NKG2A–HLA-E and TIM3–galectin 9 pathways promote immune inhibition and represent therapeutic vulnerabilities in pancreatic ductal adenocarcinoma

Abstract

Background

The roles of NKG2A–HLA-E and TIM3–galectin 9 immune checkpoint pathways in pancreatic ductal adenocarcinoma (PDAC) progression remain incompletely characterized. This study investigates their contributions to PDAC and therapeutic potential.

Methods

The expressions of ligands HLA-E and galectin 9 and receptors NKG2A and TIM3 were analyzed through bioinformatics, immunohistochemistry, and flow cytometry. Effects of HLA-E and galectin 9 overexpression on pancreatic epithelial cells were assessed by Transwell, wound healing, and CCK-8 assays. Ligand–receptor interactions and their impact on lymphocyte function were examined by multiplex immunofluorescence, single-cell RNA-sequencing, and functional assays.

Results

The expression of ligands HLA-E and galectin 9 was elevated in PDAC cancer tissues compared with both normal tissues and benign lesions. The overexpression of HLA-E enhanced the migratory and invasive ability of pancreatic epithelial cells. Moreover, the high expression of HLA-E and galectin 9 correlated with worse overall survival in PDAC patients. Mechanistically, increased NKG2A expression in both tumor stroma and parenchyma was associated with impaired function of tumor-infiltrating T cells and NK cells. Spatial analyses further revealed colocalization of NKG2A⁺ T cells with high HLA-E-expressing tumor regions, indicating an active and localized immunosuppressive circuit. Single-cell profiling showed that NKG2A⁺ and TIM3⁺ tumor-infiltrating T cells exhibited exhausted signatures, particularly when co-expressing PD-1. Importantly, dual blockade of NKG2A or TIM3 with PD-1 enhanced the anti-tumor response of CD8⁺ and CD4⁺ T cells, mediated by SHP-1 inhibition and ERK activation.

Conclusions

This study identifies NKG2A–HLA-E and TIM3–galectin 9 pathways as key mechanisms in PDAC progression and immune inhibition and provides a mechanistic rationale for combining their blockade with PD-1-PD-L1 blockade as a potential therapeutic strategy.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00262-026-04479-9.

Introduction

Pancreatic tumors encompass both malignant and benign neoplasms, with pancreatic ductal adenocarcinoma (PDAC) representing the most prevalent and aggressive form, accounting for the majority of pancreatic cancer cases [1]. In contrast, serous cystadenoma (SCA) is a benign lesion with minimal malignant potential, as only about 3% of cases process to malignancy [2]. Other precursor lesions of PDAC include pancreatic intraepithelial neoplasia (PanIN) and intraductal papillary mucinous neoplasm (IPMN), both of which are well-established risk factors for PDAC development [3]. PanIN is the most common precancerous lesion leading to PDAC [4], while invasive carcinoma is detected in approximately 40% of surgically resected IPMN cases [5]. Additionally, pancreatobiliary-type ampullary adenocarcinoma (AAC), a malignancy arising from the duodenal ampulla, exhibits histological and clinical similarities to PDAC [6, 7], warranting its inclusion in this study [8].

With the incidence of PDAC continuing to rise, the development of novel therapeutic approaches represents an urgent unmet clinical need [9]. Although immune checkpoint inhibitors (ICI) targeting CTLA-4, PD-1, and PD-L1 have revolutionized cancer treatment, their application in pancreatic cancer has been largely ineffective [10, 11], underscoring the necessity of identifying and validating alternative immune checkpoint targets for this aggressive malignancy.

Emerging evidence highlights NKG2A as a key inhibitory receptor on NK cells and CD8⁺ T cells. Studies suggest that the NKG2A–HLA-E axis enables PDAC circulating tumor cells to evade NK cell surveillance, correlating with poor patient outcomes [12, 13]. Furthermore, NKG2A blockade has been shown to enhance IFN-γ production by T cells co-cultured with autologous pancreatic tumor organoids [14]. However, the differential expression of NKG2A and HLA-E in PDAC tumor versus normal tissues and in pancreatic malignant versus benign tumors, as well as the impact of the NKG2A–HLA-E axis in combination with the PD-1-PD-L1 axis on T cell and NK cell function, remains unexplored.

Similarly, TIGIT and TIM3 contribute to immune exhaustion and immunosuppression in the tumor microenvironment [15]. Their ligands CD155 and galectin 9 are overexpressed in human PDAC tumor tissue [16–19]. However, there is still a lack of comparison in the expression of TIM3 between pancreatic tumor tissue and normal tissue. Preclinical studies demonstrate that the TIGIT-CD155 pathway promotes tumor immune escape in mouse models [18], while disruption of the TIM3–galectin 9 pathway restricts pancreatic tumor progression in mouse models by restoring anti-tumor immunity [20]. However, further validation and investigation in human samples are needed.

To address the critical gaps in understanding the roles of the NKG2A–HLA-E, TIGIT–CD155, and TIM3–galectin 9 axes during the transition from precursor lesions to invasive PDAC, we here present the first systematic and functional characterization of these pathways. Our study provides three major advances: (1) the first comparative analysis of immune checkpoint ligand expression across the pancreatic benign-to-malignant spectrum, demonstrating that HLA-E overexpression directly enhances the migration and invasion of pancreatic epithelial cells; (2) the first spatial evidence revealing localized immunosuppression mediated by NKG2A–HLA-E interaction within the PDAC tumor microenvironment; and (3) the first investigation of PD-1 co-expression with NKG2A or TIM3 in PDAC patient-derived T cells, showing that dual blockade inhibits SHP-1 and activates ERK signaling pathway to restore anti-tumor function. Collectively, this work elucidates the roles of these checkpoints in PDAC progression and provides a mechanistic rationale for their therapeutic targeting.

Methods

Patients

Patients (n = 55) with PDAC, pancreatobiliary-type AAC, IPMN, PanIN, or SCA who underwent surgical resection or tumor biopsy were registered at Department of General Surgery, in the Seventh Affiliated Hospital of Sun Yat-sen University. The study period of this research was from 2022 to 2026. Patients with intestinal-type or mixed-type ampullary adenocarcinoma, metastases from other tumors, or inflammatory pseudotumor were excluded based on pathological evaluation. From some of the patients, formalin-fixed and paraffin-embedded (FFPE) samples were obtained from the archive of Department of Pathology. From some of the patients, fresh tissues from tumor and adjacent normal pancreas and peripheral blood were collected, and mononuclear leukocytes were isolated. The clinical characteristics and pathologic information of the patients are summarized in Supplementary Table 1. All patients are Asians, and tumor staging was performed according to the AJCC 9th edition classification system. The study was performed in accordance with the 1964 Declaration of Helsinki and its later amendments and approved by the local medical ethics committee (KY-2023-120-01, KY-2025-379-01, and KY-2026-198-01) of the Seventh Affiliated Hospital of Sun Yat-sen University, and informed consent from all patients was obtained before tissue and blood donation.

Immunohistochemistry (IHC)

FFPE tissue slides (4 µm) were incubated overnight at 4 °C with the primary antibodies (Supplementary Table 2): mouse anti-human HLA-E (1:200), rabbit anti-human CD155 (1:300), or goat anti-human galectin 9 (1:200). Two independent researchers scored the slides blinded to clinical outcome based on the intensity of staining (0 [none], 1 [weak], 2 [intermediate], 3 [strong]) and the percentage of positive cells (0 [none], 1 [1–25%], 2 [26–50%], 3 [51–75%], 4 [76–100%]), by five randomly and averagely selected 200× magnification fields in two representative areas, respectively: Within the tumor and within the adjacent pancreas, the expression levels of immune checkpoints were calculated by multiplying the value of staining intensity by the value of positive cell percentage. The final score per sample was calculated by taking the averages of two independent researchers to obtain an H-score ranging from 0 to 12.

