Identifying cell-type-resolved gene expression changes in pancreatic ductal adenocarcinoma from bulk RNA-seq data

胰腺导管腺癌（Pancreatic ductal adencarcinoma， PDAC）含有少量但数量可变的肿瘤细胞。它们被嵌入肿瘤微环境（TME）中，除了肿瘤细胞外，肿瘤微环境还含有多种具有促抗肿瘤功能的基质细胞和免疫细胞。肿瘤细胞的低数量混淆了从大量转录组学数据中识别肿瘤特异性基因表达亚型和差异表达基因。然而，这些大量的转录组学数据也包含了关于TME细胞基因表达模式的信息。揭示这些额外的信息可能有助于更好地理解肿瘤细胞与其TME之间复杂的相互作用，从而促进免疫系统逃避和对治疗的抵抗。计算方法能够从大量转录组学数据中解剖肿瘤以及基质和免疫细胞基因表达模式。然而，大多数已建立的方法只对一小部分特征基因起作用，并且/或者需要将数据上传到网络服务器。我们最近开发了一种新的方法，可以在转录组范围内操作。它基于概率非负矩阵分解方法，允许将基因表达变化分配给细胞类型。我们计划将这种方法应用于CRU的PDAC RNA-seq数据，以a)推断肿瘤细胞、基质细胞和免疫细胞的丰度；b)识别依赖于临床相关终点的细胞类型分解基因表达变化，如生存，或基于生物标志物的分组，如KRAS中功能获得性突变的存在，这存在于大多数pdac中。我们利用来自pdac的单细胞RNA-seq数据，获得与该CRU中子项目相关的细胞类型的合适的全转录组细胞类型特异性基因表达模式，例如CD8+细胞毒性T细胞或炎性癌相关成纤维细胞。这些细胞类型特异性基因表达模式作为种子来推断和统计评估细胞类型解决的基因表达变化。通过这种方式，我们希望揭示肿瘤细胞与其TME之间复杂的相互作用，这将有助于更好地了解支持TME的肿瘤，并为这种致命疾病的新治疗提供可能的途径。

Pancreatic ductal adenocarcinoma (PDAC) contain a low but variable number of tumor cells. These are embedded in a tumor microenvironment (TME) that contains in addition to the tumor cells a variety of stromal- and and immune cells with bot pro- anti-tumorigenic functions. The low number of tumor cells confounds the the identification of tumor-specific gene expression subtypes and differentially expressed genes from bulk transcriptomic data. However, such bulk transcriptomic data also harbors information about the gene expression patterns of the cells of the TME. Uncovering this additional information may lead to better understanding of the complex interplay between tumor cells and their TME which facilitates immune system evasion and the resistance to treatment. Computational methods enable the dissection of tumor as well as stromal- and immune cell gene expression patterns from bulk transcriptomic data. However, most established method operate on a small subset of signature genes and/or require uploading the data to a webserver. We recently developed a novel method that operates transcriptome wide. It is based on a probabilistic non-negative matrix factorization approach, which allows to assign gene expression changes to cell types. We plan to apply this approach to the CRU’s PDAC RNA-seq data to a) infer the abundances of tumor-, stromal-, and immune cells; b) identify cell-type-resolved gene expression changes dependent on clinically relevant endpoints, such as survival, or grouping based on biomarkers, such as the presence of a gain-of-function mutations in KRAS, which are present in most PDACs. We leverage on single cell RNA-seq data from PDACs to obtain suitable full transcriptome cell-type-specific gene expression patterns for cell types relevant for the subprojects in this CRU, e.g. CD8+ cytotoxic T cells or inflammatory cancer associated fibroblasts. These cell-type-specific gene expression patterns serve as seed to infer and statistically assess cell-type-resolved gene expression changes. In this way we hope to shed light into the complex interplay between tumor cells and their TME, which should lead to a better understanding of the tumor supporting TME and possible avenues for novel treatments of this deadly disease.