Molecular Subtype Not Immune Response Drives Outcomes in Endometrial Carcinoma Free

The recent success of immunotherapy has prompted a need to better understand the features of cancers that impart susceptibility to immune recognition and control. Among the best evidence for immunologic control is the phenomenon of tumor-infiltrating lymphocytes (TILs), which are strongly associated with patient survival in almost all human cancers (1–4). Moreover, the presence of TIL is associated, albeit imperfectly, with clinical response to immunotherapies targeting the CTLA-4 and PD-1 pathways, referred to as “checkpoint blockade” (5–7). TIL responses generally show a positive association with mutation load, presumably due to an increased number of mutated antigens (neoantigens) available for T-cell recognition (8, 9). In addition, DNA damage can promote TIL responses by activating mechanisms such as the cGAS/STING pathway (10), leading to an immunologically permissive tumor microenvironment. Conversely, TIL responses can be inhibited by oncogenic signaling pathways, including the Wnt/β-catenin and PI3K signaling pathways (11–14). For a given patient, the magnitude of the TIL response likely reflects a balance of these factors, which in turn strongly influences the patient's likelihood of survival and response to checkpoint blockade. A better understanding of the interplay between these factors is required for the development of cancer immunotherapies that benefit a broader spectrum of patients.

Endometrial cancer provides a unique opportunity to study the influence of the tumor genome on antitumor immunity while also presenting an urgent need for improved therapies. Endometrial cancer is the most common gynecologic malignancy in North America, and the fourth most common cancer in women overall, with 150,000 new cases in Europe and North America each year and an increasing trend in incidence and mortality (15, 16). A woman diagnosed with endometrial cancer first undergoes surgical staging, after which histomorphologic classification and risk group stratification directs further adjuvant therapies or observation only. While the majority of women with endometrial cancer have early-stage disease and excellent outcomes, those with metastases or aggressive subtypes typically experience a modest and/or short-lived response to conventional therapies and frequently succumb to their disease (16).

Endometrial cancer pathologic classification and risk stratification has advanced with the recent development of novel molecular classification tools. These tools are objective, biologically based, and clinically relevant, with implications for prognosis, treatment prediction, and genetic heritability. The Cancer Genome Atlas (TCGA) characterized the transcriptional, proteomic, and genomic landscape of over 370 serous and endometrioid cancers and identified 4 genomic subgroups with distinct prognostic outcomes (17). From this, pared down, pragmatic classifiers have emerged, that can use standard formalin-fixed paraffin embedded (FFPE) tissues obtained from hysterectomy (18–22) or diagnostic biopsies (22, 23). ProMisE (Proactive Molecular Risk Classifier in Endometrial cancer) has been developed (18), confirmed (19) and validated (22) according to the Institute of Medicine Guidelines for the development of new biomarker tests (24). ProMisE stratifies endometrial cancer into 4 groups: (i) a mismatch repair–deficient subtype (MMRd, analogous to TCGA and TransPORTEC “MSI”); (ii) a subtype with mutations in the exonuclease domain of polymerase epsilon, which is responsible for DNA proofreading repair and characterized by excellent outcomes (POLE, analogous to TCGA “ultramutated” and “POLE-mutant” TransPORTEC subgroups (20–22, 25); (iii) a poor-prognosis subtype characterized by p53 abnormalities [p53abn, analogous to TCGA “copy-number high” and TransPORTEC “p53-mutant” subgroups (17, 20, 21)]; and (iv) an intermediate prognosis subtype, with wild-type p53 and no specific mutational profile (p53wt, analogous to TCGA “copy-number low,” TransPORTEC “NSMP” or no specific molecular profile; refs. 18–22).

