Statistical Genetics of Outcomes and Drug Response in Patients with Type 2 Diabetes.

在一个项目中，我们调查了糖尿病周围神经病变(DPN)的病因学中涉及的遗传因素。这项研究的目的是使用两个特征良好的队列来系统地寻找影响DPN风险的遗传变异。在控制糖尿病心血管风险行动(ACCORD)临床试验的参与者中进行了一项全基因组关联研究(GWAS)，测试了680万个单核苷酸多态。研究对象包括4384名患有2型糖尿病(T2D)和常见或偶发DPN(定义为密歇根神经病变筛查工具临床检查评分2.0)的白人患者，以及784名患有T2D且在基线或随访期间没有DPN证据的白人对照组受试者。在2型糖尿病旁路血管成形术血管重建研究(BARI 2D)试验中，在患有T2D的白人受试者(791名DPN阳性病例受试者和158名DPN阴性对照受试者)中寻找有意义的基因重复。在基因类型-组织表达(GTEx)数据库中评估了外周神经中显著的变异和基因表达之间的相关性。染色体2q24上的28个SNPs达到显著水平(P&lt；5×10~(-8))。铅SNP的次要等位基因(rs13417783，次要等位基因频率=0.14)使DPN的OR值降低36%(OR0.64，95%CI 0.55~0.74，P=1.9×10-9)。这种效果不受ACCORD治疗分配的影响(交互作用P=0.6)，也不受已知DPN危险因素的影响。该基因座在BARI 2D中被成功验证(OR 0.57，95%CI 0.42-0.80，P=9 10-4；汇总P=7.9 10-12)。在GTEx中，该基因座的次要保护性等位基因与编码人类电压门控钠通道NaV1.2的邻近基因(SCN2A)的胫神经高表达相关(P=9 10-4)。综上所述，我们已经确定并成功验证了一个先前未知的基因座，该基因座对T2D中DPN的发生具有强大的保护作用。这些结果可能为DPN的发病机制提供新的见解，并为新的干预措施指出一个潜在的靶点。 和我的博士生Jon Lierer一起，我们正在开发传导方法，以帮助理解疾病中的异质性。2型糖尿病(T2D)是一种高度异质性疾病。这种异质性表明，导致这种疾病的潜在潜在特征可能存在差异。发展对这些潜在特征的更精细的理解可以导致对疾病病因的重要洞察，更个性化的治疗方法，以及更好的临床结果。在这里，我们改进了拓扑数据分析(TDA)的使用，以在这种异质性疾病中发现潜在的亚群。我们将我们的TDA工作流程应用于糖尿病心血管风险控制行动(ACCORD)临床试验中的不同T2D患者队列，以1)根据源自临床特征的患者拓扑网络确定聚集性患者亚组2)测试亚组临床结果的丰富性3)确定推动亚组成员资格的特征，以及4)根据已识别的特征预测临床结果。我们开发了一个健壮的工作流程和相应的代码来实现TDA，通过内部模型验证最大限度地减少了过度拟合，并引入了新的参数优化度量。然后，我们将此工作流应用于ACCORD数据。我们确定了T2D患者的八个亚组群，其中三个包含至少一个临床结果明显偏高或偏低的患者。第2组患者出现ACCORD主要结局和主要冠状动脉事件的风险增加。第5组患者的主要结果、充血性心力衰竭和大血管结果扩大的风险增加。第6组患者的主要结局、心血管死亡、充血性心力衰竭、大血管扩大结局和主要冠状动脉事件的风险降低。对于三个重要的集群，我们确定了驱动集群成员资格的临床变量。这项研究既有方法论上的见解，也有应用上的见解。在这里，我们扩展了一种改进的TDA方法，使用数据驱动的调优和保持方法进行测试和验证，这是对广泛应用的重要发展。从应用的角度来看，在ACCORD试验中，我们在T2D患者中确定了8个集群，其中3个显著丰富了临床结果，并确定了几个驱动集群成员资格的变量。我们还比较了我们的套索回归方法，并建立了集群驱动变量选择的可比性，同时提供了许多独特的优势。这些发现提供了对T2D异构性背后的结构的洞察，并展示了一种有效的数据驱动的方法，用于从数据的潜在拓扑特征中提取洞察。 我们也有使用ACCORD数据的正在进行的项目。Jon Lierer正在对罗格列酮的反应以及体重增加的不良后果进行传导关联分析。他发现了有趣的潜在关联，我们目前正寻求复制这些关联。我们还在研究血红蛋白糖化指数的遗传学。HGI量化了糖基化倾向的个体间差异，是糖尿病并发症和强化降糖不良反应的预测因子。再一次，我们发现了有希望的结果，我们正在积极寻求复制。我们还参与了几个与药物反应有关的联盟，并帮助该领域的其他人使用ACCORD队列复制他们的结果。

In one project, we have investigator genetic factors involved in the etiology of diabetic peripheral neuropathy (DPN). The aim of this study was to conduct a systematic search for genetic variants influencing DPN risk using two well-characterized cohorts. A genome-wide association study (GWAS) testing 6.8 million single nucleotide polymorphisms was conducted among participants of the Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trial. Included were 4,384 white case patients with type 2 diabetes (T2D) and prevalent or incident DPN (defined as a Michigan Neuropathy Screening Instrument clinical examination score >2.0) and 784 white control subjects with T2D and no evidence of DPN at baseline or during follow-up. Replication of significant loci was sought among white subjects with T2D (791 DPN-positive case subjects and 158 DPN-negative control subjects) from the Bypass Angioplasty Revascularization Investigation in Type 2 Diabetes (BARI 2D) trial. Association between significant variants and gene expression in peripheral nerves was evaluated in the Genotype-Tissue Expression (GTEx) database. A cluster of 28 SNPs on chromosome 2q24 reached GWAS significance (P < 5 10-8) in ACCORD. The minor allele of the lead SNP (rs13417783, minor allele frequency = 0.14) decreased DPN odds by 36% (odds ratio OR 0.64, 95% CI 0.55-0.74, P = 1.9 10-9). This effect was not influenced by ACCORD treatment assignments (P for interaction = 0.6) or mediated by an association with known DPN risk factors. This locus was successfully validated in BARI 2D (OR 0.57, 95% CI 0.42-0.80, P = 9 10-4; summary P = 7.9 10-12). In GTEx, the minor, protective allele at this locus was associated with higher tibial nerve expression of an adjacent gene (SCN2A) coding for human voltage-gated sodium channel NaV1.2 (P = 9 10-4). To conclude, we have identified and successfully