Dedifferentiation in endometrial cancer is associated with HLA‐I silencing that may be restored by IFN‐γ

Abstract

Dedifferentiated endometrial carcinoma occurs when a prognostically favorable low‐grade endometrial carcinoma transforms into a highly aggressive undifferentiated carcinoma following genomic inactivation of core SWItch/Sucrose Non‐Fermentable (SWI/SNF) complex protein(s). This typically occurs in a microsatellite‐instable molecular context. This study aims to identify the immune evasion mechanisms associated with dedifferentiation. Whole‐slide immunohistochemistry‐based image analysis was used to compare spatial cancer immunophenotypes and human leukocyte antigen class I (HLA‐I) expression between the differentiated and undifferentiated components of 14 SWI/SNF‐inactivated dedifferentiated endometrial carcinomas. Flow cytometry and in vivo xenograft studies were performed to gain functional insights. Dedifferentiation was associated with changes in spatial cancer immunophenotypes, most frequently from an inflamed immunophenotype in the differentiated component to an excluded immunophenotype in the undifferentiated component, with loss of tumor HLA‐I expression, decreased intraepithelial CD8⁺ T‐cell, and increased stromal PD‐L1⁺ immune infiltrates. Loss of tumor surface HLA‐I expression was also observed in the undifferentiated carcinoma from two patient‐derived models of dedifferentiated endometrial carcinomas – DDEC1 and DDEC2. By flow cytometry, IFN‐γ treatment restored HLA‐I expression in DDEC2 (3.5‐fold increase, p < 0.05) but not in DDEC1, which harbored inactivating JAK1 mutations. In the in vivo xenograft model of DDEC2, intratumoral injection of IFN‐γ restored HLA‐I expression in the undifferentiated tumors. These findings implicate the loss of tumor HLA‐I expression as an immune evasion mechanism acquired during dedifferentiation in a subset of dedifferentiated endometrial carcinomas. More importantly, IFN‐γ can restore tumor HLA‐I expression, providing a rationale for combining it with immune checkpoint inhibitors in the treatment of this clinically aggressive cancer type. © 2026 The Author(s). The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.

Introduction

Dedifferentiated endometrial carcinomas (DDECs) are an uncommon but highly aggressive type of endometrial cancer that frequently presents with advanced stage disease and rapid clinical progression despite standard of care therapy [1, 2]. DDEC occurs when a prognostically highly favorable low‐grade endometrial carcinoma abruptly dedifferentiates into a phenotypically undifferentiated malignancy showing minimal to no evidence of epithelial differentiation. Genomic inactivation of core SWItch/Sucrose Non‐Fermentable (SWI/SNF) complex proteins, resulting in loss of SMARCA4, loss of SMARCB1 or co‐loss of ARID1A and ARID1B, has been implicated as an important driver for dedifferentiation [3, 4], with SWI/SNF‐inactivated DDEC behaving significantly more aggressively than SWI/SNF‐proficient DDEC [1, 2]. Moreover, dedifferentiation appears to occur most frequently in a hypermutating molecular context, as about two‐thirds of DDECs are mismatch repair protein‐deficient (MMRd) with high microsatellite instability (MSI‐H) [3, 4, 5, 6]. Despite the high mutation burden and consequently high neo‐antigen load, DDECs can metastasize widely, seemingly unimpeded by the host immune system [1]. It therefore appears that these neo‐antigen rich DDEC tumor cells must have adopted effective immune evasion mechanism(s). A number of previous studies have examined specific aspects of the immune landscape in DDECs and have noted PD‐L1 expression in tumor or stromal immune cells in about half of cases, which correlates with MMR deficiency status [5, 6]. While these findings point to important tumor–immune cell interactions in DDECs, further investigations are warranted to more comprehensively determine how DDECs evade immune surveillance and potentially how the immune escape mechanism(s) involved can be more precisely targeted therapeutically through tailored immunomodulatory strategies.

The spatial architecture of the tumor–immune microenvironment is increasingly recognized as a critical determinant of cancer behavior, influencing tumor progression, metastatic potential, and response to therapy. The classification into three major spatial cancer immunophenotypes (SCIs) – inflamed, immune excluded, and immune desert – captures the degree and distribution of immune infiltration relative to tumor cells with prognostic and predictive value across multiple tumor types [7, 8, 9]. Tumors with an inflamed phenotype typically exhibit abundant intratumoral cytotoxic T lymphocytes and heightened IFN‐γ signaling, features often associated with favorable responses to immune checkpoint blockade [8, 10]. In contrast, immune‐excluded and immune‐desert phenotypes reflect distinct immune evasion strategies, including stromal barriers, immunosuppressive cytokine milieus, and impaired antigen presentation. This study aims to characterize immune escape mechanisms during dedifferentiation with a focus on the SCI and human leukocyte antigen class I (HLA‐I) expression in a series of DDECs that have been molecularly defined with respect to their core SWI/SNF status and molecular subtypes.