The molecular features of these subtypes influence the immune landscape of endometrial cancer. Owing to intrinsic DNA repair deficiencies, the POLE and MMRd subtypes give rise to ultramutated (mean = 232 × 10⁻⁶ mutations/Mb) and hypermutated (18 × 10⁻⁶ mutations/Mb) tumors, respectively (17). Accordingly, these subtypes have been associated with high neoantigen loads and robust TIL responses (9, 26–29). Given the general association between TIL and favorable prognosis across cancers, it has been suggested that the good to excellent prognosis associated with MMRd and POLE tumors is attributable to the antitumor immune response (30, 31). However, prior studies of TILs in endometrial cancer have not had sufficient sample size to test this hypothesis. Moreover, relatively less is known about the immune response within the p53wt and p53abn molecular subtypes, which encompass the majority of endometrial cancer cases (12, 31). Notably, p53wt cases have a similar prognosis as MMRd cases (18), despite an almost 10-fold lower mutation rate (2.9 vs. 18 × 10⁻⁶ mutations/Mb; ref. 17). Indeed, with mutation rates spanning 2 orders of magnitude, endometrial cancer provides a powerful setting to investigate the paradigm that high mutation load drives favorable prognosis via the antitumor immune response.

As with other cancers, there is intense interest in the use of checkpoint blockade to treat recurrent endometrial cancer (6, 32, 35). Although relapse is exceedingly rare among POLE cases, there are anecdotal reports of objective responses to anti-PD-1 antibody (36–38), which fits the paradigm that high mutation load predicts response to such agents. Anti-PD-1 antibody is an FDA-approved treatment option for MMRd cases (39), again based on the predictive value of high mutation load. However, it remains unclear whether the p53wt and p53abn subtypes of endometrial cancer might also benefit from checkpoint blockade, as we have a relatively poor understanding of how the immune system responds to these characteristically low mutation load tumors.

To define the relationship between molecular subtype, TIL response, and prognosis in endometrial cancer, we profiled the major TIL subsets (including CD4 and CD8 T cells and B-lineage cells) across a 460-case, molecularly subtyped endometrial cancer cohort with full clinical data. To inform future immunotherapy strategies, we also evaluated immunosuppressive factors, including regulatory T cells (Tregs), IDO-1, and the PD-1/PD-L1 pathway. We report that TIL responses and immunosuppressive features vary in frequency but not magnitude across the molecular subtypes of endometrial cancer. Moreover, the prognostic effect of molecular subtype outweighs TIL response, challenging the paradigm that high mutation load drives prognosis by promoting antitumor immunity. Our findings further suggest that molecular subtype alone may be inadequate to stratify patients with endometrial cancer for immunotherapy.

Our cohort consisted of 460 endometrial cancer cases from 2 previously published series that were used to develop and confirm the ProMisE molecular classifier (18, 19). Clinical, pathologic, molecular, and outcome parameters have previously been described (18, 19) and are detailed in the Supplementary Methods. The study was approved by the Research Ethics Board of BC Cancer and the University of British Columbia.

Representative samples of pathology-reviewed endometrial cancer from our Vancouver center were cored (0.6 mm) in duplicate and arrayed as described previously (18, 19). TMAs were cut at 4-μm thickness onto Superfrost+ glass slides for IHC analysis. Single and multiplex IHC was used to profile immune cells and associated factors in both the epithelial and stromal compartments of tumors. Four antibody panels were used: (i) CD3/CD8/CD79a, (ii) CD138/CD79a, (iii) PD-1/CD8/PD-L1, and (iv) FoxP3/PD-1/CD3. IDO-1 was stained separately (Supplementary Methods). Slides were scanned in brightfield using the Vectra Multispectral Imaging System (PerkinElmer). InForm 2.1 software (PerkinElmer) was used to quantify signals. PD-L1 and IDO-1 positivity was scored visually as a binary variable (whereby >5% staining of immune cells and/or tumor cells was considered positive) or by image analysis as a continuous variable. Additional information is provided in Supplementary Methods.

For each tumor core, raw cell counts and marker values were normalized to epithelial and stromal area to yield a density value. For several markers, a substantial proportion of samples had scores of zero (i.e., zero-inflated/right-skewed data). When log (base 2) transformed, the data followed a delta log normal distribution (Supplementary Fig. S1) and were analyzed using a 2-step approach (40, 41). In the first step, the proportion of cases having no marker expression was modeled separately using a χ² test. Conditional on a nonzero value, the second step followed a lognormal distribution and was assessed using a 1-way ANOVA.

