以非目標性代謝體方法學搜尋與第二型糖尿病發生風險相關之血清代謝產物

Abstract

背景: 傳統評估個體罹患第二型糖尿病風險時，多採用年齡、性別、家族罹病史、空腹血糖值、糖化血色素及代謝症候群相關因子。然而使用這些傳統風險因子之評估模型，對於糖尿病未來風險之預測，並不完善。隨著代謝體學的發展與進步，其能提供低分子量(<1000 Da)代謝體詳細的背景資訊，以協助了解代謝物與目標疾病之致病機轉間的相關性。在本論文研究，我們以使用非目標性方法搜尋與第二型糖尿病罹病相關的代謝物，試圖改善第二型糖尿病的風險預測模型。 材料與方法: 本研究為嵌入型病例對照設計，研究對象來自於CVDFACTS (the CardioVascular Disease risk FACtor Two-township Study，竹東及朴子地區心臟血管疾病長期追蹤研究)，為一個以探討及評估心臟血管相關疾病為研究目的之社區型長期追蹤研究。我們選取50個第二型糖尿病之新罹病患者，並以年齡、性別做配對選出50個健康對照組。使用單變數羅吉斯迴歸分析，搭配共變量校正(年齡、性別、身體質量指數及血清血糖值)，篩選出可能之第二型糖尿病預測因子。多變數羅吉斯迴歸分析及正向逐步選取法，則用來產生第二型糖尿病風險評估之最佳模型。 結果: 共有39個代謝物被篩選為與第二型糖尿病有關之可能因子。經正向逐步選取法，挑出6個候選代謝物，並搭配2個傳統風險因子(身體質量指數及血清血糖值)，為最佳的第二型糖尿病風險預測組合。與傳統風險因子預測模型(AUCs= 0.89 (95% CI = 0.82-0.96)比較，本研究所找出的模型(AUCs= 0.99, 95%CI = 0.98 -1.00)具有顯著較好的第二型糖尿病預測能力(p-value=0.0018)。 結論: 經代謝體資料庫比對後，本研究發現5種代謝物(包含: 陽性電場之馬尿酸、陰性電場之馬尿酸、C16H27O6P、烷烴類衍生物、植物生物鹼及植物類黃酮)，組合此5種代謝物以及2個傳統風險預測因子(包含: 身體質量指數及血清血糖值)可提供良好的第二型糖尿病預測能力(AUCs= 0.99, 95%CI = 0.98-1.00)，且相對於傳統風險預測模型有顯著的改善(p-value =0.0018)。

Background: Traditional risk assessment to estimate of type 2 diabetes (T2D) risk for individual often include predictors, such as age, gender, family history, fasting plasma glucose, glycated hemoglobin (HbA1C) and metabolic syndrome components. However, it is not altogether satisfactory in term of predictability. With the development of metabolomics technology, it is likely to discover low-molecular-weight (<1000 Da) metabolites associated with disease etiology. Therefore, we aimed to identify novel metabolites for T2D with non-targeted metabolomics approach and with the goal to improve T2D risk prediction panel. Materials and Methods: We designed a nested case-control study, taking advantage of the data from CVDFACTS (the CardioVascular Disease risk FACtor Two-township Study), a community-based longitudinal cohort study designed to study risk factors and evaluation of cardio-metabolic diseases in Taiwan. We selected 50 new-onset T2D and 50 age and gender matched controls who were chosen from those who did not develop T2D. Their stored baseline fasting serums were used for metabolomics study. Univariate logistic regression with covariates adjustment (age, sex, BMI and serum glucose) was used to screen potential determinants. Multivariate logistic regression was used to generate risk assessment model for predicting T2D risk. Results: A total of 39 peaks were initially screened out as potential metabolites. Then by the forward stepwise selection, 6 candidate peaks in combination with 2 traditional risk factors (BMI and serum glucose) were selected into the T2D risk prediction panel. With the comparison to traditional risk factors model (AUCs=0.89 (95% CI = 0.82-0.96), our model performed significantly better in terms of ROC result (AUCs= 0.99, 95%CI=0.98-1.00) for T2D prediction (p-value =0.0018). Conclusion: After the identification on metabolomics database, we have identified 5 metabolites: hippuric acid in positive mode, hippuric acid in negative mode, C16H27O6P, alkane derivative, plant alkaloid and plant flavonoid together with 2 traditional risk factors (BMI and serum glucose). This diabetes risk prediction panel can distinguish future T2D cases from healthy controls with an AUC of 0.99 (95%CI=0.98-1.00). A significant improvement is achieved compared to the traditional risk factor panel (p-value =0.0018).