High-Efficiency Cell-Type Proteomics Strategy Deciphers Cholangiocarcinoma Fibrosis-Associated Pathological Heterogeneity

Cholangiocarcinoma (CCA) is the second most common primary liver cancer and is characterized by huge heterogeneity, difficult diagnosis, and poor prognosis. Fibrosis-associated heterogeneity in CCA serves as an indicator of the malignant progression of cancer; however, a precise approach to deciphering fibrosis heterogeneity is still lacking. Typically, the tissue proteome is profiled by analyzing bulk tissues, which gives average results of different cell types, especially for CCA tissues in which cancer cells occupy a very small proportion. Laser microdissection (LMD) can precisely dissect CCA cell clusters, but the required manual, time-consuming annotation limits its efficiency. Herein, we develop π-CellSeg-CCA, a pathological image analysis algorithm based on Mask R-CNN and ResNet-18, to enable automated annotation of CCA and normal bile duct regions for LMD and achieve an enhanced recognition accuracy of ∼90%. Driven by π-CellSeg-CCA, we develop a new strategy by integrating a machine learning algorithm, LMD, simple and integrated spintip-based proteomics technology (SISPROT), and high-sensitivity mass spectrometry to decipher CCA fibrosis-associated pathological heterogeneity. We identify over 8000 proteins, including marker proteins specifically expressed in CCA from only 1 mm² samples. A protein specifically upregulated in fibrosis CCA, MUC16, is further investigated to reveal its association with worse prognosis and its contribution to the progression of CCA. We expect that the algorithm-assisted cell-type proteomics strategy is promising for studying the tumor microenvironment with limited clinical materials.