predict preeclampsia risk, generate, using the machine learning model, an output indicative of the likelihood of preeclampsia, and provide the output to a user or system. In some embodiments, the computer-readable medium may further store instructions for preprocessing the input data, such as normalization, batchcorrection, or transformation (e.g., centered log-ratio transformation for microbial abundances), and for generating a report or alert if the predicted risk exceeds a predefined threshold.

[0070] In certain embodiments, microbiome data used by the prediction model undergoes a two stage preprocessing workflow that improves robustness in low biomass clinical samples.

[0071] First, an optional contamination handling stage identifies and removes exogenous microbial signal by detecting clonal strain profiles recurring across otherwise unrelated samples and not explained by biological co-occurrence or by negative control profiles (e.g., laboratory reagent contaminants). Second, a processing bias correction stage estimates taxonspecific measurement biases introduced by experimental pipelines and applies a correction constrained by phylogenetic relatedness, such that closely related taxa are regularized to have similar inferred biases. The decontaminated and bias corrected abundances are then supplied to the downstream feature engineering and modeling steps described herein.

[0072] To enhance clinical translatability, the model can be trained and evaluated with a rebalanced leave one out cross validation scheme and assessed for generalization across collection sites and fully independent cohorts. This framework can further incorporate (i) automated reduction of the feature space to a minimal assay panel suitable for deployment, (ii) sensitivity and fairness analyses to assess stability across demographic and clinical subgroups, and (iii) integrative modeling of mechanistic interactions among vaginal microbes, local immune factors, and physicochemical context (e.g., pH), optionally including metabolite measurements, to generate hypotheses for precision therapeutic strategies.Examples

[0073] The present disclosure is further illustrated by the following examples that should not be construed as limiting. The contents of all references, patents, and published patent applications cited throughout this application, as well as the Figures, are incorporated herein by reference in their entirety for all purposes.Example 1. Study Design

[0074] The discovery of the new preeclampsia risk assessment and diagnosis strategy arose from examination of the vaginal microbiome, host immunity, and their interaction with preeclampsia featuring severe features, as well as related clinical factors, in 124 individuals. Subjects were enrolled between 6 weeks, 0 days and 13 weeks, 6 days of gestation (ultrasound-dated), and provided self-collected cervicovaginal swabs at the first-trimester visit. A case-control subset included 62 individuals who developed preeclampsia with severefeatures (sPEC) and 62 matched controls. The individual’s clinical data (e.g., first-trimester BMI, blood pressure, medical history) was obtained from their records. Consistent with prior research, individuals who developed sPEC had higher BMI, higher 89 first-trimester blood pressure, and gave birth earlier in pregnancy. Characteristics of the sample group are presented in Table 1 below.Table 1 : Cohort characteristics

[0075] Vaginal swabs were self-collected between 6.4-13.9 weeks of gestation, 2.4-7.6 months before sPEC diagnosis. The individuals’ samples were analyzed with immune factor profiling, microbiome profiling, and multi-omic analyses. A multiplex immunoassay (Luminex) measured the levels of 20 pro- and anti-inflammatory immune factors: EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL- 10, IL- 18, IL- IRA, IL- la, IL-ip, IL-27, IL-4, IL-6, IL-8, 95 MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. Each measurement was obtained in duplicate. Batch standardization was performed and values below LOD are imputed as LOD / 2.

[0076] The vaginal microbiome was profiled using shotgun metagenomics sequencing to a mean±sd depth of 43.23±23.4 million reads, yielding an average of 5.83±7.6 million microbial reads after processing and discarding reads classified as human DNA. Microbiome profiling may include collecting microbial DNA from the sample. As an example, a 200 pL sample is extracted (e.g., Qiagen QIAcube HT or DNeasy PowerSoil HTP 96), libraries are prepared (e.g.,Nextera DNA), and sequenced 2^ 100 bp on NovaSeq (mean depth ~ 43M reads). Reads are quality-filtered (Trimmomatic), host-depleted (Bowtie2), and samples with <500,000 non-human reads are excluded. Here, that left 110 samples (57 sPEC). Taxonomic assignment was performed using Kraken2 / Bracken against the Human Vaginal Microbiome Genome Collection (VMGC) with centered log-ratio transformation for analysis.

[0077] Multi-omic analyses may include association testing such as Mann-Whitney U, two-sided t-tests with FDR control, PERMANOVA, and robust PCA for microbiome profiles. Multi-omic integration uses DIABLO (sparse discriminant analysis) to identify correlated immune-microbe-clinical feature sets. Predictive modeling uses L2-regularized logistic regression; evaluation uses rebalanced leave-one-out cross-validation with inner 5-fold hyperparameter tuning, and stratification on BMI >25 as a selectable parameter. External validation uses the MOMS-PI cohort with DEBIAS-M processing-bias correction for cross-study generalization.

[0078] FIG. 1 illustrates an overview of the study design and analytical workflow or method for early prediction of preeclampsia using first-trimester vaginal samples. The timeline at the top of the figure shows key gestational milestones: pregnancy initiation, the first-trimester visit and / or sample collection (approximately 6+0 to 13+6 weeks), and the later onset of clinical symptoms leading to diagnosis of preeclampsia with severe features (sPEC).

[0079] FIG. 1 also depicts components of the method, according to some embodiments. The method may include collecting a sample. This may be a self-collected cervicovaginal swab during the first trimester. In some embodiments. The cervicovaginal swab may be collected by a medical professional during a medical appointment. The collected cervicovaginal swab may be subjected to laboratory analyses, including multiplex immunoassays for quantifying local immune factors (e.g., Flt-3L, EGF, IL-IRA) and shotgun metagenomic sequencing for profiling the vaginal microbiome. a) Identifying clinical features for association and prediction

[0080] nuMoM2b study participants provided extensive sociodemographic and clinical data at a screening visit that occurred between 6 and 14 weeks of pregnancy. To construct a dataset of clinical features for multi-omic association analysis as well as for prediction, we relied on previous work that used clinical data to predict sPEC in the nuMoM2b cohort86 as well as ACOG guidelines regarding PEC risk factors. We used the following variables in our multi- omic association analysis with DIABLO: history of liver disease, gallbladder disease, diabetesprior to pregnancy, autoimmune disease, endocrine disease, blood clots or thromboembolic disease, heart disease, kidney disease, anemia, or hypertension, as well as first trimester blood pressure readings, maternal age, maternal BMI, maternal education level and maternal socioeconomic status. For predictive modeling we used the following variables: history of liver disease, gallbladder disease, diabetes prior to pregnancy, autoimmune disease, endocrine disease, blood clots or thromboembolic disease, heart disease, kidney disease, anemia, or hypertension, as well as first trimester blood pressure readings, maternal age, and maternal BMI. All clinical data was ascertained or measured at first trimester visits. b) Subject and sample selection for immune factor and microbiome profiling

[0081] Sixty-two individuals who developed sPEC were randomly chosen for immune factor and vaginal microbiome profiling. These 62 cases were matched with 62 participants who did not develop sPEC, eclampsia, or superimposed preeclampsia, during their pregnancy, although these individuals may have developed gestational hypertension or mild preeclampsia. These individuals were frequency matched based on a combination of maternal age and self-identified race, as well as enrollment site. c) Immune factor profding

[0082] The levels of the following immune factors were quantified in 150 ul aliquots of the samples intended for protein and metabolite analysis via a multiplexed Luminex assay (Luminex Corporation) using Millipore kits (HCYTA-60K-PX48 and HCYTA-60K): EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL- 10, IL- 18, IL- IRA, IL- la, IL-ip, IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L. All measurements were obtained in duplicate and data were in units of pg / ml. Immune factor profiles for the 124 participants were obtained on two separate days, with 84 samples profiled (42 sPEC cases, 42 controls) on one day, and 40 samples (20 sPEC cases, 20 controls) on another. There was no confounding association between processing batch and sPEC (Fisher’s Exact p=L0). To account for technical variation between batches, data was batch-standardized by subtracting the mean and dividing by the standard deviation of each feature in each batch. Concentrations and limits of detection were determined following the construction of a standard curve for each analyte. Values below the limit of detection were imputed with the limit of detection specific to each immune factor of the corresponding batch and then divided by 2. Values that were obtained with a high coefficient of variation (>20%) were replaced and imputed with the mean value of the corresponding batch for each immune factor.d) Microbiome profiling

[0083] Microbial DNA was extracted from 200 ul aliquots of the sample collected for microbial DNA analysis using the QIAcube HT and either the QIAmp 96 Virus QIAcube HT kit (Cat. No. 57731) or the Qiagen DNeasy PowerSoil HTP 96 kit (Cat. No. 12955-4-5D). The former kit was used to extract DNA from 27 sPEC cases and 27 controls, while the latter kit was used to extract DNA from all other samples. Extracted DNA was then prepared for sequencing via the Nextera DNA Library Prep Kit (Cat. No. 20060059). Metagenomic sequencing data was generated for each sample via paired end sequencing (2* 100bp) using a NovaSeq 6000 to a mean±std depth of 43.23 ± 23.4 million reads. Samples were processed in batches, which were explicitly balanced with respect to sPEC, clinical enrollment site, small for gestational age, maternal BMI, maternal age, maternal self-identified race, smoking status, maternal socioeconomic status, spontaneity of birth, chori oamnionitis, short cervix, cesarean birth, fetal sex, hypertensive disorders, live birth, gestational age, and prelabor membrane rupture, to ensure a lack of batch confounding. Therefore, there was no association between study group and kit or batch status in our experimental design (Fisher’s exact p=1.0 for kit, p=0.99 for batch).