Materials and methods

Ethical approval and patient consent

This study was conducted with approval from institutional research ethics boards CE‐AVEC (Comitato Etico Area Vasta Emilia Centro; approval numbers 27/2019/Sper/AOUBo and 10/2023/Sper/AOUBo) and HREBA (CC‐22‐0202, the University of Calgary and the University of Alberta).

The in vivo assessment of IFN‐γ on xenograft tumors was approved under institutional animal ethics approval (AUP00002496, University of Alberta).

Study cohort and clinicopathologic data collection

The study included 14 SWI/SNF‐inactivated DDECs. DDEC cases were obtained from Calgary Laboratory Services (Calgary, Canada), Royal Alexandra Hospital (Edmonton, Canada), and IRCCS Azienda Ospedaliero‐Universitaria di Bologna (Bologna, Italy).

Molecular classification and POLE exonuclease domain sequencing

Tumors were classified into POLE‐mutated, MMRd, no special molecular type (NSMP), or p53‐abnormal (p53abn) subtypes according to the WHO 2020 algorithm [11]. Subtyping was assigned hierarchically: (i) presence of pathogenic POLE mutation; (ii) MMR status by immunohistochemistry (IHC); (iii) p53 status; with cases negative for all three features classified as NSMP. For POLE exonuclease domain sequencing, DNA was extracted from formalin‐fixed paraffin‐embedded tissue starting from 2 to 4 10‐μm thick sections, depending on the amount of tissue present in the paraffin block. The areas of interest were marked on the control H&E‐stained slide and manually dissected under microscopic guidance (BX41 Microscope, Olympus Corp., Tokyo, Japan) using a sterile blade. DNA was extracted using the Quick Extract Kit (LGC Biosearch Technologies, Petaluma, CA, USA) and quantified using the ‘Qubit’ fluorometer (ThermoFisher Scientific, Waltham, MA, USA). Samples were analyzed using a laboratory‐developed multigene next‐generation sequencing (NGS) panel of genomic regions and sequenced using the Gene Studio S5 Prime sequencer (ThermoFisher Scientific), following the manufacturer's protocol as previously described [12]. Template preparation was performed using the Chef Machine instrument (ThermoFisher Scientific) and then sequenced using an Ion 530 chip (ThermoFisher Scientific). The panel covered a total of 169 amplicons that included POLE (exons 9–14; human reference sequence hg19/GRCh37, 12.74 kb). Only nucleotide variations observed in at least 5% of the total number of reads analyzed were considered for mutational call. POLE variants were evaluated according to Leon‐Castillo et al [13].

Immunohistochemistry and digital image analysis for spatial cancer immunophenotype

All IHC and molecular analyses were performed on whole‐tissue sections as previously described [1, 7]. Detailed antibody clones and staining protocols are listed in the supplementary material, Table S1. All antibodies were from Ventana Medical Systems (Tucson, AZ, USA) other than CD163 (Cell Marque, Rocklin, CA, USA), CD68 (Diagnostic Biosystems, Pleasanton, CA, USA), HLA‐I (Abcam, Cambridge, UK), ARID1A (Abcam), ARID1B (Abnova, Taipei, Taiwan), SMARCA4 (Abcam), and SMARCB1 (Cell Marque). PD‐L1/CD68 and CD20/CD3 double staining was conducted using OptiView DAB detection kit (Ventana Medical Systems) first (brown color), then UltraView RED AP detection kit (Ventana Medical Systems) (red color). The single staining tests were performed using the DAB kit alone (Ventana Medical Systems). All sections were then counterstained using Hematoxylin and bluing reagent following the Ventana Medical Systems protocols. For DDECs, the differentiated component and the undifferentiated component were evaluated separately. For the IHC portion of the molecular classification, p53 expression was classified as wild type or mutant‐like (overexpression, complete absence, or cytoplasmic staining). Tumors were considered MMR deficient (MMRd) when showing complete loss of at least one MMR protein. Immune cell markers (CD20, CD3, CD8, PD‐L1, CD68, CD163) were assessed by IHC and quantified by digital image analysis using Image Pro Plus 5.1 software (Media Cybernetics Inc., Rockville, MD, USA) in operator‐guided mode as previously described [7] to define the Spatial Cancer Immunophenotype (SCI) into the following groups: inflamed: high intraepithelial CD8⁺ tumor‐infiltrating lymphocyte (TIL) density; excluded: low intraepithelial CD8⁺ TILs but high peritumoral total inflammatory infiltrate; desert: low CD8⁺ TILs and low total inflammatory infiltrate. Each immune cell marker (CD68, CD20, CD3, CD8, and PD‐L1) was quantified using IMAGE Pro Plus V5.1 (Media Cybernetics Inc.). The percentage index of positive staining was calculated as the ratio of pixels of immunostained area to pixels of the area of interest. The total inflammatory component was obtained by adding the percentages of CD68, CD20, and CD3. The intraepithelial tumor‐infiltrating CD8⁺ lymphocytes (CD8⁺ iTILs) index was obtained by identifying five hotspots in the entire tumor area at ×200 fields. The percentage index of CD8⁺ iTILs was expressed as the ratio of CD8⁺ intraepithelial lymphocytes to the neoplastic population. All markers were binarized into low and high using the respective median using cutoffs determined using a previously published single institutional and consecutive series of 213 endometrial carcinomas [7]. IHC for HLA‐I, SMARCA4 (BRG1), SMARCB1 (INI1), and ARID1A/ARID1B was performed as described (Table 1). HLA‐I expression was categorized as follows, retained – staining in > 90% of tumor cells; subclonal loss – staining in 10–90% of tumor cells, with juxtaposed positive and negative areas; complete loss – staining in < 10% of tumor cells [9]. SMARCA4 (BRG1), SMARCB1 (INI1), ARID1A, and ARID1B expression was scored as intact or deficient as previously described [4, 14].

