探討復發性頭頸癌患者免疫治療結果和微生物相和腫瘤基因體之間的關係

Abstract

免疫檢查點抑制劑是近年治療癌症的最新突破，其中吉舒達 (Pembrolizumab)不但是復發性頭頸癌患者的處方藥物，更少見的成為跨癌腫處方藥物。免疫檢查點抑制劑固然提升了無惡化存活率和整體存活率，其費用高昂且常有免疫相關的副作用產生，為了能夠有效的預測免疫檢查點抑制劑治療反應，許多研究針對免疫檢查點抑制劑的作用位置免疫檢查點或者針對病人的免疫系統狀態進行多方位的評估。已有研究報導之免疫系統的狀態評估項目包含腫瘤組織中免疫抑制配體表現量、腫瘤組織微環境中T細胞浸潤程度以及腫瘤突變負荷。另外，腸道菌相也因為與全身免疫系統狀態息息相關而成為免疫檢查點抑制劑的治療預測因子研究中重要的一環。此研究透過比較免疫檢查點抑制劑治療有效與無效的復發性頭頸癌病之間的基因體差異與腸道菌相差異來辨識可能的基因體標記以及腸道菌相標記，並利用這些治療有效的基因標記和腸道菌相標記各自建立一套二項式廣義線性模型來觀察該標記的治療預後預測效力，最後，再將兩類不同的標記整合為一套二項式廣義線性模型，並討論個別以及整合後的預後預測效力。基因體標記主要以兩種方法分析由腫瘤組織石蠟切片中所得核糖核酸而得，第一種方法為差異表現基因分析 (Differential expression of RNA-Seq, DESeq)，藉此鑑別治療有效個案中表現量較高的基因作為基因體標記，此分析結果得到112個校正後仍顯著在治療有效與無效個案間有差異表現的基因，最後透過篩選得到了20個已有文獻報導且在治療有效個案中表現顯著較多的基因，依據文獻報導將20個基因分為四大類，腫瘤組織特異性、疾病進展標記、預後標記以及腫瘤調控相關之非編碼核糖核酸。第二種基因體標記則是透過基因組分析工具 (Genome analysis toolkit, GATK)將核糖核酸的定序結果進行變異型判讀 (variant calling)，並且計算造成功能異常的變異在排除種族特有變異 (變異型頻率 >0.01) 在每一千鹼基對中所有的變異數量，得到腫瘤突變負荷的數值，雖然在此研究中所得到的治療有效與無效兩組間腫瘤變異負荷並無顯著差異，但兩組間的腫瘤突變負荷平均值可以觀察到治療無效的個案平均值較高，代表治療無效病人中的腫瘤突變負荷較高。以上兩個透過核糖核酸定序所得到的基因體分析結果皆用於建立最後的二項式廣義線性模型。腸道菌相的探討則透過16S rRNA的基因全長進行定序以及細菌物種的比對，並完成個體間的微生物多樣性、豐富度和治療有效與無效者之間的差異菌分析。差異菌最後分析得四個細菌屬為治療有效者富有的特異標記，分別為伊格爾茲氏菌屬(Eggerthella) 、多瘤胃球菌屬(Ruminococcus) 、顫螺旋菌屬(Oscillibacter) 和Soleaferrea菌數。此四類菌屬皆作為預測因子來建立腸道菌相標記之二項式廣義線性模型。在基因體標記所建立的預測模型中，考量預測因子間的相關性後，最後僅使用疾病進展標記和預後標記兩類作為預測因子，所得的曲線下面積 (Area under curve, AUC)為1.00，依據各項獨立分類與治療結果的相關性加權後總和各分類成為單一獨立變相後的預測結果則有0.9028的AUC。腸道菌相標記的預測模型則因所辨識的四個治療有效組中富含之菌屬以不同組合進行了模型的建立，四個菌屬皆納入預測模型的預測曲線下面積為0.875，而僅使用文獻中已有佐證之多瘤胃球菌屬單一屬建立的預測模型則有0.75的曲線下面積，將四個菌屬總和為一項腸道菌相變相則得到 0.7841 AUC。將上述的兩類基因體標記與腸道菌多瘤胃球菌屬所整合建立的預測模型得到了0.8667。合併以上的基因體標記與腸道菌相標記發現基因體標記有較佳的預測效力，如使用整合的預測模型則可以提昇單獨由腸道菌相標記所建立之預測模型效力。整體而論，此研究雖然僅以有限的個案數進行預測模型的建立以及評估，其預測效力與其他針對各種癌腫的免疫治療預後預測文獻中相較有較佳的預測效力。