Immune markers were further investigated using heatmaps, overlaying the previously validated ProMisE subtypes. Immune signatures were inferred using an unsupervised ensemble clustering approach implemented in the R package diceR (42). Briefly, several algorithms (Supplementary Methods) were used to cluster 100 random subsets of the data, each of which included 80% of samples. The results were combined to form a final consensus clustering using the k-modes method (43). Several numbers of clusters were tested, and the proportion of ambiguously clustered cases (PAC; ref. 44) within the ensemble was used to determine the optimal number of clusters. Univariable associations between immune clusters, ProMisE subtypes, and other clinicopathologic parameters were evaluated. Associations between immune clusters and clinical outcomes (overall, disease-specific, and progression-free survival) were assessed individually in univariable analyses using log-rank tests and Kaplan–Meier analyses and collectively by adjusting for known risk factors in multivariable Cox proportional hazard models. Survival was also evaluated within each ProMisE subtype. Multivariate analysis was performed, correcting for the immune clusters, ProMisE subtype, age, BMI, grade, histotype, stage, lymph vascular space invasion (LVSI), myometrial invasion, nodal status, or treatment (see Supplementary Statistical Methods). All statistical tests were 2-sided, and statistical significance was set at 0.05 with no adjustment for multiple comparisons.

ProMisE subtype distribution within the 460 endometrial cancer cases included 105 MMRd, 42 POLE, 202 p53wt, and 111 p53abn cases. Clinicopathologic parameters across ProMisE subgroups are shown in Supplementary Table S1. Patients were diagnosed between 1983 and 2013 and received treatment according to standard of care at our center over this time period (18, 19). Adjuvant treatment, when given, consisted primarily of standard radiotherapy (most commonly pelvic external beam radiotherapy or vaginal brachytherapy) with or without platinum/taxane chemotherapy. No patients were known to have received immune checkpoint inhibitors or other immunotherapies.

The densities of various TIL subsets and associated factors were assessed by multiplex IHC and automated image analysis with visual confirmation. Representative images are shown in Fig. 1. The number of missing values (due to missing or uninterpretable cores) ranged from 10 to 28 for individual markers and were not significantly associated with subtype (Supplementary Table S2).

The epithelial and stromal densities of immune markers are shown in Fig. 2 and Supplementary Table S3. Consistent with prior reports (2, 31), the density of epithelial and stromal CD3⁺CD8⁺ TIL (putative cytotoxic T cells) was highest in POLE tumors, followed closely by MMRd tumors, with p53abn and p53wt tumors showing the lowest levels (P < 0.01; Supplementary Table S2). A similar, statistically significant pattern was seen for CD3⁺CD8⁻ TIL (putative CD4⁺ TIL), although the differences between molecular subtypes were less pronounced (P < 0.01). Similar patterns were seen in the stromal compartment (P < 0.01), although the densities of CD3⁺CD8⁺ and CD3⁺CD8⁻ T cells were generally higher compared with epithelium.

In contrast to T cells, neither B cells (CD138⁻CD79a⁺) nor plasma cells (CD138⁺CD79a⁺) were prevalent in tumor epithelium, and neither subset showed a significant difference across ProMisE subtypes (P = 0.74 and P = 0.22, respectively). However, in tumor stroma, higher densities of B cells and plasma cells were seen and varied by ProMisE subtype (P < 0.01 for both), with POLE tumors showing the highest density. Generally, elevated densities of epithelial and stromal T cells in POLE and MMRd tumors were accompanied by higher levels of B cells and plasma cells in the stromal compartment, with the next highest expression levels seen in MMRd (B cells) or p53abn (plasma cells) tumors (Fig. 2; Supplementary Table S2).

