STATISTICAL METHODS FOR CONTROLLING POPULATION STRATIFICATION AND GENE-BASED ASSOCIATION STUDIES

Polygenic risk scores (PRSs) have been among the leading advances in biomedicine in recent years. As a proxy of genetic liability, PRSs are utilised across multiple fields and applications. While numerous statistical and machine learning methods have been developed to optimise their predictive accuracy, these typically distil genetic liability to a single number based on aggregation of an individual's genome-wide risk alleles. This results in a key loss of information about an individual's genetic profile, which could be critical given the functional sub-structure of the genome and the heterogeneity of complex disease. In this manuscript, we introduce a 'pathway polygenic' paradigm of disease risk, in which multiple genetic liabilities underlie complex diseases, rather than a single genome-wide liability. We describe a method and accompanying software, PRSet, for computing and analysing pathway-based PRSs, in which polygenic scores are calculated across genomic pathways for each individual. We evaluate the potential of pathway PRSs in two distinct ways, creating two major sections: (1) In the first section, we benchmark PRSet as a pathway enrichment tool, evaluating its capacity to capture GWAS signal in pathways. We find that for target sample sizes of >10,000 individuals, pathway PRSs have similar power for evaluating pathway enrichment as leading methods MAGMA and LD score regression, with the distinct advantage of providing individual-level estimates of genetic liability for each pathway -opening up a range of pathway-based PRS applications, (2) In the second section, we evaluate the performance of pathway PRSs for disease stratification. We show that using a supervised disease stratification approach, pathway PRSs (computed by PRSet) outperform two standard genome-wide PRSs (computed by C+T and lassosum) for classifying disease subtypes in 20 of 21 scenarios tested. As the definition and functional annotation of pathways becomes increasingly refined, we expect pathway PRSs to offer key insights into the heterogeneity of complex disease and treatment response, to generate biologically tractable therapeutic targets from polygenic signal, and, ultimately, to provide a powerful path to precision medicine.

ObjectiveTo determine whether the polygenic risk score (PRS) derived from MEGASTROKE is associated with ischemic stroke (IS) and its subtypes in an independent tertiary health care system and to identify the PRS derived from gene sets of known biological pathways associated with IS.MethodsControls (n = 19,806/7,484, age ≥69/79 years) and cases (n = 1,184/951 for discovery/replication) of acute IS with European ancestry and clinical risk factors were identified by leveraging the Geisinger Electronic Health Record and chart review confirmation. All Geisinger MyCode patients with age ≥69/79 years and without any stroke-related diagnostic codes were included as low risk control. Genetic heritability and genetic correlation between Geisinger and MEGASTROKE (EUR) were calculated using the summary statistics of the genome-wide association study by linkage disequilibrium score regression. All PRS for any stroke (AS), any ischemic stroke (AIS), large artery stroke (LAS), cardioembolic stroke (CES), and small vessel stroke (SVS) were constructed by PRSice-2.ResultsA moderate heritability (10%–20%) for Geisinger sample as well as the genetic correlation between MEGASTROKE and the Geisinger cohort was identified. Variation of all 5 PRS significantly explained some of the phenotypic variations of Geisinger IS, and the R2 increased by raising the cutoff for the age of controls. PRSLAS, PRSCES, and PRSSVS derived from low-frequency common variants provided the best fit for modeling (R2 = 0.015 for PRSLAS). Gene sets analyses highlighted the association of PRS with Gene Ontology terms (vascular endothelial growth factor, amyloid precursor protein, and atherosclerosis). The PRSLAS, PRSCES, and PRSSVS explained the most variance of the corresponding subtypes of Geisinger IS suggesting shared etiologies and corroborated Geisinger TOAST subtyping.ConclusionsWe provide the first evidence that PRSs derived from MEGASTROKE have value in identifying shared etiologies and determining stroke subtypes.

