Molecular prediction of pregnancy-related disorders

Abstract

Pre-eclampsia (PE), gestational hypertension (GHT), gestational diabetes (GDM), small or large for gestational age (SGA, LGA) and preterm birth (PTB) are common pregnancy-related disorders, associated with adverse maternal and perinatal outcomes. Being able to predict which women are likely to experience these pregnancy disorders would enable improved antenatal care, such as treatment allocation and more intense monitoring in those at the highest risk. Current
methods for prediction need improvement to better identify high and low risk women. In my thesis, I investigate improving prediction using metabolomics and epigenomics.

Chapter 1 introduces these pregnancy-related disorders, the case for improving prediction, the rationale for omics for improved prediction, and a summary of systematic reviews regarding prediction of these pregnancy-related disorders. It presents the aims and objectives of my thesis.

Chapter 2 describes in detail the metabolomic profiling of pregnant women in the Born in Bradford (BiB) cohort. In this chapter, I describe the metabolomic data available in BiB, collected during pregnancy, at birth, and the early life in the offspring. These data were generated using two platforms of metabolomic profiling: nuclear magnetic resonance (NMR) and mass spectrometry (MS), discussed in detail.

Chapters 3 and 4 report investigations of metabolomic prediction of pregnancy-related disorders: GDM, hypertensive disorders of pregnancy (HDP), SGA, LGA and PTB. In Chapter 3, I utilise NMR profiling in two birth cohorts: the BiB study for prediction model training and testing and the UK Pregnancies Better Eating and Activity Trial (UPBEAT) for external validation. I found that NMR metabolomics improves the prediction of GDM, HDP, SGA and LGA when compared with
maternal risk factors (age, smoking, parity, body mass index (BMI), ethnicity). Prediction was poor for PTB. External validation in UPBEAT showed similar, but weaker, results.

Chapter 4 reports MS profiling for metabolomic prediction of GDM, HDP, SGA, LGA and PTB. I used two birth cohorts: the BiB study for prediction model training and testing and the Pregnancy Outcome Prediction study (POPs) for external validation. I found that metabolites improve the prediction of GDM, HDP, and LGA when compared with maternal risk factors (age, smoking, parity, BMI, ethnicity). Prediction was improved, but modest, for SGA and poor for PTB. External
validation showed very similar results in POPs.

Chapters 5 and 6 report investigations of DNA methylation (DNAm) prediction of GDM and HDP, respectively. In Chapter 5, I compared models of National Institute of Health and Care Excellence (NICE) guideline risk factors for GDM, with models of all profiled CpG sites on the Illumina HumanMethylation EPIC BeadChip array (EPIC), and models comprised of CpG sites from epigenome-wide association studies (EWAS) of factors known to associate with GDM. The models are trained and tested in BiB. I found that a model of 13 CpG sites improved upon models
of maternal risk factors. No suitable cohorts were available for external validation.
In Chapter 6, I compared NICE guideline risk factors for GDM with prediction models of all profiled CpG sites on the Human Methylation 450k (HM450) BeadChip array and models comprised of CpG sites from epigenome-wide association studies (EWAS) of factors known to associate with HDP. The models are trained and tested in BiB and externally validated in the Avon Longitudinal
Study of Parents and Children (ALSPAC). Risk factors were better predictors of HDP than DNAm.

Chapter 7 discusses how the work described in this thesis meets my aims and objectives including overall results, strengths, limitations, implications for clinical use and further research. I discuss the extent to which they support my original hypothesis that molecular data can improve the prediction of pregnancy-related disorders.

Date of Award	28 Sept 2021
Original language	English
Awarding Institution	University of Bristol
Supervisor	Debbie A Lawlor (Supervisor), Paul D Yousefi (Supervisor), Caroline L Relton (Supervisor) & Matthew J Suderman (Supervisor)