適性整合法聯結多組織之預測轉錄體以進行基因環境交互作用分析

Abstract

現今有許多方法可以檢測基因與環境因子之間的交互作用效應。全基因組關聯研究 (Genome-wide association studies, GWAS) 已偵測出某些染色體區域與可能導致疾病或性狀的環境因子產生交互作用。但是，目前尚不清楚大多數已發表的基因-環境因子交互作用的生物學機制。 為了測試單核苷酸多態性 (Single-Nucleotide Polymorphisms, SNPs) 與環境因子在感興趣之性狀上的交互作用，我們提出了一種新的統計方法，即適性整合法 (Adaptive Integration Method, AIM) 。首先，我們通過使用PrediXcan 和基因型組織表達庫 (Genotype-Tissue Expression ,GTEx) 通過不同組織的推測表達定量性狀基因座 (expression quantitative trait loci, eQTLs) 預測轉錄組。接下來，我們將預測的轉錄組轉換為「轉錄組風險評分」 (Transcriptome Risk Score , TRS) ，並測試了感興趣之性狀與每個組織中TRS之間的關聯。因為並非所有組織都與感興趣之性狀相關，所以只有某些組織的轉錄組可以作為基因與環境因子交互作用的證據。因此，我們採用適性整合方法 (AIM) 偵測適性聯結來自多個組織的基因-環境因子交互作用的信號。 例如，我們評估基因和腰臀比 (Waist-to-Hip Ratio, WHR) 對於空腹血糖的關聯性。我們的分析包括來自「臺灣人體生物資料庫」 (Taiwan Biobank, TWB) 中，在TWB2的58,774名參與者，使用密西根插補伺服器插補大約600萬個SNPs位點。通過使用我們的AIM，使用顯著性水準為2×〖10〗^(-6)共偵測出51個基因，為了呈現可複製性，在TWB1的25,460名參與者進行同樣的分析步驟，使用顯著水準0.05，一共檢測出10個顯著基因。從DNA角度來看，根據目前以基因為單位的分析方法，在TWB 2.0中的51個基因有18個被偵測出與WHR對於空腹血糖中有交互作用。例如，PPP1R3C (protein phosphatase 1 regulatory subunit 3C) 基因在DNA水平和RNA水平上都共定位以顯示與WHR的顯著交互作用。

There are bountiful methods for detecting the interactions between genes and environmental factors. Genome-wide association studies (GWAS) have identified some chromosomal regions interacting with environmental factors that may be responsible for diseases or traits. However, the biological mechanisms underlying most published gene-environment interactions have been still unclear. To test for the interactions of single-nucleotide polymorphisms (SNPs) with environmental factors on traits of interest, we proposed a new statistical method, Adaptive Integration Method (AIM). Firstly, we predicted transcriptome by putative expression quantitative trait loci (eQTLs) of different tissues by using PrediXcan and the Genotype-Tissue Expression (GTEx) library. Next, we transformed the predicted transcriptome into transcriptome risk score (TRS) and tested associations between the trait of interest and TRS in each tissue. Because not all tissues are related to the trait of interest, only transcriptome from some tissues may serve as evidence of gene-environment interactions. Therefore, we adaptively integrated the gene-environment interaction signals from multiple tissues with our Adaptive Integration Method (AIM). As an example, we evaluated the interactions between genes and waist-to-hip ratio (WHR) on fasting glucose. Our analysis included 58,774 participants from the Taiwan Biobank 2.0 (TWB), with approximately 6 million imputed SNPs by the Michigan Imputation Server. By using our AIM, a total of 51 genes were identified at the significance level of 2 × 10-6. To replicate the results, we performed the analysis for 25,460 participants from the Taiwan Biobank 1.0. We replicated 10 significant genes with p-value<0.05. From the DNA perspective, 18 out of the 51 genes identified from TWB 2.0 were also confirmed to have interactions with WHR on fasting glucose according to current gene-based analysis methods. For example, the PPP1R3C (protein phosphatase 1 regulatory subunit 3C) gene was colocalized to exhibit significant interaction with WHR, from both the DNA level and the RNA level.