validated a previously unknown locus with a powerful protective effect on the development of DPN in T2D. These results may provide novel insights into DPN pathogenesis and point to a potential target for novel interventions. With my PhD student Jon Lierer, we are conduction methods develop to help understand the heterogeneity in the disease. Type 2 diabetes (T2D) is a highly heterogenous disease. This heterogeneity suggests that there may be differences in latent underlying features that are driving the disease. Developing a more refined understanding of these underlying features could lead to important insights into disease etiology, more personalized treatment approaches, and better clinical outcomes. Here we refine the use of topological data analysis (TDA) to find latent subgroups in such heterogeneous diseases. We apply our TDA workflow on a diverse cohort of patients with T2D from the Action to Control Cardiovascular Risk in Diabetes (ACCORD) clinical trials to 1) identify clustered subgroups of patients based on the patient topology network derived from clinical characteristics 2) test the subgroups for enrichment in clinical outcomes 3) identify the features driving subgroup membership and 4) predict clinical outcomes based on identified features. We developed a robust workflow and corresponding code to implement TDA that minimizes overfitting through internal model validation, and introduces new metric for parameter optimization. We then applied this workflow in the ACCORD data. We identified eight subgroup clusters of T2D patients, three of which contained patients who were significantly over- or under-represented in at least one clinical outcome. Patients in Cluster 2 showed increased risk for the ACCORD primary outcome and major coronary events. Patients in Cluster 5 showed increased risk for primary outcome, congestive heart failure and expanded macrovascular outcome. Patients in Cluster 6 showed decreased risk for primary outcome, cardiovascular death, congestive heart failure, expanded macrovascular outcome, and major coronary events. For the three significant clusters, we identified the clinical variables driving cluster membership. There are both methodological and applied insights gleaned from this study. Here, we extend a modified TDA method by using data-driven tuning and holdout methods for testing and validation, which is an important development for a broad range of applications. From an applied perspective, we identified 8 clusters in patients with T2D in the ACCORD trial, 3 of which are significantly enriched for clinical outcomes, and identified several variables driving membership in clusters. We also compare our approach to LASSO regression, and establish that the cluster driven variable selection is comparable, while offering a number of unique advantages. The findings provide insight into structure behind the heterogeneity in T2D and demonstrate a valid data-driven method for extracting insight from the underlying topological characteristics of data. We have ongoing projects using the ACCORD data as well. Jon Lierer is working on conduction association analyses of response to rosiglitazone, as well as the adverse outcome of weight gain. He has identified interesting potential associations that we currently looking to replicate. We are also working the genetics of the hemoglobin glycation index. HGI quantifies the interindividual variation in the propensity for glycation and is a predictor of diabetes complications and adverse effects of intensive glucose lowering. Again, we have found promising results that we are actively pursuing for replication. We also participate in several consortia related to drug response, and help others in the field replicate their results using the ACCORD cohort.