We also evaluated PD-1⁺ cells, PD-L1⁺ cells, and Tregs (defined as FoxP3⁺PD-1⁺CD3⁺ cells; Fig. 2; Supplementary Table S2). Consistent with overall TIL densities, POLE tumors expressed the highest densities of epithelial PD-1⁺CD8⁻ (putative CD4⁺) and PD-1⁺CD8⁺ TILs, followed by MMRd tumors; significantly lower levels of these cells were seen in p53abn and p53wt tumors (P = 0.01 and P < 0.01, respectively). The stromal compartment showed a similar ranking, except that the p53abn subtype ranked second for PD-1⁺CD8⁻ TILs (P < 0.01). The proportion of PD-1⁺CD8⁺ cells of all PD-1⁺ cells was similar in POLE, MMRd, and p53abn tumors (78.4%, 74.7%, and 77.9%, respectively; Supplementary Fig. S2); however, this value dropped to 33.1% in p53wt tumors. A similar though less-pronounced pattern was seen in stroma. This suggests that CD8⁺ TILs have a lower level of activation and/or exhaustion in p53wt tumors compared with the other 3 subtypes.

To compare the extent of PD-L1 expression across subtypes, we measured the total area of PD-L1 positivity in tumor cores (i.e., irrespective of tumor versus lymphocyte expression or stromal vs epithelial location). Using a binary scoring method (cores with >5% of cells expressing PD-L1 were considered positive), the proportion of PD-L1⁺ cases was highest for POLE tumors (70%), followed by MMRd (56%), p53abn (32%), and p53wt tumors (16%; P < 0.01). Likewise, when evaluated as a continuous variable, PD-L1 expression differed significantly across ProMisE subtype (P < 0.01) and followed a similar ranking as seen with binarized data.

Similar to other immune markers, Tregs were highest in the stromal compartment in POLE and MMRd tumors, followed in order by p53abn and p53wt tumors (P < 0.01). When the 4 subtypes were considered together, the ratio of CD3⁺CD8⁺ T cells to Tregs was approximately 4:1 in epithelium and 2:1 in stroma (Supplementary Table S4). Within subtypes, these ratios (in epithelium/stroma, respectively) were 5.25/2.72 for POLE, 4.46/2.49 for MMRd, 3.80/1.72 for p53abn, and 3.46/2.28 for p53wt. Thus, POLE cases had the highest ratio of CD3⁺CD8⁺ T cells to Tregs.

Within the MMRd group, we explored differences between patients with loss of MLH1/PMS2 versus MSH2/MSH6. MLH1 loss is usually somatic or epigenetic and generally occurs in older women, raising the possibility it may be associated with a weaker immune response. Of the 105 MMRd cases in our cohort, 71 showed MLH1/PMS2 loss, and 13 showed MSH2/MSH6 loss. Although we were insufficiently powered to make strong conclusions, within the stromal compartment there was a trend toward decreased density of plasma cells in MLH1/PMS2-loss versus MSH2/MSH6-loss tumors (P = 0.07). Moreover, MLH1/PMS2-loss tumors showed significantly lower stromal density of Tregs (FOXP3⁺PD-1⁺CD3⁺ cells; P = 0.05) compared with MSH2/MSH6-loss tumors (Supplementary Fig. S3).

An ensemble clustering approach was used to identify TIL patterns present in endometrial cancer based on the log-transformed IHC results for all TIL subsets (i.e., CD8⁺ T cells, putative CD4⁺ T cells, Tregs, B cells, and plasma cells) and the PD-1/PD-L1 pathway in both the epithelial and stromal compartments. We tested different cluster sizes and found that PAC was lowest in the 2-cluster solution (Supplementary Table S5); nonetheless, we also explored the 3-cluster solution. The 2-cluster solution separated tumors into 2 major groups: TIL^high versus TIL^low (Fig. 3A). TIL^high tumors showed significantly higher densities of all 5 TIL subsets, including CD8⁺ and CD4⁺ T cells, B cells, plasma cells, and Tregs. High levels of PD-1⁺ and PD-L1⁺ cells also characterized this group. In contrast, TIL^low tumors showed low or negligible levels of all 5 TIL subsets, as well as PD-1⁺ and PD-L1⁺ cells. In the 3-cluster solution (Fig. 3B), the TIL^high cluster remained largely the same, whereas the TIL^low cluster split into 2 subgroups referred to as TIL^medium and TIL^negative. Relative to TIL^high tumors, TIL^medium tumors exhibited lower levels of all 5 TIL subsets and notably lower levels of PD-1⁺ and PD-L1⁺ cells. The majority of TIL^medium tumors (75%) belonged to the p53wt subtype. TIL^negative tumors exhibited negligible levels of all 5 TIL subsets and PD-1⁺ and PD-L1⁺ cells.

