Indiana University-Purdue University Indianapolis (IUPUI) / Determining an individual’s genetic susceptibility in complex diseases like Alzheimer’s disease (AD) is challenging as multiple variants each contribute a small portion of the overall risk. Polygenic Risk Scores (PRS) are a mathematical construct or composite that aggregates the small effects of multiple variants into a single score. Potential applications of PRS include risk stratification, biomarker discovery and increased prognostic accuracy. A systematic review demonstrated that methodological refinement of PRS is an active research area, mostly focused on large case-control genome-wide association studies (GWAS). In AD, where there is considerable phenotypic and genetic heterogeneity, we hypothesized that PRS based on endophenotypes, and pathway-relevant genetic information would be particularly informative. In the first study, data from the NIA Alzheimer’s Disease Neuroimaging Initiative (ADNI) was used to develop endophenotype-based PRS based on amyloid (A), tau (T), neurodegeneration (N) and cerebrovascular (V) biomarkers, as well as an overall/combined endophenotype-PRS. Results indicated that combined phenotype-PRS predicted neurodegeneration biomarkers and overall AD risk. By contrast, amyloid and tau-PRSs were strongly linked to the corresponding biomarkers. Finally, extrinsic significance of the PRS approach was demonstrated by application of AD biological pathway-informed PRS to prediction of cognitive changes among older women with breast cancer (BC). Results from PRS analysis of the multicenter Thinking and Living with Cancer (TLC) study indicated that older BC patients with high AD genetic susceptibility within the immune-response and endocytosis pathways have worse cognition following chemotherapy±hormonal therapy rather than hormonal-only therapy. In conclusion, PRSs based on biomarker- or pathway- specific genetic information may provide mechanistic insights beyond disease susceptibility, supporting development of precision medicine with potential application to AD and other age-associated cognitive disorders.

Genome-wide Association Studies (GWAS) is the study used to associate common variants and phenotypes and has uncovered thousands of disease-associated variants. However, there is limited research on the contribution of a rare variant. The UK Biobank (UKB) contains detailed medical records and genetic information for nearly 500,000 individuals and offers a great opportunity for genetic association studies on rare variants. Here we focused on the role of rare protein-coding variants on UKB phenotypes. We selected three diseases for analysis: breast cancer, hypothyroidism and type II diabetes. We defined criteria for qualifying variants and pruned the control group to reduce interference signals from similar phenotypes. We identified the most known biomarkers for those diseases, such as BRCA1 and BRCA2 gene for breast cancer, TG and TSHR gene for hypothyroidism and GCK for type II diabetes. This result supports the model validity and clarifies the contribution of rare variants to diseases. Moreover, we also tried the geneset based collapsing method to aggregate information across genes to strengthen the signal from rare variants and build a diagnosis model that only relies on the genetic information. Our model could achieve great performance with an AUC of more than 20% improvement for type II diabetes and breast cancer and more than 90% accuracy for hypothyroidism.

Posttraumatic stress disorder (PTSD) is a psychiatric condition that presents with 3 main symptoms're-experiencing, avoidance/numbing, and hyper arousal'after an individual experiences a traumatic event. Recent evidence suggests a potential genetic basis for PTSD and a sub symptom of hyper arousal, sleep, as a potential pathway for PTSD development, but no study has identified candidate genes associated with specific symptoms such as sleep difficulty. Based on a conceptual framework in which specific genes are associated with the onset of PTSD, this study used a genome-wide association study (GWAS) method with a case control study design to compare the genomes of individuals with and without PTSD. A secondary GWAS dataset from a study on alcohol dependence in European and African Americans was obtained from the National Center for Biotechnology Information. PTSD cases and controls were analyzed using PLINK software. Signals from 2 single nucleotide polymorphisms (SNPs), which have not been previously associated with PTSD, exceeded the established genome-wide threshold: SNP rs13160949 on chromosome 5 (p = 7.33x10-9, OR: 1.565) and SNP rs2283877 on chromosome 22 (p = 2.55x10-8, OR: 1.748). Neither SNP, though, maintained genomewide significance following corrected tests for multiple testing, population stratification, and false discovery, so the planned analysis for possible associations with PTSD by symptom category then by the sub symptom of sleep could not be completed. The results of this study suggest that PTSD may be the result of polygenic SNPs with weak effects, which supports a recent study indicating the disease may be highly polygenic. Positive social change implications include bringing attention to the clinical and research community that PTSD may involve complex polygenic factors in need of further study.