Multiplex immunohistochemistry (mIHC)

Multiplex immunohistochemistry was performed in the FFPE tissue samples. There were three panels examined in this study. The primary antibodies of Panel 1 were used in the following order (Supplementary Table 2): rabbit anti-human CD3 (1:150), rabbit anti-human CD56 (1:1000), and rabbit anti-human IFN-γ (1:500), which were labeled with IF647, IF594, and IF546, respectively, and the cell nucleus was labeled with DAPI. After staining, the sections were scanned using a KFBIO device (KF-FL-120; China). Subsequently, the same sections were subjected to HE staining. Tumor and adjacent normal areas were identified based on the HE staining results, and representative regions (0.25 mm² each) were selected from the corresponding mIHC images for cell counting. The primary antibodies of Panel 2 were used in the following order (Supplementary Table 2): rabbit anti-human NKG2A (1:1000), rabbit anti-human CD3 (1:200), rabbit anti-human CD56 (1:200), mouse anti-human PanCK (1:200), and mouse anti-human HLA-E (1:200), which were labeled with TSA570, TSA480, TSA520, TSA700, and TSA620, respectively. The primary antibodies of Panel 3 were used in the following order (Supplementary Table 2): rabbit anti-human CD45 (1:200), rabbit anti-human CD3 (1:200), rabbit anti-human CD56 (1:200), and mouse anti-human PanCK (1:200), which were labeled with TSA570, TSA480, TSA520, and TSA700, respectively. Nuclei were counterstained with DAPI for all the panels. Image analysis was performed using HALO Image Analysis software (Indica Labs, version v3.3.2541). We conducted quantitative analysis of immune cell infiltration and molecular expression by selecting rectangular regions (approximately 0.11 mm² each) from tumor parenchyma, stroma, and adjacent normal tissues. HLA-E-high-expressing and HLA-E-low-expressing regions (approximately 1–2 mm² each) were selected within tumor areas. For spatial analysis, we selected ≥ 3 mm² rectangular regions from tumor areas.

Cell preparation

Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll density gradient centrifugation. Single-cell suspensions from tumors and adjacent normal pancreas were obtained by tissue digestion with 0.125 mg/mL collagenase IV (Sigma-Aldrich, Cat: C5138-100 mg), 0.2 mg/mL of DNAse I (Roche, Cat: 10104159001), and 1000 units/mL hyaluronidase (Sigma-Aldrich, Cat: H6254-500MG) in Hanks’ balanced salt solution with Ca²⁺ and Mg²⁺ (Servicebio, Cat: G4204-500ML). Viability was determined by trypan blue exclusion.

Flow cytometry analysis

Fresh or cryopreserved PBMC and mononuclear leukocytes isolated from tissues were analyzed for expression of surface and intracellular markers using specific antibodies (Supplementary Table 3). Dead cells were excluded by using Zombie UV™ Fixable Viability Kit (BioLegend, Cat: 423107). For intracellular staining, cells were fixed and permeabilized using the FoxP3 staining buffer set (eBioscience). Cells were measured using a CytoFLEX LX flow cytometer (Beckman Coulter) and analyzed using FlowJo software (version 10.0, LLC, BD).

Ex vivo T cell activation assay

Cryopreserved PBMC or tumor-derived mononuclear leukocytes from 13 patients with PDAC or AAC were stimulated with 25 ng/mL anti-human CD3 antibody (BioLegend, Cat: 317326, clone OKT3) and 1 µg/mL anti-human CD28 antibody (BioLegend, Cat: 302934, clone CD28.2). Cell cultures were performed in the presence or absence of 10 µg/mL monoclonal antibodies against human PD-1 (pembrolizumab provided by Department of Oncology), NKG2A (monalizumab), TIM3, TIGIT, or combinations of these antibodies (Supplementary Table 4). After four days, culture supernatant was collected and frozen at − 20 °C until secretion of IFN-γ was quantified by enzyme-linked immunosorbent assay (ELISA, Thermo Fisher, Cat: 88–7316). Cultured cells were harvested and stained with antibodies (Supplementary Table 3).

Co-culture assay of PDAC tumor cells and patient-derived immune cells

HLA typing of patients was performed with HLA-A2 and HLA-A24 antibodies (Supplementary Table 3) to select HLA-matched pancreatic cancer cell lines for co-culture. Cryopreserved PBMC were preactivated for 3 days with 25 ng/m: anti-human CD3 antibody, 1 µg/ml anti-human CD28 antibody, and 4 ng/m: IL-2 (Gibco, Cat: 200-02-50UG). PANC-1 cells (Procell system, Cat: CL-0184) and MIA PaCa-2 cells (Procell system, Cat: CL-0627) were pre-treated with 100 ng/mL IFN-γ for 2 days. On the day of co-culture, pre-treated PBMC were co-cultured with pre-treated PANC-1 or MIA PaCa-2 cancer cells at an effector cell: target cell (E/T) ratio of 40:1 in the absence or presence of anti-PD-1, anti-NKG2A, anti-TIM3, anti-TIGIT, or combinations of these antibodies (Supplementary Table 4). After 4 days of co-culture, cells were harvested and stained with antibodies (Supplementary Table 3). Counting beads (Thermo Fisher, Cat: 01–1234-42) were added to the cell suspension for calculating the number of tumor cells. Dead cells were excluded using DAPI.

Detection of phosphoproteins by flow cytometry

To investigate the signaling pathways downstream of NKG2A, TIM3, and PD-1 blockade, Jurkat cells (3 × 10⁵ cells/well) were seeded into 96-well plates and stimulated with anti-CD3 (25 ng/mL) and anti-CD28 (1 µg/mL) antibodies, in the presence or absence of single or combined blocking antibodies against human PD-1, NKG2A, or TIM3, along with recombinant human HLA-E, galectin-9, and PD-L1 proteins (1 µg/mL each; MCE, Cat# HY-P77758, HY-P70535, HY-P73361). Cryopreserved PBMC from PDAC patients were similarly stimulated with anti-CD3/CD28 antibodies, with or without the indicated blocking antibodies. After three days, cells were washed with PBS. Dead cells were excluded using the Zombie UV™ Fixable Viability Kit. Cells were fixed with FluoroFix Buffer (BioLegend, Cat: 420801) for 20 min at 37 °C, followed by surface staining with anti-CD3 and anti-CD45 antibodies to identify T cells. After surface staining, cells were permeabilized with True-Phos Perm Buffer (BioLegend, Cat: 425401) on ice for 10 min and then incubated with anti-pSHP-1, anti-pERK1/2, and anti-pAKT antibodies for 1 h in the dark. Flow cytometric acquisition and analysis were performed as described above.

Cell transfection

Overexpression RNA (oeRNA) vectors targeting HLA-E and galectin 9, along with corresponding control vectors, were purchased from MiaoLing plasmid platform and Guangzhou IGE Biotechnology Co., Ltd. These vectors were transfected into hTERT‑HPNE and MIA PaCa-2 cells using the Lipofectamine™ 3000 kit (Invitrogen, Cat: L3000-008). Subsequent experiments were performed 24–72 h post-transfection.

Western blot

Cells were lysed using RIPA lysis buffer (Beyotime, Cat: P0013C) for protein extraction, and protein concentrations were determined using a BCA protein assay kit. Proteins were separated by SDS-PAGE using a rapid gel preparation kit (Beyotime, Cat: P0697S), with β-tubulin serving as a loading control. Following electrophoresis, proteins were transferred onto PVDF membranes (Millipore, Cat: ISEQ00010) and blocked with non-fat dry milk (Servicebio, Cat: GC310001) for 4–6 h. The membranes were then incubated overnight at 4 °C with primary antibodies against HLA-E, galectin 9, and β-tubulin (Supplementary Table 2). After three washes with TBST (5 min each), the membranes were incubated with horseradish peroxidase HRP-conjugated goat anti-rabbit or goat anti-mouse secondary antibodies. Following secondary antibody incubation, chemiluminescent substrate was applied, and signals were visualized using the ChemiDoc system (Bio-Rad).

Cell proliferation assay

MIA PaCa-2 and hTERT‑HPNE cells in the logarithmic growth phase were seeded onto 96-well plates at a density of 5000 cells per well. After 24 h of culture, transfection reagents and corresponding plasmids were added to perform transfection. Cell viability was assessed at four time points (0, 24, and 48 h after transfection) using a cell viability detection kit (Accurate Biology, Cat: AG51006) for the CCK-8 assay. Absorbance was measured at 450 nm using a microplate reader (BioTek).

Cell scratch assay

Several parallel lines were pre-drawn on the bottom of six-well plates as reference marks. MIA PaCa-2 and HPNE cells were seeded into the plates and cultured until approximately 90% confluence. Linear scratches were then made along the reference lines using a pipette tip. After washing with PBS, serum-free medium containing transfection reagents and plasmids was added. The same scratched areas were photographed at 0 and 12 h post-scratching, and the wound area was quantified using Fiji software [21].

Transwell invasion assay

Matrigel was diluted with water at a ratio of 1:7, and 70 μL of the diluted Matrigel was added to each Transwell chamber (Corning, Cat. No. 353090) and allowed to dry before use. MIA PaCa-2 and hTERT‑HPNE cells, which had been transfected for 24 h in advance, were seeded into the upper chamber at a density of 10,000 cells per well in serum-free medium. The lower chamber was filled with medium containing 20% fetal bovine serum. After 48 h of culture at 37 °C, the Transwell inserts were removed, and the cells were fixed with 4% paraformaldehyde, stained with 0.1% crystal violet, and quantified under a microscope.