PurposeTo determine genetic correlations between common myopia and primary open-angle glaucoma (POAG).MethodsWe tested the association of myopia polygenic risk scores (PRSs) with POAG and POAG endophenotypes using two studies: the Australian & New Zealand Registry of Advanced Glaucoma (ANZRAG) study comprising 798 POAG cases with 1992 controls, and the Rotterdam Study (RS), a population-based study with 11,097 participants, in which intraocular pressure (IOP) and optic disc parameter measurements were catalogued. PRSs were derived from genome-wide association study meta-analyses conducted by the Consortium for Refractive Error and Myopia (CREAM) and 23andMe. In total, 12 PRSs were constructed and tested. Further, we explored the genetic correlation between myopia, POAG, and POAG endophenotypes by using the linkage disequilibrium score regression (LDSC) method.ResultsWe did not find significant evidence for an association between PRS of myopia with POAG (P = 0.81), IOP (P = 0.07), vertical cup–disc ratio (P = 0.42), or cup area (P = 0.25). We observed a nominal association with retinal nerve fiber layer (P = 7.7 × 10−3) and a significant association between PRS for myopia and disc area (P = 1.59 × 10−9). Using the LDSC method, we found a genetic correlation only between myopia and disc area (genetic correlation [RhoG] = −0.12, P = 1.8 × 10−3), supporting the findings of the PRS approach.ConclusionsUsing two complementary approaches we found no evidence to support a genetic overlap between myopia and POAG; our results suggest that the comorbidity of these diseases is not influenced by common variants. The association between myopia and optic disc size is well known and validates this methodology.

Inflammation is endemic to obesity and type 2 diabetes (T2D) but it is unclear if it is a cause or a consequence. If the latter, then inflammation may increase the risk of obesity and diabetes complications. Large genome-wide association studies (GWAS) of inflammation and cardiometabolic traits can be used to investigate causal mechanisms. Summary statistics were assembled from GWAS of 49 cytokine (N=840-80,000) and 20 cardiometabolic traits (N=2,447-465,333). Shared genetic background was measured using LD score regression. Polygenic scores (PGS) were generated from GWAS data for 49 cytokines (representing inflammation) and 20 cardiometabolic traits and used in two-way Mendelian Randomisation (MR) analyses. Causal relationships were identified using summary statistics for all traits, and individual data (UK Biobank, N=465,333) to test the association of cytokine PGS with body mass index (BMI) and T2D. LD score regression analysis demonstrated that BMI and T2D shared a genetic background with acute phase cytokines. However, MR analyses showed that genetically-driven variation in cytokine levels were not associated with increased BMI or T2D risk. This argues against a causal role for inflammation (as represented by these cytokines) in the development of obesity and T2D. In contrast, genetically driven obesity was associated with increased inflammation: a PGS for increased BMI was associated with increased C-reactive protein (Beta [95% CI], 0.40 [0.34,0.46], p=1.3×10-36) and activated plasminogen inhibitor (Beta [95% CI], 0.39 [0.19, 0.59], p=7.8×10-5) levels. Genetically-driven variation in cytokine levels were causally linked with chronic kidney disease in subjects with T2D: fibroblast growth factor 2 (OR[95% CI], 0.58[0.46,0.72], p=1.8×10-6) and interleukin 13 (OR[95% CI], 1.28[1.15,1.43], p=7.0×10-6) We conclude that chronic inflammation is a consequence, rather than a cause, of T2D and obesity, and may contribute to some of their complications. Disclosure N. van Zuydam: None. M. Wielscher: None. M. McCarthy: Research Support; Self; Eli Lilly and Company, Takeda Pharmaceuticals U.S.A., Inc., Novo Nordisk A/S. Advisory Panel; Self; Novo Nordisk A/S, Pfizer Inc.. Research Support; Self; Pfizer Inc., Sanofi, Boehringer Ingelheim GmbH, AstraZeneca, Merck Sharp & Dohme Corp.. M. Jarvelin: None.

A polygenic score (PGS) or polygenic risk score (PRS) is an estimate of an individual's genetic liability to a trait or disease, calculated according to their genotype profile and relevant genome-wide association study (GWAS) data. While present PRSs typically explain only a small fraction of trait variance, their correlation with the single largest contributor to phenotypic variation-genetic liability-has led to the routine application of PRSs across biomedical research. Among a range of applications, PRSs are exploited to assess shared etiology between phenotypes, to evaluate the clinical utility of genetic data for complex disease and as part of experimental studies in which, for example, experiments are performed that compare outcomes (e.g., gene expression and cellular response to treatment) between individuals with low and high PRS values. As GWAS sample sizes increase and PRSs become more powerful, PRSs are set to play a key role in research and stratified medicine. However, despite the importance and growing application of PRSs, there are limited guidelines for performing PRS analyses, which can lead to inconsistency between studies and misinterpretation of results. Here, we provide detailed guidelines for performing and interpreting PRS analyses. We outline standard quality control steps, discuss different methods for the calculation of PRSs, provide an introductory online tutorial, highlight common misconceptions relating to PRS results, offer recommendations for best practice and discuss future challenges.

