Distinct Immunological Landscapes Characterize Inherited and Sporadic Mismatch Repair Deficient Endometrial Cancer

Abstract

Around 30% of endometrial cancers (EC) are mismatch repair (MMR) deficient, mostly as a consequence of mutations acquired during tumorigenesis, but a significant minority is caused by Lynch syndrome (LS). This inherited cancer predisposition syndrome primes an anti-cancer immune response, even in healthy carriers. We sought to explore the intra-tumoral immunological differences between genetically confirmed LS-associated MMR-deficient (MMRd), sporadic MMR-deficient, and MMR-proficient (MMRp) EC. Endometrial tumors from women with known LS were identified (n = 25). Comparator tumors were recruited prospectively and underwent microsatellite instability (MSI) testing, immunohistochemistry (IHC) for MMR expression and MLH1 methylation testing. Those found to have MLH1 hypermethylation formed the sporadic MMR-deficient group (n = 33). Those found to be mismatch repair proficient and microsatellite stable formed the MMR-proficient group (n = 35). A fully automated monoplex IHC panel was performed on sequential formalin-fixed paraffin-embedded tumor sections to identify CD3+, CD8+, CD45RO+, FoxP3+, and PD-1+ immune cells, and PD-L1 expression by tumor/immune cells. Two independent observers quantified immune marker expression at the tumor center and invasive margin. Mean and overall compartmental T-cell counts generated standard (binary: Low/High) and higher resolution (quaternary: 0-25, 25-50, 50-75, 75-100%) immune scores, which were used as explanatory features in neural network, support vector machine, and discriminant predictive modeling. Overall T-cell counts were significantly different between the three cohorts: CD3+ (p = <0.0001), CD8+ (p = <0.0001), CD45RO+ (<0.0001), FoxP3+ (p = <0.0001), and PD1+ (p = <0.0001), with LS-associated MMR-deficient tumors having highest infiltrations. There were significant differences in CD8+ (p = 0.02), CD45RO+ (p = 0.007), and PD-1+ (p = 0.005) T-cell counts at the invasive margin between LS-associated and sporadic MMR-deficient tumors, but not between sporadic MMR-deficient and MMR-proficient tumors. Predictive modeling could accurately determine MMR status based on CD8+ T-cell counts within the tumor center alone. This study shows that LS-associated and sporadic MMR-deficient EC are distinct immunological entities, which has important implications for treatment and prognosis.

Keywords: Lynch Syndrome; PD-1; endometrial cancer; immune checkpoint; immune microenvironment; microsatellite instability; mismatch repair; predictive modeling.

Figure 1

Study flow schema. EEC, Endometrioid Endometrial Cancer; CC, Clear Cell Endometrial Cancer; CCS, Carcinosarcoma of the uterus; G, Grade; NK, Not known; MMRd, mismatch repair deficient; MMRp, mismatch repair proficient.

Figure 2

The components of immune scores I1:I10 and Q1:Q10. Each immune score is defined by the density of a set panel of immune markers (CD3, CD8, CD45RO, FoxP3, PD-1) at the tumor center (CT) +/– invasive margin (IM) relative to the corresponding densities across all tumors. The I-scores define each lymphocyte population using the median threshold methodology (0:Low, 1:High), while the Q-scores are defined using quartile ranges (0: 0–25%, 1: 25–50%, 2: 50–75%, 3: 75–100%). Therefore, for any given I-score, the corresponding Q-score presents a higher resolution scoring system. Highlighted boxes indicate inclusion of that particular lymphocyte population within an immune score. I-score, binary immune score; Q-score, quaternary immune score; CT, Tumor Center; IM, Invasive margin.

Figure 3

Representative immunohistochemistry images of immune densities across the three molecular cohorts at the Tumor Center (CT) and Invasive Margin (IM).

Figure 4

Immune cell counts as per their molecular profile. LS-associated MMRd, Lynch Syndrome-associated mismatch repair deficient; Sporadic MMRd, Sporadic mismatch repair deficient; MMRp, mismatch repair proficient.

Figure 5

Heatmap outlining the clustering of the molecular groups by immune score. This figure clearly illustrates the broad immune profile of sporadic MMRd loss ECs, as they are seen to infiltrate around both MMRp and LS-associated MMRd groups. LS-associated MMRd, Lynch Syndrome-associated mismatch repair deficient; Sporadic MMRd, Sporadic mismatch repair deficient; MMRp, mismatch repair proficient.

Figure 6

Neural network modeling for two-class response variables of mismatch repair deficient vs. mismatch repair proficient. Models were trained on a 50 observation random subset and tested on a 20 observation validation data subset as described in methods and described further in Supplementary Appendixes 6–9. (A) Immune score predictive accuracy from the tumor core compartment only generated from neural network modeling. Receiver operating characteristic (ROC) curves are shown with notable AUCs of 0.899 (I1), 0.846 (Q2), 0.965 (Q7), and 0.882 (Q8) depicted with asterisks. (B) Neural network models constructed using feature counts instead of immune scores. Select ROC curves are shown with AUCs of 0.913 (CD8 CT); 0.786 (CD3 CT, CD3 IM); 0.784 (CD3 CT, CD8 CT, CD8 IM).