[0084] Metagenomic sequencing reads from each sample were quality filtered using Trimmomatic (v0.39) in order to remove reads containing Illumina adapter sequences, bases with quality scores below 25, and reads shorter than 50 bases. Metagenomic reads were then aligned to the human genome and PhiX genome via Bowtie2. Any reads where either end mapped were subsequently removed. Following the removal of host DNA, all samples were subsampled to a consistent depth of 500,000 microbial reads to normalize the data across our dataset. Any sample with less than 500,000 reads was discarded from further analysis, leaving 110 samples (N sPEC = 57). Microbial abundances were then estimated using Kraken2 (v2.1.3) followed by Bracken (v2.9). In order to ensure the most accurate and relevant results, we utilized the Human Vaginal Microbiome Genome Collection (VMGC) as the reference database, which is specifically curated to include a comprehensive and diverse array of vaginal microbial genomes. As the VMGC classified G. vaginalis, G. piotii, G. leopoldii, and G. swidsinskii as species within the genus Bifidobacterium, we reclassified these species as members of the Gardnerella genus. All unmapped reads were accounted for in the final read count matrix in an ‘unmapped’ column. For all analyses, the microbiome data was preprocessed by converting to relative abundances followed by a center log transform with a pseudocount of10 to the power of the smallest whole number that keeps the pseudocount below the smallest observed relative abundance value across the dataset. e) Association of immune factors and microbes with sPEC

[0085] To evaluate associations between the microbiome data and sPEC, we performed two- sided Mann-Whitney U tests comparing the relative abundance of each microbe with a minimum of 10 detected reads in a minimum of 10 samples per group between the sPEC and non-sPEC groups, among samples within which a given microbe was detected. Additionally, we performed a robust principal component analysis to compare the entire microbial communities between the two groups. We investigated the relationships among microbes, immune factors, sPEC, and BMI via DIABLO. We performed the analysis for 1) the entire cohort; 2) the 43 individuals with a BMI<25; and 3) the 67 individuals with a BMI>25. Because nonrestrictive presence thresholds produced a high stochastic variation in results, we ensured consistency for the DIABLO analysis by reducing the number of features; only the microbes present in abundance of 10-3 in a minimum of 5% of samples were considered. The relevance network for the fitted DIABLO models were plotted using a ‘cutoff parameter of 0.5. f) Predicting risk of preeclampsia

[0086] To determine whether first trimester immune factor, microbiome, and clinical data is predictive of later diagnosis of sPEC, we trained and evaluated models using L2 regularized logistic regression. All predictive model pipelines were evaluated using a rebalanced leave- one-out cross-validation scheme, with an inner (nested) 5-fold cross-validation to tune hyperparameters (FIG. 6). The hyperparameters tuned in this analysis were the number of features to select via ANOVA F-value ([2, 5, 10]), implemented via scikit-learn’s ‘SelectKBesf , and the L2 regularization strength in the linear model which followed scikit- learn’s default cross-validation range from 1 to 108. The nested tuning also considered models stratified by BMI above and below 25, which, like other hyperparameters, was performed only on the inner folds. To apply the rebalanced leave-one-out approach with stratification for BMI, the additional held-out point was randomly selected from the same BMI group as the held-out test point. Combined models were constructed by averaging the predictions of each individual predictor across the three data types - microbiome, immune factor, and clinical. To determine the statistical significance of our models’ auROCs vs. a random null, we used the Mann- Whitney U test to compare the predictions made for sPEC and non-sPEC121.g) Assessing generalizability of microbiome, immune factor, and clinical models

[0087] To assess the ability of our models to generalize to data from the Multi-Omic Microbiome Study -Pregnancy Initiative (MOMS-PI) cohort, raw microbial reads and immune factor measurements were downloaded from dbGaP, study no. 20280 (accession ID phs001523.vl.pl). MOMS-PI samples were included from study participants who responded either yes or no during their delivery visit as to whether they experienced preeclampsia or eclampsia during pregnancy. Metagenomics data was processed as described above, and microbial abundances were also estimated using the VMGC reference database. Prior to validation, the two microbiome datasets were preprocessed using log-additive DEBIAS-M89 with default parameters, while only observing metadata from the nuMoM2b samples. Models based on clinical and immune factors data were trained and evaluated using only features shared by both cohorts. For the clinical models, this included only maternal age, BMI, weight, systolic blood pressure, diastolic blood pressure, and mean arterial pressure. For the immune factors, this included only 11 immune factors: IL-8, IL-4, IL-10, RANTES, TNFa, IL-6, IL- ip, IL-IRA, MCP-1, G-CSF, and IFNy. A single validation model was selected from within nuMoM by Rebalanced Leave-One-Out cross-validation to tune the same set of hyperparameters listed above. Prior to validation, the immune factor and clinical measurements from the MOMS-PI cohort were standardized using scikit-learn’s StandardScaler class. When combining multiple models together, we first standardized each model’s predictions using StandardScaler.Example 2. Early pregnancy immune factors are associated with sPEC

[0088] The overall profile of individuals who developed sPEC were compared to those who did not to determine whether the vaginal immune profile in early pregnancy is associated with the development of sPEC. Overall immune factor levels were weakly associated with the development of sPEC (PERMANOVA p=0.029; FIG. 2A), driven by their first principal component (PCI), which captured 32% of the variance (Mann-Whitney U p=0.013; FIG. 2B). No individual immune factor dominated the composition of this first principal component, with all immune factors having coefficients between 0.13 and 0.33, except EGF which had a negative coefficient (FIG. 2H). These results demonstrate that the association of the immune system with the early pathogenesis of sPEC is attributable to variation in multiple immune factors.

[0089] Intriguingly, a significant reduction in the total concentration of immune factors was measured in sPEC group (Mann-Whiteny U p=0.005, FIG. 2C). When next investigating theassociations of specific immune factors with sPEC, this pattern was consistent across most immune factors, as nearly all had lower levels in individuals who developed sPEC (17 out of 20; FIG. 2D). Similar reductions in both pro- and anti-inflammatory immune factors in PEC were previously reported in the placenta, and a similar observation of lower levels was also reported in the vaginal ecosystem for other adverse outcomes, such as spontaneous preterm birth.

[0090] Three specific immune factors were significantly reduced in individuals who subsequently developed sPEC (t-test p<0.05, q<0.2; FIG. 2D): Flt-3L (p=0.0087, q=0.095; Fig. 2E), a pro-inflammatory cytokine that activates dendritic cells in response to infection; EGF (p=0.024, q=0.16, FIG. 2F), a growth factor that stimulates the proliferation of epithelial cells and fibroblasts; and IL-IRA (p=0.0095, q=0.095; FIG. 2G), a competitive inhibitor of the IL- 1 receptor with a powerful anti-inflammatory effect. While it has similarly been reported that placental EGF is depleted in individuals with PEC, in the serum, levels of IL-IRA and Flt-3L late in pregnancy have previously been shown to be positively associated with PEC, suggesting that associations between immune factors and sPEC may be influenced by gestational age or biological niche. Thus, the local immune system in the vagina may be dysregulated early in pregnancy, months before symptoms become evident, among individuals who develop sPEC.

[0091] The association of immune factors significantly reduced in the vaginal ecosystem of individuals who developed sPEC may be due to compromised immune function within the vaginal ecosystem and perhaps along the entire reproductive tract, as it is known that immune cells are integral to maintaining the protective vaginal epithelium and play a role in spiral artery remodeling. This is echoed in recent investigations of placentas of women with preeclampsia, which exhibit reduced numbers of macrophages during delivery, and lower local expression of innate immune factors. The reduction in placental immune cells is also associated with decreased spiral artery remodeling in mice. At the same time, some of the vaginal immune factors that we found to be reduced in individuals who developed sPEC have been positively associated with preeclampsia when measured in serum late in pregnancy.

[0092] FIGs. 2A-2I show information related to immune factors relation to sPEC. In all box and swarm plots, the box represents the interquartile range (IQR), the line indicates the median, and the whiskers extend to the nearest point within 1.5 times the IQR.

[0093] FIG. 2A shows a principal component analysis (PCA) of vaginal immune factor profiles, with each sample colored according to whether the subject later developed severe preeclampsia (sPEC). This analysis demonstrates a separation of immune profiles by future sPEC status. FIG. 2B shows a box and swarm plot illustrating the distribution of the first principal component(PCI) of immune factor profiles, stratified by sPEC status. The plot demonstrates a significant association between immune factor PCI and the subsequent development of sPEC. FIG. 2C shows a box and swarm plot of the total concentration of the 20 measured vaginal immune factors, comparing subjects who developed sPEC to those who did not. The total immune factor concentration is significantly lower in individuals who developed sPEC. FIG. 2D shows a scatter plot with significance (y-axis; two-sided t-test p-value) and effect size (x-axis; Cohen’s d) for the association between each individual immune factor and sPEC status. Most immune factors tested have lower vaginal levels in individuals who developed sPEC.