Background:C-reactive protein (CRP) is a well-established biomarker of systemic inflammation. In pregnancy, several studies show association of elevated CRP with gestational diabetes mellitus (GDM). However, the genetic contributions of CRP levels during early pregnancy and their potential association with GDM remain largely understudied.Methods:We conducted maternal and offspring genome-wide association studies (GWAS) of first-trimester CRP levels in the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-Be (nuMoM2b) cohort. Genetic correlation between CRP and GDM was assessed using linkage disequilibrium score regression (LDSC), and polygenic risk scores (PRS) of CRP were further evaluated for associations with GDM.Results:In the European maternal sub-cohort of nuMoM2b, three genome-wide significant loci (CRP, LEPR, and HNF1A) were associated with early pregnancy CRP, consistent with previous GWAS. Multi-ancestry maternal GWAS revealed two additional associated loci (ENSG00000257703 and APOC1). Offspring GWAS did not identify any genome-wide significant associations. LDSC using the nuMoM2b European maternal CRP GWAS showed no significant genetic correlation with GDM. In contrast, a significant correlation was observed using the large population-based CRP GWAS, suggesting context-specific genetic architecture. However, PRS of CRP, based on GWAS summary statistics from either study, was not significantly associated with GDM risk in the nuMoM2b cohort.Conclusions:Our findings validate known genetic loci regulating systematic inflammation in a cohort of pregnant individuals, correlating with serum CRP levels, and highlight the value of pregnancy-specific GWAS in uncovering unique biological pathways relevant to maternal-fetal health.

The aim is to investigate the evidence for shared genetic architecture between each of asthma, allergic rhinitis and eczema with gastro-esophageal reflux disease (GERD). Structural equation models (SEM) and polygenic risk score (PRS) analyses are applied to three Swedish twin cohorts (n = 46,582) and reveal a modest genetic correlation between GERD and asthma of 0.18 and bidirectional PRS and phenotypic associations ranging between OR 1.09-1.14 and no correlations for eczema and allergic rhinitis. Linkage disequilibrium score regression is applied to summary statistics of recently published GERD and asthma/allergic disease genome wide association studies and reveals a genetic correlation of 0.48 for asthma and GERD, and Genomic SEM supports a single latent factor. A gene-/gene-set analysis using MAGMA reveals six pleiotropic genes (two at 12q13.2) associated with asthma and GERD. This study provides evidence that there is a common genetic architecture unique to asthma and GERD that may explain comorbidity and requires further investigation.

This dissertation includes three papers with each distributed in one chapter. In chapter 1, we proposed an Adaptive Weighting Reverse Regression (AWRR) method to test association between multiple traits and rare variants in a genomic region. AWRR is robust to the directions of effects of causal variants and is also robust to the directions of association of traits. Using extensive simulation studies, we compared the performance of AWRR with canonical correlation analysis (CCA), Single-TOW, and the Weighted Sum Reverse Regression (WSRR). Our results showed that, in all of the simulation scenarios, AWRR is consistently more powerful than CCA. In most scenarios, AWRR is more powerful than Single-TOW and WSRR. In chapter 2, we proposed an “optimal” maximum heritability test (MHT-O) to test the association between multiple traits and a single variant. MHT-O includes a procedure of deleting traits that have weak or no association with the variant. Using extensive simulation studies, we compared the performance of MHT-O with MHT, Trait-based Association Test uses Extended Simes procedure (TATES), SUM_SCORE and MANOVA. Our results showed that, in all of the simulation scenarios, MHT-O is either the most powerful test or comparable to the most powerful test among the five tests we compared. In chapter 3, we developed a statistical method by testing an optimally weighted combination of variants with multiple traits (TOWmuT) to test the association between multiple traits and a weighted combination of variants (rare and/or common) in a genomic region. TOWmuT is robust to the directions of effects of causal variants and is applicable to different types of traits. Using extensive simulation studies, we compared the performance of TOWmuT with the following five existing methods: gene association with multiple traits (GAMuT), multiple sequence kernel association test (MSKAT), adaptive weighting reverse regression (AWRR), single-TOW, and MANOVA. Our results showed that, in all of the simulation scenarios, TOWmuT has correct type I error rates and is consistently more powerful than the other five tests. We also illustrated the usefulness of TOWmuT by analyzing a whole-genome genotyping data from a lung function study.

