生物晶片資料分析與肺腺癌存活之預測

Abstract

中文摘要 近年來肺癌高居國內及全球癌症相關死因首位，其死亡率至今仍然居高不下。非小細胞肺癌乃發生率最高之肺癌，其中又以肺腺癌最為普遍。研究指出，肺癌病患的治療方式不只取決於腫瘤類型，而不同肺癌分期亦應適當選擇給予不同治療方式。因此，為幫助病患選取最有利之治療方式，建立更準確之新肺癌診斷方法，有其重要性及迫切性。其中，因前期肺癌病患仍有數種治療方式可選擇，故對於前期肺癌病患之診斷最為重要。 生物晶片技術的發展使得研究人員得以同步測量數以萬計之基因表現量，並提供了新的研究平台。一項大型的肺腺癌研究建立了豐富的肺腺癌病患之基因表現量資料和臨床資料，用以建立及驗證數種以基因表現量導出之肺癌診斷方法。然而，對於前期病患的存活預測，尚未能找到一種基因訊號對於所有的驗證資料皆能達到百分之五之顯著水準。我們的研究目的即是重新研究這份資料，以期建立一個新的基因訊號對於所有的驗證資料皆能有顯著的存活預測力。 我們引用一個兩階段之維度縮減方法佐以部份的修正來導出肺癌診斷之基因訊號。在第一階段中，我們以基因和存活時間的相關性與配對基因和存活時間的流動關聯性來選出重要的候選基因。第二階段，我們則使用了改良的切片逆迴歸分析方法導出最後的肺癌診斷基因訊號。 分析結果指出，以相同的驗證流程檢驗我們導出的基因訊號，在所有的驗證資料都能達到百分之五之顯著水準的存活預測能力，更進一步的，在另一個包含肺腺癌及鱗狀細胞癌的非小細胞肺癌資料上，我們導出的基因訊號也同樣達到百分之五之顯著水準的存活預測能力。因此，我們認為以TMEM66、CSRP1、BECN1、FOSL2、ERO1L、SRP54及PAWR七個基因所導出的基因訊號對於非小細胞肺癌病患有好的存活預測能力。

Abstract Purpose Recently, several new gene expressions based signatures were proposed to predictive the survival of Non Small Cell Lung Cancer (NSCLC) patients. However, for stage I patients, the task is more difficult and no signatures had been found from a large study of lung adenocarcinoma. We reanalyzed this large sample data and tried to construct a gene signature, which had significant prediction power for all stage and early stage patients in all the validation sets. We also used an external independent cohort data set containing both adenocarcinomas and squamous cell carcinomas to test if our gene signature still had significant prediction power for all stage and early stage NSCLC patients. Materials A total of 442 lung adenocarcinoma gene expression profiles from Shedden et al. (2008) containing four independent data sets were reanalyzed in our study. Two of the data sets were combined as a training data set to derive our gene signature. The other two data sets were used for validation. An external NSCLC data from Duke lung caner cohort was used for additional validation. Methods We modified a two-steps dimension reduction method proposed by Wu et al. (2008) to derive our gene signature. In the first step, both correlation and liquid association methods were used to select the candidate genes. In the second step, we applied the modified sliced inverse regression proposed by Li et al. (1998) to derive a gene signature from the candidate genes. Results Five genes TMEM66, CSRP1, BECN1, FOSL2 and ERO1L were selected by correlation methods. SRP54 and PAWR (as a LA pair) were selected by liquid association method. The final signature gave significant prediction power for samples with all stage patients and for samples with stage I patients only in all the validation sets. Conclusion The gene signature derived from the seven genes (TMEM66, CSRP1, BECN1, FOSL2, ERO1L, SRP54 and PAWR) had good prediction power for all stage and early stage NSCLC patients.