Public database analysis

“Differential Visualization” module of Xiantao tool (https://www.xiantaozi.com), “Expression Analysis” module of GEPIA2 (http://gepia2.cancer-pku.cn/), [22] “Gene_Corr” module, and “Immune” module in TIMER 2.0 (http://timer.cistrome.org/) based on TCGA-PAAD RNA-seq data [23], “Survival Curve” module of Xiantao tool with automated log-rank testing and optimal cutoff determination and “Immune_Outcome” module of TIMER 3 (https://compbio.cn/timer3/) [24] were utilized. Specifically, TIMER2.0 estimates immune cell abundance by integrating six algorithms spanning two complementary categories—deconvolution-based methods (TIMER, CIBERSORT, EPIC, quanTIseq) and gene signature-based methods (xCell, MCP-counter)—applied to bulk RNA-seq data. Then, Spearman correlation was used to evaluate the relationship between estimated immune cell abundance and expression of the target genes. The output represents the relative proportion or estimated abundance of a given immune cell type within the bulk RNA-seq sample.

Single-cell RNA-sequencing data analysis

Single-cell RNA-sequencing (scRNA-seq) data from PDAC patient samples were obtained from the Gene Expression Omnibus (GEO) database under accession number GSE254250 (https://www.ncbi.nlm.nih.gov/geo/). Among the nine tumor samples, five (#249, #262, #282, #309, and #418) were selected for further analysis based on data availability and the presence of major T cell subsets. Data processing was performed using the SeekSoul Online cloud platform (version 1.2, Beijing SeekGene BioScience Co., Ltd.) following standardized protocols. The analytical pipeline comprised the following steps: (1) Quality control (QC): Low-quality cells were filtered based on the following thresholds: unique molecular identifier (UMI) counts < 200 or > 10,000, gene counts (nFeature_RNA) < 100 or > 30,000, and mitochondrial gene expression > 20%. After applying these QC criteria, a total of 35,287 cells were retained for downstream analyses. (2) Normalization and dimensionality reduction: Data were normalized using the default Seurat workflow (2000 features by default), and principal component analysis (PCA) was applied for dimensionality reduction. (3) Clustering and cell type annotation: Cell types were annotated based on the criteria established in the original study [25], with manual refinement conducted according to marker gene expression profiles and literature-derived signatures. (4) Visualization: Uniform manifold approximation and projection (UMAP) plots were generated using the cloud platform’s built-in tools.

Gene set scoring analysis

Gene set scoring analysis was performed using gene sets associated with T cell activation and exhaustion. The activation gene set comprised CD69, CD38, IL2RA, TNFRSF4, TNFRSF9, TBX21, MKI67, NR4A1, and HLA-DRA; the exhaustion gene set consisted of CTLA-4, LAG3, TIGIT, CD244, ENTPD1, TOX, TOX2, EOMES, PRDM1, NR4A1, HAVCR2, CXCL13, PDCD1, CXCR6, CD27, CD63, ITGAE, HLA-DRA, CD38, CD82, and STMN1. The analysis was conducted using the “Feature Analysis” module of the SeekSoul Online platform, which implements gene set scoring via the AddModuleScore function from the Seurat package.

Differential expression and enrichment analysis

Differentially expressed genes (DEGs) were identified using the “FindMarkers” function with the following criteria: |log₂(fold change)|> 0.25 and adjusted P < 0.05, as determined by Wilcoxon rank-sum tests. DEGs meeting these thresholds were subjected to overrepresentation analysis (ORA) for pathway enrichment. Enrichment analysis focused on Gene Ontology (GO) terms, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways, Reactome, and Hallmark gene sets to identify signaling pathways and biological processes enriched in specific cell populations.

Statistical analysis

Differences between matched groups of data were analyzed by either paired t test or Wilcoxon matched pairs test according to their distributions. Differences between different groups of patients were analyzed by either unpaired t test or Mann–Whitney test according to their distributions. Differences among multiple groups of data were analyzed by one-way ANOVA, two-way ANOVA, or mixed-effects analysis depending on whether there are missing values. Log rank tests were used in survival analysis. The statistical analyses were performed using GraphPad Prism software (version 10.0, GraphPad). P values lower than 0.05 were considered statistically significant (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001).

Results

Screening of immune checkpoint candidates from public databases

To identify potential immunotherapeutic targets, we systematically analyzed RNA-seq data from multiple public databases. Initial screening using the Xiantao online platform revealed significant overexpression of immune checkpoint ligands HLA-E (p = 2e−52), CD155 (PVR) (p = 7.6e−45), galectin 9 (LGALS9) (p = 1.7e−49), and PD-L1 (CD274) (p = 3.4e−15) and inhibitory receptors NKG2A (KLRC1) (p = 1.2e−19), TIGIT (p = 3.6e−49), TIM3 (HAVCR2) (p = 1.7e−47), and PD-1 (PDCD1) (p = 2e−41) in tumor tissues from patients with PDAC compared to normal pancreas from healthy individuals (Fig. 1A–H). Analytical results performed using the GEPIA2 online platform validated these findings (Supplementary Fig. 1A–H), except for NKG2A (KLRC1), PD-1 (PDCD1), and PD-L1 (CD274). Subsequently, we analyzed the correlations among these genes and their correlations with established immune checkpoints PD-1 (PDCD1) or PD-L1 (CD274) via TIMER 2.0 platform. Co-expression analysis exhibited positive associations between receptors NKG2A, TIGIT, and TIM3 (Supplementary Fig. 1I–K), as well as between ligand HLA-E and its receptor NKG2A (rho = 0.509, p = 3.53e−13) (Supplementary Fig. 1O). Notably, significant positive associations between HLA-E and PD-L1 (rho = 0.557, p = 5.73e−16) (Fig. 1I) and between NKG2A (rho = 0.625, p = 9.04e−21), TIGIT (rho = 0.808, p = 1.49e−42), TIM3 (rho = 0.613, p = 7.03e−20), and PD-1 (Fig. 1L–N) were found, suggesting potential synergistic therapeutic opportunities, especially correlation between HLA-E-NKG2A and PD-L1-PD-1 immune checkpoint pathways.

Fig. 1.

Elevated expression of HLA-E and galectin 9 is associated with adverse clinical outcome in PDAC patients. A–H Differential expression of HLA-E, B CD155 (PVR), C galectin 9 (LGALS9), D PD-L1 (CD274), E NKG2A (KLRC1), F TIGIT, G TIM3 (HAVCR2), and H PD-1 (PDCD1) between pancreatic cancer tissues (n = 179) and unpaired normal pancreatic tissues (n = 171), analyzed using the TCGA–GTEx-PAAD dataset on Xiantao tool (n = 350). I–N Correlation of expression of I HLA-E–PD-L1, J CD155–PD-L1, K galectin 9–PD-L1, L NKG2A–PD-1, M TIGIT–PD-1, (N) TIM3–PD-1, analyzed using TCGA-PAAD dataset on TIMER2.0 platform (n = 179). O–Q Correlation of expression of O NKG2A, P TIGIT, and Q TIM3 with CD8⁺ T cell abundance. R–T Correlation of expression of R NKG2A, S TIGIT, and T TIM3 with NK cell abundance. Correlations of immune checkpoints with CD8⁺ T and NK cell abundances were analyzed using TIMER and MCP-counter algorithms, respectively. U–W Kaplan–Meier survival analysis derived from Cox regression based on the TCGA-PAAD dataset (via Xiantao tool) evaluated the association between expression levels of HLA-E, CD155, or galectin 9 (high vs. low) in the tumor and overall survival in pancreatic cancer patients (n = 179). Tick marks indicate censoring. Log rank tests were used. ***p < 0.001

Furthermore, immune abundance analysis via TIMER 2.0 demonstrated that the expression of NKG2A (rho = 0.587, p = 2.47e−17; rho = 0.53, p = 7.46e−14), TIGIT (rho = 0.783, p = 6.91e−37; rho = 0.597, p = 5.26e−18), and TIM3 (rho = 0.685, p = 3.44e−25; rho = 0.451, p = 5.28e−10) showed strong positive correlations with CD8⁺ T cell (Fig. 1O–Q) and NK cell abundance (Fig. 1R–T), but no correlations with CD4⁺ T cells or B cells (data not shown), suggesting their potential immunosuppressive impact on CD8⁺ T cells and NK cells in the PDAC tumor microenvironment. Notably, HLA-E expression demonstrated significant positive correlations with CD8⁺ T cell abundance (rho = 0.655, p = 1.95e−22), NK cells (rho = 0.512, p = 6.73e−13), myeloid dendritic cells (rho = 0.679, p = 1.46e−24), and monocytes/macrophages (rho = 0.59, p = 1.63e−17), suggesting its potential multifaceted role in shaping the immune landscape of PDAC (Supplementary Fig. 1R–U), but no correlations between CD155 (PVR) or galectin 9 (LGALS9) and these immune cells were observed (data not shown). Based on these findings, we continued with immune checkpoint pathways HLA-E-NKG2A, CD155-TIGIT, and galectin 9-TIM3 for further investigation.