Table 1.

Clinicopathologic and molecular characteristics of dedifferentiated endometrial carcinomas (DDECs).

Case number	Age (years)	FIGO stage	Histotype	Molecular subtype	POLE status	SWI/SNF status	Components	MMR IHC	ARID1A, IHC	ARID1B, IHC	SMARCA4, IHC	SMARCB1, IHC
1	53	II	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G1 endometrioid	Loss of MLH1/PMS2	Loss	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
2	68	IIIC	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G2 endometrioid	Loss of MLH1/PMS2	Loss	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
3	74	1B	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G1 endometrioid	Loss of MLH1/PMS2	Loss	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
4	61	IIIC	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G2 endometrioid	Loss of MLH1/PMS2	Loss	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
5	75	IB	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G2 endometrioid	Loss of MLH1/PMS2	Loss	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
6	48	IA	DDEC	MMRd	wild‐type	ARID1A/1B‐inactivated	G1 endometrioid	Loss of MLH1/PMS2	Intact	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Loss	Loss	Intact	Intact
7	57	IA	DDEC	NSMP	wild‐type	SMARCA4‐inactivated	G1 endometrioid	Intact	Intact	Intact	Intact	Intact
							Undifferentiated	Intact	Intact	Intact	Loss	Intact
8	69	II	DDEC	NSMP	wild‐type	SMARCA4‐inactivated	G1 endometrioid	Intact	Intact	Intact	Intact	Intact
							Undifferentiated	Intact	Intact	Intact	Loss	Intact
9	54	IA	DDEC	NSMP	wild‐type	SMARCA4‐inactivated	G1 endometrioid	Intact	Intact	Intact	Intact	Intact
							Undifferentiated	Intact	Intact	Intact	Loss	Intact
10	50	IVB	DDEC	NSMP	wild‐type	SMARCA4‐inactivated	G1 endometrioid	Intact	Intact	Intact	Intact	Intact
							Undifferentiated	Intact	Intact	Intact	Loss	Intact
11	73	IA	DDEC	NSMP	wild‐type	SMARCA4‐inactivated	G2 endometrioid	Intact	Intact	Intact	Intact	Intact
							Undifferentiated	Intact	Intact	Intact	Loss	Intact
12	70	IIIC	DDEC	MMRd	wild‐type	SMARCA4‐inactivated	G2 endometrioid	Loss of PMS2	Intact	Intact	Intact	Intact
							Undifferentiated	Loss of PMS2	Intact	Intact	Loss	Intact
13	67	IIIC	DDEC	MMRd	wild‐type	SMARCB1‐inactivated	G1 endometrioid	Loss of MLH1/PMS2	Intact	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Intact	Intact	Intact	Loss
14	64	IIIC	DDEC	MMRd	wild‐type	SMARCB1‐inactivated	G1 endometrioid	Loss of MLH1/PMS2	Intact	Intact	Intact	Intact
							Undifferentiated	Loss of MLH1/PMS2	Intact	Intact	Intact	Loss

DDEC, dedifferentiated endometrial carcinoma; G1, FIGO grade 1; G2, FIGO grade 2; IHC, immunohistochemistry; MMRd, mismatch repair protein‐deficient molecular type; NSMP, no special molecular profile molecular type; SWI/SNF, SWItch/Sucrose Non‐Fermentable.

