以差異表現模組探討二十二種腫瘤型態的腫瘤異質性

Abstract

癌症長久以來都是造成人類死亡的主要疾病之一。近年來的研究指出，不同癌症間及相同癌症不同病人間都存在著高度的異質性，其生化機制、轉移風險、治療策略皆存在不少差異性，而這種異質性的現象，也是導致癌症難以治癒的原因之一。為了全面性了解癌化現象與腫瘤異質性（tumor heterogeneity），高通量全基因組資料（如：微陣列圖譜（Microarray）及次世代定序（next generation sequencing））已被廣泛應用於探討正常細胞跟癌細胞之間的差異，表現量顯著差異的基因也往往被當作潛在生物標記（biomarker）。 在細胞中，模組（module）是由一群具有高度連結並執行特定生物功能的蛋白質所組成。過去已有相關研究透過基因表現資料辨識模組，並結合實驗辨識的生物網路或KEGG資料庫的生化途徑來探討癌症異質性，然而這些方法大多受限於已知的生化網路及途徑。在我們先前的研究中，已結合實驗所辨識的蛋白質交互作用（protein-protein interactions, PPIs）及同源預測的蛋白質交互作用（homologous predicting PPIs）來建立同源映射網路（homologous mapping network），並透過演化保留性於此網路上辨識全面性模組（comprehensive modules）。在此研究中提出新的分析策略：「以模組為基礎的差異表現（module-base differential expression）」，我們從GEO（Gene Expression Omnibus）資料庫中蒐集了二十二種腫瘤類型共三十八個資料集（含一千一百八十八樣本，六百六十六個正常細胞的樣本，五百二十二個癌細胞樣本）。透過分析二十二種不同腫瘤中具有顯著差異表現之模組及在各個腫瘤形態中建立模組-蛋白質差異表現網路（module-protein differential expression network）以探討腫瘤異質性。 為了驗證模組是否適合用來分析腫瘤異質性，我們發現這些模組中包含了79.24%已知的癌症相關基因，在KEGG資料庫所記載的癌症相關的生化途徑中也包含了79.67%的基因。在預測癌症相關基因的分析中，相較於以基因為基礎的分析方法，我們的這套方法能擁有更高的精確率，同時也更能將相似的腫瘤類型分群（cluster）在一起。此外，我們發現在大部分的腫瘤類型中會正向調控的模組，主要參與細胞週期調控之生化途徑，而負向調控的模組主要參與細胞間訊息傳遞與運輸。我們相信這套以模組為基礎的差異表現分析法對於探討癌症異質性及辨識特定腫瘤類型之潛在生物標記是十分有用的。

Cancer is one of the most common causes of death in human. Recent studies have indicated that phenotypic and genetic heterogeneity, including disease mechanisms, metastasis risk, and therapy, exist not only between tumors but also between individuals with same tumor. This heterogeneity is one of the reasons for the difficulties in developing cancer therapy. For comprehensive understanding tumor heterogeneity, high-throughput genomic data (e.g., microarray and next generation sequencing ) has been widely used to explore the difference between normal and cancer cells. These significantly expressed genes were treated as potential biomarkers. In cells, a module is a group of proteins that are often highly connected and perform a certain kind of biological functions. To explore tumor heterogeneity, previous studies used gene expression data to identify modules, and described their involved biological processes using the biological network derived from experimental data or KEGG pathways. However, most of these studies are still limited to the use of well-known pathways and biological networks. In our previous studies, the homologous mapping networks are constructed through experimental protein-protein interactions (PPIs) and homologous predicting PPIs. Next, we have proposed that evolutionary conservation could be used to identify comprehensive modules in homologous mapping networks. Here, we proposed a new strategy, “module-base differential expression”, to integrate gene expression data sets across 22 different tumor types (666 normal samples and 522 tumor samples) from Gene Expression Omnibus database into comprehensive modules. For exploring tumor heterogeneity in 22 tumor types, significant differential expression modules are identified for each tumor types, and are used to establish differential expression network. We examined the analysis applicability of tumor heterogeneity for comprehensive modules, and found that the comprehensive modules contained 79.24% of known cancer-related genes and 79.67% genes that involved in KEGG cancer-related pathway. For predicting cancer-related genes, the precision of our strategy is higher than one of gene-based strategy. In addition, for clustering 22 different types of tumors, our strategy can categorize different types of cancers with same tissue, but gene-based strategy not. Moreover, the results show that up-regulated modules in most of tumor types are involved in the regulation of cell cycle; down-regulated modules are involved in cell-cell signaling and transport. We believe that the strategy of module-based differential expression is useful for exploring tumor heterogeneity and identifying biomarkers in specific tumor type.