Developmental Dyslexia (DD), Attention Deficit Hyperactivity Disorder (ADHD), and Educational Attainment (EA) are highly prevalent conditions that have a significant impacton individuals' academic and social functioning. These conditions have a complex genetic basis and are often comorbid. To assess the polygenic architecture of these traits, psychiatric genetics researchers utilize a sophisticated tool known as polygenic scores (PGS). By combining numerous genes of individually modest effects, PGS summarizes an individual's genetic risk as a single score. In this study, we aimed to examine the association between PGS for ADHD and EA with typical DD traits. Using a meta-analytic approach, we analyzed data from earlier studies and found that PGS-ADHD accounts for 1.2% of the variance in DD, with a pooled effect size of r = -0.11 (95% CI = [-0.171, -0.050]). Similarly, PGS-EA accounted for 3.2% of the variance in DD, with a pooled effect size of r = 0.18 (95% CI = [0.070, 0.288]). Although these effect sizes are relatively small, it is important to note that PGS are not typically strong predictors on their own, but rather capture a small portion of the genetic variation that contributes to a trait or outcome. My findings suggest that PGS for ADHD and EA are associated with DD, indicating that DD has a complex genetic basis. However, these findings also raise questions about the impact of PGS on psychiatric research moving forward. To address these questions, I provide recommendations for future researchdirections.

BACKGROUND: Polygenic risk scores (PRS) have emerged as a promising tool for predicting the risk of developing a variety of illnesses, including psychiatric disorders. PRS are calculated by analyzing the genetic variants across the genome to assess an individual’s risk for developing a disorder. Family history (FHx) of psychiatric disorders has long been recognized as a valuable tool in assessing an individual’s risk in lieu of a genetic blood-based biomarker, like PRS. However, the accuracy of self-reported family history remains limited as a consequence of incomplete or unreliable information collected during a clinical interview. Existing risk factors for developing psychiatric disorders such as FHx tend to be non-specific in their prediction of outcome. Few research studies have evaluated the possibility of using PRS as a complement to FHx across psychosis spectrum disorders. The present study seeks to examine the relationship between the current standard indirect measure of inherited susceptibility being used, FHx, and an individual’s PRS to more directly predict risk of familial susceptibility in those diagnosed with schizophrenia (SZ) by comparing SZ probands based on their FHx of psychotic disorders diagnosis. METHODS: 396 SZ Probands with FHx data were identified. Data on polygenic risk scores for SZ (PRSSCZ) and FHx were obtained from the Bipolar-Schizophrenia Network on Intermediate Phenotypes consortium (B-SNIP 1). Genetic susceptibility was identified using PRSSCZ. FHx was established from detailed family interviews. SZ probands with only an affected first-degree relative (n= 42) or only an affected second-degree relative (n= 55) with history of a psychotic disorder diagnosis were included in the analyses. SZ probands without any affected relative (n=179) were used as a comparison group. Demographic information for all participant groups were compared using Chi-square for categorical variables, and ANOVA for continuous variables. ANCOVA was used to identify differences among relative proximity and PRSSCZ while accounting for covariates (age, sex, race). Multiple comparisons were adjusted for using Bonferroni correction. Healthy controls were added as a reference only. The significance level was set at p < 0.05. RESULTS: In SZ probands, there was a significant difference between those with an affected first-degree relative with non-affective psychosis and those without any affected relatives (p< 0.05). No significant difference was observed between those with an affected second-degree relative with non-affective psychosis and those without any affected relatives. Having only an affected first-degree relative with non-affective psychosis carries significantly more risk than having only an affected second-degree relative with non-affective psychosis (p< 0.05). CONCLUSIONS: These findings a) support the validity of taking careful family history of non-affective psychosis diagnosis when evaluating individuals with a psychotic disorder, b) suggest that PRSSCZ may be a useful complement to taking family history, and c) relative proximity is important in risk for SZ. The limitations of this study include lack of direct interviews of affected first- and second-degree relatives, and the lack of complete pedigree information that might allow for calculation of familial load.

