胰腺癌相关糖尿病的血清代谢组学分析及诊断模型建立的初探

Serum metabolomics analysis and establishment of diagnostic model of pancreatic cancer associated diabetes

目的通过检测胰腺癌相关糖尿病的血清代谢标志物,拟构建诊断模型。方法纳入2013年6月至2014年7月在上海交通大学医学院附属瑞金医院被诊断为胰腺癌伴新发糖尿病和新发2型糖尿病的患者各30例。采用代谢组学液相色谱-质谱(LC-MS)检测对纳入患者的血清标本进行代谢组学分析。原始数据经正交偏最小二乘法方法(OPLS)分析,获得两组间最具差异性表达的代谢产物。以每组各前15例为训练样本,其余为验证样本进行验证。在训练样本中,采用逐步法将差异代谢产物和临床数据纳入logistic回归分析建立模型。再于验证样本中验证该模型的诊断效率。结果在胰腺癌伴新发糖尿病组和新发2型糖尿病组中筛选出10个最具差异性的代谢产物。正离子模式下差异最显著的代谢产物分别为3-酮基鞘氨醇、花生酰多巴胺、磷脂酰乙醇胺(18∶2)、辅酶Q1和缬氨酸,负离子模式下差异最显著的代谢产物分别为C16-鞘胺醇-1-磷酸、酮棕榈酸、异亮氨酸、N-琥珀酰-L-二氨基庚二酸和尿苷。在训练样本中建立诊断模型:p=e(Xβ)/(1+e(Xβ)),(Xβ)=-158.975-1.891(年龄)+0.309(磷脂酰乙醇胺18∶2)+1.035 (C16-鞘胺醇-1-磷酸)+0.084(异亮氨酸)+1.114 5(N-琥珀酰-L-二氨基庚二酸)。此模型在验证样本的ROC AUC值为0.982,灵敏度和特异度均为93.3%。结论基于血清代谢产物的诊断策略是具有前景的新发糖尿病筛选胰腺癌的手段。

Objective To establish the diagnostic model based on detection of serum biomarkers in pancreatic cancer (PC) associated diabetes. Methods From June 2013 to July 2014, at Ruijin Hospital, School of Medicine, Shanghai Jiao Tong University, 30 patients diagnosed with PC companied with new onset diabetic mellitus and 30 patients with new onset type 2 diabetic mellitus, were enrolled. Serum samples were examined by liquid chromatography-mass spectrometry (LC-MS) for metabolomics analysis. Orthogonal partial least square (OPLS) was performed for raw data analysis to obtain the differentially expressed metabolites between two groups. The first 15 cases of each group were taken as training samples and the left as validation samples. The model was established using logistic regression via stepwise differentially expressed metabolites and clinical data input in training samples. The diagnostic efficiency of the model was verified in validating samples. Results Ten differentially expressed metabolites were identified in PC companied with new onset diabetic mellitus group and new onset type 2 diabetic mellitus group. The differentially expressed metabolites identified in positive ion mode were 3-ketosphingosine, arachidonoyl dopamine, phosphatidylethanolamine (18∶2), ubiquinone-1 and valine. The differentially expressed metabolites identified in negative ion mode were C16 sphingosine-1-phosphate, keto palmitic acid, isoleucine, N-succinyl-L-diaminopimelic acid and uridine. The diagnostic model was established in training samples: p=e(Xβ)/(1+ e(Xβ)),(Xβ)=-158.975-1.891 (age)+ 0.309 (phosphatidylethanolamine 18∶2)+ 1.035 (C16 sphingosine-1-phosphate)+ 0.084 (isoleucine)+ 1.114 5 (N-succinyl-L-diaminopimelic acid). The area under curve (AUC) of receiver operating characteristic (ROC) of this model was 0.982 in validation samples, the sensitivity and specificity were both 93.3%. Conclusion Serum metabolomics-based diagnostic approach is a promising method for screening PC from new onset diabetic mellitus.