探討臺灣族群胰臟癌與卵巢癌的相關基因

Abstract

研究背景： 根據衛福部國人癌症相關死因統計，胰臟癌與卵巢癌的死亡率分別位居第七名與第十名。發生率有種族上的差異，兩種癌症都是白人的發生率最高，其次為亞洲人與黑人，其中胰臟癌死亡率(8.9/100,000人)跟發生率(9.5/100,000人)接近，尤其研究目標的胰臟管腺癌的五年存活率更未達10%，致死率相當高。透過流行病學和基因的研究，發現胰臟癌和卵巢癌之間存在一些共通性。有胰臟癌家族史的親屬罹患卵巢癌的風險為2.25倍，而有遺傳性卵巢癌的人罹患胰臟癌的風險是一般人的2到3.5倍，流病上的證據表明了胰臟癌與卵巢癌有相關性。基因研究方面，這兩種癌症有些致病基因是重疊的，例如常見的BRCA2。研究發現除了生活習慣等危險因子，基因變異約可以解釋5-6%的胰臟癌與10-15%的卵巢癌，在某種程度上影響一個人的終生風險。大多數的胰臟管腺癌與卵巢癌在早期皆無明顯症狀，然而全部人做篩檢並不符合經濟效益，因為胰臟癌與卵巢癌的發生率低，因此找到特定的風險變異位點來區分高低風險人群，並針對高危險人群做定期追蹤與詳細檢查更有效益。胰臟癌與卵巢癌的全基因組關聯性分析研究多以西方為主，東方的驗證性分析發現，重疊的顯著相關基因有限，代表不同種族的結果無法很好地外推，因此臺灣需有自己種族的研究來找出胰臟癌與卵巢癌的相關變異位點，並建立多基因風險評分模型。 研究方法： 研究對象的病例組來自台大醫院，胰臟癌有180人，卵巢癌有91人。基因資訊是從血液檢體萃取出DNA後，用Axiom Genome-Wide TWB 2.0 Array Plate定序。對照組來自臺灣人體生物資料庫，共有4,999人，會從中以年齡及性別匹配1：4的人當作對照組。基因資訊是從血液檢體萃取出DNA後，用Axiom Genome-Wide Array Plate定序。本研究有3個模型，模型一有176為胰臟癌病人與704位對照組、模型二有85位卵巢癌病人與340位對照組、模型三有261為胰臟癌或卵巢癌病人與1044位對照組。第一部分為全基因組關聯分析(Genome-wide association study, GWAS)。在經過資料前處理和品質控制之後，根據既有的單倍型，判斷與千人基因組計畫的東亞參考族群中最相似的部分，加以插補未定序到的位點，插補完並經過位點的品質控制後得到約9百多萬個位點。樣本以年齡與性別來配對1：4的病例組與對照組。接著對單核甘酸變異位點進行關聯性分析，將找到的顯著基因與資料庫或研究比較。第二部分是建構多基因風險分數(polygenic risk score, PRS)模型。將插補後的資料分為80%的訓練集與20%的測試集。透過上一步全基因組關聯分析得到的摘要統計量，選擇顯著的位點，再用訓練集對這些位點進行關聯性分析。分數計算考慮了納入的p值與連鎖不平衡，透過訓練集得到的權重來估計出測試集的分數，最後依據AUC、準確率、F1指標來評估所建立的模型。 研究結果： 模型一找到878個顯著位點(p<2.69E-8)，模型二達顯著水準的位點有815個(p<2.71E-8)，模型三有4039個位點達顯著(p<2.63E-8)，將鄰近的基因與GWAS catalog比對，沒有相關基因與卵巢、胰臟疾病或是癌症被報導過，有數個基因被報導與乳癌、肺癌、攝護腺癌、大腸癌、膀胱癌相關。模型一得到胰臟癌的風險有37.6%受到7個位點的影響，AUC為0.7564，準確率為0.6648。模型二得到卵巢癌的風險有8%受到68個位點的影響，AUC為0.5753，準確率為0.5412。模型三用18個位點解釋得到胰臟癌或卵巢癌風險變異的能力高達41.7%，AUC為0.8654，準確率為0.7548。然而用一個英國生物樣本庫建立的胰臟癌多基因風險分數模型來驗證本研究，AUC下降至0.52。 結論： 透過全基因關聯性分析找到了許多與胰臟癌與卵巢癌相關的基因，且不管是分開或合併分析，都有重疊的基因或位點，像是TNFRSF14、LINC00970、GPATCH1、PSMD14、LINC00954、DDX6P1、LOXL2、AHCYP2，因此有共通性的胰臟癌與卵巢癌是可以合併分析的。將國外的研究與本研究比較，相同的位點等位基因頻率差距大，甚至風險方向性不一致，以國外建立的風險模型套用在臺灣族群也是表現不佳。這些都表明胰臟癌與卵巢癌在臺灣族群有不同於其他族群的相關基因。