Background: Canada continues to face an opioid epidemic with 5,368 opioid apparent related deaths occurring between January and September of 2021. Methadone Maintenance Treatment (MMT), a form of Medication Assisted Treatment used to treat Opioid Use Disorder (OUD), has been reported to decrease opioid cravings and opioid use, however, individual differences exist in the effective dose of methadone. Further, individuals living with an OUD have higher rates of substance use including cannabis. A genetic component has been suggested to exist for both cannabis use and MMT outcomes, however inconsistent findings have been reported. Methods: Knowledge synthesis and primary genetic association studies were conducted. A protocol was prepared for the planning of a systematic review for Genome-Wide Association Studies (GWASs) of cannabis use. The full systematic review was then conducted, providing an assessment of the literature and a description of studies quality. A GWAS and Polygenic Risk Score (PRS) was then conducted for cannabis use and MMT outcomes, separately, in Europeans only. The top Single Nucleotide Polymorphisms (SNPs) were then analyzed separately by sex and sex interactions were conducted. Results: The systematic review included 6 studies, identifying 96 genetic variants associated with cannabis use. The GWASs for both cannabis use and MMT outcomes did not identify any significant results. A significant PRS was found for regular cannabis use and methadone dose. No sex-specific results were identified. Discussion: This thesis summarised the evidence on the genetics of cannabis use as well as employed GWASs and PRSs to investigate cannabis use and MMT outcomes within a European population. We were able to highlight gaps within the genetic literature of cannabis and MMT outcomes as well as identify areas of interest for future research. / Dissertation / Doctor of Philosophy (PhD) / Cannabis use rates in Canada are increasing, with Opioid Use Disorder (OUD) patients having high rates of cannabis use despite inconsistent findings on the impacts. To combat the opioid crisis, Methadone Maintenance Treatment (MMT) is utilized to reduce opioid cravings and use. However, individuals on MMT are likely to use other substances, including cannabis. This thesis explores the genetic literature on cannabis use and conducts a Genome-Wide Association Study (GWAS) and a Polygenetic Risk Score (PRS). The GWAS investigates genetic variants throughout the whole genome associated with a trait, while the PRS creates a genetic weight risk score. GWAS and PRS methods were used to investigate cannabis use and MMT outcomes within Europeans with OUD. While no significant GWAS results were found, a statistically significant PRS was found for regular cannabis use and methadone dose, suggesting each respective score can estimate an individual’s risk of that trait.

Polygenic risk scores are scores used in precision medicine in order to assess an individual's risk of having a certain quantitative trait based on his or her genetics. Previous works have shown that machine learning, namely Gradient Boosted Regression Trees, can be successfully applied to calibrate the weights of the risk score to improve its predictive power in a target population. Neural networks are a very powerful class of machine learning algorithms that have demonstrated success in various elds of genetics, and in this work, we examined the predictive power of a polygenic risk score that uses neural networks to perform the weight calibration. Using a single neural network, we were able to obtain prediction R2 of 0.234 and 0.074 for height and BMI, respectively. We further experimented with changing the dimension of the input features, using ensembled models, and varying the number of splits used to train the models in order to obtain a nal prediction R2 of 0.242 for height and 0.0804 for BMI, achieving a relative improvement of 1.26% in prediction R2 for height. Furthermore, we performed extensive analysis of the behaviour of the neural network-calibrated weights. In our analysis, we highlighted several potential drawbacks of using neural networks, as well as machine learning algorithms in general when performing the weight calibration, and o er several suggestions for improving the consistency and performance of machine learning-calibrated weights for future research. / Thesis / Master of Science (MSc)

