Patterns of cytotoxic T-cell densities in immunogenic endometrial cancers reveal a potential mechanism for differences in immunotherapy efficacy

Introduction

Endometrial cancer (EC) is the most common gynaecological malignancy in the Western world; the incidence of EC continues to rise globally year-on-year, primarily due to rising obesity levels and an ageing population.1 With an increasing incidence and an associated rise in recurrent disease, more women are dying of EC cancer than at any other time.2 This is further complicated by an increasing degree of complexity seen within EC patients, with far higher levels of class III obesity related sequelae and a lower performance status, making surgical treatment problematic.3

Historically, the treatment of EC was directed by histological features such as grade and stage of disease.2 Recently there has been movement towards a personalised molecular approach, with the recent literature looking to the genomic landscape of the disease to inform clinical management.4–6 Since the publication of The Cancer Genome Atlas’ report on EC, the molecular aetiology of the disease has been viewed in four distinct groups: POLE mutated (both somatic and germline), mismatch repair deficient (MMRd) (both somatic and germline), copy number high and copy number low variants.7 These groups are distinct both in their biology but also in their clinical outcome, a finding that has been ratified in independent cohorts.8 9 Therefore, the identification and incorporation of molecular classification is of prognostic importance and helps direct treatment, especially the use of immunotherapy.10

There is growing evidence that distinct molecular entities in EC illicit different immune responses.11 12 EC that arise from a defective mismatch repair system (MMR), either through germline or somatic pathogenic variants in MLH1, MHS2(EPCAM), MSH6, PMS2 or their epigenetic controls (such as hypermethylation of the MLH1 promotor region, which is by far the most frequent) are known to have a high number of tumour infiltrating lymphocytes (TILs).11 Furthermore, there are differences in the degree of numbers of TILs dependent on the mechanism of MMRd, be it somatic or as a result of Lynch syndrome (LS); an autosomal dominant inherited pathogenic mutation affecting one of the MMR genes.13 14 ECs with mutations in the DNA Polymerase Epsilon, Catalytic Subunit (path_POLE) are also more immunogenic then those without a path_POLE.15 16 Immune profiling solid tumours also has a prognostic function, with CD8+T cell derived ‘immune-scores’ being highly predictive of clinical outcome.13 14 17 Furthermore, it has implications for treatment.

The immune response in EC with a high mutational burden is thought to derive from the production of immune active frame-shift proteins that act as neo-antigens.18 This in turn applies a selective pressure on the tumour, encouraging immune-editing and ultimately immune escape often mediated through the PD-1/PD-L1 axis.19 This is a druggable pathway and pharmacological treatments blocking the axis have shown promise in MMRd ECs.20 21 However, these studies had an overrepresentation of LS tumours (up to 48%) which does not reflect the clinical situation whereby the vast majority of MMRd tumours arise through somatic events.22 It is therefore important to explore the immune landscape of MMRd tumours by their aetiology (somatic vs inherited) to ensure they are biologically comparable and composite treatment predictions are meaningful. In addition, not all MMRd ECs respond to immune therapy; therefore, it is important to establish a detailed picture of the immune landscape within these cancers to see if variance could help explain the clinically observed treatment heterogeneity.14 21

The aim of this study was to explore the density of CD8+T TILs in different molecular groups of EC. Furthermore, by describing these densities by tumour compartment we aim to explore the underlying immune environment of these tumours. By using a potentially fully automatable processes we looked at the feasibility of generating automated CD8+T cell counts in a way that would have minimal implications for clinical workflows. Overall, we believe these results will inform the optimisation of immunotherapy in EC.

Materials and methods

Image acquisition and analysis

All slides were scanned within 14 days of immunofluorescence staining. They were imaged using the Vectra 3.0 Automated Quantitative Pathology Imaging System (Perkin Elmer) at 40× magnification as per manufacturer’s instructions. Image analysis and spectral separation was performed using InForm Cell Analysis Software (Perkin Elmer). Spectral separation was informed by the single marker spectral libraries and optimised by eye. Image analysis workflows were then created for high volume analysis. This was achieved manually with DAPI and keratin to segment images into tumour, stroma, necrosis and ‘no tissue’ areas. Cell segregation was optimised with the use of the DAPI stain to detect cell contours. Cell phenotype (CD8+T or CD8−) was trained in 40 tumours with 40 cells per tumour (see figure 1).