To elucidate the prognostic value of NKG2A–HLA-E, TIGIT–CD155, and TIM3–galectin 9 interactions in PDAC, we analyzed RNA-seq data from TCGA-PAAD dataset through Xiantao online platform. Survival analysis indicated unfavorable prognostic associations for higher expression of HLA-E (HR = 2.03, p = 0.01) and galectin 9 (HR = 1.69, p = 0.012) (Fig. 1U and W), with NKG2A (HR = 1.44, p = 0.148) and TIM3 (HR = 1.49, p = 0.096), showing a similar trend in PDAC patients (Supplementary Fig. 2 A). In contrast to the aforementioned immune checkpoint pairs, patients exhibiting higher expression of CD155 (HR = 0.6, p = 0.015) demonstrated an improved clinical outcome (Fig. 1V). Moreover, utilizing the TIMER3 online platform, we analyzed the patient survival related to immune cell abundance levels at different molecular expression thresholds in the tumor microenvironment.

(1) First, within patients exhibiting high HLA-E (HR = 0.415, p = 0.0235), high CD155 (HR = 0.224, p = 0.00855), high galectin 9 (HR = 0.543, p = 0.0432), high NKG2A (HR = 0.532, p = 0.047), high TIGIT (HR = 0.404, p = 0.049), or high TIM3 (HR = 0.404, p = 0.0387) expression in the tumor, increased CD4⁺ T cell abundance was associated with longer overall survival (Supplementary Fig. 2B–G). (2) Second, within patients displaying low CD155 (HR = 0.319, p = 0.00355), high galectin 9 (HR = 0.511, p = 0.0312), high NKG2A (HR = 0.473 p = 0.0404), low TIGIT (HR = 0.307, p = 0.00306), or low TIM3 (HR = 0.38, p = 0.00672) expression in the tumor, elevated NK cell abundance was correlated with better prognosis (Supplementary Fig. 2H–M). (3) Third, within patients showing low HLA-E (HR = 0.235, p = 0.0485), low CD155 (HR = 0.361, p = 0.0328), low galectin 9 (HR = 0.361, p = 0.0328), low NKG2A (HR = 0.197, p = 0.0303), or low TIGIT (HR = 0.188, p = 0.0222) expression in the tumor, enhanced monocyte abundance was associated with prolonged overall survival (Supplementary Fig. 2N–S). (4) Last, within patients revealing low HLA-E (HR = 0.436, p = 0.0427), high CD155 (HR = 0.271, p = 0.0292), low NKG2A (HR = 0.432, p = 0.0432), low TIGIT (HR = 0.447, p = 0.043), or low TIM3 (HR = 0.42, p = 0.0457) expression in the tumor, raised myeloid dendritic cell abundance was correlated with improved prognosis (Supplementary Fig. 2T–Y). These findings highlight the context-dependent influence of different immune cell abundances on patient survival, which varies with specific molecular expression thresholds in the PDAC tumor microenvironment, and demonstrate the prognostic significance of HLA-E-NKG2A, CD155-TIGIT, and galectin 9-TIM3 immune checkpoints in PDAC.

Expression of checkpoint ligands HLA-E, CD155, and galectin 9 is elevated in the PDAC tumor microenvironment

To validate the expression levels of immune checkpoint ligands HLA-E, CD155, and galectin 9 in PDAC and AAC patients, we measured their protein expression on tumor cells (Fig. 2A). Quantitative analysis demonstrated significant elevation of the levels of HLA-E (p = 0.0001), CD155 (p < 0.0001), and galectin 9 (p < 0.0001) in PDAC tumor tissue compared to adjacent normal tissue (Fig. 2B–D). This pattern was conserved for HLA-E (p = 0.0078) and galectin 9 (p = 0.0039) in AAC tumor tissue (Fig. 2E and G), though the differential expression of CD155 (p = 0.1687) was less pronounced (Fig. 2F). Accordingly, we consolidated the two tumor types into a malignant tumor category for comparative analysis. Pooled analysis confirmed consistently elevated expression of these three molecules in the tumor versus adjacent normal tissues across both cancer types (Fig. 2H–J), suggesting their potential immunosuppressive roles in the tumor microenvironment of PDAC and AAC.

Fig. 2.

Expression of checkpoint ligands HLA-E, CD155, and galectin 9 is elevated in the PDAC tumor microenvironment. A Representative images of immunohistochemistry (IHC) staining of HLA-E, CD155, and galectin 9 in tumor versus adjacent normal tissues from patients with PDAC or AAC; scale bar: 150 µm. B–D Quantitative IHC scores of B HLA-E, C CD155, and D galectin 9 in PDAC tumors versus paired adjacent tissues. E–G Quantitative IHC scores of E HLA-E, F CD155, and G galectin 9 in AAC tumors versus paired adjacent tissues. H–J Combined quantitative IHC scores of H HLA-E, I CD155, and J galectin 9 in both PDAC and AAC tumors versus paired adjacent tissues (n = 37). Each line represents individual patient data. Differences were analyzed by paired t test or Wilcoxon matched pairs test. **p < 0.01, ***p < 0.001, ****p < 0.0001

Upregulation of HLA-E and galectin 9 is specific to malignant tumors relative to benign tumors

To understand the roles of HLA-E, TIGIT, and galectin 9 in the development from precursor lesions to invasive cancers, we compared PDAC and AAC malignant tumors with PanIN, IPMN, and SCA benign tumors (Fig. 3A). Across the three benign tumor types, because of their low prevalence and limited sample size, none of the three molecules reached statistically significant differences in the expression between tumor and adjacent normal tissues (Fig. 3B–M), except for galectin 9, which showed tumor-enriched expression specifically in SCA (p = 0.0083) and merged sample analyses (p = 0.0005) (Fig. 3J and M). Moreover, galectin 9 showed higher expression in PDAC compared to AAC, PanIN, IPMN, and SCA, respectively (Fig. 3P). HLA-E expression was higher in PDAC compared to PanIN lesions (Fig. 3N). Aggregate analysis across disease subtypes demonstrated markedly increased expression of HLA-E (p = 0.0021) and galectin 9 (p < 0.0001) in malignancies overall versus benign neoplasms (Fig. 3T and V). CD155 revealed similar expression trend but failed to reach statistical significance (p = 0.0538) (Fig. 3U). These results indicate that the HLA-E-NKG2A and galectin 9-TIM3 immune checkpoint interactions may exert progressive function during the transition from precancerous lesions to PDAC, suggesting a mechanism by which the nascent tumor microenvironment evolves to actively inhibit anti-tumor immunity. Specifically, the upregulation of these pathways appears to create an increasingly immunosuppressive niche that cripples effector lymphocytes, ultimately permitting the emergence of invasive carcinoma.

Fig. 3.

Upregulation of HLA-E and galectin 9 is specific to malignant tumors relative to benign tumors. A Representative images of immunohistochemistry (IHC) staining of HLA-E, CD155, and galectin 9 in tumor tissues from patients with PDAC, AAC, PanIN, IPMN, or SCA; scale bar: 150 µm. B–D Quantitative IHC scores of HLA-E, CD155, and galectin 9 in PanIN tumors versus paired adjacent tissues. E–G Quantitative IHC scores of HLA-E, CD155, and galectin 9 in IPMN tumors versus paired adjacent tissues. H–J Quantitative IHC scores of HLA-E, CD155, and galectin 9 in SCA tumors versus paired adjacent tissues. K–M Combined quantitative IHC scores of HLA-E, CD155, and galectin 9 in PanIN, IPMN, and SCA tumors versus paired adjacent tissues. Each line represents individual patient data, and paired samples were unavailable for some cases, resulting in unpaired points or missing values. N–P Comparative IHC scores of HLA-E, CD155, and galectin 9 in tumor tissues across different pancreatic neoplasms. Q–S Comparative IHC scores of HLA-E, CD155, and galectin 9 in adjacent normal tissues across different pancreatic neoplasms. T–V Combined IHC scores of HLA-E, CD155, and galectin 9 in tumor tissues from all malignant (n = 37) versus all benign neoplasms (n = 15). W–Y Combined IHC scores of HLA-E, CD155, and galectin 9 in adjacent normal tissues from all malignant versus all benign neoplasms. Differences were analyzed by paired t test or Wilcoxon matched pairs test. *p < 0.05, **p < 0.001, ***p < 0.001, ****p < 0.0001