Recent genetic studies showed that polygenic risk score (PRS) could be used to identify individuals at high risk of Alzheimer's disease (AD). Despite this success, early detection and intervention of AD still remain challenging. In this study, we suggest a novel PRS approach that is weighted by predicted tissue-specific gene expression levels. We used whole-genome sequencing and phenotypic data of 446 European participants (207 late-onset AD cases and 239 cognitively normal controls) from the Alzheimer's disease Neuroimaging Initiative (ADNI) (Table 1). We performed transcriptome-wide association studies (TWAS) in 13 brain regions by using MetaXcan (weights from GTEx V8) and GWAS summary statistics (IGAP stage 1 excluding ADNI participants) and integrated these 13 TWAS results using MultiXcan. To generate transcriptome-based weighting (TW)-PRS, expression weights in each gene were mapped to SNPs in the gene, and these were applied as additional weights to the SNPs' beta coefficients in the GWAS summary statistics. Then, PRS was derived based on the weighted GWAS summary statistics using PRSice-2. Finally, we evaluated prediction performance of a set of different models that incorporated demographic information and TW-PRS. A total of 17,588 gene expression weights were obtained from 13 brain tissues. These gene expression weights covered about 220,000 SNPs. Among them, expression weights were applied to about 140,000 SNPs overlapped with GWAS summary statistics. An AD prediction using conventional PRS yielded a pseudo-R2 of 0.0462 (P=1.30E-04). Compared with the conventional PRS, the TW-PRS improved the performance and statistical power with a pseudo-R2 of 0.0574 (P=1.74E-05). As shown in Table 2, a fully adjusted model achieved an AUC of 0.794 in which TW-PRS were significant even when APOE ε4 status and demographic information were adjusted (P=2.33E-05). This study proposes a novel TW-PRS approach that combines predicted tissue-specific transcriptomic weights and PRS. Expression weights of brain regions critical to AD progression enhanced the performance of conventional PRS. Our finding suggests that tissue-specific transcriptomic factors may be independent and complementary to conventional PRS and provide additional information for tissue-specific regulatory effects in AD.

M88 - ASSOCIATION OF GENETIC RISK SCORES FOR SCHIZOPHRENIA WITH LYMPHOBLAST AND CORTICAL GENE EXPRESSION NETWORKS

Polygenic risk score (PRS) has been recently developed for predicting complex traits and drug responses. It remains unknown whether multi-trait PRS (mtPRS) methods, by integrating information from multiple genetically correlated traits, can improve prediction accuracy and power for PRS analysis compared with single-trait PRS (stPRS) methods. In this paper, we first review commonly used mtPRS methods and find that they do not directly model the underlying genetic correlations among traits, which has been shown to be useful in guiding multi-trait association analysis in the literature. To overcome this limitation, we propose a mtPRS-PCA method to combine PRSs from multiple traits with weights obtained from performing principal component analysis (PCA) on the genetic correlation matrix. To accommodate various genetic architectures covering different effect directions, signal sparseness and across-trait correlation structures, we further propose an omnibus mtPRS method (mtPRS-O) by combining P values from mtPRS-PCA, mtPRS-ML (mtPRS based on machine learning) and stPRSs using Cauchy Combination Test. Our extensive simulation studies show that mtPRS-PCA outperforms other mtPRS methods in both disease and pharmacogenomics (PGx) genome-wide association studies (GWAS) contexts when traits are similarly correlated, with dense signal effects and in similar effect directions, and mtPRS-O is consistently superior to most other methods due to its robustness under various genetic architectures. We further apply mtPRS-PCA, mtPRS-O and other methods to PGx GWAS data from a randomized clinical trial in the cardiovascular domain and demonstrate performance improvement of mtPRS-PCA in both prediction accuracy and patient stratification as well as the robustness of mtPRS-O in PRS association test.

Chronic pain and sleep disturbance are highly comorbid disorders, which leads to barriers to treatment and significant healthcare costs. Understanding the underlying genetic and neural mechanisms of the interplay between sleep disturbance and chronic pain is likely to lead to better treatment. In this study, we combined 1206 participants with phenotype data, resting-state functional magnetic resonance imaging (rfMRI) data and genotype data from the Human Connectome Project and two large sample size genome-wide association studies (GWASs) summary data from published studies to identify the genetic and neural bases for the association between pain and sleep disturbance. Pittsburgh sleep quality index (PSQI) score was used for sleep disturbance, pain intensity was measured by Pain Intensity Survey. The result showed chronic pain was significantly correlated with sleep disturbance (r = 0.171, p-value < 0.001). Their genetic correlation was rg = 0.598 using linkage disequilibrium (LD) score regression analysis. Polygenic score (PGS) association analysis showed PGS of chronic pain was significantly associated with sleep and vice versa. Nine shared functional connectivity (FCs) were identified involving prefrontal cortex, temporal cortex, precentral/postcentral cortex, anterior cingulate cortex, fusiform gyrus and hippocampus. All these FCs mediated the effect of sleep disturbance on pain and seven FCs mediated the effect of pain on sleep disturbance. The chronic pain PGS was positively associated with the FC between middle temporal gyrus and hippocampus, which further mediated the effect of chronic pain PGS on PSQI score. Mendelian randomization analysis implied a possible causal relationship from chronic pain to sleep disturbance was stronger than that of sleep disturbance to chronic pain. The results provided genetic and neural evidence for the association between pain and sleep disturbance, which may inform future treatment approaches for comorbid chronic pain states and sleep disturbance.

