The tumor microenvironment (TME) is composed of malignant and non-malignant cells, each contributing to spatial intratumoral heterogeneity (ITH) and heterocellular communication altering the composition and architecture of the TME. A high degree of ITH is correlated to metastatic progression and therapeutic response. Previous studies investigating spatial ITH have been limited due to a steep trade-off between cellular resolution, spatial context, and dimensionality of biomarkers. A recent explosion of multi to hyperplexed imaging modalities (e.g., fluorescence imaging, mass-spec imaging) enable the quanti?cation of greater than 7 and up to > 100 biomarkers through sequentially multiplexed imaging of 2 to 3 biomarkers using iterative cycles of label-image-dye inactivation. The generation of this new type of data poses both unique opportunities and challenges. There are no state-of-the-art methods for harnessing the complexity of spatial data to infer tumor biology with a high dimensionality of biomarkers. In this project, we will probe the spatial complexity of a TME in hyperplexed immunofluorescence (HxIF) based spatial proteomics colorectal carcinoma (CRC) data (51 biomarkers + DAPI, 356 patient samples) to elucidate the heterocellular communication networks promoting spatial ITH through cellular phenotyping, microdomain extraction, and network biology inference algorithms. We will demonstrate the applicability of our algorithms to cancer types beyond CRC with multiplexed immunofluorescence breast cancer tissue samples In Aim 1, we will continue to develop unsupervised learning algorithm for cellular phenotypic heterogeneity (LEAPH) to identify specialized, rare, and transitional cell populations. Initial results applying LEAPH on the HxIF CRC data have revealed cellular heterogeneity patterns consistent with CRC literature (STEM cell differentiation, immune evasion, macrophage evolution). We will incorporate machine learning- based methods into LEAPH to measure spatial distribution patterns of each phenotype and correlate them with CRC progression (e.g., recurrence).
In Aim 2, we will quantify spatial ITH in greater detail by identifying differentially expressed pair- or group-wise spatial relationships based on outcome data (e.g., recurrence vs no-recurrence within 5 years) to reveal phenotypic domains, microdomains, with prognostic potential. We expect improvement of prognostic power with pair- or group-wise spatial interactions in comparison to the single-phenotype based spatial ITH characterization of Aim 1.
In Aim 3, we will dissect the microdomain- specific heterocellular communication dynamics with causal inference network models. We expect to identify emergent signaling networks conferring malignant phenotypes, such as known features from CRC consensus molecular subtypes. The algorithms constructed in this project will be implemented and disseminated through the Tumor Heterogeneity Research Interactive Visualization Environment (THRIVE), an open source tool to assist cancer researchers in interactive hypotheses testing and guiding the design of therapeutic strategies.
The recent explosion of next-generation, high-content, high-throughput spatial imaging technologies for intact tissues measuring protein expressions, DNA and RNA probes has attracted the interest of NIH and other international agencies in funding precision medicine efforts (e.g., HTAN, IOTN, HuBMAP, HCA, HPA). The overarching goal of this proposed project is to probe the spatial complexity of the tumor microenvironment in hyperplexed image datasets of tumor samples to elucidate the heterocellular communication networks promoting spatial ITH and related to disease progression through cellular phenotyping, microdomain extraction, and spatial network biology inference algorithms. The algorithms will be disseminated through the Tumor Heterogeneity Research Interactive Visualization Environment (THRIVE), an open source tool to assist cancer researchers in interactive hypotheses testing and guiding the rational design of therapeutic strategies.