Background: According to the Ministry of Health and Welfare of Taiwan, the mortality rates of pancreatic cancer and ovarian cancer among Taiwanese patients were ranked seventh and tenth respectively. There are racial differences in the incidence of both cancer types. The incidence of both cancers is highest in whites, followed by Asians and blacks. The mortality rate (8.9 per 100,000 people) of pancreatic cancer is close to the incidence (9.5 per 100,000 people), especially the five-year survival rate of patients with pancreatic ductal adenocarcinoma (PDAC) is less than 10%. According to previous epidemiological and genetic studies, there are some commonalities between pancreatic and ovarian cancers. For example, individuals with a family history of pancreatic cancer are 2.25 times more likely to develop ovarian cancer, and those with hereditary ovarian cancer are 2 to 3.5 times more likely to develop pancreatic cancer. Epidemiological evidence suggests that there exists an association between pancreatic cancer and ovarian cancer. Genetic studies have reported overlapping causative genes from the two cancers, such as BRCA2. Prior studies have demonstrated that in addition to risk factors such as lifestyle, genetic variation can explain about 5-6% of the risk of occurrence of pancreatic cancer and 10-15% of ovarian cancer. Most patients with PDAC and ovarian cancer are asymptomatic in the early stage. However, screening all people is not cost-effective, because of the low incidence, so it is necessary to find specific risk variant loci that can provide potential risk assessment. Currently, GWAS of pancreatic cancer and ovarian cancer are mostly be done on patients from the western countries. However, the results have seldom been validated on patients from the east-Asian ancestry. Therefore, it is required to analyze and provide a genetic model for both cancer types based on the Taiwanese patients. Methods: 180 PDAC patients and 91 ovarian cancer patients for this study were enrolled from National Taiwan University Hospital. The SNP genotyping was done using Axiom Genome-Wide TWB 2.0 Array Plate after the DNA was extracted from the peripheral blood specimens. The control group was obtained from the Taiwan biobank healthy samples. A total of 4,999 people were included, and a matching of 1:4 ratio was used to identify healthy control group for the analysis, which were genotyped using the Axiom Genome-Wide Array Plate. 3 models were developed in this study. Model 1 consists of 176 PDAC patients and 704 controls, the model 2 was developed using 85 ovarian cancer patients and 340 controls, and model 3 has 261 PDAC or ovarian cancer patients and 1044 controls. For each of the three models, a genome-wide association study (GWAS) was performed, respectively. After data preprocessing and quality control, the East Asian samples from the 1000 Genomes Project was used as a reference panel for imputation, and about 9 million loci were obtained after imputation. Genome wide association analysis were then performed after quality control. Next, polygenic risk score (PRS) models were constructed to do the prediction of genetic risk for both cancer types. The samples were divided into 80% training set and 20% testing set. Based on the summary statistics obtained in the previous step, significant loci were selected, and the correlation analysis was performed on the remaining loci using the training set. The score calculation took significant loci with correlation less than 0.1 into account, and estimates the score of the testing set through the weight obtained from the training set. Finally, the established model was evaluated based on the area under the ROC curve (AUC), accuracy, and F1-score. Results: Model 1 found 878 genome-wide significant loci with P-value <2.69E-8, model 2 had 815 loci with P-value <2.71E-8, and model 3 had 4039 significant loci with P-value <2.63E-8. The nearest genes were compared with the GWAS catalog, of which none were prior reported for both the cancers. Several genes had been reported in previous study, including breast cancer, lung cancer, prostate cancer, colon cancer and bladder cancer. The phenotypic variance explained by PRS of model 1 was 37.6% using 7 loci, with an AUC of 0.7564 and an accuracy of 0.6688. The phenotypic variance explained by PRS of model 2 was 8% using 68 loci, with an AUC of 0.5753 and an accuracy of 0.5412. The phenotypic variance explained by PRS of model 3 was 41.7% using 68 loci, with an AUC of 0.8654 and an accuracy of 0.7548. However, using a polygenic risk score model for pancreatic cancer established by UK Biobank to validate this study, the AUC dropped to 0.52. Conclusion: Through genome-wide association analysis, many genes associated to PDAC and ovarian cancer were found, and irrespective of separate or combined analysis, there were overlapping genes or loci reported in this study, such as TNFRSF14、LINC00970、GPATCH1、PSMD14、LINC00954、DDX6P1、LOXL2、AHCYP2, implying existence of overlap. PDAC and ovarian cancer can be combined for analysis. When compared with European population, the allele frequencies of the same locus are very different, and even the risk direction is different. The polygenic risk score model established using the European patient set did not perform well when used on Taiwanese patients. It indicates that pancreatic cancer and ovarian cancer have different risk genes in Taiwanese from that of other populations.