Request permission

Figure 1

Workflow for image preparation and analysis. (A) Whole slide scans reviewed for quality and four regions of interest (ROI) selected manually for high resolution scan. (B) ROI individually visually inspected. Cell contouring (C), tissue segregation (D), cell phenotyping solutions run (E) generating cell phenotype counts by tissue segmentation (F). These data are exported to spreadsheet format. Composite images quality assured manually against ROI to assure accuracy. If errors identified, manual count conducted.

For each tumour, four regions of interest (ROI) were manually selected covering an area of 20 μm×20 μm. Two ROI were selected (by NR) at the tumour/stromal interface and two covered the tumour core. In those cases where there was no tumour/stromal interface (such as endometrial biopsies) only two ROI’s were taken within the tumour core. Chromogenic histology was visually checked to ensure accurate tissue segregation and CD8+/−T cell identification for each ROI; in cases of inaccurate counts a manual count was performed (see online supplemental figure S1). If imaging or segregation/CD8+T cell fluorescence failed, preventing manual counting, the ROI was excluded from analysis.

Results

The confirmed LS cohort from BOLT numbered 60 ECs. These were combined with ECs from PORTEC 1 and 2, which had all undergone prior molecular classification. In total 706 ECs were stained, however, 99 tumours had either a scanning or ROI acquisition failure, meaning 607 ECs contributed to the analysis. The study schema is outlined in online supplemental figure S2. Overall, 2303 ROI were analysed, of these 328 (15%) had to be re-scored manually due to cell phenotyping inaccuracies, namely non-CD8+T cells (eg, erythrocytes) being incorrectly counted as CD8+T cells. The demographics of the analysed EC are outlined in table 1. Tumours were predominantly low stage due to the inclusion criteria of the PORTEC trials.

It was not possible to acquire all four ROI in all tumours. All samples from the LMB, path_POLE, Lynch Like and MLH1-methylation cohorts were cut from hysterectomy slides. In the confirmed LS cohort, 7 samples were cut from endometrial biopsies with no tumour/myometrial interface and therefore only two ROI could be generated within the tumour core. In several hysterectomy derived samples, it was not possible to secure four ROIs due to insufficient tumour bulk, regions of poor fixation or tumour that was confined to the endometrial cavity. In the LMB cohort 56 tumours had less than four ROI (one ROI n=2, two ROI n=23, three ROI n=31). For the Lynch Like cohort there were three samples with less than four ROI (3 ROI n=3), MLH1-methylation group four samples were less than four ROI (two ROI n=3, three ROI n=1), path_POLE group four samples were less than four ROI (one ROI n=1, three ROI n=3) and in the confirmed LS group 17 were less than four ROI (two ROI n=7, three ROI n=10).

Comparative statistics are sumerised in table 2 and figure 2. CD8+T cell counts in confirmed LS ECs were around threefold higher than in the MLH1-methylated cohort, excluding the tumour stroma. Confirmed LS, Lynch Like, MLH1-mehtylated and path_POLE ECs had significantly higher CD8+T cell counts across all tumour locations when compared with LMB. This difference was less pronounced in CD8+T counts between path_POLE versus confirmed LS ECs. Interestingly, CD8+T cell counts were significantly higher in the tumour compartments of confirmed LS ECs when compared with those of the Lynch Like group. However minimal differences were observed across all locations between the Lynch Like versus MLH1-methylated ECs (figure 3). Looking at all Lynch (confirmed LS+Lynch Like) ECs versus MLH1-methylated, all had significantly higher CD8+T cell count averages in all compartments excluding stromal compartment within the tumour core (online supplemental figure S3 and table S2), however, this result was weighted by the confirmed LS cohort. When all MMRd (confirmed LS+Lynch Like+MLH1-methylated) versus path_POLE ECs were analysed, path_POLE ECs had significantly higher CD8+T cell count averages in the stromal compartments (online supplemental figure S4). When looking at variation in CD8+T cell counts within LS itself, there was no significant difference between the average stomal or tumour CD8+T cell counts as per the gene affected (online supplemental figures S5 and S6). However, it should be noted that this analysis was impacted by small numbers.

