Jie Zheng, Qian Yang, Haoyu Liu, Huiling Zhao, Shuangyuan Wang, Yi Liu, Xueyan Wu, Yilan Ding, Hui Ying, Yifan Ye, Xi Huang, Lei Ye, Ruizhi Zheng, Hong Lin, Mian Li, Tiange Wang, Zhiyun Zhao, Min Xu, Yi Duan, Hao Guo, Zhongshang Yuan, Philip Haycock, George Davey Smith, Richard M. Martin, Guang Ning, Fang Hu, Weiqing Wang, Tom R. Gaunt, Jieli Lu, Yufang Bi
Single-cell expression quantitative trait loci data offer promising opportunities to inform immune-related drug development in cancer. However, pleiotropy can complicate causal inference. We introduce MR-DEG, a framework that integrates Mendelian randomization (MR) and differential expressed gene (DEG) to strengthen causal inference. Using eight conventional MR and colocalization methods, we estimated effects of 11 021 dynamic gene expression profiles during CD4+ T cell activation on the risk of six cancer types. This identified 1000 gene-cancer pairs with putative effects (https://www.omicsharbour.com/sc-eqtl-mr). Of these 1000 pairs, 517 involved 205 unique genes that were differentially expressed in relevant cancer tissues based on single-cell RNA-sequencing data. Of these 517 pairs, 265 were classified as likely causal using the conventional MR methods. After applying MR-DEG to the remaining 252 potentially pleiotropic pairs, an additional 89 were classified as likely causal. Sixty-four and 391 of the 1000 original pairs exhibited time- and non-time dependent effects on cancer risk, respectively. Integrating the 1000 gene-cancer pairs of MR findings and clinical trial evidence, we identified 200 pairs corresponding to 33 unique genes that encode drug targets under clinical investigation. These results demonstrate how combining genetic, transcriptomic and clinical trial evidence can reduce pleiotropic bias, and prioritize immune-related drug targets for cancer prevention.