[0094] FIG. 2E shows a box and swarm plot of Flt-3L, a pro-inflammatory cytokine that activates dendritic cells in response to infection, concentrations in vaginal samples, demonstrating that Flt-3L is significantly depleted in individuals who developed sPEC. FIG. 2F shows a box and swarm plot of EGF, a growth factor that stimulates the proliferation of epithelial cells and fibroblasts, concentrations in vaginal samples, demonstrating that EGF is significantly depleted in individuals who developed sPEC. FIG. 2G shows a box and swarm plot of IL-IRA, a competitive inhibitor of the IL-1 receptor with a powerful anti-inflammatory effect, in vaginal samples, demonstrating that IL- IRA is significantly depleted in individuals who developed sPEC.

[0095] FIG. 2H presents a bar plot of the coefficients for each immune factor in the first principal component (PCI) derived from the vaginal immune factor profiles. The figure demonstrates that the association between immune factor PCI and sPEC is not driven by a single immune factor, but rather by contributions from multiple immune factors. Most coefficients fall between 0.13 and 0.33, with EGF being the only factor with a negative coefficient. This supports the conclusion that variation across a broad panel of immune factors, rather than a single dominant marker, underlies the observed association with sPEC.

[0096] FIG. 21 shows the effect sizes and significance (p-values) for the association between individual vaginal immune factors and sPEC, specifically in individuals with BMI < 25. The plot indicates that, in this lower BMI group, only a few immune factors show notable associations with sPEC, and the overall pattern of immune factor reduction is less pronounced compared to the higher BMI group. This suggests that the relationship between vaginal immune factor levels and sPEC is weaker in individuals with BMI < 25.

[0097] FIG. 2J displays the effect sizes and significance (p-values) for the association between individual vaginal immune factors and sPEC, specifically in individuals with BMI > 25. In contrast to the lower BMI group, this figure shows that a greater number of immune factors are significantly reduced in individuals who develop sPEC, and the effect sizes are generally larger.This highlights that the association between reduced vaginal immune factor concentrations and sPEC is stronger and more consistent in individuals with higher BMI.

[0098] The immune profiles of the individuals who developed sPEC were compared to those who did not. Overall, immune factor levels are weakly associated with the development of sPEC, driven by their first principal component (PCI), which captured 32% of the variance. Similar reductions in both pro- and anti-inflammatory immune factors in PEC were previously reported in the placenta, and a similar observation of lower levels was also reported in the vaginal ecosystem for other adverse outcomes, such as spontaneous preterm birth. Therefore, the results disclosed herein suggest a phenomenon of reduced local immune factor concentrations in sPEC and potentially other pathologies.

[0099] Individuals with PEC, in the serum, levels of IL-IRA and Flt-3L late in pregnancy have previously been shown to be positively associated with PEC, suggesting that associations between immune factors and sPEC may be influenced by gestational age or biological niche. Overall, our results suggest that the local immune system in the vagina may be dysregulated early in pregnancy, months before symptoms become evident, among individuals who develop sPEC.

[0100] As shown, analysis of first-trimester vaginal immune factor profiles reveals that subjects who subsequently develop preeclampsia with severe features (sPEC) exhibit a significant reduction in the total concentration of measured immune factors compared to controls. Specifically, Flt-3L, EGF, and IL- IRA are among the immune factors most consistently reduced in these subjects. The association is attributable to variation in multiple immune factors, rather than dominance by a single marker (see FIG. 2H, Principal Component Analysis of Immune Factors).Example 3. Vaginal Microbiome Profiling and BMI-Conditioned Associations

[0101] Strain level abundances (ANI=99%) were quantified using a reference derived from a recent large-scale assembly effort. Similarly to analysis of immune factors, a robust principal component analysis was first used to examine the difference between the entire microbiome profiles of the cohort. However, a significant association between the vaginal microbiome and sPEC (PERMANOVA p=0.12; FIG. 3 A) or between the first microbiome robust principal component (RPC1) and sPEC (Mann-Whitney U p=0.21; FIG. 3B) was not detected. As BMI is associated with both PEC risk and the vaginal microbiome, microbiome-sPEC associations stratified by first trimester BMI were examined. We identified a relatively stronger association between the vaginal microbiome and sPEC in individuals with higher BMI (BMI>25; N=67,N sPEC=37), both when examining the entire microbiome profile (PERMANOVA p=0.01; FIG. 3 A) and when examining only RPC1 (Mann-Whitney U p=0.021; FIG. 3B). While we did not detect statistically significant associations in individuals with lower BMI (BM I<25; N=43, N sPEC=20; PERMANOVA p = 0.46; FIG. 3A; p=0.39; FIG. 3B), we noted opposite trends in the first principal coordinate between the two groups (higher values for sPEC in BMI <25, while lower in BMI >25; FIG. 3B). When examining whether community state types (CSTs), measured as dominance with >30% abundance (or lack thereof), we found a weak association between Gardnerella dominance and non-sPEC among individuals with BMI >25 (Fisher’s exact test p=0.03, q=0.14), but not in other groups or CSTs (FIGs. 3H and 31). Thus, BMI, which is associated with metabolic health and systemic inflammation, may impact the associations between vaginal microbes and sPEC.

[0102] Investigation of associations between specific microbes and the development of sPEC only identified weak associations in the full cohort (no taxa with Mann-Whitney U q<0.2; FIG. 3C). Additionally, while we did not find any differentially abundant or dominant taxa in individuals with BMI<25 (Mann-Whitney U p>0.01, q>0.2 for all taxa; FIGs. 3J and 3K), likely due in part to smaller sample size, we identified 24 taxa that were significantly associated with the development of sPEC in the high BMI group (Mann-Whitney U p<0.05, q<0.1 for 9 taxa, 0. l<q<0.2 for 15; FIG. 3D). The top three taxa with the most significant associations with sPEC were two G. vaginalis strains (p = 0.0007 overall, p = 0.0018 within BMI >25 for VMGC50 strain SGB020 and p=0.0027 overall, p=0.0014 within BMI >25 for SGB182) and an unidentified Bifidobacterium species (p=0.0014 overall, p=0.0016 within BMI >25; VMGC strain SGB065; FIG. 2E and 2F), which were all negatively associated with sPEC.

[0103] We observed that the majority of taxa strongly associated with lower sPEC risk in this group belong to the species Gardnerella vaginalis (8 of 20, 0.001<p<0.012, 0.06<q<0.16; FIG. 3C), which has been frequently associated with bacterial vaginosis, or to the genus Bifidobacterium (7 of 20, 0.002<p<0.017, 0.06<q<0.16; FIG. 3C), a Gram positive anaerobe that has various immunomodulatory effects and is the genus most closely related to G. vaginalis. We next observed that 3 of the 4 taxa positively associated with the development of sPEC among individuals with higher BMI belong to the species P. timonensis (0.005<p<0.016; 0.1 l<q<0.16). Prevotella species were previously shown to be associated with preterm birth and are major producers of sialidases.

[0104] Results of the vaginal microbiome profiling and BMI-conditioned associations are discussed below. In all box and swarm plots, the box represents the interquartile range (IQR),the line indicates the median, and the whiskers extend to the nearest point within 1.5 times the IQR.

[0105] FIG. 3 A shows a principal component analysis (PCA) of vaginal microbiome profiles, with each sample colored according to whether the subject later developed severe preeclampsia (sPEC). In the full cohort, there was no significant association between overall microbiome composition and sPEC. However, when stratified by BMI, a stronger association was detected in individuals with BMI >25.

[0106] FIG. 3B shows a box and swarm plot of the first robust principal component (RPC1) of microbiome profiles, stratified by BMI and sPEC status. The plot demonstrates a significant association between RPC1 and sPEC in the BMI >25 group, with opposite trends observed in BMI <25. FIGs. 3C and 3D present taxon-level associations between specific vaginal microbes and sPEC. In the full cohort, associations were generally weak (no taxa with Mann-Whitney U q<0.2). In individuals with BMI >25, 24 taxa were significantly associated with sPEC (Mann- Whitney U p<0.05, q<0.1 for 9 taxa, 0.1<q<0.2 for 15 taxa).

[0107] FIGs. 3E-3G show box and swarm plots for representative taxa. Two G. vaginalis strains (SGB020 and SGB182) and an unidentified Bifidobacterium species (SGB065) were negatively associated with sPEC (p=0.0018, p=0.0014, and p=0.0016, respectively), indicating lower risk. Conversely, P. timonensis was positively associated with sPEC, indicating increased risk. FIGs. 3H-3K display dominance analyses, where dominance is defined as >30% relative abundance. Gardnerella dominance was weakly associated with non-sPEC status among individuals with BMI > 25 (Fisher’s exact test p=0.03, q=0.14), but not in other groups or CSTs. Differential abundance results in BMI <25 were not significant.

[0108] Overall, the majority of taxa strongly associated with lower sPEC risk in the high BMI group belonged to G. vaginalis (8 of 20 taxa, p=0.001-0.012, q=0.06-0.16) o Bifidobacterium (7 of 20 taxa, p=0.002-0.017, q=0.06-0.16), both of which have known immunomodulatory effects. In contrast, three of the four taxa positively associated with sPEC in individuals with BMI >25 belonged to P. timonensis (p=0.005-0.016, q=0.11-0.16), a species previously linked to adverse pregnancy outcomes and known for its sialidase activity. Thus, G. vaginalis, Bifidobacterium, and P. timonensis may play a role in the development of sPEC by regulating the inflammatory milieu of the reproductive tract.

