Beyond the ivory tower

Type 2 diabetes mellitus (T2D) is a leading cause of disability and mortality in the United States, and is responsible for $327 billion annually in economic damage. The presence of sustained hyperglycemia is used to screen for, diagnose and manage diabetes, with the diagnostic threshold above that which predisposes to microvascular complications. However, hyperglycemia is the end product of several pathological processes that eventually converge on the inability of the pancreatic beta-cells to produce enough insulin to meet the demands of the target tissues. The onset of these various pathological processes occurs years prior to clinical presentation, during which time complications have already occurred in a significant fraction of individuals. While hypoglycemic agents acutely delay some of the complications, thus far, they have been unable to change the progressive course of T2D or adequately reduce the risk of late-stage diabetic complications. Due to this inadequacy, a proactive vs reactive approach is the best strategy to reduce the disease and economic burden. Be that as it may, our inability to identify and adequately understand the etiological heterogeneity of T2D underpins the diagnostic and prognostic challenges faced in clinical practice. As highlighted in Chapter 1, the risk of developing T2D increases with age, obesity and lack of physical activity. It occurs more frequently in those with hypertension or dyslipidemia, family history and certain racial/ethnic groups. Similar to dysglycemia, several of these anthropometric and clinical risk factors reflect early disease manifestations or shared upstream mechanisms rather than true predictive antecedents. From this perspective, the advent of the Human Genome Project held the most promise for the development of a truly predictive risk assessment for T2D. Large-scale sequencing studies revealed that T2D is in fact significantly less genetically driven than previously hypothesized and the heritable architecture was mostly due to common variants of small effects. Consequently, the polygenic risk score for T2D did not significantly improve discrimination, calibration or risk reclassification compared with conventional predictors. Other attempts to elucidate the disease etiology include identifying epigenetic alterations which mediate environmental influences on aberrant gene expression. DNA methylation (DNAm) is the most widely studied and characterized epigenetic modification, due to its scalability and its potential for clinical translation. Several investigators have demonstrated that the DNAm status at cg19693031, a CpG site located within the 3’UTR of the TXNIP gene, is highly correlated with both HbA1c values and T2D. However, these prior studies failed to account for potential confounding genetic influences. Given the dynamic cross-talk between genetic variation, DNA methylation and environmental exposures, these factors should not be considered as self-standing layers of pathophysiological regulation but should only be interpreted in light of each other. For example, several genetic variants for common complex diseases are not detected in GWAS analyses when the environmental influences are not included. Therefore, this partially-informed approach of traditional GWAS/EWAS analyses fail to adequately explain the interconnected network that underlies the development and progression of diabetes which limits our ability to prevent and treat the disease. Thus, using resources from the Family and Community Health Studies (FACHS), I sought to develop a robust integrated genetic-epigenetic biosignature that could be used to prevent or forestall clinical T2D. My goal was to determine the feasibility for translation and potential clinical utility of DNA methylation for T2D medicine. In Chapter 2, I demonstrate the importance of the genetic and hyperglycemia effects on cg19693031 methylation in T2D. I found that confounding genetic variation acting both in cis and trans on cg19693031 occur both independent of and in combination with HbA1c levels. Correcting for genetic confounding improved the ability of cg19693031 to distinguish T2D from normoglycemic and prediabetic subjects but not between normoglycemic and prediabetic subjects. Unfortunately, this suggests that the methylation status of cg19693031 may not be able to detect the early stages of T2D. Enrichment and functional genomic analyses suggest the crosstalk between genetic variants and the demethylation of cg19693031 methylation may contribute to the upregulation of TXNIP in diabetes. These analyses identified several well-established TXNIP regulators including transcription factors and modulators of mRNA stability. In addition, several factors known to influence genome and epigenome plasticity were also identified. A significant body of literature has implicated the upregulation of TXNIP in diabetes, cancer, neurodegenerative and cardiovascular diseases. Based on these observations, I performed functional SNP and gene mapping using the confounding genetic variants. Enrichment in both Type 1 and Type 2 diabetes and biological pathways related to