Cell lines

DDEC1 and DDEC2 were developed in our laboratory and 3D spheroid models were cultured as previously described [15]. The whole exome sequencing data described for DDEC1 and DDEC2 are available in Sequence Read Archive (https://www.ncbi.nlm.nih.gov/sra). Non‐synonymous single nucleotide variation (SNV) and small insertion/deletion (indel) findings on genes of interest (B2M, HLA‐A, HLA‐B, HLA‐C, IFNGR1, IFNGR2, JAK1, JAK2, STAT1, TAP1, TAP2, TAPBP) are shown in the supplementary material, File S1.

Western blotting analysis

To prepare cell lysates for western blotting analysis, 100 μl of 5× sample buffer composed of 10% dithiothreitol and 2% benzonase was added to each well. To prepare samples for SDS‐PAGE, samples were heated to 50 °C and centrifuged at 14,000 rpm for 30 s. Approximately 10 μl of sample was loaded per lane, using a PageRuler Prestained Ladder (Thermo Fisher Scientific) as the molecular‐weight marker. Samples were run on a 10% SDS‐PAGE gel and then transferred to a nitrocellulose membrane. The membrane was blocked in a 5% BSA in TBS‐Tween20 solution for 1 h prior to incubation with primary antibodies. The following antibodies were used for western blotting analysis: pSTAT1 (Tyr701) anti‐rabbit mAb (Cell Signaling Technology, Danvers, MA, USA; Cat. no. 9167S) or β‐tubulin anti‐rabbit mAb (Cell Signaling Technology, Cat. no. 2146S) at a 1:1,000 dilution. The membrane was then analyzed using the LI‐COR Odyssey CLx imaging system (LI‐COR Biosciences, Lincoln, NE, USA) using a goat anti‐rabbit (LI‐COR Inc., Cat. no. 926‐32,212) IgG secondary antibody at 1:10,000 dilution.

Flow cytometry and western blotting in vitro IFN‐γ study

Cells were seeded in 12‐well dishes at a seeding density of 150,000 cells/well and left to grow in 1 ml of media for 24 h. Cells were then treated with media containing a final concentration of 100 ng/ml. IFN‐γ (Cell Applications Inc., San Diego, CA, USA, Cat. no. RP1077/RP1077AF) prepared in sterile water for 24 h before analysis via flow cytometry. Flow cytometry analysis was performed by staining cells with a PE‐conjugated W6/32 antibody (Santa Cruz Biotechnology, Mississauga, ON, Canada, Cat. no. sc‐32235) to detect HLA‐I. Each sample of cells was obtained from one well of a 12‐well dish. The cells were washed using physiologic buffer solution (PBS) and then stained for 30 min (in the dark at 4 °C) using 100 μl of a 1:100 dilution of antibody in FACS buffer solution containing 1%BSA/PBS/0.05% sodium azide. Following staining, cells were washed and then resuspended in 400 μl of FACS buffer solution containing 1% BSA/PBS/0.05% sodium azide. Cells were then analyzed using a Fortessa SORP (BD Biosciences, Mississauga, ON, Canada), using gates to exclude dead cells/debris and doublets.

In vivo IFN‐γ xenograft tumor study

The in vivo assessment of IFN‐γ on xenograft tumor was performed following a previously published protocol [16]. In brief, DDEC2 xenograft tumor was generated by subcutaneously injecting the tumor lines into both flanks of adult female NOD scid gamma (NSG) mice to generate xenograft tumors in six mice. Once the tumors grew to a palpable size of 5–10 mm, mice were randomized to the treatment or control group (n = 3 mice per group). Intratumoral injection was performed using a 27‐G needle, administering either 100 μl of diluted IFN‐γ (R&D systems, Minneapolis, MN, USA, Cat. no. 285‐IF) at 125,000 units in PBS or 100 μl of PBS, followed by another injection 24 h later. Mice were sacrificed 48 h after the initial injection. The tumors were harvested for histologic and HLA‐I immunochemical evaluation.

Statistical analyses

Analyses were conducted with Stata v17 (StataCorp LLC, College Station, TX, USA) and R version 4.4.3 (https://cran.r-project.org/bin/windows/base/old/4.4.3/), with statistical significance set at p < 0.05. Categorical variables were presented as counts and percentages. Wilcoxon test was used to evaluate for differences between the immune infiltration in the two DDEC components (differentiated and undifferentiated) and these paired analyses were corrected for multiple testing using the Benjamini–Hochberg false discovery rate method.