[0109] These findings demonstrate that the association between the early pregnancy vaginal microbiome and sPEC is modified by maternal BMI, with stronger and more specific microbial signals observed in individuals with BMI >25.Example 4. Combined Multi-omic Risk Assessment

[0110] The relationships among immune dysregulation and inflammation, both locally and systemically, was investigated starting with comparing the principal components of the vaginal immune factors with the robust principal components of the vaginal microbiome. We identified a significant correlation between microbiome RPC1 and the first two immune factor PCs (Pearson R=-0.38, p=4.5<I O5and R=0.27, p=0.0038; FIGs. 4A and 5 A), as well as between immune factor PCI and microbiome RPC3 (R=0.26 p=0.0026; FIG. 5 A). Accordingly, the vaginal microbiome and local host immune profile in early pregnancy are partially associated, although there are independent components to each.[OHl] FIGs. 4A-4E and 5A-5H show information related to the integration of vaginal microbiome, immune factors, and clinical features in relation to sPEC. The relationships among these data types were explored using principal component and correlation analyses, as well as integrative modeling approaches.

[0112] FIG. 4A shows a correlation analysis between the first robust principal component (RPC1) of the vaginal microbiome and the principal components (PCI and PC2) of the immune factor profiles. Specifically, RPC1 is negatively correlated with immune PCI and positively correlated with immune PC2 (Pearson R=0.27, p=0.0038), indicating partial association between the vaginal microbiome and local immune status in early pregnancy.

[0113] To further investigate these patterns and identify interactions among vaginal microbes, immune factors, and sPEC, we ran a sparse discriminant analysis using DIABLO, an integrative multi-omic analysis tool that identifies groups of features across multiple omics datasets that distinguish phenotypic groups. This analysis was stratified by maternal BMI. Even though DIABLO considers the associations between immune factors and sPEC in the context of how these features also relate to the vaginal microbiome and clinical characteristics, this analysis replicates results from Examples 2 and 3.

[0114] The majority of immune factors were found to be reduced in individuals who developed sPEC (FIGs. 4C, 4C, and 3 J). While most of these immune factors are secreted by immune cells, sCD40L is largely released by platelets, which often become dysfunctional in preeclampsia, so its association with sPEC here may reflect subclinical platelet dysfunction early in pregnancy. DIABLO also identified three pro-inflammatory immune factors in individuals with BMI <25, TGFa, MCP-1, and RANTES, which were elevated in sPEC (FIG. 4B). MCP-1 and RANTES both recruit immune cells to sites of inflammation and to the uterus during pregnancy and were shown to be elevated the serum of individuals with PEC. Thesehigher levels in sPEC suggest that the vaginal ecosystem in individuals with BMI <25 may better reflect the systemic proinflammatory milieu that is characteristic of preeclampsia.

[0115] To obtain better insights into the interactions among microbes, immune factors, and clinical features associated with sPEC, the network of multivariate correlations inferred by DIABLO was investigated. In individuals with BMI <25, we found four Prevotella strains which were negatively associated with blood pressure (FIG. 4D). Conversely, while the role of vaginal Prevotella in hypertension is unknown, studies have suggested that high levels of Prevotella in the gut are associated with increased risk of developing hypertension, suggesting an intriguing difference between Prevotella strains and the niche they occupy. Positive correlations were also found between blood pressure and the proinflammatory immune factors MCP-1 and RANTES, which have previously been associated with hypertension and atherosclerosis. We also observed that anemia was negatively correlated with a P. colorans strain and positively correlated with RANTES in this network (FIG. 4D). While anemia itself was not associated with sPEC (FIGs. 5C-5H), its correlations with microbes and immune factors that are associated with sPEC may reflect the higher incidence of anemia during pregnancy in individuals with BMI <25, which we also observed in our data (21% of individuals with BMI <25 had anemia, compared to 12% in BMI >25). It may also suggest indirect mechanisms by which anemia may be involved in the early pathogenesis of sPEC.

[0116] In the DIABLO correlation network for individuals with BMI >25, we identified several negative correlations between blood pressure and vaginal immune factors and microbes. Some of these associations, including a Lactobacillus crispatus strain that was positively associated with sPEC (albeit with a small coefficient in the DIABLO model; FIG. 5G) and blood pressure, and two S. vaginalis strains that were negatively associated with sPEC and blood pressure (FIG. 4E), are surprising in light of the view of L. crispatus as a highly protective species and S. vaginalis as a pathobiont associated with chori oamnionitis and preterm birth. We also identified several positive associations between S. vaginalis and immune factors that were lower in individuals who developed sPEC, including IL- 18, IL-27, Flt-3L, and sCD40L. The positive associations between S. vaginalis strains and immune factors may reflect the highly immunogenic response that S. vaginalis can elicit during pregnancy. We also identified strong correlations in this network between blood pressure and several immune factors, including Flt- 3L, sCD40L, and IL-18. This similar involvement of multiple immune factors that otherwise belong to different signaling pathways may reflect a single underlying process with broad effects on vaginal immunity in sPEC. The stark differences observed between the associationsidentified in different BMI groups strongly suggest that underlying systemic inflammation and metabolic health may exert a strong influence on the vaginal ecosystem in sPEC.

[0117] FIGs. 4B and 4C present DIABLO loadings that identify immune factors contributing to group separation by BMI stratum. In individuals with BMI <25, pro-inflammatory immune factors TGFa, MCP-1, and RANTES are elevated in sPEC. In contrast, in individuals with BMI >25, a broader set of immune factors are reduced in sPEC, highlighting BMI-dependent differences in immune response. FIGs. 4D and 4E display network graphs (relevance networks with correlation cutoff 0.5) that highlight BMI-specific multiomic associations. In the BMI <25 group, multiple Prevotella strains are negatively correlated with blood pressure, while blood pressure is positively correlated with MCP-1 and RANTES. Anemia is negatively correlated with P. colorans and positively correlated with RANTES. In the BMI >25 group, Sneathia vaginalis strains are negatively associated with sPEC and blood pressure, and positively correlated with immune factors IL- 18, IL-27, Flt-3L, and sCD40L. L. crispatus shows a positive association with sPEC and blood pressure in this multivariate context.

[0118] FIG. 5A details the correlation matrix between microbiome RPCs and immune PCs, providing a comprehensive view of multi-omic relationships. FIG. 5B presents a network analysis for all individuals, regardless of BMI. FIGs. 5C-5H show DIABLO coefficients for microbes, immune factors, and clinical features, split by BMI strata, further illustrating the contribution of each modality to sPEC risk prediction.

[0119] These multi-omic analyses highlight complex interactions among the vaginal microbiome, immune factors, and clinical features that vary with maternal BMI. The findings suggest that underlying systemic inflammation and metabolic health exert a strong influence on the vaginal ecosystem in sPEC, and that integrating data across these domains may improve early risk prediction.

[0120] To identify groups of features across multiple omics datasets that distinguish sPEC status, sparse discriminant analysis was performed using the DIABLO framework. Given the observed differences in associations by BMI, analyses were stratified by maternal BMI. In both BMI <25 and BMI >25 groups, the majority of immune factors were found to be reduced in individuals who developed sPEC. Notably, in the BMI <25 group, three pro-inflammatory immune factors — TGFa, MCP-1, and RANTES — were elevated in sPEC. MCP-1 and RANTES are known to recruit immune cells to sites of inflammation and have been previously reported to be elevated in the serum of individuals with preeclampsia. These findings suggest that, in individuals with lower BMI, the vaginal immune profile may better reflect the systemic pro-inflammatory environment characteristic of preeclampsia.

[0121] Network analysis of multi-omic correlations revealed distinct patterns by BMI group. In individuals with BMI <25, several Prevotella strains were negatively associated with blood pressure (FIG. 4D). Blood pressure was positively correlated with pro-inflammatory immune factors MCP-1 and RANTES, both of which have been linked to hypertension and atherosclerosis. Anemia was negatively correlated with a P. colorans strain and positively correlated with RANTES, suggesting indirect mechanisms by which anemia may be involved in the early pathogenesis of sPEC.

[0122] In the BMI >25 group, negative correlations were observed between blood pressure and both vaginal immune factors and specific microbes (FIG. 4E). For example, S. vaginalis strains were negatively associated with sPEC and blood pressure, while L. crispatus showed a positive association with sPEC and blood pressure in this context. Sneathia vaginalis also exhibited positive correlations with immune factors that were lower in individuals who developed sPEC, including IL- 18, IL-27, Flt-3L, and sCD40L. These associations may reflect the immunogenic response elicited by S. vaginalis during pregnancy.

[0123] Overall, these multi-omic analyses highlight complex interactions among the vaginal microbiome, immune factors, and clinical features that vary with maternal BMI. The findings suggest that underlying systemic inflammation and metabolic health exert a strong influence on the vaginal ecosystem in sPEC, and that integrating data across these domains may improve early risk prediction.Example 5. Vaginal immune factors and microbes predict sPEC months before diagnosis

[0124] Preeclampsia is typically not diagnosed until late in pregnancy. To assess the potential for early prediction of severe preeclampsia, predictive models were developed using clinical, immune factor, and microbiome data collected during the first trimester. Logistic regression models were trained and evaluated using a nested, rebalanced leave-one-out cross-validation scheme, with hyperparameters optimized within each training set and performance assessed on held-out samples. Given the observed interactions with BMI, stratification by BMI was included as a model parameter.