Expression levels of HLA-E and galectin 9 influence the motility of pancreatic epithelial cells

To further investigate the potential roles of HLA-E and galectin 9 in the malignant progression of pancreatic tumors, we established MIA PaCa-2 (pancreatic cancer) and hTERT‑HPNE (immortalized normal human pancreatic epithelial) cell lines overexpressing these molecules (Fig. 4A, B). Subsequently, CCK-8, wound healing, and Transwell invasion assays were performed. Notably, the endogenous expression levels of HLA-E and galectin 9 in MIA PaCa-2 and hTERT-HPNE cells were not sufficiently high to enable reliable knockdown experiments. The results demonstrated that overexpression of HLA-E and galectin 9 did not affect the proliferative capacity of either normal pancreatic epithelial cells or cancer cells (Fig. 4C, D). In wound healing assays, the high expression of HLA-E and galectin 9 enhanced the migratory ability of hTERT‑HPNE cells (Fig. 4E) but had no significant effect on the migratory capacity of MIA PaCa-2 cells (Fig. 4F). In Transwell invasion assays, the overexpression of HLA-E significantly enhanced the invasive ability of hTERT‑HPNE cells (Fig. 4G), while again showing no effect on MIA PaCa-2 cells (Fig. 4H). The overexpression of galectin 9 did not markedly alter the invasive capacity of either cell line (Fig. 4G–H). Collectively, these results demonstrate that HLA-E may promote pancreatic epithelial cells to acquire malignant features by enhancing their migration and invasion capacities, suggesting a direct role in malignant transformation beyond immune regulation, which is a previously unrecognized function that highlights the importance of HLA-E in early pancreatic carcinogenesis.

Fig. 4.

Expression levels of HLA-E and galectin 9 influence the motility of pancreatic epithelial cells. A–B hTERT‑HPNE and MIA PaCa-2 cells were transfected with overexpression plasmids (oe control, oe HLA-E, or oe galectin 9). Overexpression of A HLA-E and B galectin 9 was validated by Western blotting. (C-D) Proliferative capacity of the transfected C hTERT‑HPNE and D MIA PaCa-2 cells was assessed at the indicated time points using the CCK-8 assay. Relative optical density (OD) values were calculated by subtracting the OD value of blank wells (no cells) from original OD values. Statistical analysis was performed using two-way ANOVA. E–F Migratory capacity of the transfected E hTERT‑HPNE and F MIA PaCa-2 cells was evaluated at the indicated time points using the wound healing assay (under 100 × microscope). The wound healing rate was calculated as (wound area at 0 h – wound area at 12 h)/wound area at 0 h. Statistical analysis was performed using unpaired t test or Mann–Whitney U test. G–H Invasive capacity of transfected G hTERT‑HPNE and H MIA PaCa-2 cells was determined using the Transwell invasion assay. The number of invaded cells was counted under 200 × microscope. Differences were analyzed by unpaired t test or Mann–Whitney U test. *p < 0.05, **p < 0.001

Upregulation of checkpoint receptor NKG2A on tumor-infiltrating lymphocytes is linked to their dysfunction in PDAC

To investigate the functional impact of NKG2A–HLA-E, TIGIT–CD155, and TIM3–galectin 9 pathways on T cells and natural killer (NK) cells in PDAC, we measured the expression profiles of these inhibitory receptors (i.e., NKG2A, TIGIT, TIM3, PD-1) and the functional profiles (i.e., HLA-DR, Ki-67, granzyme B, perforin) of tumor-infiltrating lymphocytes (TIL) (Supplementary Fig. 3). Due to the typically small tumor size of PDAC, the yields of mononuclear leukocytes from both tumor and adjacent tissues were very low. Consequently, comprehensive tumor-adjacent tissue-peripheral blood comparisons were precluded.

CD8⁺ T cells showed significant upregulation of NKG2A (p = 0.0170) and PD-1 (p = 0.0026) in TIL versus PBMC (Fig. 5B and K), while exhibited modest but nonsignificant increases in expression of TIGIT and TIM3 in TIL (Fig. 5E and H). Tumor-infiltrating CD4⁺ T cells upregulated PD-1 expression (p = 0.0474) (Fig. 5J), while NK cells did not show significant differences (Fig. 5L). Notably, expression of HLA-DR as an activated T cell and NK cell marker was elevated in tumor-infiltrating CD4⁺ T cells (p = 0.0016), CD8⁺ cells (p = 0.0104), and NK cells (p = 0.0167) (Fig. 5M–O), indicating T cell activation probably caused by tumor antigens within the tumor microenvironment. However, lower expression of Ki-67 as a proliferative marker demonstrated decreased proliferation of tumor-infiltrating CD4⁺ T cells (p = 0.0135) and CD8⁺ T cells (p = 0.0283) (Fig. 5P, Q). Reduced production of cytotoxic effector molecules granzyme B (p_NK = 0.0301, p_CD8 = 0.0387 or 0.0002) or perforin (p_CD8 = 0.0317, p_CD4 = 0.0286) in tumor-infiltrating NK, CD8⁺, and CD4⁺ T cells (Fig. 5T–W), together with decreased proliferation, suggested a functionally compromised status of TIL. Consistently, effector cytokine IFN-γ expression was significantly decreased in tumor-infiltrating NK cells (p_NK = 0.0029), but not in tumor-infiltrating T cells (Fig. 5Y, Z), which is possibly related to the immunosuppressive function of the abovementioned immune checkpoint interactions.

Fig. 5.

Upregulation of checkpoint receptor NKG2A on tumor-infiltrating lymphocytes is linked to their dysfunction in PDAC. A–L Flow cytometry analyses show proportions of A–C NKG2A, D–F TIGIT, G–I TIM3, and J–L PD-1-positive cells in CD3⁺CD4⁺ T cells, CD3⁺CD8⁺ T cells, and CD3⁻CD56⁺ NK cells from TIL, adjacent normal tissues (NIL), or PBMC of PDAC patients (n = 10). M–X Flow cytometry analyses show proportions of M–O HLA-DR, P–R Ki-67, S–U granzyme B, and V–X perforin-positive cells in CD3⁺CD4⁺ T cells, CD3⁺CD8⁺ T cells, and CD3⁻CD56⁺ NK cells from TIL, NIL, or PBMC of PDAC patients (n = 10). Each line represents individual patient data. Differences were analyzed by paired t test, unpaired t test, or Mann–Whitney U test. Y Representative images of mIHC Panel 1 and the corresponding HE stained images, showing IFN‑γ expression in CD3⁺ T cells (upper) and CD3⁻CD56⁺ NK cells (lower) in PDAC tissues. White arrows indicate CD3⁺IFN‑γ⁺ or CD3⁻CD56⁺IFN‑γ⁺ cells. (Z) Proportions of IFN-γ-positive cells in CD3⁺ T cells or CD3⁻CD56.⁺ NK cells within tumor or adjacent tissues (n = 23). Scale bar: 20 µm. Differences were analyzed by Wilcoxon matched pairs test. *p < 0.05, **p < 0.001, ***p < 0.001

The spatial interaction between NKG2A and HLA-E mediates localized immunosuppression of tumor-infiltrating lymphocytes in the PDAC microenvironment

Building upon our aforementioned analytical results and considering the current scarcity of HLA-E research in pancreatic cancer, we performed mIHC analysis to further investigate the spatial distribution characteristics of the NKG2A–HLA-E molecular pair in pancreatic cancer tissues (Fig. 6A). The average fluorescence intensity of HLA-E was significantly higher in tumor parenchyma than in adjacent normal tissues (p = 0.0219) (Fig. 6B), corroborating our previous immunohistochemistry findings. Comparative analysis of matched tumor and adjacent normal tissues revealed significantly higher NKG2A-expressing T cell infiltration in both tumor parenchyma (p = 0.0416) and particularly in tumor stroma (p = 0.0015) (Fig. 6C). A similar trend was observed in NK cells, though the differences did not reach statistical significance, likely attributable to their relatively low abundance in the tumor microenvironment (Fig. 6F). Notably, the proportion of NKG2A⁺ T cells was higher in the tumor stroma than the adjacent normal tissues (p = 0.0358) (Fig. 6D). These data confirm that both HLA-E and NKG2A were upregulated in the tumor microenvironment of PDAC and AAC, either in the parenchyma or in the stroma. The specificity of CD3 and CD56 staining was validated by CD45 staining, as shown by all CD3⁺ or CD56⁺ cells co-expressing CD45 (Supplementary Fig. 4H–J).

Fig. 6.

