Dual inhibition of tumor cells and tumor-associated neutrophils by anti-CXCR2 antibodies suppresses tumor growth and augments immunotherapy efficacy in ARID1A-deficient pancreatic cancer

Abstract

Pancreatic ductal adenocarcinoma (PDAC) is largely resistant to immune checkpoint blockade (ICB) due to its highly immunosuppressive tumor microenvironment. How to switch the “immune-cold” microenvironment to “immune hot” one is a clinical unmet need. We recently generated a genetically engineered PDAC mouse model (KAR) by combining K-Ras mutation and AT-rich interactive domain-containing protein 1A (ARID1A) inactivation in the pancreas. Compared to the KPC (K-Ras mutation + p53 inactivation) tumors, KAR tumors express higher C-X-C chemokine receptor 2 (CXCR2) and have more CXCR2⁺ neutrophils infiltration. Our previous study demonstrated that activation of CXCR2 by its cognate ligands promotes proliferation and drug resistance in PDAC cells. In addition, increase in tumor-associated neutrophils and high neutrophil-to-lymphocyte ratio are associated with worse prognosis in PDAC patients. Here, we developed a monoclonal antibody against human CXCR2 (huCXCR2-Ab) and tested its therapeutic effect in cells and in animals. Treatment with huCXCR2-Ab inhibited IL-8–induced downstream signaling, and the proliferation of human PDAC cancer cells. In addition, huCXCR2-Ab suppressed PDAC growth in an orthotopic nude mouse model. We next developed a monoclonal antibody against mouse CXCR2 (muCXCR2-Ab) and investigated its immune regulatory effect in vivo. We showed that muCXCR2-Ab reduced tumor burden, increased CD8⁺ T-cell infiltration, and decreased immunosuppressive neutrophils in the KAR mice. Moreover, muCXCR2-Ab augments the therapeutic efficacy of anti-PD-L1. Thus, dual targeting of CXCR2⁺ tumor cells and tumor-associated neutrophils remodels tumor microenvironment and enhances immunotherapy response in ARID1A-deficient PDAC, providing a new strategy for precision therapy of PDAC.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00262-026-04498-6.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal malignancy characterized by a fibrotic and immunosuppressive tumor microenvironment (TME) that renders it largely refractory to ICB therapy [1–3]. The paucity of responses to anti-PD-1 or anti-PD-L1 monotherapy reflects an immune-evasive niche dominated by myeloid-derived suppressor cells (MDSCs), tumor-associated macrophages, and neutrophils [4–6]. Among these, tumor-associated neutrophils (TANs) and granulocytic MDSCs are increasingly recognized as major mediators of immune suppression and tumor promotion in PDAC. High neutrophil infiltration, particularly at invasive tumor fronts, correlates with worse patient survival, whereas tumors enriched in CD8⁺ T cells exhibit more favorable outcomes [7, 8].

The chemokine receptor CXCR2, a master regulator of neutrophil trafficking, orchestrates the recruitment of TANs and MDSCs through ligands such as IL-8 (CXCL8), CXCL1, and CXCL5, which are abundantly expressed in PDAC [9–11]. CXCR2⁺ myeloid cells accumulate at tumor margins and suppress cytotoxic T-cell infiltration via the secretion of proteases, reactive oxygen species, and cytokines, establishing an immunosuppressive tumor microenvironment [12]. Notably, Steele et al. reported that CXCR2⁺ neutrophils and MDSCs clustering at PDAC tumor margins were strongly associated with poor prognosis and early metastatic relapse [3]. Inhibition or genetic deletion of Cxcr2 in genetically engineered PDAC mouse models markedly reduces neutrophil infiltration, enhances intratumoral CD8⁺ T-cell access, and limits tumor growth and metastasis [3, 13]. These findings highlight CXCR2 as a critical driver of PDAC immune evasion. In addition to neutrophil recruitment, CXCR2 is also expressed in different cells including cancer cells, tumor-associated fibroblasts and endothelial cells in PDAC tumors [10]. Our previous results also showed that CXCR2 is expressed in human PDAC cells and is involved in the enhancement of gemcitabine resistance [53].