[0125] As a benchmark, we first devised models predicting sPEC using only clinical information, such as pre-pregnancy history of blood clots, hypertension, and heart disease. These models obtained reasonable accuracy on par with previous research. Training similar models using microbiome data obtained a stronger performance, and similar results were obtained using immune factor levels. We observed that 99% of microbiome-based models that were ultimately selected by cross validation within the training data utilized BMI stratification,compared to only 2% of immune factor-based models. These predictive associations, which generalized to held-out samples with conservative evaluation, further demonstrate the robust association between the development of sPEC and the early pregnancy vaginal microbiome and immunity.

[0126] FIGs. 7A and 7B show models using only clinical information, such as pre-pregnancy history of blood clots, hypertension, and heart disease, achieved moderate accuracy (area under the receiver operating characteristic curve [auROC] = 0.62; area under the precision-recall curve [auPR] = 0.64). Models trained on microbiome data alone performed better (auROC = 0.71, auPR = 0.67, p = l. l x lO-4), as did models using immune factor levels (auROC = 0.72, auPR = 0.73, p = 7.9* 10“5). Notably, nearly all microbiome-based models selected by cross- validation utilized BMI stratification, while only a small fraction of immune factor-based models did so.

[0127] Combining predictions from different data modalities further improved performance. Models integrating both microbiome and immune factor data achieved an auROC of 0.78 and auPR of 0.76. Adding clinical data provided a slight additional improvement. These results demonstrate that microbiome and immune factor data each contribute independent information regarding sPEC risk, while clinical data beyond BMI does not substantially enhance prediction accuracy. Despite being evaluated in a conservative nested cross-validation framework, the final performance of our model, with an auROC of 0.78, is superior to models based on blood proteins. While this performance is comparable to models based on other molecular data, such as serum cell-free RNA or P1GF and sFlt-1, which each achieve an auROC of 0.82, our models notably use data collected substantially earlier in pregnancy (first trimester compared to second trimester for cell-free RNA and late pregnancy for Pl GF and sFlt-1).Example 6. Validation

[0128] In some aspects, models trained on the primary cohort generalize to independent data. The MOMS-PI subcohort used for validation includes 17 individuals (5 with PEC / eclampsia by self-report), with significant differences vs. nuMoM2b in BMI, parity, and race, and differences in immune panels and clinical features available. Microbiome data are harmonized with DEBIAS-M; immune / clinical features are standardized.

[0129] To further demonstrate the potential of vaginal microbiome and immune factor data for sPEC diagnosis early in pregnancy, we next tested if models trained on data from our cohort can generalize to data from an independent external cohort. To this end, we used data from the MOMS-PI study, which generated metagenomic sequencing and immune factor data fromsamples collected early in pregnancy. Thirty-three participants with available early pregnancy data in this cohort self-reported whether they developed eclampsia or preeclampsia. While multiple samples were collected per participant in the MOMS-PI cohort, we used the earliest sample collected before week 17 of gestation. We processed metagenomic sequencing data from this cohort using the same pipeline we used to process data from the nuMoM2b cohort, retaining 17 individuals with >500,000 non-human reads, with 5 of them reporting a positive diagnosis. Of note, these participants differed significantly from our nuMoM2b cohort in their BMI (Mann-Whitney U p=0.0091) and parity (p=7.2>I O7), two factors known to impact the vaginal ecosystem52,88, as well as in self-identified race (p=3.1 >106; Table 2). Additionally, only 11 of the 20 immune factors measured in nuMoM2b were available for this cohort, and the only clinical features available were maternal age, blood pressure, BMI, and weight. We corrected processing bias in the microbiome data with DEBIAS-M, to which we applied models that were trained and selected using cross-validation within the nuMoM2b cohort to the MOMS-PI dataset without retraining (FIGs. 8A and 8B).Table 2

[0130] FIGs. 8 A and 8B show models trained on data from the nuMoM2b cohort and tested on data from the multi-omic Microbiome Study-Pregnancy Initiative (MOMS-PI). Despite differences in population characteristics, measurement techniques, and outcome assessment,the combined microbiome and immune factor model maintained strong performance in the external cohort. Models based solely on clinical data did not generalize well, while microbiome-only and immune factor-only models achieved auROC values of 0.68 and 0.77, respectively.

[0131] These findings demonstrate that predictive models leveraging first-trimester vaginal microbiome and immune factor data can accurately identify individuals at increased risk for sPEC months before clinical diagnosis. The robustness and generalizability of these models across cohorts underscore their potential utility for early risk assessment and intervention in pregnancy.

[0132] In some embodiments, a prediction model receives as inputs the quantified immune factor concentrations, microbial abundances, and optionally maternal BMI, and outputs a numerical risk score indicative of preeclampsia risk. The model may be trained using logistic regression, random forest, or other suitable algorithms, and validated using cross-validation and external cohort data.

[0133] When the risk score exceeds a predetermined threshold, clinical actions may be initiated, including enhanced surveillance, prophylactic therapy, counseling, or additional diagnostic testing. The timing of sample collection and risk assessment enables prediction at least two months prior to the typical symptom-based diagnosis window for preeclampsia.

[0134] More specifically, this disclosure, its aspects and embodiments, are not limited to the specific material types, components, methods, or other examples disclosed herein. Many additional material types, components, methods, and procedures known in the art are contemplated for use with particular implementations from this disclosure. Accordingly, for example, although particular implementations are disclosed, such implementations and implementing components may comprise any components, models, types, materials, versions, quantities, and / or the like as is known in the art for such systems and implementing components, consistent with the intended operation.

[0135] Many additional implementations are possible. Further implementations are within the CLAIMS.

[0136] It will be understood that implementations of the preceding disclosure include but are not limited to the specific components disclosed herein, as virtually any components consistent with the intended operation may be utilized. Accordingly, for example, it should be understood that, while the drawings and accompanying text show and describe particular implementations, any such implementation may comprise any shape, size, style, type, model, version, class,grade, measurement, concentration, material, weight, quantity, and / or the like consistent with the intended operation.