Request permission

Request permission

Discussion

This study reports a comprehensive description of CD8+T cell distribution in ECs by molecular subtype. This work explored the application of automated cell counting as a means of high-volume TILs phenotyping. Our findings indicate that MLH1-methylated ECs are immunologically distinct from ECs in confirmed LS carriers. Furthermore, Lynch Like ECs, a group that will include both somatic and inherited MMRd, are not immunologically comparable ECs found in women with confirmed LS. Finally, path_POLE ECs are as immunogenic, if not more so, than LS associated ECs indicating that they may be ideal targets for immunotherapy. These data could have significant implications for the application of immune treatments in ECs and our understanding of the immune environment of EC.

LS carriers are known to survive multiple, normally lethal, primary cancers.28 29 This is despite cancers in LS being associated with locally aggressive and poor prognostic features when compared with MMR proficient cancers.14 This may be explained by the association between MMRd ECs and high numbers of TILs; a phenomenon that is well established.11 Furthermore, higher numbers of TILs are associated with better survival.13 17 MMRd therefore seems to generate an immune response which in turn confers a survival benefit for the patient. The accumulation of frame shift mutations that occur in a MMRd cancer leads to the production of frame-shift peptides, de facto neo-antigens, which in turn elicit a tumour specific immune response.18 This mechanism is also seen in path_POLE cancers.30 Such immunogenic cancers invoke immune-editing and escape mechanisms often via upregulation of the PD-1/PD-L1 pathway.10 However, PD-1 and PD-L1 expression have been found to be poor proxies of immunotherapy response.31 Furthermore, recent murine data has indicated that somatically driven MMRd leads to heterogenous immunological response in part due to the timing of MMR inactivation and the variation in clonal expansion.32

Case reports have described favourable treatment outcomes from the application of PD-1 blockade in path_POLE cancers.15 33 Furthermore, recent trials have explored the application of PD-1 blockade in MMRd cancers, reporting positive results.20 21 34 35 The best treatment response was seen in those tumours with the highest CD8+T cell numbers; specifically, high CD8+T cell densities (cells/mm²) within the tumour core. However, traditionally trials heavily rely on cancers in LS patients with up to 48% carrying germline path_MMR.20 34 This does not reflect clinical practice where the majority of MMRd cancers arise from somatic hyper-methylation of the MLH1 promotor region; this is especially the case in EC.22 36 No subgroup analysis was performed in these studies exploring the differences in response between somatic versus germline derived MMR loss as it was assumed their immune response was homogenous. However, there are distinct differences between these groups.11 13 Furthermore, not all MMRd ECs respond to immunotherapy, and therefore the use of CD8+T cell density could be a means to better stratify treatments.17 21 Certainly, the variation in treatment response that has been noted in clinical trials can potentially be explained by the variation in MMRd aetiology.35 Our results confirm that CD8+T cell numbers is significantly different between LS confirmed and MLH1-methylated ECs.

This observation is in keeping with the literature exploring the biology of these tumours. LS carriers have a defective MMR gene from conception. Although they carry a wild type allele, which for the most part preserves MMR system function, the mutant allele is expressed and this halplosufficiency may lead to low levels of MSI.37 In addition, they are more likely to periodically develop MMRd cells which do not develop into a malignancy.38 These two mechanisms explain the presence of a frame-shift peptide specific immunity in healthy LS carriers.39 Therefore, LS carriers develop a pre-primed immune system with a memory of MMRd generated frame-shift peptides, so that when a MMRd EC develops, they mount a more pronounced and cytotoxic immune response. As somatic hypermethylation of the promotor region of MLH1 is an acute event, that is not present from conception, the period of haplosufficiency is less than in LS. In addition, the change is clonal and not constitutional. Therefore, the degree of frame-shift peptide production is likely to be lower than in LS healthy carriers as far fewer cells are affected in the pre-cancerous stage. This may limit priming of the immune system and reduce the magnitude of a subsequent immune response to sporadic EC.32