While myeloid-targeted strategies can remodel the TME, effective antitumor immunity also depends on tumor-intrinsic immunogenicity. Recent evidence identifies deficiency of ARID1A, a chromatin-remodeling gene in the SWI/SNF complex, as a determinant of increased tumor immunogenicity and immune checkpoint blockade (ICB) responsiveness. ARID1A is inactivated in approximately 5–8% of PDAC cases [15]. Notably, genomic profiling studies of human PDAC have identified concurrent KRAS and ARID1A alterations in approximately 5–6% of cases, supporting the clinical relevance of our KAR model, which combines the constitutively active oncogenic Kras^G12D allele with pancreas-specific Arid1a deletion [16, 17]. ARID1A loss causes genomic instability and defective DNA damage repair, elevating mutational burden and expanding the neoantigen repertoire [18, 19]. Notably, ARID1A-deficient tumors have been shown to exhibit microsatellite instability or a “mutator” phenotype in some settings, due to ARID1A’s role in recruiting DNA mismatch repair proteins like MSH2. Shen et al. demonstrated that ARID1A inactivation in cancer cells causes a mismatch repair defect and a higher tumor mutation burden, thereby unleashing antitumor immunity that can be further amplified by PD-1 blockade. ARID1A-deficient tumors mounted durable responses to anti-PD-1 in mouse models, whereas ARID1A-proficient tumors did not, directly linking ARID1A loss to increased ICB sensitivity [18]. Recently, we generated a novel genetically engineered PDAC mouse model (named as KAR) by combining K-Ras mutation and ARID1A inactivation in the pancreas. Single-cell transcriptomic analysis and immunohistochemical staining demonstrated that KAR tumors exhibited higher CXCR2 expression and have more CXCR2⁺ neutrophils infiltration when compared to the KPC tumors [20]. These results suggest that the KAR mouse provides an excellent model for in vivo testing whether targeting CXCR2 may simultaneously suppress the proliferation of PDAC cells and the infiltration of CXCR2⁺ neutrophils to create an “immune hot” microenvironment to enhance ICB response.

Currently, several small-molecule CXCR2 antagonists, including AZD5069, Reparixin, Danirixin, and Navarixin, have been clinically tested in cancer and inflammatory diseases [13, 21–23]. Although these agents demonstrated strong preclinical activity, their clinical efficacy has been limited; for example, AZD5069 reduced neutrophil infiltration in airway inflammation without improving symptoms, and Danirixin trials were halted due to adverse events [23]. The modest outcomes of these inhibitors underscore the need for alternative approaches. Therapeutic antibodies offer advantages of target selectivity, prolonged half-life, and improved safety. In this study, we developed a novel humanized anti-CXCR2 monoclonal antibody (huCXCR2-Ab) and demonstrated its anti-cancer activity on human PDAC cells in vitro and in vivo. In addition, we also developed a mouse anti-CXCR2 monoclonal antibody (muCXCR2-Ab) and confirmed its effect on enhancing ICB response in the KAR mice. We hypothesized that ARID1A deficiency defines a neutrophil-enriched PDAC subtype that is particularly susceptible to CXCR2 blockade combined with immune checkpoint inhibition.

Materials and methods

Genetically engineered mouse models

Pdx1-Cre, LSL-Kras^G12D, and Arid1a^tm1.1Zhwa mouse strains were obtained from the Mouse Models of Human Cancers Consortium (MMHCC) under material transfer agreements and from The Jackson Laboratory. Mice were genotyped according to PCR protocols provided by MMHCC and The Jackson Laboratory. ARID1A-deficient pancreatic cancer mice (Pdx1-Cre; Kras^G12D; Arid1a^L/L, referred to as KAR) and control KPC mice (Pdx1-Cre; Kras^G12D; Trp53^R172H) were maintained under specific pathogen-free conditions. All animal studies were approved by the Institutional Animal Care and Use Committee (IACUC) of the National Health Research Institutes (NHRI). Experiments were performed using male offspring mice.

