動態同源映射網路揭示細胞行為及疾病機制

Abstract

隨著基因體、蛋白質體、生化訊息及代謝學資料的快速增加，建立蛋白質交互作用網路將有助於探討生理及病理狀況的分子機制。近年來，為了建立蛋白質交互作用網路，許多大規模實驗方法被開發用來辨識蛋白質交互作用（protein-protein interaction, 簡稱PPI）。然而，相關研究指出大規模實驗所辨識的PPI資料中，部分物種的PPI存在著高偽陽性率。此外，在相同時間和空間尺度下，細胞生化網路中蛋白質間動態地協同作用，並聚集成蛋白質模組（Module）對於生物程序是不可或缺的。因此，透過建立動態網路（dynamic networks）及動態模組來揭示細胞行為與疾病機制乃是趨勢所需。 針對這些議題，我們提出蛋白質交互作用家族（PPI family）與模組家族（Module family）之概念，透過多物種映射在目標物種中建立其同源映射蛋白質交互作用網路（homologous mapping network）。舉例來說，相對於實驗所辨識的人類蛋白質交互作用網路僅具有12,121蛋白質及66,057蛋白質交互作用，我們的蛋白質交互作用網路則較為完整，包含了17,343蛋白質及100,788蛋白質交互作用。為了分析複雜的蛋白質交互作用網路，我們提出模組的基本組成原則，用來辨識模組、同源模組及模組交互作用網路（module-module interaction network）以協助分析蛋白質交互作用網路。同時，我們也開發了MoNetFamily伺服器來提供網路註解，如模組家族之蛋白質交互作用組成分析、模組家族及鄰近模組之功能性註解。透過我們的模組交互作用網路，推測兩個模組（JAK2-PAFR-TYK2及IL-6/IL6Rα/gp130）可用來探討HIV誘發牛皮癬的疾病機制。 進一步，我們將21種不同癌症的人類全基因表現資料整合進網路中，用來探討尼古丁乙醯膽鹼受體α9（α9-nAChR）在尼古丁誘導之癌症中的調控機制。目前，所有的PPI資料庫都缺乏尼古丁乙醯膽鹼受體α9（α9-nAChR）的交互作用蛋白質。針對α9-nAChR，我們的方法成功地找到9個α9-nAChR的交互作用蛋白質，包括EGFR、ERBB2、CSNK1D、COPS6及NCLN等，我們的網路顯示出EGFR、FGFR1及INSR會透過MAPK訊息傳遞路徑來調控細胞增生及凋亡。透過結合58個乳癌腫瘤組織及147正常組織之基因表現資料所建立的動態網路，我們發現腫瘤形成過程中這些交互作用蛋白質會與α9-nAChR分開來增加其活性，然而，ERBB2在正常組織及類基底型（basal-like）腫瘤組織皆呈現結合關係。因此，我們推測當病人具有吸菸習慣或接觸二手菸時，會誘導ERBB2與α9-nAChR分開而活化ERBB2訊息路徑，進而促進乳癌細胞（特別是類基底型腫瘤）之形成。此外，我們收集並結合了超過7,000個人類微陣列資料集（microarray datasets）至模組及網路中建立動態模組及動態網路，並用來反映在不同細胞及狀態下驅動表型變異的細胞生化途徑，例如在21種癌症（含983個臨床檢體）中探討腫瘤異質性（tumor heterogeneity）。總結上述結果，我們認為在同源映射網路中模組的動態行為對於探討疾病機制及細胞行為是十分有用。

As an increasing number of genomic, proteomic, signaling and metabolomic data become available, the construction of networks is valuable for understanding the molecular mechanisms and disease mechanism. Many large-scale experimental methods have been proposed to identify the protein-protein interactions (PPIs) for constructing PPI networks, but these experimental PPIs were often studied on some species with high false-positive rates. In addition, dynamic cooperation between proteins for assembly of protein modules in time and space is essential to execute biological processes in networks of cells. Therefore, the construction of dynamic networks and dynamic modules are urgently required to reveal cell behaviors and disease mechanisms. To address these issues, we first proposed PPI and module family to construct (enlarge) the PPI networks through homologous mapping (across multiple species). For example, we derived homologous mapping human network, 17,343 proteins and 100,788 PPIs, which is much more complete than the previous works, ~12,121 proteins and 66,057 PPIs. For analysis of complex PPI networks, we proposed the organizational principles of modules for identifying the modules, homologous modules (called module family), and module-module interaction (MMI) networks. Moreover, we also developed a server (MoNetFamily) to provide the network annotations; the profiles of proteins and PPIs in module families; GO annotations of neighboring modules and module families. Our MMI network inferred two modules, JAK2-PAFR-TYK2 and IL-6/IL6Rα/gp130, to explore the mechanism of HIV-associated psoriasis. Next, we used these networks integrated with human gene expression data across 21 diverse cancer types to explore nicotine-induced cancer mechanisms regulated by nicotinic acetylcholine receptor subunit alpha-9 (α9-nAChR). Currently, all databases lack the records of interaction partners of α9-nAChR. For the α9-nAChR, our strategies successfully identified 9 novel interacting proteins, including EGFR, ERBB2, CSNK1D, COPS6, and NCLN. Our network shows that EGFR, ERBB2, CSNK1D, and COPS6 are involved in the MAPK signaling pathway to regulate cell proliferation and apoptosis. Based on the dynamic networks combined with gene expression profiling of 58 breast tumor tissues and 147 corresponding normal tissues, we observed the activations of these interacting proteins may increase when they disassociated from the interactions with α9-nAChR during tumorigenesis. Interestingly, ERBB2 seems to associate with α9-nAChR both in normal tissues and tumor tissues of basal-like subtype. Therefore, patients with active and passive smoking would induce disassociation between ERBB2 and α9-nAChR to increase activation of ERBB2 signalling, and result in promotion and progression of breast cancer cells, especially in basal-like cancers. In addition, we collected over 7,000 human gene expression data sets from Gene Expression Omnibus and integrated them into modules/networks to construct dynamic modules/networks for reflecting the cellular processes, driving phenotypic variations, in different cells and states; for example, exploring tumor heterogeneity in 21 diverse cancer types among 983 clinical samples. We believe that the dynamic modules in homologous mapping networks are useful for discovering cell behaviors and disease mechanisms.