Immune checkpoint inhibitors (ICI) are the latest breakthrough in cancer treatment, and Pembrolizumab was prescribed for recurrent head and neck squamous carcinoma (recurrent HNSCC) as well as becoming a prescription available for more than 1 type of cancer. Despite the promising prolonged progression free and overall survival rate, ICI treatments are high in cost and also with a certain degree of side effects. In order to identify patients who will benefit the most from immune checkpoint inhibitors, immune status of a patient is broadly recognized as an indicator with a wide variety of specific prediction targets. Such as, tumor expression level of immune checkpoint inhibiting ligands, tumor microenvironment considering T cell infiltration and tumor mutation burden (TMB). Last but not least, gut microbiome has also gained focus since it is well known to modulate host systemic immune response. In this study, we aim to identify genomic based and gut microbiome-based ICI treatment response predictors and evaluate their individual and integrated prediction power. Genomic based treatment responder biomarkers were identified through 2 main aspects focusing on original HNSCC tumor tissue RNA profile. First, the differentially expressed genes between responders and progressive non-responders. Among the 112 differentially expressed genes with significant p value (FDR adjusted p<0.05), final 20 genes were identified to be expressed higher in responder group and with reference reports of either enriched in cancer, acts as a progression marker or plays a role in prognostic prediction. Secondly, the tumor mutation burden of responder and progressive non-responders were also generated based on RNA sequencing data. Although the difference of 2 groups were not significant, progressive non-responders showed a generally higher TMB compared with responders. Based on previous evidence, significant difference between responder and progressive non-responders based on DESeq were further used as genomic based predictors for the establishment of prediction model. For gut microbiome, 16S rRNA full length DNA sequencing was used to identify bacterial species and comparison of treatment responder and progressive non-responder for the diversity, richness and differentially composed species were conducted. Gut microbiome biomarkers were identified with 4 genus that were more abundant in responders, Eggerthella, Ruminococcus, Oscillibacter and Soleaferrea, while no significant difference was observed between responder and progressive non-responders regarding diversity or richness. Prediction model was established by generalized linear model (GLM) with logit link to not only individually evaluate genomic and gut microbiome biomarker-based prediction, but also to combine the 2 different aspects of biomarkers for a more comprehensive prediction scope. The genomic biomarker based GLM prediction resulted with AUC 1.0000 by including the progression markers and prognostic markers identified through responder and non-responder differentially expressed genes analysis, and an AUC of 0.9028 for GLM based on the same 20 genes but summed by gene expression levels as in 1 preditor. The gut microbiome based GLM prediction resulted in a higher AUC at 0.8750 if all 4 genera were used as individual variables. A lower AUC of 0.7841 was found for a summed relative abundance of the 4 genera. The integrated GLM prediction model using both the singular genomic based predictor with the addition of singular gut microbiome-based predictor each with summed expression level or relative abundance gave an AUC of 0.8667. In sum, genomic biomarkers found in recurrent HNSCC were highly predictable for ICI treatment response. The gnomic biomarker-based model combined with gut microbiome improved the prediction solely established by gut microbiome biomarkers. In general, the prediction models were established and evaluated through limited samples, but with better performance compared with currently reported models which were not aimed for predicting HNSCC patient ICI treatment response alone.