Isolation of pancreatic tumor cells and tissue dissociation

To isolate pancreatic tumor cells, KPC and KAR mice at 8 and 16 weeks of age were euthanized by inhalation of 5% CO₂. Pancreatic tissues were harvested, washed in ice-cold phosphate-buffered saline (PBS), and manually minced into small fragments using sterile scalpels. Tissue fragments were transferred into 3–5 mL of dissociation buffer from the Tumor Dissociation Kit (Miltenyi Biotec, Cat. No. 130–095-929) and further processed into submillimeter-sized particles.

Enzymatic and mechanical dissociation was performed using the gentleMACS Octo Dissociator (Miltenyi Biotec, Cat. No. 130–095-937) for 25–30 min according to the manufacturer’s instructions. The resulting cell suspensions were filtered through 40 μm cell strainers to remove undigested debris. Cells were quenched with ice-cold RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS), treated with ACK lysis buffer (3–5 mL) to remove red blood cells, and washed again with RPMI 1640 containing 10% FBS. Isolated single-cell suspensions were used for downstream single-cell RNA sequencing or flow cytometry analyses. Single-cell suspensions from three mice per group were pooled to generate one scRNA-seq library.

Single-cell RNA sequencing and data processing

Single-cell RNA sequencing (scRNA-seq) library preparation and sequencing were performed as previously described [20]. Briefly, viable single-cell suspensions were isolated from three biologically independent mice per genotype (KAR and KPC) and pooled within each group to generate one scRNA-seq library using the 10 × Genomics Chromium platform. Sequencing was performed to a depth sufficient for robust transcriptome coverage. Raw gene–cell matrices were filtered to exclude low-quality cells (< 200 detected transcripts per cell or > 10% mitochondrial gene expression) and low-abundance genes (< 10 cells per gene). Filtered matrices were imported into the Seurat R package (v3.1.2) for downstream analysis. Gene expression values were normalized such that the total number of unique molecular identifiers (UMIs) per cell was scaled to the median UMI count across all cells. A total of 2,000 highly variable genes were identified for principal component analysis (PCA). Significant principal components were determined using the JackStraw method, and the top principal components (PCs 1–10) were used for clustering and visualization by Uniform Manifold Approximation and Projection (UMAP).

Differential gene expression analyses between cell populations were performed using the Wilcoxon rank-sum test implemented in Seurat. To account for multiple hypothesis testing, P values were adjusted using the Benjamini–Hochberg false discovery rate (FDR) correction, and genes with an adjusted P value < 0.05 were considered significantly differentially expressed. Gene Ontology (GO) overrepresentation analysis was performed using the clusterProfiler R package, and enrichment significance was determined based on Benjamini–Hochberg-adjusted P values. Unless otherwise specified, transcriptomic analyses were performed using default parameters implemented in Seurat and associated R packages.

Functional enrichment and pathway analysis

Differentially expressed gene clusters were analyzed for functional enrichment using the clusterProfiler R package. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed to characterize biological processes and signaling pathways associated with specific immune and myeloid cell populations. A cutoff value of p < 0.05 and q < 0.05 was applied to determine statistical significance.

Manufacturing and production of anti-CXCR2 antibodies

A peptide-based immunization strategy was employed to generate murine monoclonal antibodies against human CXCR2. The immunogen peptide corresponded to the N-terminal region of CXCR2 was selected as the target epitope for antibody generation. Following immunization, antibody-producing hybridomas were generated using standard hybridoma technology. Five high-affinity clones were initially selected for further characterization, and a monoclonal antibody-producing hybridoma, NICR2-5, was chosen as the lead clone for subsequent development.

