探討癌症中基因體調節網絡

Abstract

差異網絡生物學是一門相當新興的學門，其主要探討不同細胞中「基因體調控網絡（genomic regulatory network）」之動態變化。由「基因體調節（genomic modulation）」的觀點切入，細胞中差異調控可由「調節基因（modulator gene）」之調節所造成。由於這樣的受調節網絡在分析上複雜度較高，目前尚無發表全基因體層次之系統性分析方法與研究。有鑑於此，本論文目標為全面的探討受調節之基因體調控網絡與其在細胞功能之整體面貌，並藉此進一步研究癌症中複雜的互動體（interactome）。具體而言，本研究根據調節基因的數值特質分為三部分：二階段性（two-state）、多階段性（multi-state）、及連續性（continuous）調節，並分別發展了可進行整合性分析之生物資訊方法。 在第一部分的二階段性調節研究中，我們首先探討最被廣為研究的調節基因─雌激素受體（estrogen receptor；ER）。我們建立了一套新穎的統計模型，藉由整合相關係數（correlation coefficient）與費雪轉換（Fisher transformation），該統計模型大幅改進了先前生物資訊方法之運算效率與表現。將此統計模型應用於分析公開之乳癌病患基因體資料，我們發現了受ER調節之TGF─NFB互動關係，並且證實了此動態互動關係與乳癌病患預後之關聯。我們接著探討急性骨髓性白血病（acute myeloid leukemia；AML）中一重要基因突變─NPM1基因突變─是否具有調節微小核醣核酸（microRNA）與基因間調控之能力。我們設計了一套系統性分析方法，在台大醫院之AML病患的基因體資料中，找出了數百對僅在NPM1基因正常病患中具有顯著調控關係的微小核醣核酸─基因對。我們並利用兩組獨立病患群的基因體資料及兩個細胞株的生物實驗，驗證了NPM1基因突變於調節微小核醣核酸─基因調控網絡之能力。此外，我們更從中發現了九對可做為獨立預後預測因子的微小核醣核酸─基因調控關係，顯示此受動態調控機制在AML之重要性。 在多階段性調節研究中，我們主要針對內生性競爭核醣核酸機制（competing endogenous RNA）進行研究，該機制描述藉由競爭微小核醣核酸而達成之基因共同表現（coexpression）關係。我們首先系統性尋找會影響內生性競爭核醣核酸調控強度的因子，並發現微小核醣核酸之表現量具有調節內生性競爭核醣核酸調控強度的能力。因此，在後續的研究中，我們探討在多型性神經膠母細胞瘤（glioblastoma multiforme；GBM）中受微小核醣核酸調節之內生性競爭核醣核酸調控網絡，並發現其參與突觸訊號傳遞和癌症相關生物功能。此外，我們的分析結果指出，三個免疫反應相關基因間之內生性競爭核醣核酸調控強弱，而非這些基因之表現量，具有預測GBM病患預後的能力，並於兩個獨立病患群資料中獲得驗證。 於本論文的最後一部分研究─連續性調節，我們提出了一套可用以分析受連續性基因體特徵調節的基因調控網絡之迴歸模型，並分析了微小核醣核酸與轉錄因子（transcription factor）等連續性基因體特徵。在GBM中，此部分之研究再度顯示基因體調節與重要的細胞功能有關。我們進一步發展了一個多元迴歸（multiple regression）模型，用以分析多個調節基因間互動關係。分析乳癌病患的基因體資料之結果顯示，乳癌中兩個重要基因─ESR1與ERBB2─間具有密切且複雜的共同調節（co-modulation）機制，參與調節相當廣泛的基因調控網絡。 總結而言，本論文全面的探討癌症中受調節之基因體調控網絡，並證實其在細胞功能之重要性，以及成為癌症中預後預測因子的潛力，進一步幫助了解複雜的癌症基因體與互動體。

Differential network biology, an emerging field in regulatory biology, studies the dynamics of genomic regulatory networks among cellular conditions. From the viewpoint of genomic modulation, such differential regulation can be modulated by modulator genes. Due to computational complexity, a systematic exploration into genomic modulation on a genome-wide scale has not been carried out. The thesis aims to comprehensively investigate the landscape of modulated genomic regulation and to pave the way to better understanding complex interactomes of cancers. Specifically, we proposed several bioinformatics methods to study three forms of genomic modulation characterized by the quantitative features of modulators: two-state, multi-state, and continuous modulation. In the study of two-state modulation, we started by investigating the best known modulator, estrogen receptor (ER). We devised a novel statistical inference that greatly improved computation efficiency from previous methods. Analyzing ER modulated gene and gene set (representing biological functions) interactions, we identified a ER+ dependent interaction between TGF and NFB, which was associated with patients’ prognosis. In acute myeloid leukemia, we investigated a novel role of NPM1 gene mutation as a modulator in microRNA (miRNA)-mRNA regulation (MMR). Significant pairwise correlation of hundreds of MMR pairs was seen specifically in NPM1-wild type patients, corroborated in independent cohorts and in vitro models. We further showed that the dynamic regulations of nine MMR pairs were independent predictors of prognosis. As for the multi-state genomic modulation, we studied a layer of gene-gene coexpression through the competition for common targeting miRNAs, namely the competing endogenous RNA (ceRNA) regulation. Our analysis indicated that ceRNA regulation was modulated by miRNAs. We investigated the modulation of miRNAs in ceRNA regulation in GBM and showed its involvement in synaptic transmission and tumor-related functions. Furthermore, we identified that the regulatory strength, rather than expressional abundance, of three immune response genes was predictive of survival. We proposed a regression model to analyze the gene regulatory networks modulated by continuous-state genomic features, including miRNAs and transcription factors. Applying the method to expression profiles of GBM, our results, again, suggested modulated regulation is involved in essential cellular processes. Our extended analysis to study the interactions among multiple modulator genes by a multiple regression further revealed the complex co-modulation between ESR1 and ERBB2 genes in breast cancer. In conclusion, we comprehensively investigated genomic modulation in cancers and illuminated its significance in cellular functions and potential as prognostic biomarkers, contributing to a better understanding of complex cancer genomes and interactomes.