Type 2 diabetes (T2D) is known as an inflammatory disease. NRF2 (Nuclear Factor Erythroid 2 Like2) encodes a transcription factor that binds to antioxidant response elements (AREs) and regulates the expression of genes involved in many antioxidant responses. This study aimed to gain insight into individual anti-inflammatory activity to prevent T2D development in humans. We performed a genome-wide association study (GWAS) to identify genetic variants influencing NRF2 expression in LCLs (lymphoblastoid cell lines) generated from 74 different individuals. Association analyses between T2D or its related traits and genetic risk score (GRS) calculated by combining genetic variants detected from GWAS for cellular NRF2 expression were performed using data from 8715 subjects. The T2D prediction model using GRS was evaluated by measuring the area under the curve (AUC) of the receiver operating characteristics (ROC) curve. Our GWAS identified six genetic variants (SNP) showing suggestive evidence of associations with cellular NRF2 expression (P < 10- 6). Logistic regression analysis demonstrated that GRS was associated with an increased risk of T2D (P value = 0.003, OR = 1.13). In addition, linear regression analyses showed positive associations between GRS and fasting glucose (P value = 0.028, β = 0.62), 2-h glucose (P value = 0.0004, β = 1.13) and HbA1C (P value = 0.033, β = 0.03). In the T2D prediction model using GRS, the AUC of the ROC curve was 0.69. This study highlights genetic variants associated with cellular NRF2 expression and suggests that the GRS of NRF2 expression-associated variants is likely to be a useful indicator of T2D development in the human population.

Schizophrenia (SZ) and HIV infection are serious disorders with a complex phenotypic relationship. Observational studies have described their comorbidity; their genetic correlation is not well studied. We performed extensive analysis in search of common genetic factors for SZ and HIV, and their relationship with risky sexual behavior (RSB). Summary statistics from genome-wide association studies of HIV infection and schizophrenia were obtained and 2379 European Americans were genotyped and assessed for RSB score. Genetic relationships between traits were analyzed in three ways: linkage disequilibrium (LD) score regression to estimate genetic correlation; GPA (Genetic analysis incorporating Pleiotropy and Annotation) to test pleiotropy and identify pleiotropic loci; polygenic risk scores (PRS) of SZ and HIV to predict RSB using linear regression. We found significant pleiotropy (p=5.31E-28) and a positive genetic correlation (cor=0.17, p=0.002) for SZ and HIV infection. Pleiotropic SNPs with opposite effect directions (antagonistic) and SNPs with the same effect direction (synergistic) were enriched for distinctly different biological functions. SZ PRS computed with antagonistically pleiotropic SNPs consistently predicted RSB score with nominal significance, but SZ PRS based on either synergistically pleiotropic SNPs or all SNPs did not predict RSB. The epidemiologic correlation between schizophrenia and HIV can partly be explained by overlapping genetic risk factors, which are related to risky sexual behavior.

To take advantage of the increasingly available high-density SNP maps across the genome, various tests that compare multilocus genotypes or estimated haplotypes between cases and controls have been developed for candidate gene association studies. Here we view this two-sample testing problem from the perspective of supervised machine learning and propose a new association test. The approach adopts the flexible and easy-to-understand classification tree model as the learning machine, and uses the estimated prediction error of the resulting prediction rule as the test statistic. This procedure not only provides an association test but also generates a prediction rule that can be useful in understanding the mechanisms underlying complex disease. Under the set-up of a haplotype-based transmission/disequilibrium test (TDT) type of analysis, we find through simulation studies that the proposed procedure has the correct type I error rates and is robust to population stratification. The power of the proposed procedure is sensitive to the chosen prediction error estimator. Among commonly used prediction error estimators, the .632+ estimator results in a test that has the best overall performance. We also find that the test using the .632+ estimator is more powerful than the standard single-point TDT analysis, the Pearson's goodness-of-fit test based on estimated haplotype frequencies, and two haplotype-based global tests implemented in the genetic analysis package FBAT. To illustrate the application of the proposed method in population-based association studies, we use the procedure to study the association between non-Hodgkin lymphoma and the IL10 gene.