ELISA screening, antibody sequencing, and humanization

The binding specificity of NICR2-5 was evaluated by enzyme-linked immunosorbent assay (ELISA) against the CXCR2 antigen peptide, a CXCR1 control peptide, and a non-specific control peptide. NICR2-5 showed selective binding to the CXCR2 antigen peptide without detectable cross-reactivity to CXCR1 or non-specific peptides, across a serial dilution range of 0–1000 ng/mL. To enable antibody engineering, next-generation sequencing (NGS) was performed on NICR2-5 hybridomas to determine full-length heavy- and light-chain sequences and identify complementarity-determining regions (CDRs). Multiple NICR2-5 hybridoma clones exhibited > 99% sequence homology, supporting monoclonality. For therapeutic development, NICR2-5 was humanized based on its CDR sequences via gene synthesis and cloned into an expression vector for small-scale recombinant antibody production. The resulting humanized anti-CXCR2 antibody (huCXCR2-Ab) retained specific binding to the CXCR2 antigen peptide as confirmed by ELISA.

Affinity characterization and hybridoma-based antibody production

The binding affinity of huCXCR2-Ab to the CXCR2 immunogen was determined by surface plasmon resonance (SPR) using a Biacore platform, yielding an equilibrium dissociation constant (Kd) of approximately 0.3 nM (3 × 10⁻¹⁰ M). Murine anti-CXCR2 (muCXCR2-Ab) monoclonal antibodies were generated using standard hybridoma-based manufacturing. Hybridoma cells were expanded under optimized culture conditions, and a post-thaw viability of > 85% was required prior to production. An initial small-scale test production (50 mL) was performed to evaluate antibody yield, followed by scale-up to a 3 L production volume.

Antibody purification, quality control, and formulation

Monoclonal antibodies were purified from hybridoma culture supernatants using affinity chromatography. Purified antibodies were formulated in phosphate-buffered saline (PBS, pH 7.3) without preservatives and supplied as liquid formulations under cold-chain conditions (2–4 °C). Final antibody concentrations ranged from 2.5 to 3.1 mg/mL, as determined by bicinchoninic acid (BCA) protein assay. Antibody purity exceeded 90%, as assessed by SDS–PAGE. Endotoxin levels were measured using a chromogenic endotoxin quantification assay and were below 0.1 EU/µg, ensuring suitability for in vivo applications. Antibody identity, specificity, and functional integrity were further validated by ELISA using the CXCR2 antigen peptide. The final product consisted of a mouse IgG monoclonal antibody with a total yield of approximately 160 mg.

In vivo antibody treatment

For therapeutic studies, KAR tumor-bearing mice (n = 5 per group) were randomly assigned to four treatment groups receiving anti-PD-L1 antibody, anti-CXCR2 antibody, combination therapy, or vehicle control. Anti-PD-L1 antibody (InVivoMAb anti-mouse PD-L1, B7-H1; Bio X Cell, Cat. #BE0101) was administered at a dose of 200 μg per mouse. Anti-CXCR2 antibody was administered at 30 mg/kg per mouse. Antibodies were delivered by intraperitoneal injection according to the following schedule: anti-PD-L1 was administered twice per week, and anti-CXCR2 was administered once per week, for a total treatment duration of 2 weeks. Tumor burden was assessed at endpoint by tumor size and weight measurements. Tumor tissues were collected for the evaluation of immune cell infiltration, proliferation, apoptosis, and target protein expression.

β-arrestin recruitment assay

IL-8–induced β-arrestin recruitment to CXCR2 was measured using a LiveBLAzer™ FRET-based β-arrestin assay (Thermo Fisher Scientific), which detects β-arrestin recruitment through β-lactamase reporter complementation and fluorescence resonance energy transfer (FRET). CXCR2-expressing cells were pre-incubated with humanized anti-CXCR2 antibody (huCXCR2-Ab) at a final concentration of 80 µg/mL (approximately 3.2 nM) for 30 min at 37 °C, followed by stimulation with recombinant human IL-8 at a concentration of 300 ng/mL. Cells were subsequently loaded with the LiveBLAzer™ FRET substrate CCF4-AM and incubated according to the manufacturer’s protocol. β-arrestin recruitment was quantified by measuring the blue/green fluorescence emission ratio, and signals were normalized to IL-8–stimulated control groups.