[0137] In places where the description above refers to particular implementations, it should be readily apparent that a number of modifications may be made without departing from the spirit thereof and that these implementations may be applied to other implementations disclosed or undisclosed. The presently disclosed are, therefore, to be considered in all respects as illustrative and not restrictive.REFERENCES CITED AND INCORPORATED BY REFERENCE o Duley, L. The global impact of pre-eclampsia and eclampsia. Semin. Perinatal. 33, (2009). o Pribadi, A. et al. Assessing the Impact of the Zero Mother Mortality Preeclampsia Program on Maternal Mortality Rates at a Single Center in Bandung, West Java (2015-2022): A Retrospective Study. Med. Sci. Monit. 29, e941097-l (2023). o Hodgins, S. Pre-eclampsia as Underlying Cause for Perinatal Deaths: Time for Action. Glob Health Sci Pract 3, 525-527 (2015). o Chappell, L. C., Cluver, C. A., Kingdom, J. & Tong, S. Pre-eclampsia. Lancet 398, 341— 354 (2021). o Wright, D. et al. Aspirin for Evidence-Based Preeclampsia Prevention trial: influence of compliance on beneficial effect of aspirin in prevention of preterm preeclampsia. Am. J. Obstet. Gynecol. 217, 685.el-685.e5 (2017). o Phipps, E. A., Thadhani, R., Benzing, T. & Karumanchi, S. A. Pre-eclampsia: pathogenesis, novel diagnostics and therapies. Nat. Rev. Nephrol. 15, 275-289 (2019). o American College of Obstetricians and Gynecologists’ Committee on Practice Bulletins- Obstetrics. Gestational hypertension and preeclampsia: ACOG Practice Bulletin, number 222. Obstet. Gynecol. 135, e237-e260 (2020). o Zeisler, H. et al. Predictive Value of the sFlt-l :PlGF Ratio in Women with Suspected Preeclampsia. N. Engl. J. Med. 374, 13-22 (2016). o Rasmussen, M. et al. RNA profiles reveal signatures of future health and disease in pregnancy. Nature 601, 422-427 (2022). o Black, K. D. & Horowitz, J. A. Inflammatory Markers and Preeclampsia: A Systematic Review. Nurs. Res. 67, 242-251 (2018). o Harmon, A. C. et al. The role of inflammation in the pathology of preeclampsia. Clin. Sci. 130, 409- 419 (2016).Spence, T. et al. Maternal Serum Cytokine Concentrations in Healthy Pregnancy and Preeclampsia. J. Pregnancy 2021, 6649608 (2021). Prins, J. R. et al. Preeclampsia is associated with lower percentages of regulatory T cells in maternal blood. Hypertens. Pregnancy 28, 300-311 (2009). Murray, E. J., Gumusoglu, S. B., Santillan, D. A. & Santillan, M. K. Manipulating CD4+ T Cell Pathways to Prevent Preeclampsia. Frontiers in Bioengineering and Biotechnology 9, (2022). Aggarwal, R. et al. Association of pro- and anti-inflammatory cytokines in preeclampsia. J. Clin. Lab. Anal. 33, e22834 (2019). Weel, I. C. et al. Association between Placental Lesions, Cytokines and Angiogenic Factors in Pregnant Women with Preeclampsia. PLoS One 11, e0157584 (2016). Milosevic-Stevanovic, J. et al. Number of decidual natural killer cells & macrophages in pre-eclampsia. Indian J. Med. Res. 144, 823-830 (2016). Kell, D. B. & Kenny, L. C. A Dormant Microbial Component in the Development of Preeclampsia. Front. Med. 3, 60 (2016). Beckers, K. F. & Sones, J. L. Maternal microbiome and the hypertensive disorder of pregnancy, preeclampsia. Am. J. Physiol. Heart Circ. Physiol. 318, H1-H10 (2020). Conde-Agudelo, A., Villar, J. & Lindheimer, M. Maternal infection and risk of preeclampsia: systematic review and metaanalysis. Am. J. Obstet. Gynecol. 198, 7-22 (2008). Goldenberg, R. L., Hauth, J. C. & Andrews, W. W. Intrauterine infection and preterm delivery. N. Engl. J. Med. 342, 1500-1507 (2000). Fettweis, J. M. et al. The vaginal microbiome and preterm birth. Nat. Med. 25, 1012-1021 (2019). Berard, A. R. et al. Vaginal epithelial dysfunction is mediated by the microbiome, metabolome, and mTOR signaling. Cell Rep. 42, 112474 (2023). Kindinger, L. M. et al. Relationship between vaginal microbial dysbiosis, inflammation, and pregnancy outcomes in cervical cerclage. Sci. Transl. Med. 8, 350ral02 (2016). Param el Jayaprakash, T., et al. High diversity and variability in the vaginal microbiome in women following preterm premature rupture of membranes (PPROM): A prospective cohort study. PLoS One 11, e0166794 (2016). Brown, R. G. et al. Establishment of vaginal microbiota composition in early pregnancy and its association with subsequent preterm prelabor rupture of the fetal membranes. Transl. Res. 207, 30- 43 (2019).Lin, C.-Y. et al. Severe preeclampsia is associated with a higher relative abundance of Prevotella bivia in the vaginal microbiota. Sci. Rep. 10, 18249 (2020). Geldenhuys, J. et al. Diversity of the gut, vaginal and oral microbiome among pregnant women in South Africa with and without pre-eclampsia. Front Glob Womens Health 3, 810673 (2022). Haas, D. M. et al. A description of the methods of the Nulliparous Pregnancy Outcomes Study: monitoring mothers-to-be (nuMoM2b). Am. J. Obstet. Gynecol. 212, 539. el- 539. e24 (2015). Fingar, K. R. et al. Delivery hospitalizations involving preeclampsia and eclampsia, 2005- 2014. in Healthcare Cost and Utilization Project (HCUP) Statistical Briefs (Agency for Healthcare Research and Quality (US), Rockville (MD), 2006). Kongwattanakul, K., Saksiriwuttho, P., Chaiyarach, S. & Thepsuthammarat, K. Incidence, characteristics, maternal complications, and perinatal outcomes associated with preeclampsia with severe features and HELLP syndrome. Int. J. Womens Health 10, 371— 377 (2018). Yang, Y. et al. Preeclampsia Prevalence, Risk Factors, and Pregnancy Outcomes in Sweden and China. JAMA Network Open 4, e218401 (2021). O’Brien, T. E., Ray, J. G. & Chan, W.-S. Maternal Body Mass Index and the Risk of Preeclampsia: A Systematic Overview. Epidemiology 14, 368 (2003). Bodnar, L. M., Ness, R. B., Markovic, N. & Roberts, J. M. The risk of preeclampsia rises with increasing prepregnancy body mass index. Ann. Epidemiol. 15, 475-482 (2005). Beigi, R. H., Yudin, M. H., Cosentino, L., Meyn, L. A. & Hillier, S. L. Cytokines, pregnancy, and bacterial vaginosis: comparison of levels of cervical cytokines in pregnant and nonpregnant women with bacterial vaginosis. J. Infect. Dis. 196, 1355-1360 (2007). Marangoni, A. et al. New Insights into Vaginal Environment During Pregnancy. Frontiers in Molecular Biosciences 8, (2021). Zanotta, N. et al. Cervico-vaginal secretion cytokine profile: A non-invasive approach to study the endometrial receptivity in IVF cycles. Am. J. Reprod. Immunol. 81, el3064 (2019). Imseis, H. M. et al. Characterization of the inflammatory cytokines in the vagina during pregnancy and labor and with bacterial vaginosis. J. Soc. Gynecol. Investig. 4, 90-94 (1997).Sanchez, A. et al. Variations in cytokine profiles between parous and nulliparous adult women. GREM - Gynecological and Reproductive Endocrinology & Metabolism 039-044 (2022). Kumar, M. et al. Vaginal Microbiota and Cytokine Levels Predict Preterm Delivery in Asian Women. Front. Cell. Infect. Microbiol. 11, (2021). Guermonprez, P. et al. Inflammatory Flt3L is essential to mobilize dendritic cells and for T cell responses during Plasmodium infection. Nat. Med. 19, 730-738 (2013-6). Karsunky, H., Merad, M., Cozzio, A., Weissman, I. L. & Manz, M. G. Flt3 ligand regulates dendritic cell development from Flt3+ lymphoid and myeloid-committed progenitors to Flt3+ dendritic cells in vivo. J. Exp. Med. 198, 305-313 (2003). Holgate, S. T. Epithelial damage and response. Clin. Exp. Allergy 30 Suppl 1, 37-41 (2000). Cao, S. et al. Epidermal growth factor receptor activation is essential for kidney fibrosis development. Nat. Commun. 14, 7357 (2023). Arend, W. P. The balance between IL-1 and IL-IRa in disease. Cytokine Growth Factor Rev. 13, 323- 340 (2002). Gabay, C., Lamacchia, C. & Palmer, G. IL-1 pathways in inflammation and human diseases. Nat. Rev. Rheumatol. 6, 232-241 (2010). Armant, D. R. et al. Reduced expression of the epidermal growth factor signaling system in preeclampsia. Placenta 36, 270-278 (2015). Stefanska, K. et al. Cytokine Imprint in Preeclampsia. Front. Immunol. 12, 667841 (2021). Greer, I. A., Lyall, F., Perera, T., Boswell, F. & Macara, L. M. Increased concentrations of cytokines interleukin-6 and interleukin- 1 receptor antagonist in plasma of women with preeclampsia: a mechanism for endothelial dysfunction? Obstet. Gynecol. 84, 937-940 (1994). Huang, L. et al. A multi-kingdom collection of 33,804 reference genomes for the human vaginal microbiome. Nat Microbiol 9, 2185-2200 (2024). Martino, C., et al. A novel sparse compositional technique reveals microbial perturbations. mSystems 4, (2019). Allen, N. G. et al. The vaginal microbiome in women of reproductive age with healthy weight versus overweight / obesity. Obesity 30, 142-152 (2022). Emanuela, F. et al. Inflammation as a link between obesity and metabolic syndrome. J. Nutr. Metab. 2012, 476380 (2012).Qing, W., Shi, Y., Zhou, H. & Chen, M. Gut microbiota dysbiosis in patients with preeclampsia: A systematic review. Medicine in Microecology 10, 100047 (2021). Miao, T. et al. Decrease in abundance of bacteria of the genus Bifidobacterium in gut microbiota may be related to pre-eclampsia progression in women from East China. Food Nutr. Res. 65, (2021). Schwebke, J. R., Muzny, C. A. & Josey, W. E. Role of Gardnerella vaginalis in the pathogenesis of bacterial vaginosis: a conceptual model. J. Infect. Dis. 210, 338-343 (2014). Riedel, C.-U. et al. Anti-inflammatory effects of bifidobacteria by inhibition of LPS- induced NF-kappaB activation. World J. Gastroenterol. 12, 3729-3735 (2006). Ruiz, L., Delgado, S., Ruas-Madiedo, P., Sanchez, B. & Margolles, A. Bifidobacteria and Their Molecular Communication with the Immune System. Front. Microbiol. 