Results

Study cohort

The study examined 14 DDECs composed of a spatially distinct differentiated endometrioid carcinoma component and an undifferentiated carcinoma component. The clinicopathologic features are summarized in Table 1. In terms of molecular types, nine (64%) were MMRd and five (36%) were NSMP; none of the tumors harbored a pathogenic POLE exonuclease domain mutation or showed a mutated p53 staining pattern by immunohistochemistry. All 14 DDECs showed core SWI/SNF inactivation with six (43%) showing co‐loss of ARID1A and ARID1B, six (43%) showing loss of SMARCA4, and two (14%) showing loss of SMARCB1 exclusively in the undifferentiated carcinoma component of the tumor.

Dedifferentiation is associated with changes in spatial cancer immunophenotype with diminished HLA‐I expression

The SCI of the differentiated endometrioid carcinoma and the undifferentiated carcinoma components from the 14 DDECs were analyzed separately, with detailed results shown in Table 2. The endometrioid carcinoma component showed an inflamed SCI in 10 of 14 cases (71.5%), followed by a desert SCI in four cases 21% (28.5%), while the undifferentiated carcinoma component showed an excluded SCI in eight of 14 cases (57%) and inflamed SCI in five of 14 cases (36%) followed by desert SCI in one case (7%) (p = 0.005). Dedifferentiation was accompanied by changes in SCI in nine of 14 DDECs (64%), most commonly from an inflamed to an excluded SCI (six cases), followed by two cases from desert to excluded SCI and one case from desert to inflamed SCI (Figure 1).

Table 2.

Tumor immune landscape analysis of dedifferentiated endometrial carcinomas (DDECs).

Case number	Tumor component	CD20+ cells	CD3+ cells	PT CD8+ cells	IE CD8+ cells	PD‐L1+ and CD163+ cells	SCI pattern	Tumor HLA‐I
1	DC	Low	Low	Low	High	Low	Inflamed	Subclonal loss
	UC	Low	High	High	High	High	Inflamed	Subclonal loss
2	DC	High	High	High	High	High	Inflamed	Complete loss
	UC	Low	High	High	Low	High	Excluded	Complete loss
3	DC	Low	Low	Low	High	High	Inflamed	Present
	UC	High	High	High	High	High	Inflamed	Complete loss
4	DC	Low	Low	Low	Low	Low	Desert	Subclonal loss
	UC	Low	High	High	High	High	Inflamed	Complete loss
5	DC	High	High	High	High	High	Inflamed	Complete loss
	UC	Low	High	High	Low	High	Excluded	Complete loss
6	DC	Low	Low	Low	Low	Low	Desert	Complete loss
	UC	Low	High	Low	Low	High	Excluded	Complete loss
7	DC	High	High	High	High	Low	Inflamed	Complete loss
	UC	High	High	High	Low	High	Excluded	Complete loss
8	DC	Low	Low	Low	Low	Low	Desert	Subclonal loss
	UC	Low	Low	Low	Low	High	Excluded	Complete loss
9	DC	High	High	High	High	Low	Inflamed	Present
	UC	High	High	High	Low	Low	Excluded	Complete loss
10	DC	Low	Low	Low	Low	Low	Desert	Present
	UC	Low	Low	Low	Low	Low	Desert	Subclonal loss
11	DC	Low	Low	Low	High	Low	Inflamed	Present
	UC	Low	High	High	Low	High	Excluded	Complete loss
12	DC	Low	Low	High	High	Low	Inflamed	Present
	UC	Low	Low	High	High	High	Inflamed	Present
13	DC	Low	Low	Low	High	Low	Inflamed	Present
	UC	Low	High	High	High	High	Inflamed	Present
14	DC	High	High	High	High	Low	Inflamed	Complete loss
	UC	Low	High	High	Low	Low	Excluded	Subclonal loss

DC, differentiated component; IE, intraepithelial; PT, peri‐tumoral; SCI: spatial cancer immunophenotype; UC: undifferentiated component.

Figure 1.

Dedifferentiation in endometrial carcinoma is associated with changes in spatial cancer immunophenotypes (SCI), most frequently from an inflamed SCI in the well‐differentiated component (DC) to an excluded SCI in the undifferentiated component (UC). Created with BioRender.com.

