Spatial-BrTME: Multicellular spatial dynamics of immunotherapy response in breast cancer

免疫疗法给癌症治疗带来了革命性的变化，但它在乳腺癌中的作用尚不清楚。为了让乳腺癌患者受益，我们必须理解为什么一些人有反应，而另一些人没有，并确定一个实用的生物标记物来区分他们。免疫治疗依赖于肿瘤微环境(TME)的空间组织，因为它以T细胞相互作用为目标，但TME组织的原理鲜为人知。免疫疗法如何在治疗过程中重塑这种结构也是未知的，但可能解释了为什么反应不同。为了了解乳腺癌的免疫治疗反应，并发现一个可靠的鉴别生物标志物，我建议通过组织的高度多元化成像来原位解剖多细胞TME结构。成像质量细胞术(IMC)使用飞行时间质谱仪来定位组织中亚细胞分辨率的44种蛋白质的表达。使用IMC，我们将分析数百名乳腺癌患者的数千个样本，这些样本被招募到免疫治疗的随机试验中，其中收集了纵向样本(基线、治疗中和治疗后)。使用自动图像分析、图论和空间统计，我们将识别在肿瘤中复发的多细胞配置，并绘制这些配置在治疗有效和无效的情况下如何演变。然而，从这些分析中得出的结果与常规的临床病理学是脱节的，因为在那种情况下不可能进行同等的分析，这阻碍了翻译。我们将利用这些试验积累的大量数字病理学资源，开发新的机器学习工具，将在高维空间中学习的特征转移到常规的数字病理染色，从而弥合这一差距。总之，这个项目将阐明免疫治疗反应的病理学基础，并朝着增强病理学的新临床学科迈出第一步。

Immunotherapy has revolutionised cancer treatment, but its role in breast cancer is unclear. For breast cancer patients to benefit, we must understand why some respond whereas others don't, and identify a pragmatic biomarker to distinguish between them. Immunotherapy depends on spatial organisation of the tumour microenvironment (TME) because it targets T cell interactions, but the principles of TME organisation are poorly understood. How immunotherapy remodels this structure during treatment is also unknown but may explain why responses differ. To understand immunotherapy response in breast cancer, and to uncover a reliable discriminatory biomarker, I propose to dissect multicellular TME structure in situ by highly multiplexed imaging of tissues. Imaging mass cytometry (IMC) uses time-of-flight mass spectrometry to localise the expression of 44 proteins at subcellular resolution in tissues. Using IMC, we will analyse thousands of samples from hundreds of breast cancer patients recruited to randomised trials of immunotherapy where longitudinal samples (at baseline, on-treatment, and post-treatment) have been collected. Using automated image analysis, graph theory and spatial statistics we will identify multicellular configurations that recur across tumours and chart how these evolve under therapy in responders versus non-responders. Findings arising from these analyses are disconnected from routine clinical pathology however, because equivalent assays are not possible in that setting, frustrating translation. We will bridge this gap by using the large digital pathology resource accrued for these trials to develop novel machine-learning tools to transfer features learned in high-dimensional space to routine digital pathology stains. Together, this programme will elucidate the pathologic basis of immunotherapy response and take first steps toward a new clinical discipline of augmented pathology.