The spatial interaction between NKG2A and HLA-E mediates localized immunosuppression of tumor-infiltrating lymphocytes in the PDAC microenvironment. A Representative images of mIHC Panel 2; scale bar: 20 µm or 100 µm. Green arrows indicate NKG2A⁺CD56⁺CD3⁻PanCK⁻DAPI⁺ cells, and blue arrows indicate NKG2A⁺CD3⁺PanCK⁻DAPI⁺ cells. B Average fluorescence intensity of HLA-E in tumor parenchyma, tumor stroma, and adjacent tissues from PDAC patients. C and F Cell counts of NKG2A-positive T cells (CD3⁺PanCK⁻) and NK cells (CD3⁻CD56⁺PanCK⁻) in tumor parenchyma, tumor stroma, and adjacent tissues. D and G Proportions of NKG2A-positive cells in T cells and NK cells in tumor parenchyma, tumor stroma, and adjacent tissues. E and H Comparison of cell numbers of NKG2A-positive T cells and NK cells in HLA-E-high-expressing versus HLA-E-low-expressing regions within tumor tissues, which were different from the regions analyzed in C–D and F–G. All values with error bars represent mean with SEM. I–P Cell counts of NKG2A-positive T cells and NK cells within 0–10, 10–20, 0–20, or 0–30 µm radial distance from HLA-E positive PanCK⁺ cells versus HLA-E-negative PanCK⁺ cells. Each line represents individual patient data. Differences were analyzed by paired t test or Wilcoxon matched pairs test. *p < 0.05, **p < 0.01, ****p < 0.0001

Tumor regions with high HLA-E expression displayed comparable NKG2A⁺ T cell and NKG2A⁺ NK cell infiltration levels compared to areas with low HLA-E expression (Fig. 6E and H). These findings suggest suppressed immune function in T cells, potentially mediated through the NKG2A–HLA-E interaction. Furthermore, we analyzed the spatial distribution of NKG2A⁺ T cells and NK cells at varying radial distances from PanCK⁺ tumor cells. Quantitative assessment revealed significantly enriched infiltration of NKG2A⁺ T cells but not NKG2A⁺ NK cells within the 0–10 µm (p_T < 0.0001), 10–20 µm (p_T = 0.0290) and 0–20 µm (p_T = 0.0017) radius of HLA-E⁺ tumor cells compared to HLA-E⁻ tumor cells (Fig. 6I–K, M–O). The representative images visually demonstrate these spatial interactions (Supplementary Fig. 4 A). However, this spatial pattern dissipated beyond the 20 μm radius threshold, as HLA-E⁺ tumor cells did not recruit more NKG2A⁺ T cells than HLA-E⁻ tumor cells within the distances of 20–30 µm and 30–50 µm radius (Supplementary Fig. 4B, C). Notably, within the 50–100 μm radial interval, NKG2A⁺ T cells exhibited a reversal of infiltration trend (p = 0.0044) (Supplementary Fig. 4D). Given the typical dimensions of immune cells, we propose that the 0–20 µm radius more accurately reflects the functional interplay range between lymphocytes and tumor cells. Consequently, these observations support the hypothesis that NKG2A⁺ T cells preferentially engage with HLA-E⁺ tumor cells. These results provide spatial evidence for NKG2A–HLA-E-mediated immune cell–tumor cell interactions, suggesting a mechanism by which tumor cells may facilitate immune escape via inhibiting T cells.

Single-cell characterization of tumor-infiltrating T cells expressing NKG2A, TIM3, or PD-1 in PDAC

To provide the first single-cell characterization of the NKG2A-HLA-E and TIM3-galectin 9 pathways in PDAC, we performed in-depth computational analyses using a public PDAC single-cell RNA-sequencing (scRNA-seq) dataset (GSE254250) [25]. We first examined the expression profiles of NKG2A (KLRC1), TIM3 (HAVCR2), and PD-1 (PDCD1) across all tumor-infiltrating T cell subsets (Fig. 7A). NKG2A and TIM3 were predominantly expressed in exhausted CD8⁺ T cells and proliferating CD8⁺ T cells (Fig. 7B–C), whereas PD-1 was mainly detected in exhausted CD8⁺ T cells, proliferating CD8⁺ T cells, and exhausted CD4⁺ T cells (Fig. 7D). Notably, T cells expressing NKG2A or TIM3 were largely composed of exhausted CD8⁺ T cells, and this enrichment was particularly pronounced when these cells co-expressed PD-1 (Fig. 7F and H).

Fig. 7.

Single-cell characterization of tumor-infiltrating T cells expressing NKG2A, TIM3, or PD-1 in PDAC. A Single-cell RNA-sequencing analysis on T cells from PDAC tumors using a public dataset (GSE254250), UMAP plot shows T cell subset clusters. B–D Single-cell RNA-sequencing data show the proportions of NKG2A (KLRC1), TIM3 (HAVCR2), and PD-1 (PDCD1)-positive and PDCD1-negative cells in different T cell subsets. E and G Distribution of T cells grouped by E NKG2A and PD-1 expression or G TIM3 and PD-1 expression across different T cells subsets. F and H Distribution of different T cell subsets within four subgroups defined by F NKG2A and PD-1 expression or H TIM3 and PD-1 expression. I–M Activation and exhaustion scores of T cells based on gene sets related to T cell activation and exhaustion, within T cells grouped by I NKG2A, J TIM3, or K PD-1 expression, and within four subgroups defined by L NKG2A and PD-1 expression, or M TIM3 and PD-1 expression. (N–O) Pathway enrichment analysis of differentially expressed genes showing N upregulated or O downregulated signaling pathways, comparing T cells and T cell subsets grouped by NKG2A (KLRC1), TIM3 (HAVCR2), or PD-1 (PDCD1) expression. The X axis represents different T cell subsets, the Y axis represents signaling pathways, the bubble size indicates the proportion of statistically significant differentially expressed genes relative to the total genes in each pathway, and the bubble color represents the adjusted P value determined by the hypergeometric test

We next defined the transcriptional signatures of tumor-infiltrating T cells expressing these inhibitory receptors. T cells positive for NKG2A, TIM3, or PD-1 exhibited comparable activation scores but higher exhaustion scores (enriched for established exhaustion-related genes) compared with their negative counterparts (Fig. 7I–K, Supplementary Fig. 5A-C and F–H). More importantly, T cells co-expressing PD-1 with either NKG2A or TIM3 displayed higher exhaustion scores than those expressing only a single molecule (Fig. 7L–M, Supplementary Fig. 5D, E and I, J), suggesting a synergistic inhibitory effect of these checkpoint receptors on tumor-infiltrating T cells in PDAC.

To further delineate the functional consequences of NKG2A, TIM3, and PD-1 expression, we performed pathway enrichment analysis of differentially expressed genes (DEG) between receptor-positive and receptor-negative T cell subsets. Compared with NKG2A⁻ T cells, NKG2A⁺ T cells exhibited upregulation of pathways associated with PD-1 signaling, IL-10 signaling, lymphocyte apoptosis, and negative regulation of cell killing, adaptive immune responses, immune effector processes, T cell-mediated cytotoxicity and immunity, T cell activation, and lymphocyte activation and immunity (Fig. 7N). Conversely, NKG2A⁺ T cells showed downregulation of MAPK3 (ERK1) activation, IFN-γ signaling, TCR signaling, and CD28 family costimulation pathways (Fig. 7O). Similarly, when comparing TIM3⁺ T cells with TIM3⁻ T cells, we observed upregulation of lymphocyte and T cell apoptotic processes, along with negative regulation of cell killing, cytokine production, adaptive immune responses, immune effector processes, T cell- and lymphocyte-mediated immunity, and T cell and lymphocyte proliferation and activation (Fig. 7N), whereas the MAPK cascade, ERK1/2 signaling, T cell activation involved in immune responses, lymphocyte proliferation, regulation of T cell activation, and inflammatory responses were downregulated (Fig. 7O). In PD-1⁺ versus PD-1⁻ T cells, upregulation of PD-1 signaling, IL-10 signaling, and apoptotic processes and negative regulation of cell killing, cytokine production, adaptive immunity, immune effector process, T cell proliferation and activation, and lymphocyte-mediated immunity were observed (Fig. 7N), while pathways related to T cell activation and lymphocyte proliferation were downregulated (Fig. 7O). Collectively, these transcriptional and pathway-level findings further support a synergistic inhibitory role for PD-1 together with NKG2A or TIM3 in constraining T cell function within the PDAC tumor microenvironment.

