基于机器学习的结直肠癌肝转移免疫微环境分型及预后预测研究

Colorectal cancer has a high incidence in China with liver metastasis as the most common metastasis pattern. In previous studies, we found that there were significant differences in immune microenvironment among paracancerous, primary and metastatic lesions in colorectal cancer patients with liver metastasis. Recent studies have shown that the degree and phenotype of immune infiltration in tumor microenvironment are not only important for the resolution of tumor heterogeneity and therapeutic resistance, but also can predict the efficacy of immunotherapy and predict new druggable targets. In order to accurately quantify the cell composition in highly heterogeneous immune microenvironment such as colorectal cancer, it is necessary to develop new bioinformatics methods and biomarkers for immune phenotyping and predicting prognosis. In this project, we will at first utilize microarray sequencing and flow cytometry to extract the cell expression characteristics from the collected paired surgical samples of colorectal cancer patients with liver metastasis. Machine learning algorithms will be applied to select and generate gene expression signatures of immune and stroma cell composition in tumor lesions by integrating publicly available database and our own measurements. Bioinformatic methods such as deconvolution method and gene set enrichment analysis will be applied to solve the inverse problem that is to infer specific cell proportions from expression data from cell mixtures using expression signature matrices. We aim to explore and establish the immune subtype classification of colorectal cancer with liver metastases. We plan to utilize the public database with clinical information, gene expression signatures and immune subtype characteristics to model the prognosis of colon cancer patients with liver metastases. Independent follow-up information collected in this study will be used to validate prognostic model. The results obtained in this study are expected to predict the prognosis of colorectal cancer by combining molecular markers and immune phenotyping, and thus lay theoretical foundation for the prediction of relevant immunotherapy targets.

结直肠癌是我国高发恶性肿瘤，肝转移是其最常见的转移形式，目前缺乏有效的预后标志物和治疗手段。研究结直肠癌肝转移患者的癌旁、原发灶和转移灶之间显著的免疫微环境差异不仅对解决肿瘤异质性和治疗抵抗性至关重要，还能预测免疫疗法疗效和预测新靶点。本项目拟结合高通量基因芯片和流式细胞技术准确测定搜集的成对手术样本中不同部位基因表达谱和肿瘤微环境中的免疫及间质细胞组成，整合公开数据库和测量结果采用机器学习算法提取细胞的基因表达特征，针对这一反演问题采用反卷积算法和基因集富集分析开发生物信息学工具估算不同细胞亚群的绝对比例，探讨并建立结直肠癌肝转移的免疫亚型分类。在此基础上，利用公开数据库的临床信息结合基因表达谱和免疫分型特征对结肠癌肝转移进行预后建模，利用研究搜集的随访队列检验预后预测模型。本研究所获得的成果可望通过特征性的分子标志和免疫分型判断结直肠癌预后，并为预测相关免疫治疗靶点奠定基础。

通过研究结直肠癌不同风险病人之间的微环境差异以及肝转移灶的免疫特征，可以更加全面细致地了解其复发和转移的原因。因此，建立更敏感的预后预测模型和探索更有效的治疗药物具有重要的临床意义。同时，通过药物重定位研究可以挖掘已有药物的治疗价值，加速新药研发。本研究通过项目搜集及整合GEO数据库中结直肠癌肝转移数据，通过差异表达分析，识别肝转移相关差异基因，通过生存分析进一步筛选具有预后价值的差异基因。使用机器学习算法，分别以总生存期和无进展间隔为临床终点构建了分别包含10基因（AUC = 0.693）与11基因（AUC = 0.722）的预后模型（命名为MAOS与MAPS）。在独立数据集GSE39582和GSE17536中对模型进行了验证，表现优于其他预后模型，并具有独立预测价值。利用CIBERSORTx算法通过基因表达谱量化了结直肠癌原发灶与肝转移灶肿瘤微环境中22类免疫细胞的浸润情况，构建了肿瘤微环境细胞相互作用网络。研究发现，MAOS与MAPS评分识别的高风险病人具有类似的肿瘤微环境特征，表现为低免疫活性、高免疫抑制和CD8+ T细胞耗竭状态。MAOS评分与免疫治疗响应显著相关（P = 0.003）。最后，基于与预后特征相关的信号通路，我们利用药物敏感性数据库进行了全面的计算机分析，以确定潜在的治疗CRLM的药物，并通过MTT和Transwell试验验证其体外效力。在候选药物中，Obatoclax在CRLM小鼠模型中表现出相当大的抑制肝转移的作用。本研究建立的预后预测模型有望对结直肠癌病人进行风险分类和肿瘤微环境特征评估。利用药物敏感性数据库和机器学习算法进行药物重定位研究，针对高风险病人筛选潜在治疗药物，并通过体外和体内实验验证了结直肠癌肝转移治疗药物，为结直肠癌转移病人预后诊断及治疗指导提供一种新型研究思路。

内容获取失败，请点击重试