Pathomic Predictors of Prostate Cancer Progression

摘要 最近的研究表明，在美国，前列腺癌被过度发现和过度治疗，导致 严重的发病率和财务成本。这些问题是敏感度和特异度差的结果。 血清前列腺特异性抗原(PSA)作为筛查工具，导致不必要的活检，发现细小和 主要是惰性前列腺癌。虽然许多前列腺癌应该积极治疗 监测，围绕可用临床侵袭性工具的不确定性(如PSA、Gleason评分 和临床阶段)往往会驱使患者和医生进行治疗。改善预测的尝试 使用候选生物标记物，主要是从大片癌症的基因组分析中发现的，已经 几乎没有成功，而且可用的分子工具充其量只能提供适度的预测。 基因组驱动焦点的另一种选择是分子事件的组合，在分子事件的影响下 肿瘤微环境，驱动肿瘤的分子进化和进展。因此，分析 病理数据中可检测到的肿瘤特征，如表达亚型的异质性，数量 间质、微环境异质性的程度、免疫渗透的程度或缺氧的程度可能 最终导致更好的患者分层。我们的建议从根本上讲是围绕最关键的 作为临床决策基础的早期前列腺癌的临床问题：我们能否确定 蛋白质组、成像和/或微环境特征区分那些侵袭性癌症 进展是否会导致良性癌症的危害，这可以通过观察等待来安全地监测？ 研究筛查发现的早期前列腺癌的异质性与患前列腺癌的可能性之间的联系。 进展，我们将询问一组225名前列腺摘除患者的回顾性研究。在目标1中，我们将使用GE的 复杂的免疫病理平台(细胞潜水)，在细胞和亚细胞中分析50多种蛋白质 水平以及定义微环境的基质成分，这些微环境存在于该基质中。在……里面 目标2，我们将专注于单细胞水平的数据，并系统地提取不同细胞的患病率 在这些肿瘤中发现的亚型。细胞将沿着传统的轴线进行分类(例如，上皮细胞、CD4T细胞)。在……里面 此外，我们将使用分子和结构特征来定义新的亚型。关联的功能 细胞类型(例如，存在、流行、多样性)将单独使用，并与格里森结合使用 分级以区分可能进展的侵袭性肿瘤患者。目标3将重点放在 社区和区域分析，特别是关于开发提取肿瘤的方法的分析 已证明与进展有关的微环境特征(缺氧、间质 反应性、免疫细胞构型)。使用一组不同的这些功能，以及深度学习技术 在原始图像上，我们将开发区分侵袭性和良性肿瘤的分类器。最后，在目标4中 我们将在大型队列中验证分类器。

Abstract Recent studies suggest that in the U.S. prostate cancer is over-detected and over-treated resulting in significant morbidity and financial costs. These problems are the product of poor sensitivity and specificity serum Prostate Specific Antigen (PSA) as a screening tool, leading to unnecessary biopsies that find small and predominantly indolent prostate tumors. While many prostate cancers should be managed with active surveillance, uncertainties surrounding available clinical tools of aggressiveness (such as PSA, Gleason score and clinical stage) will often drive patients and physicians to treatment. Attempts to improve prognostication using candidate biomarkers, mostly discovered from genomic analyses of large pieces of cancers, have had few successes, and available molecular tools provide only modest prediction, at best. An alternative to the genomic driver focus is that a combination of molecular events, under the influence of the tumor microenvironment, drive tumor’s molecular evolution and progression. Consequently, analysis of tumor characteristics detectable in pathomic data, such as heterogeneity of expression subtypes, amount of stroma, extent of microenvironmental heterogeneity, extent of immune infiltration, or extent of hypoxia, may ultimately lead to better patient stratification. Our proposal fundamentally centers around the most critical clinical question in early prostate cancer that is the basis for clinical decision making: Can we identify proteomic, imaging, and/or microenvironment features that distinguish those aggressive cancers that will progress to cause harm from benign cancers that can be safely monitored by watchful waiting? To examine the links between the heterogeneity of early, screen-detected prostate cancers and likelihood of progression, we will interrogate a retrospective set of 225 prostatectomy patients. In Aim 1, we will use GE’s hyperplexed immune-pathology platform (Cell DIVE) to profile over 50 proteins at the cellular and subcellular level along with matrix components that define the microenvironments with the cells present in this matrix. In Aim 2, we will focus on single-cell level data and systematically extract the prevalence of the diverse cell subtypes found within these tumors. Cells will be typed along traditional axes (e.g. epithelial, CD4+ T-cells). In addition, we will use molecular and structural characteristics to define novel subtypes. Features associated with cell types (e.g. existence, prevalence, diversity) will be used alone and in combination with Gleason grading to distinguish patients with aggressive tumors that are likely to progress. Aim 3 will focus on neighborhood and regional analyses, particularly on developing approaches to extract tumor microenvironmental characteristics that have demonstrated linkages to progression (hypoxia, stromal reactivity, immune cell patterning). Using a diverse set of these features, alongside deep learning techniques on primary images, we will develop classifiers distinguishing aggressive and benign tumors. Finally, in Aim 4 we will validate classifiers in large cohorts.