With respect to the individual immune microenvironment markers, dedifferentiation was associated with increased stromal M2 macrophage infiltrates (p = 0.008), increased peri‐tumoral CD8⁺ T‐cell infiltrates (p = 0.026), increased PD‐L1 expression in the stromal mononuclear immune cell (p = 0.002), and decreased tumor HLA‐I expression (p = 0.027), with a trend towards a decrease in intraepithelial CD8+ T‐cell infiltrates (p = 0.216) (Figure 2). Complete or subclonal loss of HLA‐I expression was observed in 86% of the undifferentiated carcinoma component compared with 57% in the differentiated carcinoma component. An association between SCI and HLA‐I status was observed: differentiated or undifferentiated carcinoma components showing inflamed SCI were more likely to retain HLA‐I (47%), whereas differentiated or undifferentiated carcinoma components showing excluded SCI consistently showed complete or subclonal loss of HLA‐I. Overall, these changes indicate that dedifferentiation is associated with an immune‐cold shift characterized by downregulation of antigen presentation, alongside an increase in immunosuppressive stromal macrophages and PD‐L1 expressing mononuclear immune cells. Among the nine MMRd DDECs that were eligible for immune checkpoint inhibitors (ICIs), four showed excluded SCI with a complete or subclonal loss of tumor HLA‐I expression and the presence of stromal PD‐L1 expressing mononuclear immune cells in the undifferentiated carcinoma.

Figure 2.

Dedifferentiation in endometrial carcinoma is associated with changes in tumor human leukocyte antigen class I (HLA‐I) expression and stromal immune infiltrates. (A) Representative dedifferentiated carcinoma showing complete loss of HLA‐I in the undifferentiated component (UC) compared with intact HLA‐I expression in the differentiated component (DC) that corresponds to a decrease in intraepithelial CD8⁺ T‐cell infiltrates and increased stromal PD‐L1 expression. Scale bar, 50 μm. (B) Boxplot graph illustrating the changes in immune marker expression between differentiated component and undifferentiated component of DDECs. The boxplot shows the IQR with median while the whiskers' length was set to 1.5 × interquartile range. Mac, macrophage. Bold p values indicate significance (Wilcoxon signed rank test).

Patient‐derived DDEC cancers show loss of HLA‐I expression

With the observed downregulation of HLA‐I accompanying tumor dedifferentiation in MMRd DDEC, we subsequently evaluated HLA‐I expression in two established patient‐derived cell and xenograft tumor models of MMRd DDEC—DDEC1 (SMARCA4 inactivated) and DDEC2 (ARID1A/ARID1B co‐inactivated). By IHC, similar to the findings from the respective parental tumors, DDEC1 and DDEC2 xenograft tumors showed subclonal and complete loss of HLA‐I expression (Figure 3A). Based on the published whole exome sequencing analyses for DDEC1 and DDEC2, no known pathogenic or likely pathogenic mutations involving class I heavy chain genes (HLA‐A, HLA‐B, HLA‐C), B2M, and the TAP‐encoding genes (TAP1, TAP2, and TAPBP), which regulate HLA‐I expression, were identified (supplementary material, File S1) [15]. At the epigenetics level, there was no evidence of hypermethylation involving class I heavy chain genes by DNA methylation array analysis in either DDEC1 or DDEC2 [15]. These findings suggest that the attenuated HLA‐I expression observed in these DDEC models was not due to genomic inactivation or promoter hypermethylation of antigen presentation machinery (APM) genes.

Figure 3.

Loss of human leukocyte antigen class I (HLA‐I) expression in dedifferentiated carcinoma can be restored by IFN‐γ treatment. (A) Dedifferentiated endometrial carcinoma (DDEC1) parental and xenograft tumors both show subclonal loss of HLA‐I expression by immunohistochemistry, while DDEC2 parental and xenograft tumors both show a complete loss of HLA‐I expression. Scale bar, 50 μm. (B) IFN‐γ resulted in significant increase in surface HLA‐I expression by flow cytometry in DDEC2 (n = 3) but not in DDEC1 (n = 4). The corresponding western blotting image shows reduced JAK1 expression in DDEC1 compared with DDEC2 (representative blot from a replicate of three blots), and a complete lack of pSTAT1 at baseline and with IFN‐γ treatment in DDEC1 whereas IFN‐γ induced an increase in pSTAT1 level above baseline level activity in DDEC2 (representative blot from a replicate of three blots). Representative images of the ex vivo 3D spheroids for DDEC1 and DDEC2 with control and with IFN‐γ treatment are also depicted. Scale bar, 50 μm. (C) IFN‐γ intratumoral injection restored tumor HLA‐I expression in vivo in DDEC2 xenograft tumor (n = 3). Scale bar, 50 μm.

