Graph theoretic spatial heterogeneity analysis of multiplexed immunofluorescence images enables quantitative differentiation of HGSC precursor lesions in the fallopian tube

Background: p53 signatures, serous tubal intraepithelial lesions (STILs), and serous tubal intraepithelial carcinomas (STICs) represent the precursor spectrum of high-grade serous carcinoma (HGSC), with STICs identified in 50-60% of HGSC cases. These lesions harbor TP53 mutations and exhibit similar genomic alterations to invasive HGSC, yet only a subset of these progress to malignancy. Studies suggest a close to a decade long latency period between STIC formation and invasive disease, creating a critical window for intervention. However, the factors determining malignant transformation remain poorly understood and quantified. Critically, the immune microenvironment of these precursors remains quantitatively uncharacterized. Specifically, we lack quantitative metrics associating lymphocytes, immunosuppressive cells, and immune checkpoint expressions in p53 signatures, STIL, and STIC lesions that are concordant with HGSC risk. Quantitative understanding of whether immune escape mechanisms are established early or develop during progression could identify microenvironment biomarkers that predict risk of individual lesions becoming invasive. Method and Results: Toward this goal, we have developed a spatial heterogeneity analysis (SHEAN) framework for multiplexed immunofluorescence (mIF) that utilizes graph-theoretic representations of fallopian precursor tissue samples to identify quantitative spatial immune signatures that (1) characterize and distinguish normal epithelium (Norm), p53 signatures, STICs, and HGSCs from each other; (2) are sensitive to TP53 mutation status; and (3) can differentiate STICs based on their flat (FLAT) and budding, loosely adherent or detached (BLAD) status. These statistically significant features, confirmed at the image level, include signatures associated with degree of infiltration into, and interaction between T-lymphocytes, M2 polarized macrophages, and epithelial cells. SHEAN also achieved a cross-validated area under the ROC curve (AUROC) of 87.4% in discriminating Norm, p53 signatures, STICs, and HGSCs based on a bootstrapped model that included 53 normal, 68 p53 signatures, 73 STIC, and 32 HGSC regions and used an XGBoost classifier. Conclusions: SHEAN allows the quantification of immune spatial metrics of the HGSC precursor microenvironment, providing an interpretable, spatially informed model capable of discriminating between lesion categories and enabling the reliable stratification of HGSC risk.