PO.BCS01.01 · 生物信息与计算
利用泛癌蛋白质组学图谱界定跨实体瘤的生存空间生物标志物
Defining spatial biomarkers of survival across solid tumors using a pan-cancer proteomics atlas
作者与单位 Authors & Affiliations
摘要 Abstract
中文摘要
肿瘤微环境（TME）的空间结构支配着癌症进展和治疗应答，然而将多细胞组织结构与患者生存联系起来的泛癌分析仍不明确。作为PROSPECTS（空间图谱与治疗靶点的泛癌重建）计划的一部分，我们构建了一个大规模空间蛋白质组学图谱，涵盖六种主要恶性肿瘤的415例患者：头颈部鳞状细胞癌（HPV阳性和HPV阴性）、肺癌（NSCLC和LUAD）、三阴性乳腺癌、高级别浆液性卵巢癌、结直肠腺癌和肝细胞癌（HCC）。我们的数据集包括超过1,000个组织芯和超过440万个单细胞，各适应证之间分布均衡：HNSCC（1,218,385个细胞，145例患者）、肺（906,537个细胞；63例患者）、乳腺（765,232个细胞；60例患者）、卵巢（590,945个细胞；53例患者）、结直肠（511,610个细胞；37例患者）和肝（474,237个细胞；57例患者）。使用30重抗体Phenocycler，我们以单细胞分辨率对15种主要细胞类型进行了空间定位。我们在AstroSuite（Stratica Biosciences）中开发了新一代多尺度计算流程，整合TACIT和Constellation算法，在细胞状态、细胞间距离以及包括成对、三联体和四联体细胞邻域在内的高阶基序等层面上量化TME结构。这些深度空间指标在校正临床协变量后与总生存期相关联。我们识别出TME结构中显著的异质性，每种癌症类型都表现出独特的预后结构。在TNBC中，肿瘤细胞-中性粒细胞相互作用成为最强的不良预后特征之一（P=0.00001）。在肺癌中，血管-免疫和基质-肿瘤界面是关键，尤其是血管内皮细胞-巨噬细胞邻接（P=0.0003）和成纤维细胞-肿瘤细胞相互作用（P=0.002）。尽管存在这些癌症特异性模式，PROSPECTS揭示了保守的空间特征。一种重现性的成纤维细胞-肿瘤细胞界面基序在多种恶性肿瘤中出现，尽管基质驱动因素各异：肌成纤维细胞-Treg相互作用在结直肠癌中具有预后意义（P=0.002），而成纤维细胞-Treg相互作用在卵巢癌中预测结局（P=0.002）。在HCC中，免疫-免疫相互作用（如Treg-B细胞串扰，P=0.002）与生存显著相关。复杂的空间语法在预后方面被证明优于简单的成对相互作用，凸显了高阶TME建模的价值。这项工作建立了首个将TME组织结构与临床结局联系起来的泛癌空间蛋白质组学比较，揭示了空间解析的细胞相互作用网络构成了一类新的临床可操作生物标志物。
查看英文原文 English abstract
The spatial architecture of the tumor microenvironment (TME) governs cancer progression and therapeutic response, yet pan-cancer analyses linking multicellular organization to patient survival remain unclear. As part of the PROSPECTS (Pancancer Reconstruction Of Spatial Profiles and Therapeutic TargETs) Initiative, we assembled a large-scale spatial proteomic atlas comprising 415 patients across six major malignancies: Head and Neck Squamous Cell Carcinoma (HPV-positive and HPV-negative), Lung Cancer (NSCLC and LUAD), Triple-Negative Breast Cancer, High-Grade Serous Ovarian Cancer, Colorectal Adenocarcinoma, and Hepatocellular Carcinoma (HCC). Our dataset includes over 1,000 tissue cores and over 4.4 million single cells with balanced representation across indications: HNSCC (1,218,385 cells, 145 patients), Lung (906,537 cells; 63 patients), Breast (765,232 cells; 60 patients), Ovarian (590,945 cells; 53 patients), Colorectal (511,610 cells; 37 patients), and Liver (474,237 cells; 57 patients). Using a 30-plex antibody Phenocycler, we spatially mapped 15 major cell types at single-cell resolution. We developed a next-generation multi-scale computational pipeline within AstroSuite (Stratica Biosciences), integrating TACIT and Constellation algorithms to quantify TME architecture at the levels of cell state, intercellular distance, and higher-order motifs including pairwise, triplet, and quartet cellular neighborhoods. These deep spatial metrics were linked to overall survival with adjustment for clinical covariates. We identified marked heterogeneity in TME structure, with each cancer type exhibiting distinct prognostic architectures. In TNBC, Tumor Cell-Neutrophil interactions emerged as one of the strongest adverse prognostic features (P=0.00001). In Lung cancers, vascular-immune and stromal-tumor interfaces were key, particularly vascular endothelial cell-Macrophage adjacency (P=0.0003) and Fibroblast-Tumor Cell interactions (P=0.002). Despite these cancer-specific patterns, PROSPECTS uncovered conserved spatial signatures. A recurrent fibroblast-tumor cell interface motif appeared across malignancies, though stromal drivers varied: Myofibroblast-Treg interactions were prognostic in Colorectal cancer (P=0.002), while Fibroblast-Treg interactions predicted outcome in Ovarian cancer (P=0.002). In HCC, immune-to-immune interactions such as Treg-B Cell crosstalk (P=0.002) were significantly associated with survival. Complex spatial syntax proved superior to simple pairwise interactions for prognosis, highlighting the value of higher-order TME modeling. This work establishes the first pan-cancer spatial proteomic comparison linking TME organization to clinical outcome, revealing that spatially resolved cellular interaction networks constitute a new class of clinically actionable biomarkers.
利益披露 Disclosure
K. Huynh, None..
J. Reyna, None..
B. Matuck, None.