Dual blockade of PD-1 with either NKG2A or TIM3 reinvigorates anti-tumor immunity of T cells and activates ERK signaling via SHP-1 inhibition in PDAC

To evaluate the effect of targeting NKG2A–HLA-E, TIGIT–CD155, or TIM3–galectin 9 pathway on the anti-tumor immune response, we assessed the anti-tumor function of CD8⁺ T cells and CD4⁺ T cells from PDAC and AAC patients (Fig. 8A and Supplementary Fig. 6). Single blockade of NKG2A increased activation marker CD137 expression of CD4⁺ T cells (Fig. 8B) and activation marker HLA-DR expression of CD8⁺ T cells (Fig. 8G), suggesting partially promoted T cell activation. Single blockade of TIGIT enhanced cytotoxic molecule production (granzyme B but not perforin) of CD8⁺ T cells (Fig. 8I), while single blockade of PD-1 increased granzyme B in both CD4⁺ and CD8⁺ T cells (Fig. 8E and I). More importantly, co-targeting of NKG2A and PD-1 further enhanced activation (CD137 but not HLA-DR) of CD4⁺ T cells, which was superior to single blockade of NKG2A (Fig. 8B), cytotoxicity (both perforin and granzyme B) of CD8⁺ T cells (Fig. 8H–I), and production of effector cytokine IFN-γ (Fig. 8J), and selectively upregulated granzyme B (but not perforin) in CD4⁺ T cells (Fig. 8E). Co-targeting of TIGIT and PD-1 increased activation of CD4⁺ and CD8⁺ T cells (CD137 but not HLA-DR) (Fig. 8B and F) and selectively upregulated perforin (but not granzyme B) in CD4⁺ T cells, which was superior to single blockade of TIGIT (Fig. 8D). Surprisingly, although single blockade of TIM3 did not show significant effect, co-targeting of TIM3 and PD-1 further enhanced activation of CD4⁺ (CD137 but not HLA-DR) (Fig. 8B) and CD8⁺ T cells (both CD137 and HLA-DR) (Fig. 8F–G) and cytotoxicity (both perforin and granzyme B) of CD8⁺ and CD4⁺ T cells, which was superior to single blockade of TIM3 (both perforin and granzyme B) or PD-1 (perforin but not granzyme B) (Fig. 8D, E, H, I).

Fig. 8.

Dual blockade of PD-1 with either NKG2A or TIM3 reinvigorates anti-tumor immunity of T cells and activates ERK signaling via SHP-1 inhibition in PDAC. A Experimental workflow schematic diagram was created using BioGDP platform [26]. Following isolation of PBMC or TIL from PDAC or AAC patients, functional assays were performed using CD3/CD28 stimulation, and co-culture experiments were established with HLA-A-matched pancreatic cancer cells, in the presence or absence of individual or combined blockade of NKG2A, TIGIT, TIM3, and PD-1. B–I Data graphs from functional assays show relative ratio of CD137, HLA-DR, perforin, or granzyme B-positive cells within B–E CD3⁺CD4⁺ T cell and F–I CD3⁺CD8⁺ T cell populations compared to the control condition (n = 12). J Secretion of IFN-γ in the culture supernatant was measured using ELISA (n = 10). All values with error bars represent mean with SEM. Differences were analyzed by paired t test or Wilcoxon matched pairs test. K The cell counts of living tumor cells (relative ratio compared to the control condition) in the co-culture experiments were presented, indicating tumor killing effect. Each line represents individual patient data. L–N Proportions of phosphorylated SHP-1, phosphorylated ERK, and phosphorylated AKT-positive cells within CD3⁺T cells (n = 7). Differences were analyzed by paired t test or Wilcoxon matched pairs test. *p < 0.05, **p < 0.01, absence of asterisks indicates no statistically significant difference (p > 0.05)

To further assess the functional implication on anti-tumor response of PDAC patient-derived lymphocytes, we established a co-culture system using HLA-A-matched human pancreatic cancer cells (Fig. 8A). Co-blockade of NKG2A or TIM3 with PD-1 showed significant effect compared to the control group (Fig. 8K). Notably, across all five individual patient experiments, co-blockade of NKG2A with PD-1 reduced tumor cell number most dramatically in three patients, demonstrating enhanced tumor cell killing, while co-blockade of TIM3 with PD-1 produced the most pronounced tumor cell elimination in the other two patients, indicating individual variations in responses to different immune checkpoint blockades. These findings confirm that dual blockade of NKG2A or TIM3 with PD-1 enhances anti-tumor immunity and support combination strategies. Integrating these results with previous mIHC, bioinformatics, and flow cytometry data, we propose that the NKG2A–HLA-E pathway may suppress anti-tumor immune response and promote tumor immune inhibition in pancreatic cancer.

To uncover previously unrecognized mechanistic features of the NKG2A–HLA-E and TIM3–galectin 9 pathways in PDAC, we performed phospho-flow cytometry to interrogate downstream signaling events following checkpoint blockade. NKG2A, TIM3, and PD-1 have each been individually implicated in the regulation of phosphorylated SHP-1 (pSHP-1), phosphorylated AKT (pAKT), and phosphorylated ERK (pERK) signaling [27–29]. However, whether these immune checkpoints exert synergistic effects on these signaling pathways, and whether pSHP-1 functionally links to pAKT and pERK in this context, remained unclear. To address this, we utilized Jurkat cells supplemented with recombinant HLA-E, galectin 9, and PD-L1 proteins, as well as PDAC patient-derived PBMC, and treated them with blocking antibodies against NKG2A, TIM3, PD-1, or their pairwise combinations. After three days of culture, phospho-flow cytometry was performed to assess the levels of pSHP-1, pERK1/2, and pAKT within CD3⁺ T cells (Fig. 8L–N).

Single blockade of PD-1 or TIM3 decreased pSHP-1 levels compared to the control condition (Fig. 8L). Pairwise combinations of NKG2A, TIM3, or PD-1 blockade similarly reduced pSHP-1, suggesting that pSHP-1 downregulation may be a shared downstream event of checkpoint inhibition rather than a synergistic effect (Fig. 8L). Notably, pairwise combinations of NKG2A, TIM3, or PD-1 blockade significantly increased pERK levels (Fig. 8M). Among these, combined NKG2A and PD-1 blockade further elevated pERK levels compared with either single blockade alone, indicating a synergistic effect on ERK activation (Fig. 8M). In contrast, pAKT levels were increased by single NKG2A blockade compared with the control, while combined TIM3 and PD-1 blockade elevated pAKT compared with either single blockade alone (Fig. 8N), suggesting pathway-specific and context-dependent regulation of AKT signaling by different checkpoint combinations. Collectively, these findings demonstrate that blockade of these immune checkpoint pathways may act synergistically, primarily through inhibition of pSHP-1 and activation of pERK signaling, to enhance T cell function. Importantly, these results validate our scRNA-seq findings, which showed downregulation of MAPK and ERK activation and cascade pathways in NKG2A⁺ or TIM3⁺ T cells compared with their negative counterparts. Together, these data support a model in which NKG2A, TIM3, and PD-1 engage non-redundant inhibitory pathways that converge on common signaling nodes, providing a mechanistic rationale for their combined therapeutic targeting in PDAC.

Discussion

In this study, we provide several lines of evidence establishing the NKG2A–HLA-E and TIM3–galectin 9 axes as critical players of immune suppression and potential therapeutic targets in PDAC. (1) We identified upregulation of HLA-E and galectin 9 in pancreatic malignancies and performed the first comparative analysis of these checkpoint ligands across the benign-to-malignant spectrum, revealing their upregulation during tumorigenesis. (2) We uncovered a previously unrecognized functional role for HLA-E in malignant transformation, demonstrating for the first time that HLA-E overexpression enhances the migration and invasion capacity of normal pancreatic epithelial cells. (3) We provide the first spatial evidence of NKG2A–HLA-E-mediated immunosuppression within the PDAC tumor microenvironment, showing preferential colocalization of NKG2A⁺ T cells with HLA-E⁺ tumor cells. (4) Through single-cell RNA-sequencing analysis, we present, to our knowledge, the first single-cell characterization of NKG2A⁺ and TIM3⁺ T cells in PDAC, revealing their exhausted transcriptional signature. (5) We demonstrate that dual blockade of PD-1 with either NKG2A or TIM3 synergistically enhances T cell activation and cytotoxicity in PDAC patient-derived T cells, a mechanism mediated by decreased pSHP-1 and increased pERK signaling. Collectively, these findings identify the NKG2A–HLA-E and TIM3–galectin 9 pathways as therapeutic vulnerabilities in PDAC and provide a mechanistic rationale for combination immunotherapy strategies targeting these checkpoints together with PD-1–PD-L1 blockade.