diabetes was observed. Furthermore, I identified enrichment in key pathways for cancer, neurodegeneration, and coronary heart disease. Taken together, this suggests that the demethylation response of cg19693031 in white blood cells is not specific to diabetes, let alone T2D. In Chapter 3, I designed a workflow using artificial intelligence to identify DNA methylation biomarkers adapted for the intended use of clinical translation. The data-mining strategy was guided by my commercial Precision Epigenetic expertise and was designed to facilitate the discovery of high utility biomarkers that demonstrate the technical merit required for translation. I identified 86 prospective DNAm markers associated with T2D status, and then performed comparative analyses using six well-established classifiers. Unfortunately; a soft-voting classifier ensemble using cg19693031 adjusted for age was the best performing feature and algorithm combination for discriminating between normoglycemic and T2D subjects. The incorporation of additional DNAm loci neither improved the accuracy nor clinical utility of cg19693031. Given the strong association between readily obtainable clinical and anthropometric variables and the affordability of obtaining HbA1c measurements, the potential for including new biomarkers into T2D diagnostics is limited. Thus, the value of a new biomarker lies in whether it adds to prediction over and above simple clinical measures. With this in mind, my research suggests that cg19693031 methylation would not provide clinicians with any additional information for T2D diagnostics. While my studies have enhanced our understanding of the role that the demethylation response of cg19693031 plays in T2D and has made evident that the predictive and diagnostic value of cg19693031 as a single indicator is too small to justify clinical translation, I recognize that cg19693031 is also associated with both metabolic syndrome and coronary heart disease. Additionally, TXNIP has been shown to impact numerous pathways involved in both of these phenotypes. In Chapter 4, I extend Chapter 2 analyses to determine the extent of cardiometabolic confounding on cg19693031 methylation. Across the entire cohort, a multivariate model incorporating systolic blood pressure, triglycerides and waist-to-hip ratio explained the methylation variance better than HbA1c alone. However, I found that the magnitude of confounding by cardiometabolic traits in leukocytes is contextual to the glycemic status of the cell. The variance of cg19693031 methylation in normoglycemic subjects was best explained by an additive model using systolic blood pressure and HDL. Conversely, the demethylation response in diabetic subjects is dependent on HbA1c and triglyceride levels. Finally, I demonstrated that the demethylation response of cg19693031 was more pronounced in diabetic subjects with uncontrolled glucose levels than those with levels at goal and that the methylation status of cg19693031 may improve CVD risk stratifications for diabetic subjects. These findings further support that the lack of specificity of cg19693031 for T2D would lead to inappropriate diagnostic decisions. In an attempt to find a more robust CpG site, I scrutinized the methylome of the TXNIP locus as a function of T2D status using both an array and digital PCR in Chapter 5. I found that across locus, the 3’UTR was the only genomic region associated with prevalent T2D; however, the use of additional 3’UTR CpG sites did not add predictive value nor out-perform cg19693031 singular assessments. Using two data-points from the same subjects over a 10-year period, I found that the degree of demethylation over time is associated with age, T2D progression and overall disease burden. However, the methylation does not revert as a function of glucose control. The findings from this chapter suggest that cg19693031 information may be more reflective of total disease burden than merely the presence of diabetes. In Chapter 6, I use cg19693031 along with 4 additional previously validated CpG sites that are specific for smoking, drinking and cardiovascular disease to build a mortality index using data from the Framingham Heart Study. The output from this algorithm can be used to provide focused feedback to patients, guide-recommendations for additional medical assessments and help monitor the effect of public health prevention interventions. T2D is clinically challenging disease to predict, largely due to its complex genetic and environmental architecture that underlies its development and clinical presentation. The work presented in this thesis reflects the workflow required to determine the clinical utility and translatability of DNA methylation biomarkers and suggests that it is inappropriate to use the methylation status of cg19693031 for T2D medicine in its current form. Together, the work presented here advances our understanding of the nuclear network regulatory effects underlying the clinical presentation and reflecting the etiological heterogeneity of T2D and provides new avenues for investigation to improve the prognostication and risk stratification for diabetic patients.