None of the ProMisE subtypes exhibited a TIL pattern that was not seen in the other subtypes; rather, they differed with respect to the prevalence of TIL^high tumors: POLE (89.5%), MMRd (78%), p53abn (60%), and p53wt (37%). Associations were noted between TIL patterns and certain clinicopathologic parameters such as BMI, LVSI, nodal status, grade, and treatment (Supplementary Table S6). Further investigation revealed that there were no statistical differences in clinical parameters between POLE and MMRd cases with TIL^high and TIL^low immune patterns.

We have previously reported on progression-free and overall survival in this cohort, finding highly favorable outcomes in women with POLE mutated ECs, intermediate outcomes in MMRd and p53wt cases, and poor outcomes in p53abn tumors (18, 19). Cox regression survival analysis with log-rank test was performed across all 4 molecular subtypes to assess the association of individual or combined immune markers with clinical outcomes (Supplementary Table S6). High densities of epithelial CD3⁺CD8⁻ or CD3⁺CD8⁺ TILs were associated with increased patient survival [overall survival (OS) P = 0.05; disease-specific survival (DSS) P = 0.02; progression-free survival (PFS) P = 0.06 and OS P < 0.01, DSS P < 0.01, PFS P < 0.01, respectively]. Likewise, FOXP3⁻PD-1⁻CD3⁺ cells were prognostically favorable in stroma (OS P < 0.01, DSS P = 0.01, PFS P = 0.01) and epithelium (PFS P < 0.01). Conversely, epithelial PD-1⁺CD8⁺ TILs were correlated with decreased OS (P = 0.01), and epithelial PD-1⁺CD8⁻ TILs were correlated with decreased PFS (P < 0.01). Stromal CD138⁻CD79a⁺ and CD138⁺CD79a⁺ TILs lacked prognostic significance as did PD-L1⁺ cells and epithelial or stromal Tregs. Expression of IDO-1 correlated with increased PFS (PFS P < 0.01, OS P = 0.16, DSS P = 0.29). TIL^high versus TIL^low immune clusters showed no association with OS, DSS, or PFS (Fig. 4A–C), even in multivariable analysis after adjusting for ProMisE subtype and other clinicopathologic parameters associated with survival, which have been assessed previously (18, 19) and include age, stage, LVSI, and myometrial invasion (Table 1). However, ProMisE molecular classifier maintained a statistically significant association with outcomes for OS, DSS, and PFS (P = 0.04, P = 0.04, and P < 0.01 respectively; Table 1).

We assessed the prognostic effect of epithelial CD3⁺CD8⁺ TILs in a multivariable model that included ProMisE subtype and these other clinicopathologic parameters (Table 2). Although both ProMisE subtype and epithelial CD8⁺ TILs showed a trend toward increased OS and DSS, only ProMisE subtype retained a significant independent prognostic effect (PFS P = 0.02).

Next, we evaluated the prognostic significance of individual immune markers within ProMisE subtypes. In general, most immune markers failed to show prognostic significance, with a few exceptions (Table 3). Within MMRd cases, CD8⁺CD3⁺ and CD3⁺CD8⁻ TILs were positively associated with DSS (P = 0.04 and 0.02, respectively), and within p53wt cases, stromal B cells (CD138⁻CD79a⁺) were associated with increased DSS (P = 0.04; Table 3). However, even when statistical significance was reached, differences in outcomes were small.