Our findings confirmed and extended previous reports by demonstrating consistent upregulation of checkpoint ligands CD155 and galectin 9 in PDAC versus adjacent normal tissues [16, 18, 30]. Notably, we provided the first evidence that HLA-E expression was significantly elevated in pancreatic malignancies not only compared to adjacent normal tissues but also relative to benign pancreatic tumors. Overexpression of HLA-E also enhanced the migratory ability and invasive ability of pancreatic epithelial cells, suggesting a potential role for HLA-E in the malignant transformation process. We observed that HLA-E overexpression enhanced the migratory and invasive capacities of normal epithelial cells (hTERT-HPNE) but had no apparent effect on established cancer cells (MIA PaCa-2). We propose a potential hypothesis that HLA-E may primarily participate in the early stages of pancreatic carcinogenesis, promoting normal cells to acquire malignant features, whereas once malignant transformation has occurred, its additional effects on migration and invasion may be limited. Intriguingly, survival analysis revealed an unexpected association between high expression of CD155 and improved patient prognosis, a finding that contrast with some published data [31]. Unfortunately, due to the limited follow-up duration of our patient cohort, we were unable to validate these observations, highlighting the need for future prospective studies with standardized clinical parameters to clarify these associations.

The compartment-specific expression pattern of inhibitory receptor NKG2A, elevated on tumor-infiltrating CD8⁺ T cells but without statistical difference on NK cells in PDAC tissues, suggests that the NKG2A–HLA-E axis may predominantly mediate immune inhibition through T cell regulation in pancreatic cancer. It may also act in concert with the PD-1-PD-L1 axis as shown by the upregulation of both NKG2A and PD-1 on tumor-infiltrating T cells. Our observation of a trend of upregulation of inhibitory receptors TIM3 and TIGIT in PDAC tissues aligned with previous studies [17, 18], but our data did not reach statistical significance. These discrepancies may stem from technical limitations including modest sample size and low immune cell infiltration, which may have reduced our power to detect subtle expression differences.

Bioinformatic analysis using TIMER2.0 database revealed positive correlations between HLA-E expression and NKG2A expression, as well as between galectin 9 expression and TIM3 expression in PDAC, suggesting regulation of these two ligand–receptor pairs in the tumor microenvironment. Additional positive correlations were observed between HLA-E and either PD-L1 or galectin 9, as well as between NKG2A and either PD-1 or TIM3, implying potential synergy between the HLA-E-NKG2A pathway and other two immune checkpoint pathways. Motivated by these findings and the relatively understudied role of HLA-E-NKG2A interaction in pancreatic cancer, we performed mIHC to examine the spatial interplay of NKG2A and HLA-E in PDAC and pancreatobiliary-type AAC. Our data demonstrated increased infiltration of NKG2A⁺ lymphocytes alongside elevated HLA-E expression in tumor tissues, consistent with both our IHC data and public database analysis. These results underscore the potential relevance of the NKG2A–HLA-E axis in pancreatic cancer. Moreover, mIHC analysis revealed preferential spatial clustering of NKG2A⁺ lymphocytes around HLA-E⁺ tumor cells, providing visual evidence of their functional interaction in pancreatic cancer.

Based on our analysis of patient-derived lymphocytes, three distinct TIL subpopulations exhibited higher proportions of HLA-DR⁺ cells compared lymphocytes from adjacent normal tissues and peripheral blood, suggesting sustained immune activation within the tumor microenvironment. However, the significantly reduced expression of granzyme B and perforin in TIL concurrently indicated impaired cytotoxic function. These contrasting observations imply the presence of an activated yet exhausted immune state in pancreatic cancer. Based on IHC and flow cytometric findings demonstrating significant overexpression of NKG2A and HLA-E in tumor tissues, along with mIHC further revealing spatial interactions between NKG2A⁺ lymphocytes and HLA-E⁺ tumor cells, we speculate that the NKG2A–HLA-E axis may contribute to this dysfunctional state. Supporting this hypothesis, data from Liu et al.[13] confirmed a strong correlation between exhausted T cells and the NKG2A–HLA-E axis in pancreatic primary lesions.

Functionally, dual blockade of PD-1 with either NKG2A or TIM3 synergistically enhanced T cell activation and cytotoxicity. Bioinformatic analysis further supported these findings, revealing positive correlations between PD-1 expression and NKG2A or TIM3 expression, and scRNA-seq data also suggest potential synergistic mechanisms among these immune checkpoints in promoting PDAC immune inhibition. A recent study reported that the efficacy of NKG2A blockade might depend on inflammatory signals induced by anti-PD-1 therapy [32]. Integrating these findings, we hypothesize that pancreatic tumor cells engage activated immune cells via the NKG2A–HLA-E or TIM3–galectin 9 interaction and collaborate with PD-1-related pathways to deliver exhaustion signals, ultimately facilitating immune escape. Notably, phospho-flow cytometry results demonstrate that this synergy may involve the downstream SHP-1 and MAPK (ERK) signaling pathway. Furthermore, bioinformatic data linking HLA-E and galectin 9 expression, but not NKG2A and TIM3 expression, with patient prognosis suggest that HLA-E and galectin 9 might play a dominant role.

This study has several limitations that warrant consideration. First, all patients being recruited from a single center and inherent challenges in obtaining surgical specimens from this typically advanced-stage disease inevitably limited the cohort size. Second, the paucity of tumor-infiltrating immune cells characteristic of PDAC may have obscured some immunologic signatures. Third, the recent establishment of our patient cohort precluded comprehensive survival outcome analyses and other long-term clinical correlation studies.

In summary, our work establishes the NKG2A–HLA-E and TIM3–galectin 9 interactions as key immunosuppressive pathways in PDAC. We provide the first comparative analysis across the benign-to-malignant spectrum, spatial evidence of localized T cell suppression, and single-cell characterization of exhausted TIL. Functionally, combined PD-1 blockade with NKG2A or TIM3 synergistically enhances T cell function through SHP-1 inhibition and ERK activation. These findings identify non-redundant immune checkpoints and support combination immunotherapeutic strategies targeting these pathways alongside PD-1–PD-L1 in PDAC.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AAC

Ampullary adenocarcinoma

CA19-9

Carbohydrate antigen 19-9

DEG

Differentially expressed genes

EDTA

Ethylenediaminetetraacetic

ELISA

Enzyme-linked immunosorbent assay

FFPE

Formalin-fixed and paraffin-embedded

GEO

Gene expression omnibus

GO

Gene ontology

ICI

Immune checkpoint inhibitors

IPMN

Intraductal papillary mucinous neoplasm

KEGG

Kyoto encyclopedia of genes and genomes

mIHC

Multiplex immunohistochemistry

MMR-d

Mismatch repair-deficient

NIL

Normal tissue-infiltrating lymphocytes

ORA

Overrepresentation analysis

oeRNA

Overexpression RNA

PBMC

Peripheral blood mononuclear cells

PBS

Phosphate-buffered saline

PCA

Principal component analysis

PDAC

Pancreatic ductal adenocarcinoma

PanIN

Pancreatic intraepithelial neoplasia

QC

Quality control

SCA

Serous cystadenoma

scRNA-seq

Single-cell RNA-sequencing

TIL

Tumor-infiltrating lymphocytes

UMI

Unique molecular identifier

UMAP

Uniform manifold approximation and projection

Author contributions

Conceptualization, GZ, GD and DT; methodology, Yuanyu L, WF, Yan L and GZ; validation, GZ; formal analysis, Yuanyu L and WF; investigation, Yuanyu L, WF, Yan L, Yisen T, XZ and YH; resources, Juping X, KL, YF, Yajun T, DL, ZH, YQ, JC, Jie X, ZY, LL, DT and GD; data curation, Yuanyu L and WF; writing—original draft preparation, Yuanyu L, WF and GZ; writing—review and editing, GZ, GD, DT, LL, Juping X, KL, YF, Yajun T, DL, ZH, YQ, JC, Jie X and ZY; visualization, Yuanyu L, WF, Yan L and XZ; supervision, GZ; project administration, GZ, GD and DT. All authors have read and agreed to the published version of the manuscript.

Funding

National Natural Science Foundation of China, 32300757, Shenzhen Medical Research Fund, A2303022, Basic and Applied Basic Research Foundation of Guangdong Province, 2024A1515010598, 2023A1515220239, Shenzhen Science and Technology Program, JCYJ20240813150415021, JCYJ20220530145006013

Data availability

All data generated or analyzed during this study are included in this published article and its supplementary information files.

Declarations

Conflict of interest

The authors declare no competing interests.

Ethical approval

The study was performed in accordance with 1964 Declaration of Helsinki and its later amendments and approved by the local medical ethics committee (KY-2023-120-01, KY-2025-379-01 and KY-2026-198-01) of the Seventh Affiliated Hospital of Sun Yat-sen University, and informed consent from all patients was obtained before tissue and blood donation.

Consent for publication

Not applicable.