8, 2345 (2017). Cornejo, O. E., Hickey, R. J., Suzuki, H. & Forney, L. J. Focusing the diversity of Gardnerella vaginalis through the lens of ecotypes. Evol. Appl. 11, 312-324 (2018). Park, S. et al. Ureaplasma and Prevotella colonization with Lactobacillus abundance during pregnancy facilitates term birth. Sci. Rep. 12, 10148 (2022). Pelayo, P. et al. Prevotella are major contributors of sialidases in the human vaginal microbiome. bioRxiv 2024.01.09.574895 (2024) doi: 10.1101 / 2024.01.09.574895. Segui-Perez, C. et al. Prevotella timonensis degrades the vaginal epithelial glycocalyx through high fucosidase and sialidase activities. bioRxiv 2024.01.09.574844 (2024) doi: 10.1101 / 2024.01.09.574844. Farr Zuend, C., et al. Increased genital mucosal cytokines in Canadian women associate with higher antigen-presenting cells, inflammatory metabolites, epithelial barrier disruption, and the depletion ofL. crispatus. Microbiome 11, 159 (2023). Anahtar, M. N. et al. Cervicovaginal bacteria are a major modulator of host inflammatory responses in the female genital tract. Immunity 42, 965-976 (2015). Escalda, C., Botelho, J., Mendes, J. J. & Machado, V. Association of bacterial vaginosis with periodontitis in a cross-sectional American nationwide survey. Sci. Rep. 11, 630 (2021). Keller, M. J. et al. Longitudinal assessment of systemic and genital tract inflammatory markers and endogenous genital tract E. coli inhibitory activity in HIV-infected and uninfected women. Am. J. Reprod. Immunol. 75, 631-642 (2016).Singh, A. et al. DIABLO: an integrative approach for identifying key molecular drivers from multi-omics assays. Bioinformatics 35, 3055-3062 (2019). Hassan, G. S., Merhi, Y. & Mourad, W. CD40 ligand: a neo-inflammatory molecule in vascular diseases. Immunobiology 217, 521-532 (2012). Wood, G. W., Hausmann, E. & Choudhuri, R. Relative role of CSF-1, MCP-l / JE, and RANTES in macrophage recruitment during successful pregnancy. Mol. Reprod. Dev. 46, 62-9; discussion 69-70 (1997). Mellembakken, J. R., Solum, N. O., Ueland, T., Videm, V. & Aukrust, P. Increased concentrations of soluble CD40 ligand, RANTES and GRO-alpha in preeclampsia— possible role of platelet activation. Thromb. Haemost. 86, 1272-1276 (2001). Kauma, S. et al. Increased endothelial monocyte chemoattractant protein- 1 and interleukin- 8 in preeclampsia. Obstet. Gynecol. 100, 706-714 (2002). Redman, C. W., Sacks, G. P. & Sargent, I. L. Preeclampsia: an excessive maternal inflammatory response to pregnancy. Am. J. Obstet. Gynecol. 180, 499-506 (1999). Li, J. et al. Gut microbiota dysbiosis contributes to the development of hypertension. Microbiome 5, 14 (2017). Dinakis, E. et al. Association between the gut microbiome and their metabolites with human blood pressure variability. Hypertension 79, 1690-1701 (2022). Mikolajczyk, T. P. et al. Role of chemokine RANTES in the regulation of perivascular inflammation, T-cell accumulation, and vascular dysfunction in hypertension. FASEB J. 30, 1987-1999 (2016). Rabkin, S. W., Langer, A., Ur, E., Calciu, C.-D. & Leiter, L. A. Inflammatory biomarkers CRP, MCP-1, serum amyloid alpha and interleukin- 18 in patients with HTN and dyslipidemia: impact of diabetes mellitus on metabolic syndrome and the effect of statin therapy. Hypertens. Res. 36, 550- 558 (2013). Tan, J. et al. Association between maternal weight indicators and iron deficiency anemia during pregnancy: A cohort study: A cohort study. Chin. Med. J. (Engl.) 131, 2566-2574 (2018). Callahan, B. J. et al. Replication and refinement of a vaginal microbial signature of preterm birth in two racially distinct cohorts of US women. Proceedings of the National Academy of Sciences 114, 9966- 9971 (2017). Stare, M., Lucovnik, M., Erzen Vrlic, P. & Jeverica, S. Protective effect of Lactobacillus crispatus against vaginal colonization with group B streptococci in the third trimester of pregnancy. Pathogens 11, 980 (2022).Nunn, K. L. et al. Enhanced trapping of HIV-1 by human cervicovaginal mucus is associated with Lactobacillus crispatus-dominant Microbiota. MBio 6, e01084-15 (2015). McCoy, Z. T. et al. Antibody Response to the Sneathia vaginalis Cytopathogenic Toxin A during Pregnancy. ImmunoHorizons 8, 114-121 (2024). Theis, K. R. et al. Sneathia: an emerging pathogen in female reproductive disease and adverse perinatal outcomes. Crit. Rev. Microbiol. 47, 517-542 (2021). Austin, G. I., Pe’er, I. & Korem, T. Distributional bias compromises leave-one-out cross- validation. arXiv [stat.ME] (2024). Tiruneh, S. A. et al. Externally validated prediction models for pre-eclampsia: systematic review and meta-analysis. Ultrasound Obstet. Gynecol. 63, 592-604 (2024). Li, S., et al. Improving preeclampsia risk prediction by modeling pregnancy trajectories from routinely collected electronic medical record data. NPJ Digital Medicine 5, (2022). Lin, Y. C. et al. Preeclampsia Predictor with Machine Learning: A Comprehensive and Bias-Free Machine Learning Pipeline. medRxiv 2022.06.08.22276107 (2022) doi: 10.1101 / 2022.06.08.22276107. Ghaemi, M. S. et al. Proteomic signatures predict preeclampsia in individual cohorts but not across cohorts - implications for clinical biomarker studies. J. Matem. Fetal. Neonatal Med. 35, 5621-5628 (2022). Romero, R. et al. The vaginal Microbiota of pregnant women varies with gestational age, maternal age, and parity. Microbiol. Spectr. 11, e0342922 (2023). Austin, G. I. et al. Processing-bias correction with DEBIAS-M improves cross-study generalization of microbiome-based prediction models. bioRxiv 2024.02.09.579716 (2024) doi: 10.1101 / 2024.02.09.579716. Moufarrej, M. N. et al. Early prediction of preeclampsia in pregnancy with cell-free RNA. Nature 602, 689-694 (2022). Lee, S. K., Kim, C. J., Kim, D.-J. & Kang, J.-H. Immune cells in the female reproductive tract. Immune Netw. 15, 16-26 (2015). Smith, S. D., Dunk, C. E., Aplin, J. D., Harris, L. K. & Jones, R. L. Evidence for immune cell involvement in decidual spiral arteriole remodeling in early human pregnancy. Am. J. Pathol. 174, 1959-1971 (2009). Broekhuizen, M. et al. The Placental Innate Immune System Is Altered in Early-Onset Preeclampsia, but Not in Late-Onset Preeclampsia. Front. Immunol. 12, (2021). Menkhorst, E. et al. IL11 activates the placental inflammasome to drive preeclampsia. Front. Immunol. 14, 1175926 (2023).Karge, A. et al. Performance of sFlt-1 / PIGF ratio for the prediction of perinatal outcome in obese pre-eclamptic women. J. Clin. Med. 11, 3023 (2022). Klein, K. O. et al. Effect of obesity on estradiol level, and its relationship to leptin, bone maturation, and bone mineral density in children. J. Clin. Endocrinol. Metab. 83, 3469- 3475 (1998). Shen, Z. et al. Effect of BMI on the value of serum progesterone to predict clinical pregnancy outcome in IVF / ICSI cycles: a retrospective cohort study. Front. Endocrinol. (Lausanne) 14, 1162302 (2023). Ingram, K., Ngalame Eko, E., Nunziato, J., Ahrens, M. & Howell, B. Impact of obesity on the perinatal vaginal environment and bacterial microbiome: effects on birth outcomes. J. Med. Microbiol. 73, 001874 (2024). McGregor, J. A. et al. Bacterial vaginosis is associated with prematurity and vaginal fluid mucinase and sialidase: results of a controlled trial of topical clindamycin cream. Am. J. Obstet. Gynecol. 170, 1048-59; discussion 1059-60 (1994). Cauci, S. & Culhane, J. F. High sialidase levels increase preterm birth risk among women who are bacterial vaginosis-positive in early gestation. Am. J. Obstet. Gynecol. 204, 142. el-9 (2011). Lin, C.-Y. et al. Severe preeclampsia is associated with a higher relative abundance of Prevotella bivia in the vaginal microbiota. Sci. Rep. 10, 18249 (2020). Freitas, A. C. & Hill, J. E. Quantification, isolation and characterization of Bifidobacterium from the vaginal microbiomes of reproductive aged women. Anaerobe 47, 145-156 (2017). Schwecht, I., Nazli, A., Gill, B. & Kaushic, C. Lactic acid enhances vaginal epithelial barrier integrity and ameliorates inflammatory effects of dysbiotic short chain fatty acids and HIV-1. Sci. Rep. 13, 20065 (2023). Romero, R. et al. Sterile and microbial-associated intra-amniotic inflammation in preterm prelabor rupture of membranes. J. Matern. Fetal. Neonatal Med. 28, 1394-1409 (2015). Srinivasan, S. et al. Bacterial communities in women with bacterial vaginosis: high resolution phylogenetic analyses reveal relationships of microbiota to clinical criteria. PLoS One 7, e37818 (2012). Ali, A. A., Rayis, D. A., Abdallah, T. M., Elbashir, M. I. & Adam, I. Severe anaemia is associated with a higher risk for preeclampsia and poor perinatal outcomes in Kassala hospital, eastern Sudan. BMC Res. Notes 4, 311 (2011). Johnson, A., Vaithilingan, S. & Avudaiappan, S. L. The interplay of hypertension and anemia on pregnancy outcomes. Cureus 15, e46390 (2023).Rabkin, S. W. The role of interleukin 18 in the pathogenesis of hypertension-induced vascular disease. Nat. Clin. Pract. Cardiovasc. Med. 6, 192-199 (2009). Ferroni, P. & Guadagni, F. Soluble CD40L and its role in essential hypertension: diagnostic and therapeutic implications. Cardiovasc. Hematol. Disord. Drug Targets 8, 194-202 (2008). Lu, X. et al. Classical Dendritic Cells Mediate Hypertension by Promoting Renal Oxidative Stress and Fluid Retention. Hypertension 75, 131-138 (2020). Stefan, N., Haring, H.-U., Hu, F. B. & Schulze, M. B. Metabolically healthy obesity: epidemiology, mechanisms, and clinical implications. Lancet Diabetes Endocrinol. 1, 152— 162 (2013). Greenland, P. et al. Large-Scale Proteomics in Early Pregnancy and Hypertensive Disorders of Pregnancy. JAMA Cardiol (2024) doi : 10.1001 / jamacardio.2024.1621. Hypertension in pregnancy. Report of the American College of Obstetricians and Gynecologists’ Task Force on Hypertension in Pregnancy. Obstet. Gynecol. 122, 1122— 1131 (2013). ACOG Committee Opinion No. 743 Summary: Low-Dose Aspirin Use During Pregnancy. Obstet. Gynecol. 132, 254-256 (2018). Bolger, A. M., Lohse, M. & Usadel, B. Trimmomatic: a flexible trimmer for Illumina sequence data. Bioinformatics 30, 2114-2120 (2014). Langmead, B. & Salzberg, S. L. Fast gapped-read alignment with Bowtie 2. Nat. Methods 9, 357-359 (2012). Li, H. et al. The Sequence Alignment / Map format and SAMtools. Bioinformatics 25, 2078- 2079 (2009). Wood, D. E., Lu, J. & Langmead, B. Improved metagenomic analysis with Kraken 2. Genome Biol. 20, 257 (2019). Lu, J., Breitwieser, F. P., Thielen, P. & Salzberg, S. L. Bracken: estimating species abundance in metagenomics data. PeerJ Comput. Sci. 3, el04 (2017). Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 12, 2825-2830 (2011). Mason, S. J. & Graham, N. E. Areas beneath the relative operating characteristics (ROC) and relative operating levels (ROL) curves: Statistical significance and interpretation. Q. J. R. Meteorol. Soc. 128, 2145-2166 (2002).