Finally, we evaluated the prognostic significance of the TIL^high versus TIL^low clusters within each of the 4 ProMisE subtypes. This analysis was not meaningful in the POLE subtype, as only 4 cases were TIL^low. Similarly, we were challenged to see a difference within p53wt where events were rare. In the remaining ProMisE subtypes, there was a nonsignificant trend toward inferior outcomes in TIL^low tumors within MMRd tumors. Collectively, these findings indicate that patient survival is more strongly influenced by ProMisE subtype than TILs.

There is increasing appreciation of the importance of the tumor microenvironment in defining outcomes and therapeutic opportunities in cancer. In particular, TILs have been shown to be both prognostic and predictive in that their presence is generally associated with improved outcomes and favorable response to conventional chemotherapies and immune-targeting agents (1–4, 45–49). We investigated the relationships between immune response, molecular subtype, and patient survival in endometrial cancer. Multiplexed, IHC-based analysis of 460 tumors spanning the 4 molecular subtypes of endometrial cancer (MMRd, POLE, p53wt, and p53abn) identified 2 major TIL patterns: TIL^high and TIL^low. As expected, POLE and MMRd subtypes were enriched for TIL^high tumors; however, 22% of MMRd tumors were TIL^low. Conversely, p53abn and p53wt cases were enriched for TIL^low tumors, yet also contained significant proportions of TIL^high tumors. Therefore, all 4 molecular subtypes of endometrial cancer can engender both immunologically “hot” and “cold” tumors, suggesting it may be inadequate to use molecular subtype alone to stratify patients for immunotherapy. As observed across many cancers, CD8⁺ TILs were associated with improved survival, but in multivariate analysis, individual immune markers and immune clusters were not prognostic. Rather, ProMisE molecular subtype retained prognostic value, demonstrating an association with both DSS (P = 0.04) and PFS (<0.01) and a trend toward association with OS (P = 0.08). Thus, molecular subtype, not immune response, was the most significant prognostic factor in this series, suggesting that the molecular/genomic features of tumors influence outcome through mechanisms apart from the immune response.

Immune markers have been assessed in endometrial cancer in both the pre- (2, 50) and post-TCGA eras (12, 30, 31, 46, 51). The latter have primarily focused on the newly recognized POLE ultramutated and MSI/MMRd-hypermutated subtypes. Particular attention has been paid to the notion that the exceptionally high mutation rates in these subtypes drive the generation of neoantigens, making tumors more immunogenic. Indeed, POLE and MMRd tumors have been shown to harbor the highest levels of not only CD8⁺ T cells but also CD4⁺ T cells, stromal CD20⁺ B cells, FOXP3⁺ cells, PD-1⁺ cells, and PD-L1⁺ cells (26, 27, 30). Model systems have further demonstrated that POLE ultramutated tumors are able to stimulate greater cytotoxic CD8⁺ T-cell responses compared with POLE-wt tumors (27). These findings are consistent with a mechanistic link between mutation load and TILs. On the other hand, our findings demonstrate that not all POLE and MMRd tumors engender a significant immune response. Conversely, significant numbers of p53abn and p53wt cases exhibited TIL patterns that, by the clustering analyses performed here, were indistinguishable from TIL^high POLE and MMRd cases. Although we did not directly assess mutation load in this study, the TCGA study (17) demonstrated minimal overlap in the mutation load between the POLE/MSI groups compared with the copy-number high (p53abn) and low (p53wt) groups. Thus, neoantigen load cannot be the only factor influencing the strength of TIL responses. Other factors that merit consideration include DNA damage–induced activation of the cGAS/STING, as well as inhibitory factors such as activation of the Wnt/β-catenin and PTEN/PI3K pathways (52–54). Notably, recurrent mutations in the PTEN, PIK3CA, PIK3R1, and CTNNB1 genes are common in endometrial cancer (55) and hence could play an important role in sculpting the immune response.