IFN‐γ can restore HLA‐I expression in DDECs

We next examined whether the HLA‐I loss observed in these DDEC cells can be restored with IFN‐γ treatment. We evaluated surface HLA‐I expression by flow cytometry in 3D spheroid models of DDEC1 and DDEC2. IFN‐γ treatment for 24 h at a concentration of 100 ng/ml resulted in about 3.5‐fold increase in surface HLA‐I expression in DDEC2 (p < 0.05) but no detectable changes in surface HLA‐I level in DDEC1 (Figure 3B). Further review of the reported sequencing data for DDEC1 showed that it harbored two different JAK1 frameshift mutations—c.2580dup (p.P861fs) and c.1289dup (p.L431fs)—whereas no known pathogenic mutations involving the JAK/STAT pathway (i.e. JAK1/2, INFGR1/2, and STAT1) were identified in DDEC2 (supplementary material, File S1). At the protein level, the two inactivating JAK1 mutations appear to result in decreased JAK1 protein expression in DDEC1 (Figure 3B), indicating that these two mutations likely affected the same allele. The lack of response to IFN‐γ as reflected by the lack of pSTAT1 expression with IFN‐γ treatment indicates defective IFN‐γ receptor (IFNGR) signaling in the JAK1‐mutated or JAK1‐deficient DDEC1 (Figure 3B).

We then evaluated whether the HLA‐I expression can be restored in vivo in DDEC2 by intratumoral IFN‐γ injection with two daily injections of 125,000 units of IFN‐γ. Diffuse and strong HLA‐I expression was observed in IFN‐γ‐treated DDEC2 xenograft tumor 48 h after injection in contrast to the absent HLA‐I expression seen in the vehicle control‐treated DDEC xenograft tumor (Figure 3C). These findings indicate the potential therapeutic utility of IFN‐γ for restoring HLA‐I expression in immune‐excluded DDECs lacking mutations in the IFN‐γ signaling pathway.

Discussion

In this study, we found that dedifferentiation in endometrial carcinoma is associated with significant changes in its tumor immune microenvironment, most frequently showing a transition from an inflamed SCI to an excluded SCI landscape. Moreover, dedifferentiation is frequently characterized by concurrent downregulation of tumor HLA‐I expression, decrease of tumor infiltrating CD8⁺ T cells, increase in peri‐tumoral CD8⁺ T cells, and increase in tumor stromal PD‐L1 expression. The impaired antigen presentation and upregulated immune check point protein (PD‐L1) by stromal macrophages appear to work in concert to suppress cytotoxic T‐cell‐mediated killing of tumor cells [17], hence resulting in diminished intraepithelial CD8⁺ T‐cell infiltrates, with the CD8⁺ T cells localized instead at the periphery of the tumor. Given that about two‐thirds of DDECs are MMRd/MSI‐H and eligible for ICI treatment, downregulated or absent HLA‐I expression in the tumor cells would be expected to compromise the effectiveness of ICI as it is a known acquired mechanism of resistance to ICI [18, 19]. Here, we show that loss of HLA‐I expression in DDECs can be reversed by IFN‐γ in a 3D spheroid model and in vivo xenograft tumor models of DDECs harboring a functional IFNGR signaling pathway. The effectiveness of IFN‐γ in restoring HLA‐I expression is comparable to that observed in a murine tumor model of spontaneous metastatic prostate cancer [20], and clinically in a phase 0 study of patients with myxoid/round cell liposarcoma and synovial sarcoma [21]. In addition to restoring HLA‐I expression in the tumor cells, there is existing evidence to suggest that IFN‐γ and ICI can also shift the stromal macrophages to a more proinflammatory M1 phenotype [22], which would also counter the increase in stromal M2 macrophages associated with dedifferentiation.

HLA‐I is an essential component of the APM that presents tumor‐specific neoantigens to cytotoxic T cells. Loss of HLA‐I expression can occur by genomic inactivation of APM component proteins, by inactivating JAK1/JAK2 mutations that block the IFNGR pathway signaling, by epigenetic silencing of APM components, or transcriptional downregulation. In MMRd/MSI‐H endometrial carcinomas, inactivating JAK1 (typically frameshift) mutations occur in 38% of cases based on data from the Cancer Genome Atlas Program (TCGA) and a more recent study by Manning‐Geist et al [23, 24], though only 16% harbor two or more inactivating mutations that may be biallelic [24], while inactivating mutations involving other components of the IFNGR signaling pathway occurs rarely. Even though two different inactivating JAK1 mutations were present in DDEC1 which are expected to result in either a complete loss of protein expression or the expression of a truncated protein, there was still a reduced level of JAK1 protein present (at the expected 130 kDa molecular weight), indicating that the two inactivating JAK1 mutations affected the same allele, rendering DDEC1 JAK1‐deficient. However, IFN‐γ was unable to activate the IFNGR pathway as reflected by the absent pSTAT1 expression, which indicates in this context that the reduced level of JAK1 impaired the IFNGR pathway signaling. This notion is supported by prior findings that partial JAK1 deficiency impairs IFNGR signaling in human macrophages, B cells and dermal fibroblasts [25]. Moreover, inactivating JAK1 mutations have been reported by Gulhan et al to be associated with de novo resistance to ICI in MMRd endometrial cancer [26], as two inactivating mutations in JAK1 were more frequently observed in ICI non‐responders. However, while they speculated that these may reflect biallelic inactivation of JAK1, protein expression analysis was not performed. It is thus unclear whether the presence of two inactivating JAK1 mutations observed in MMRd endometrial cancer results in the complete loss of JAK1 in some cases but only partial loss of JAK1 in other cases, as observed in DDEC1. Future studies are needed to provide further insights.