Background: South Asians are at an increased risk of developing dysglycemia during and after pregnancy. In pregnant women, dysglycemia often develops in the form of gestational diabetes mellitus (GDM), which may predispose their newborns to adverse health outcomes through abnormal cord blood insulin levels. However, reasons for the elevated risk of dysglycemia in South Asians have not been extensively studied. Genetic factors may contribute to the heritability of GDM and abnormal cord blood insulin levels in South Asians. Objectives: The objectives of this thesis were to test the association of: 1) A type 2 diabetes polygenic risk score with GDM in South Asian pregnant women from the South Asian Birth Cohort (START); 2) maternal and newborn insulin-based polygenic risk scores with cord blood insulin and glucose/insulin ratio in South Asian newborns from START Methods: Three polygenic risk scores were created to test their association with participant data (N=1012) from START. GDM was defined using cut-offs established by the Born in Bradford cohort of South Asian women. The type 2 diabetes polygenic risk score was created in 832 START mothers and included 35,274 independent variants. The maternal and newborn insulin-based polygenic risk scores were created in 604 START newborns and included 1128017 independent variants. Univariate and multiple logistic and linear regression models were used to test the associations between the polygenic risk scores and dysglycemia outcomes. Results: The type 2 diabetes polygenic risk score was associated with GDM in both univariate (OR: 2.00, 95% CI: 1.46-2.75, P<0.001), and multivariable models (OR: 1.81, 95% CI: 1.30-2.53, P<0.001). The maternal insulin-based polygenic risk score was not associated with cord blood insulin or cord glucose/insulin ratio. However, the newborn insulin-based polygenic risk score was associated with cord blood insulin in a multivariable model adjusted for maternal insulin-based polygenic risk score (β = 0.036, 95% CI: 0.002 – 0.069; P=0.038 among other factors. Conclusion: A type 2 diabetes polygenic risk score and a newborn insulin-based polygenic risk score may be associated with maternal and newborn dysglycemia. / Thesis / Master of Science (MSc) / Background: South Asians are approximately two times more at risk for developing gestational diabetes mellitus (GDM) compared to white Caucasians. Genetic factors may contribute to this elevated risk. Polygenic risk scores (PRSs), which combine the effects of multiple disease loci and variants associated with the disease into one variable could be useful in further understanding how GDM develops in South Asians. Methods: Data from the South Asian Birth Cohort (START) was used to test the association of three PRSs with the outcomes of interest. Results: The type 2 diabetes PRS was independently associated with GDM. The insulin-based maternal PRS was not associated with cord blood insulin but the insulin-based newborn PRS was independently associated with cord blood insulin. However, neither the insulin-based maternal nor newborn PRS was associated with cord blood glucose/insulin ratio. Conclusion: The PRSs suggests a possible genetic component, which contributes to abnormal glycemic status development in South Asian mothers and their newborns.

