Objectives: To estimate postmortem interval (PMI) by analyzing the protein changes in skeletal muscle tissues with the protein chip technology combined with multivariate analysis methods.
Methods: Rats were sacrificed for cervical dislocation and placed at 16 ℃. Water-soluble proteins in skeletal muscles were extracted at 10 time points (0 d, 1 d, 2 d, 3 d, 4 d, 5 d, 6 d, 7 d, 8 d and 9 d) after death. Protein expression profile data with relative molecular mass of 14 000-230 000 were obtained. Principal component analysis (PCA) and orthogonal partial least squares (OPLS) were used for data analysis. Fisher discriminant model and back propagation (BP) neural network model were constructed to classify and preliminarily estimate the PMI. In addition, the protein expression profiles data of human skeletal muscles at different time points after death were collected, and the relationship between them and PMI was analyzed by heat map and cluster analysis.
Results: The protein peak of rat skeletal muscle changed with PMI. The result of PCA combined with OPLS discriminant analysis showed statistical significance in groups with different time points (P<0.05) except 6 d, 7 d and 8 d after death. By Fisher discriminant analysis, the accuracy of internal cross-validation was 71.4% and the accuracy of external validation was 66.7%. The BP neural network model classification and preliminary estimation results showed the accuracy of internal cross-validation was 98.2%, and the accuracy of external validation was 95.8%. There was a significant difference in protein expression between 4 d and 25 h after death by the cluster analysis of the human skeletal muscle samples.
Conclusions: The protein chip technology can quickly, accurately and repeatedly obtain water-soluble protein expression profiles in rats' and human skeletal muscles with the relative molecular mass of 14 000-230 000 at different time points postmortem. The establishment of multiple PMI estimation models based on multivariate analysis can provide a new idea and method for PMI estimation.
目的: 利用蛋白质芯片检测技术结合多维统计方法分析骨骼肌组织的蛋白质变化，进行死亡时间推断。方法: 大鼠颈椎脱臼处死后置于16 ℃环境中，提取死后不同时间点（0 d、1 d、2 d、3 d、4 d、5 d、6 d、7 d、8 d和9 d）骨骼肌的水溶性蛋白质，获取相对分子质量为14 000~230 000的蛋白质表达谱数据，采用主成分分析（principal component analysis，PCA）和正交偏最小二乘（orthogonal partial least squares，OPLS）进行数据分析，构建Fisher判别模型及反向传播（back propagation，BP）神经网络模型对大鼠死亡时间进行分类预测。另外收集不同死亡时间的人体骨骼肌蛋白质表达谱数据，通过热图和聚类分析观测其与死亡时间之间的关系。结果: 大鼠骨骼肌蛋白质谱峰随死亡时间呈一定的时序性变化。PCA结合OPLS判别分析结果显示，除死后6 d、7 d和8 d外，各死亡时间组间差异有统计学意义（P<0.05）；经Fisher判别分析，模型内部交叉验证准确率为71.4%，外部验证准确率为66.7%；BP神经网络模型分类预测结果显示，模型内部交叉验证准确率为98.2%，外部验证准确率为95.8%。经聚类分析，人体骨骼肌蛋白质表达在死后4 d与死后25 h内差异有统计学意义。结论: 蛋白质芯片检测技术可快速、准确、高重复性地获取死后不同时间点大鼠和人体骨骼肌相对分子质量为14 000~230 000的水溶性蛋白质表达谱，建立基于多维统计方法的多种死亡时间推断模型，有望为死亡时间推断提供新的思路和方法。.
Keywords: Fisher discriminant analysis; back propagation neural network; forensic pathology; postmortem interval; protein chip; protein expression profile; rats.