This series represents the most extensive assessment to date of B-cell and plasma cell responses in endometrial cancer. Consistent with prior reports in high-grade serous ovarian cancer (HGSC) and other malignancies, B cells and plasma cells were positively associated with CD8 and CD4 TILs (48, 56). Given our use of TMA cores, we were unable to assess the extent to which B-cell and plasma cell infiltrates were associated with tertiary lymphoid structures (57); this could be evaluated in the future by analyzing whole-tumor sections. We found B-lineage cells in both high and low mutation load molecular subtypes, consistent with their lack of association with mutation load in HGSCs (48). Given the extraordinarily favorable prognosis associated with POLE tumors, the presence of dense B-cell and plasma cell infiltrates in this subset challenges the view advanced by some researchers that B-cell/plasma cell responses are detrimental to antitumor immunity. Instead, our findings are in line with the majority of studies showing a positive prognostic effect of B-lineage cells across human cancers (58).

On the surface, the generally higher TIL levels in POLE and MMRd tumors would support these subtypes as the most appropriate candidates for checkpoint blockade. Indeed, objective responses to PD-1 blockade have been demonstrated in patients harboring POLE and MMRd tumors (32, 36, 38). However, recent preclinical data as well as meta-analysis of all published cohorts of POLE endometrial cancers indicates that these tumors may not require any adjuvant therapy (59, 60), let alone costly immunotherapies. Thus, the need for checkpoint blockade in POLE endometrial cancers would be restricted to rare recurrences. The use of checkpoint blockade for relapsed MMRd tumors is well founded and indeed has recently received FDA approval (32, 39). The major outstanding question is whether checkpoint blockade has a place in the treatment of p53abn and p53wt cases. Our findings indicate that significant proportions of these cases exhibit TIL patterns that are equally robust to those seen in POLE and MMRd tumors. Indeed, in our clustering analyses, POLE and MMRd tumors did not demonstrate categorically distinct TIL patterns but rather were part of a continuum shared with the p53wt and p53abn subtypes. This provides justification for evaluating the efficacy of checkpoint blockade in the p53wt and p53abn subtypes.

A major distinguishing feature of TIL clusters was the presence of PD-1⁺ and PD-L1⁺ cells. In particular, our 3-cluster model (Fig. 2B) revealed that the TIL^low cluster contained a subgroup with intermediate levels of the major TIL subsets (i.e., T cells, B cells, plasma cells, Tregs) but negligible levels of PD-1⁺ T cells and PD-L1⁺ macrophages/tumor cells. This subgroup was dominated by p53wt tumors, which as corroborating evidence also demonstrated markedly lower percentages of PD-1⁺ T cells. Thus, with either 2- or 3-cluster statistical models, only the TIL^high subset had the PD-1⁺/PD-L1⁺ pattern that is associated (albeit imperfectly) with response to checkpoint blockade in other cancers. This TIL^high group contained the majority of POLE and MMRd tumors but also 28% of p53wt and 25% of p53abn tumors. Thus, there may be merit in comparing the TIL^high pattern versus molecular subtype as predictive biomarkers for checkpoint blockade in endometrial cancer.

With the MMRd subtype, we investigated whether the underlying genetic defect had an influence on TIL pattern. Somatic or epigenetic loss of PMS2/MLH1 is more common and usually arises in older individuals, who could conceivably mount less robust antitumor immune responses. In contrast, loss of MSH6/MSH2 is often seen in germline/Lynch Syndrome–associated endometrial cancers, which arise in younger individuals. Although our study was underpowered to address this issue definitively, cases with PMS2 loss showed a trend towards lower levels of several immune cells or markers (Tregs, B cells, plasma cells, PD-1, PD-L1), particularly in tumor stroma (Supplementary Fig. S2). Sloan and colleagues showed higher PD-L1 expression in Lynch syndrome–associated endometrial cancers as compared with endometrial cancers with MLH1 promoter methylation (61). However, as in our series, these observations were not accompanied by data on actual response rates to immune blockade, thus the predictive value of these differences within molecular subtype awaits verification through clinical trials.