Cancer is a complex disease shaped by a heterogeneous landscape of inherited genetic variants and acquired somatic aberrations. Although specific patterns of somatic aberrations within key pathways are recognized as hallmarks of many cancers, and mounting evidence suggests a significant interplay between germline and somatic variants, the intricate relationship between germline predisposition and the disruption of these pathways remains poorly understood. Here, I present an integrative approach using multi-omics data to functionally characterize germline variants and explore the heterogeneous landscape of somatic mutations, with the aim of establish mechanistic links between functional variants and the disruption of cancer-related biological processes. To enable the identification of functional variants, I initially performed a comprehensive characterization of functionally annotated transcriptional regulatory elements, establishing a hierarchy of ‘consensus’ elements across multiple levels of abstraction. This analysis generated a vast collection of consensus promoters, enhancers, and active enhancers, spanning 198 cell lines and 38 tissue types, with aggregate data providing global consensus definitions for each element type. Additionally, ‘total binding affinity’ method was employed, integrating 1000 Genomes Project genotype data and thousands of transcription factor binding motifs, to further characterize and functionally annotate these regulatory elements. The results generated from this analysis can be interactively explored and visualized through the CONREL web application. To allow effective annotation of individual’s ancestry, I developed and successfully employed an improved version of EthSEQ (version 3), an R package that provides a rapid and reliable pipeline for ancestry annotation. Accurate stratification of individual ancestry is essential for correctly interpreting the impact of genomic variations in associations studies. EthSEQ version 3 was successfully utilized to determine the genetic ancestry of over 500 pediatric patients diagnosed with 11 different tumor types, enabling further investigation into the genetic landscape of patients confidently identified as of European ancestry.To further investigate into the interplay between germline and somatic variants, I conducted genome-wide association studies across 33 cancer types characterized by The Cancer Genome Atlas, using binary traits defined by somatic aberration profiles in ten oncogenic signaling pathways. Functional links between associated variants and somatic profiles were investigated through cis-eQTL data to identify regulatory interactions with pathway-related genes. Additionally, using GWAS summary statistics I employed polygenic scores to examine the contribution of germline genetic variation to somatic molecular profiles, tumor subtypes, and clinical outcomes such as patient survival and tumor aggressiveness. Polygenic scores were validated using external data from PCAWG and CCLE datasets. Lastly, to explore the heterogeneity of somatic mutational profiles, I employed a network-based approach to propagate somatic alterations through a molecular interaction network, aiming to reveal novel patterns of somatic alteration with potential significance in cancer. I then conducted a series of GWAS analyses, utilizing traits defined by combinations of these propagated somatic scores across genes involved in well-defined DNA repair pathways. Overall, I demonstrate that germline genetics can describe patients’ genetic liability to develop specific cancer molecular and clinical profiles. Understanding the functional roles of genetic variants can provide valuable insights into the biological mechanisms underlying a disease or trait.

Sleep disturbances and insomnia are prevalent, with around 33% of adults indicating that they experience at least one main symptom of insomnia, and bidirectional relationships exist with common psychopathology, particularly major depressive disorder (MDD). However, genetic and environmental (e.g., traumatic event exposure) contributions to the etiology of these phenotypes are not yet well understood. A genetically informative sample of approximately 12,000 Han Chinese women aged 30-60 (50% with recurrent MDD) was used to address several gaps within the sleep literature. Sleep disturbances were assessed in all individuals using a general item addressing sleeplessness (GS). A sleep within depression sum score (SDS) was also created in MDD cases, combining information from the GS and two insomnia items within MDD. A total of 11 traumatic events were assessed and additional information on childhood sexual abuse (CSA) was also obtained. First, factor analyses were conducted to determine trauma factor structure. The best-fit solution included 3 factors: interpersonal, child interpersonal, and non-assaultive, and composite variables were constructed accordingly. A series of hierarchical regressions were run to examine differential effects of trauma type and timing on sleeplessness. All traumatic events predicted sleeplessness at similar magnitudes, although population models indicated that childhood interpersonal trauma may be particularly potent. An association between CSA and sleeplessness was also replicated. A series of genetic analyses demonstrated that the single nucleotide polymorphism-based heritability of sleep phenotypes did not differ significantly from zero. Further, association analyses did not identify any genome-wide significant loci. However, using a liberal false discovery rate threshold of 0.5, two genes of interest, KCNK9 and ALDH1A2, emerged for the SDS. Polygenic risk score (PRS) analyses demonstrated genetic overlap between the SDS in MDD cases and GS in MDD controls, with PRSs explaining 0.2-0.3% of the variance. A final combined model of both genetic and environmental risk indicated that both PRS and traumatic events were significant predictors of sleeplessness. While genetic results should be interpreted with caution given the lack of heritability, additional research into the genetic and environmental contributions to insomnia, utilizing more standardized phenotypes and properly ascertained samples, is clearly warranted.