整合基因與長片段非編碼RNA表現量資料預測病人之輻射線敏感度

Abstract

放射線治療被廣泛應用在癌症治療上，隨著放射技術與影像定位技術日益精進，各種癌症在適當的治療計畫規劃下，都能夠進行放射線治療。然而，在臨床上，多數的治療計畫主要依照癌症種類不同而非依照病人對於放射線敏感程度進行劑量規劃，使每位病人在治療後的預後結果與治療效果因個體而有所不同。在個人化醫療的時代逐漸到來，評估病人的放射線敏感程度，能夠幫助治療團隊提出更適切每個人的治療計劃，減少病人在治療後的副作用，並且達到治療效果的最大功效。 近年來，許多文獻對於放射線對於基因表現之影響有相當深入的研究，但探討長片段非編碼RNA在輻射線上的影響研究數量並不多。而相較於針對個體輻射線敏感程度，多數研究放射線的主題仍多主要聚焦在預測病人吸收放射線劑量。在本篇研究中，我們整合了長片段非編碼RNA的表現及基因的表現量來建立預測模型以預測接受輻射線治療後的病人之預後。 整體的研究流程可以簡單分為三個階段： 首先，藉由GSE26835中共1086個檢體，我們先定義出了對於照射輻射線後會有表現量差異的基因與長片段非編碼RNA。接著，透過廣泛被應用的標準細胞株：NCI-60做為我們的輸入資料，結合其放射線參數與基因表現量，並利用基因演算法從上一步已挑選出顯著差異的基因與長片段非編碼RNA群中計算出最能用以預測細胞株放射線敏感程度的20個基因與長片段非編碼RNA。最後，利用這20個變數與機器學習演算法(SVM)，應用在兩組臨床的腦膜癌(glioblastoma mutliforme)資料上-TCGA與GSE16011，用以評估病人的放射線敏感高低及其個別的預後結果。為證明本研究的結果實為針對放射線治療而非因為其他因子影響成果，我們也將沒有進行放射線治療的病人進行分析，其結果如我們預期的沒有達到統計上的顯著差異。 相較於其他文獻，我們提出了一個較標準的預設模型，即每位病人都可以單獨進行預測，而不需要透過累積一群病人的資料量才能進行分析，此點使本方法在應用層面上更能使用於臨床治療。而我們也是第一個將長片段非編碼RNA做為放射線敏感預測模型的研究。 簡而言之，本研究透過高通量的基因晶片資料，辨別出20個可以用於臨床預測病人放射線治療預後之模型，期望透過未來透過這樣的結果可以提供給醫生與醫學物理師更多病人資訊，用於改善病人的治療計畫並達到個人化醫療之目的。

Radiotherapy has been a standard procedure in cancer treatment. With the improvement of medical technology, the conventional radiation therapy is combined with image-guided method (processed by Computed Tomography) and treatment planning, provides high control rate in tumor region. However, most of the radiotherapy planning is based on the cancer type rather than the radiosensitivity of individuals. Thus, to identify the radiosensitivity of each patient to improve the treatment planning and to reduce the side effect becomes an important issue in the era of personal medicine. Recently, there are more and more studies that focus on the function of long non-coding RNA (LncRNA) which were once regarded as “dark matter”. Although there are several studied that discuss the relationship between gene expression and radiation, LncRNA is relatively new to the field of radiation. In this study, the process can be described into three parts: First, we identified radiation responded genes and LncRNAs which has differential expression in GSE26835 with total 1086 samples. Second, we integrate the NCI-60 cell line data, which has been regarded as standard cell lines and its radiation parameter to filter out the radiosensitivity-related pattern in the genes and LncRNAs that identified in the previous step. To select appropriate variables to build up prediction model, we used genetic algorithms to identify 20 probe sets that have the best performance in predicting the radiosensitivity of cell lines. Lastly, we applied two real glioblastoma multiforme (GBM) – TCGA and GSE16011 to validate the 20 probes sets can be used to predict the radiosensitivity of patients or not. We defined patients into radiosensitivity group and radioresistance group, and analyze their survival curve through Log rank test and cox regression test. To prove these 20 probe sets are radiotherapy-specific, we also examined the patients that did not receive radiotherapy. As expected, there is no significant difference in the survival of Non-RT patients. Compares to other research, our study first integrates the expression profile of genes and LncRNAs. Moreover, we can predict the radiosensitivity once we have the data of patients instead of collecting a series of patients, which makes our method more practically and can be used in clinical situation. To sum up, our study identifies 20 probe sets that can predict the radiosensitivity of patient with high throughput microarray. We are looking forward to our research can help doctors and medical physicians to improve the prognosis of cancer patients.