In addition to the PD-1/PD-L1 pathway, we assessed the expression of IDO-1, which represents another pharmacologically targetable mechanism of immune suppression in cancer (62–66). Similar to PD-L1, IDO-1 expression was significantly higher in TIL^high tumors and spanned the 4 molecular subtypes. Prior publications have suggested high prevalence of IDO-1 expression in MMRd tumors, particularly Lynch syndrome–associated endometrial cancer (67, 68). Our series confirmed this and revealed even higher expression in the POLE subtype. Given that IDO-1 is frequently coexpressed with PD-L1 (not specifically evaluated in this series), dual immunotherapy could be considered (69, 70).

Strengths of this study include evaluation of a large, well-annotated cohort of endometrial cancers and a wide panel of immune markers. Moreover, we used a molecular classification tool that has undergone rigorous validation according to the Institute of Medicine criteria (70). Limitations of this study include the use of tissue microarrays instead of whole sections; however, we assessed cores in duplicates and the number of uninterpretable cases was very low in this series. Moreover, prior literature supports the use of tissue microarrays for immune studies of this type (71, 72). Nonetheless, if TIL patterns were to be used as clinical biomarkers, additional studies would be required to determine whether whole sections provide more robust data for clinical decision making. A second limitation is that we evaluated only a single tumor site from each patient and hence were unable to assess the extent of intratumoral heterogeneity of TIL patterns, as we reported recently in high-grade serous ovarian cancer (56). Importantly, however, this phenomenon does not negate the clinicopathologic significance of TIL when measured across large patient cohorts. A third limitation is that mutation load and other molecular/genomic features of tumors were inferred by ProMisE subtype rather than direct measurement. Finally, given that this cohort was accrued prior to the widespread use of immune-modulatory agents in endometrial cancer, we were unable to assess whether any immune cells or markers were predictive of response to PD-1/PD-L1 or IDO-1 inhibitors.

In summary, we have profiled the immune response in endometrial cancer through the modern lens of a pragmatic molecular classification tool. Immune markers, alone or in combination, failed to match the prognostic ability of ProMisE molecular classification. We demonstrate the diversity of immune marker expression across and within molecular subtypes. Candidates for immunotherapy extend beyond the POLE and MMRd subtypes, and assessment of immune response rather than molecular subtype may be more helpful in patient selection. Continued work is needed to identify the most accurate biomarkers for stratification of patients with endometrial cancer for immunotherapy, so as to minimize toxicity and cost for women who are unlikely to respond while ensuring that opportunities for treatment are not missed.

No potential conflicts of interest were disclosed.

Conception and design: A. Talhouk, H. Derocher, B. Gilks, B.H. Nelson, J.N. McAlpine

Development of methodology: H. Derocher, K. Milne, B. Gilks, B.H. Nelson, J.N. McAlpine

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Derocher, K. Milne, B. Gilks, B.H. Nelson, J.N. McAlpine

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A. Talhouk, S. Leung, M.S. Anglesio, B.H. Nelson, J.N. McAlpine

Writing, review, and/or revision of the manuscript: A. Talhouk, H. Derocher, P. Schmidt, S. Leung, K. Milne, B. Gilks, M.S. Anglesio, B.H. Nelson, J.N. McAlpine

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A. Talhouk, S. Leung, K. Milne, B.H. Nelson, J.N. McAlpine

Study supervision: A. Talhouk, B.H. Nelson

We thank Dr. Phineas Hamilton for advice. This work was supported by grants from the BC Cancer Foundation (to J.N. McAlpine, B.H. Nelson), Terry Fox Research Institute (to B.H. Nelson), Canadian Institutes of Health Research (to A. Talhouk, B.H. Nelson, J.N. McAlpine), Canada's Networks of Centres of Excellence (to B.H. Nelson), and Canada Foundation for Innovation (to B.H. Nelson).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.