Claims

CLAIMSWhat is claimed is:

1. A method of predicting a risk or the likelihood that a pregnant patient will develop preeclampsia, the method comprising:(a) obtaining a cervicovaginal or vaginal swab from the subject during weeks 6-14 of gestation;(b) quantifying, in the cervicovaginal or vaginal sample, concentrations of local immune factors comprising at least one of Flt-3L, EGF, and IL-IRA;(c) determining, from the sample, abundances of vaginal microbial taxa comprising at least one of a Bifidobacterium species, a Prevotella species, a Gardnerella species, a Lactobacillus species, a Bergeylla species, a Megasphaera species, an Aerococcus species, a Fenollaria species, a Sneathia species, and a Mobiluncus species; and(d) generating, by a prediction model that receives the immune-factor concentrations and microbial abundances as inputs, a numerical risk score indicative of preeclampsia, wherein the prediction model utilizes a machine learning algorithm or a statistical algorithm trained to map the inputs to generate a predicted risk value.

2. The method of claim 1, further comprising quantifying concentrations of local immune factors selected from G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-la, IL-1 , IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L.

3. The method of claim 1 or 2, wherein the method further comprises: measuring a maternal body mass index (BMI) value during the first trimester, wherein the prediction model receives the maternal BMI value as an input to generate the numerical risk score for preeclampsia.

4. The method of claim 3, wherein the prediction model is parameterized by a BMI >25 stratum.

5. The method of any one of claims 1-4, wherein the abundances of a Bifidobacterium species, Prevotella limonensis. Sneathia vaginalis, and Gardnerella vaginalis are determined and the prediction model receives the abundances of a Bifidobacterium species, Prevotella timonensis, Sneathia vaginalis, and Gardnerella vaginalis as an input to generate the numerical risk score for preeclampsia.

6. The method of any one of claims 1-5, further comprising initiating a clinical action when the numerical risk score exceeds a threshold, wherein the clinical action isselected from the group consisting of: initiating or intensifying prophylactic therapy, enrolling the subject in enhanced surveillance, adjusting visit cadence, counseling, and scheduling additional diagnostic testing.

7. The method of any one of claims 1-6, wherein reduced concentration of Flt-3L, EGF, and IL- IRA relative to gestational-age-matched controls increases the risk score.

8. The method of any one of claims 1-7, wherein presence of P. timonensis and / or reduced Bifidobacterium abundance increases the risk score when BMI >25.

9. A method of predicting the likelihood of a pregnant patient developing preeclampsia comprising: providing a biological sample from a pregnant patient; and determining the concentration of an immune factor from the biological sample, wherein the immune factor is selected from the group consisting of: Flt-3L, EGF, and IL- IRA; wherein reduced concentration of the immune factor in the pregnant patient compared to a control level indicates the pregnant patient has an increased likelihood of developing preeclampsia.

10. The method of claim 9, wherein the pregnant patient has a body mass index of 25 or more, the method further comprises: identifying vaginal microbiome from the biological sample collected from the pregnant patient; and determining the presence of Prevotella species and / or the abundance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome, wherein: the presence of P. timonensis in the vaginal microbiome indicates the pregnant patient has an increased likelihood of developing preeclampsia; dominance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome indicates the subject has reduced likelihood of developing preeclampsia.

11. The method of claim 9 or 10, wherein the biological sample is a vaginal swab.

12. The method of any one of claims 9-11, wherein the biological sample is collected from the subject before 15 weeks of gestation.

13. A method of predicting the likelihood of a pregnant patient developing preeclampsia comprising:identifying vaginal microbiome from a biological sample collected from the pregnant patient, wherein the pregnant patient has a body mass index of more than 25; and at least one step from: determining the abundance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome, wherein dominance of Gardnerella species and / or Bifidobacterium species in the vaginal microbiome indicates the subject has reduced likelihood of developing preeclampsia; or determining the presence of Prevotella species in the vaginal microbiome, wherein the presence of Prevotella timonensis in the vaginal microbiome indicates the subject has increased likelihood of developing preeclampsia.

14. The method of any one of the preceding claims, wherein generating the numerical risk score for preeclampsia occurs at least two months prior to a symptom-based clinical diagnosis window for preeclampsia.

15. A computer-implemented system for early prediction of preeclampsia for use in the method of any one of claims 1-14, the system comprising:(i) a sample-processing module configured to receive a first-trimester cervicovaginal or vaginal swab and (A) quantify levels of at least one immune factor selected from EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL- 27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L, and (B) determine abundances of vaginal microbes comprising at least one of a Bifidobacterium species, Prevotella timonensis, Sneathia vaginalis, and Gardnerella vaginalis,'(ii) a memory storing a trained prediction model that outputs a risk score for preeclampsia based on immune factor levels and microbial abundances and that is optionally stratified or parameterized by a maternal BMI threshold; and(iii) a processor operatively coupled to the sample-processing module and the memory and configured to execute the trained prediction model to generate the risk score and provide, to a user interface, an indication of the risk score and a corresponding clinical action recommendation.

16. A kit for first-trimester assessment of preeclampsia risk for use in the method of any one of claims 1-14, the kit comprising:(A) assay reagents for detecting in a cervicovaginal or vaginal swab at least one immune factors selected from the group consisting of: EGF, Flt-3L, G-CSF, IFNa2, IFN-y, IL-10, IL-18, IL-IRA, IL-la, IL-1 , IL-27, IL-4, IL-6, IL-8, MCP-1, RANTES, TGFa, TNFa, VEGF, and sCD40L;(B) nucleic-acid reagents to detect or quantify, in the sample, at least one microbial taxon selected from a Bifidobacterium species, Prevotella limonensis. Sneathia vaginalis, and Gardnerella vaginalis by sequencing, target enrichment, or amplification; and(C) instructions specifying that the sample is collected during weeks 6-14 of gestation and that measured immune factor levels and microbial abundances are combined, optionally with maternal BMI information, maternal blood pressure information, maternal type II diabetes status, and / or maternal hbAlC information, to compute a risk score indicative of preeclampsia or preeclampsia with severe features.

17. A non-transitory computer-readable medium storing instructions that, when executed by a processor, cause a computer system to perform a method comprising: receiving, as input, data representing immune factor concentrations and microbial biomarker abundances from a vaginal sample of a pregnant patient; inputting the data into a machine learning model trained to predict preeclampsia risk; generating, using the machine learning model, an output indicative of the likelihood of preeclampsia; and providing the output to a user or system.

18. The non-transitory computer-readable medium of claim 17, wherein the inputs comprise a vector of immune-factor concentrations and a vector of CLR-transformed microbial abundances.

19. A system for predicting a risk that a pregnant patient will develop preeclampsia, the system comprising:(A) a sample processing module configured to: a. receive a cervicovaginal or vaginal swab obtained from the subject during weeks 6-14 of gestation; b. generate sequencing reads from the sample; and c. quantify concentrations of immune factors from the cervicovaginal or vaginal swab;(B) one or more processors and a non-transitory memory storing instructions that, when executed, cause the system to:a. estimate and correct laboratory processing bias across taxa using a model that imposes similarity constraints based on phylogenetic relatedness; b. determine abundances of vaginal microbial taxa; c. optionally receive a maternal body mass index (BMI) value and parameterize or stratify subsequent modeling by a BMI threshold; d. train, validate, and / or apply a prediction model that receives as inputs the bias corrected microbial abundances, the immune factor concentrations, and optionally the BMI value, wherein training employs a rebalanced leave one out cross validation scheme with intra training hyperparameter selection to improve generalization across collection sites and independent cohorts; e. generate a numerical risk score indicative of preeclampsia; f. provide, to a user interface, the risk score and, when the risk score exceeds a threshold, one or more clinical action recommendations; and g. optionally detect and remove contaminant microbial signal from the sequencing reads by identifying clonal strain profiles recurring across samples and inconsistent with biological co-occurrence and / or negative control profiles.

20. The system of claim 19, wherein the sample processing module quantifies concentrations of at least one immune factor selected from the group consisting of: Flt-3L, EGF, and IL- IRA.

21. The system of claim 19 or 20, wherein the non-transitory memory storing instructions that, when executed, causes the system to determine, from the decontaminated and bias corrected data, abundances of at least one species selected from group consisting of: aBifidobacterium species, Prevotella limonensis, Sneathia vaginalis, and Gardnerella vaginalis.

Images