Another finding was that path_POLE somatic ECs have a similar CD8+T cell count to LS ECs. POLE is vital in DNA synthesis and maintaining DNA fidelity.30 Path_POLE ECs have an ultra-mutated phenotype with a mutational burden that exceeds MMRd-EC.7POLE ultramutated-ECs produce a rich abundance of antigenic neoepitopes compared with MMRd cancers.40 Furthermore, path_POLE are an early event in sporadic cancers and therefore the production of neoantigens is early and persistent during tumourigenesis.40 Due to this massive and sustained immune activation, it follows the immune response seen in path_POLE ECs would be similar to that seen in LS ECs, where neo-antigens are less abundant but produced over many years. Interestingly we did not see a variation in CD8+T cell counts within LS itself, in that no specific gene (path_MLH1, path_MSH2, path_MSH6 or path_PMS2) demonstrated a significant difference in CD8+T cell density. This is expected given the dimeric relationship seen within in MMR proteins in which if one is defunct its partner protein is also often absent. Therefore, a defective MMR gene leads to global consequences with the MMR system.41

Our study has several key strengths. First, it includes over 600 ECs, giving a broad spectrum of samples from which, we derive our analysis. Second, the molecular groups were defined using diagnostic methods used clinically, and therefore reflect real world experience. We included a large cohort of ECs arising in confirmed LS carriers; to the authors’ knowledge, this is the largest number of such samples to undergo CD8+T cell phenotyping. Third, all our automated tissue segregation and counts were reviewed manually to ensure accuracy. Finally, this paper explores a crucial clinical question: are LS ECs immunologically different to other MMRd ECs? Our data suggesting they are, could inform better selection of EC patients for immunotherapy and this phenomenon should now be explored in treated cohorts.

Our study has limitations. Many of the ECs were sourced from the prospectively recruited PORTEC 1 and 2 studies, which restricted trial entry to stage 1 (PORTEC 1) and high-risk low stage ECs (PORTEC 2). It is unclear whether our findings are applicable to advanced stage ECs where immunotherapy arguably has most clinical utility and application. Women recruited to the BOLT study who contributed to the confirmed LS cohort were mostly LS ECs survivors; this could mean the cohort represented ECs with a survivable biology. However, LS ECs are known to have an excellent prognosis, more so than sporadic disease and therefore we believe these samples remain clinically representative.42 Furthermore, due to the design of our study, we cannot explain why we see a difference in CD8+T cells here. We do not know whether this is due to increased cellular recruitment/chemotaxis, cell survival or clonal expansion. In addition, due to the volume of samples analysed in this study, we did not phenotype the subsets of CD8+T cells or other immune cells within our samples. Of note, there is a considerable difference in age between those with MLH1-methylated ECs (mean age 65 years old at diagnosis) and those with confirmed LS (53 years old at diagnosis). This is expected due to the inherited nature of LS. However, it means that we cannot exclude the possibility that age is a confounding factor in our findings. Finally, our cohorts did not have meaningful representation of non-white women. This prevented us from exploring the impact of ethnicity on immune microenvironment.

In conclusion, to our knowledge, this work presents the most comprehensive phenotyping of CD8+T cells in ECs stratified by molecular subgroup. This study found that CD8+T cell counts and distribution are not equal between MLH1-methylated and confirmed LS ECs and we have outlined the potential biology underlying these differences. Our findings call for a review of current trial data looking to the application of immunotherapy in MMRd cancers. These trials have historically heavily relied on patients with LS. Our data may indicate this group is more susceptible to these treatments than the more clinically numerous, sporadic MMRd endometrial tumours. CD8+T cell density may be an important prognostic indicator of immunotherapy response and should be further explored in prospective treatment cohorts.