In immune‐rich tumors such as MMRd endometrial carcinomas, loss of HLA‐I represents a major resistance mechanism to ICI by enabling tumor cells to escape CD8⁺ T‐cell‐mediated recognition and restoration of tumor HLA‐I expression for instance by IFN‐γ can sensitize the tumor to respond to ICI [27, 28]. IFN‐γ is currently approved for the treatment of chronic granulomatous disease and severe malignant osteopetrosis [29]. A recent study has established clinically tolerable dosing for patients with advanced stage solid tumors where IFN‐γ was self‐administered subcutaneously every other day [30]. At present, there is no approved indication for the use of IFN‐γ either as a single agent or as part of a combination therapy in the treatment of cancer. Clinical studies to date that evaluated the efficacy of IFN‐γ in the treatment of advanced stage solid tumors including ovarian cancer, urinary bladder cancer, breast cancer, melanoma, and soft tissue sarcoma have yielded mixed results [31], though it is plausible that more biologically informed pre‐selection of patients based on tumor immune microenvironment characteristics may be needed to identify patients who would derive clinical benefits.

There were a number of limitations of the present study that warrant further investigation. Firstly, we focused on SWI/SNF‐inactivated DDEC given its more aggressive clinical behavior than SWI/SNF‐proficient DDEC [1]; hence it is not clear whether the observation made here is generalizable to all DDECs. Secondly, given there are multiple mechanisms for SWI/SNF inactivation (i.e. ARID1A/ARID1B co‐inactivated, SMARCA4 inactivated, and SMARCB1 inactivated) in DDEC, future studies evaluating a larger cohort are needed to determine whether dedifferentiation‐associated immune landscape changes vary across different mechanisms of SWI/SNF inactivation. Thirdly, our study used an established intratumoral IFN‐γ injection protocol in the in vivo xenograft tumor model, while clinically IFN‐γ is typically administered subcutaneously. Although subcutaneous IFN‐γ injection has demonstrated clinical efficacy in restoring tumor HLA‐I expression in sarcomas [21], this remains to be confirmed clinically in human DDECs.

In summary, our findings demonstrate that dedifferentiation is associated with immune evasion mechanisms that include a complete or subclonal loss of tumor HLA‐I expression coupled with increased stromal immune cell PD‐L1 expression and M2 macrophage infiltrates. The loss of tumor HLA‐I expression can be restored using IFN‐γ in tumors with an intact INFGR signaling pathway. This indicates that a tailored immunotherapy approach of combining ICI with IFN‐γ may be clinically more effective in eradicating tumor cells, particularly for MMRd and SWI/SNF‐inactivated DDEC with attenuated tumor HLA‐I expression and absent JAK1 mutations.

Author contribution statement

ADL, PL, MK and C‐HL were responsible for conceptualization. ADL, MW, MG, CAC, PPK, AT, JK, HK, FK and MK contributed to data curation. ADL, MW, MG, AT, FK and GT performed formal analysis. ADL, MC, CC, GT, PL and C‐HL were responsible for funding acquisition. ADL, MW, MG, CAC, PPK, AT, JK, HK, FK, MC, YF, CC, GT, PL, MK and C‐HL contributed to investigation. ADL, MW, MG, CAC, PL, MK and C‐HL were responsible for methodology. ADL, YF, PL and C‐HL contributed to supervision. ADL, MW, AT, MK and C‐HL contributed to visualization. ADL, MK and C‐HL wrote the original draft. MW, MG, CAC, PPK, AT, JK, HK, FK, MC, YF, CC, GT and PL contributed to review and editing of the manuscript. All authors approved the final version of the manuscript.

Supporting information

File S1. Mutations of antigen presentation genes and JAK/STAT pathway genes in DDEC1 and DDEC2

Table S1. List of antibodies used for immunohistochemistry analysis

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.