Gi signaling and cAMP measurements

To evaluate CXCR2-mediated Gi signaling, intracellular cAMP levels were quantified using a FRET-based cAMP detection assay following IL-8 stimulation. CXCR2-expressing HMC-1 and MiaPaCa-II cells were pre-treated with huCXCR2-Ab at 80 µg/mL for 30 min, followed by stimulation with recombinant human IL-8 at 300 ng/mL. Changes in intracellular cAMP levels were measured as HTRF ratio signals, reflecting Gi-dependent inhibition of adenylate cyclase activity. Data were normalized to untreated control or IL-8–stimulated groups as indicated.

Quantification and statistical analysis

All experiments were performed at least three times in triplicate, and representative data are shown. Statistical analyses were performed using GraphPad Prism 5.0 software. Quantitative data are presented as mean ± SEM. Comparisons between two groups were performed using unpaired two-tailed Student’s t test. Comparisons among multiple groups were analyzed using one-way or two-way ANOVA followed by Tukey’s multiple comparisons post hoc test.

Normality and homogeneity of variance were assessed prior to parametric analyses using the Shapiro–Wilk normality test and Brown–Forsythe test, respectively. A p value < 0.05 was considered statistically significant. Statistical significance was indicated as *, p < 0.05; **, p < 0.01; ***, p < 0.001; and ns, not significant.

Results

CXCR2 is upregulated in PDAC and correlates with poor clinical outcomes

To determine the clinical relevance of CXCR2 across human malignancies, we first analyzed its mRNA expression using The Cancer Genome Atlas (TCGA) pan-cancer dataset. CXCR2 transcripts were broadly elevated across multiple tumor types compared with corresponding normal tissues, with the highest expression observed in pancreatic ductal adenocarcinoma (PDAC) (Fig. 1A), suggesting a particularly prominent role for CXCR2 in pancreatic cancer.

Fig. 1.

CXCR2 is upregulated in pancreatic cancer and correlates with poor clinical outcomes. A CXCR2 mRNA expression across multiple cancer types from The Cancer Genome Atlas (TCGA) datasets, showing markedly elevated expression in pancreatic ductal adenocarcinoma (PDAC). B–C Kaplan–Meier survival analyses of PDAC patients stratified by CXCR2 expression levels. High CXCR2 expression is significantly associated with reduced overall survival (OS) (B) and disease-free survival (DFS) (C). D Comparison of CXCR2 mRNA expression between PDAC tumor tissues and normal pancreatic tissues using the GEPIA2 database. E Representative immunohistochemistry (IHC) images of human PDAC tissue microarrays showing strong CXCR2 staining in tumor epithelial cells but minimal or no staining in adjacent normal pancreatic tissues. Scale bars, 100 μm

Survival analyses further demonstrated that high CXCR2 expression was associated with significantly worse overall survival (OS) and disease-free survival (DFS) in PDAC patients (Fig. 1B–C), indicating that CXCR2 expression correlates with poor clinical outcomes. Consistent with the pan-cancer analysis, direct comparison of PDAC tumors and normal pancreatic tissues using gene expression profiling interactive analysis 2 (GEPIA2) database revealed significantly higher CXCR2 expression in tumors (Fig. 1D). To validate these findings at the protein level, we performed immunohistochemical (IHC) staining of CXCR2 on human PDAC tissue microarrays. CXCR2 protein expression was readily detected in PDAC specimens, whereas minimal staining was observed in adjacent non-tumor tissues. Notably, CXCR2 expression was detected not only in infiltrating immune cells but also within tumor epithelial compartments, highlighting its dual contributions to both tumor-intrinsic and microenvironment-mediated disease progression (Fig. 1E). These data suggest CXCR2 as a clinically relevant and prognostically unfavorable marker in PDAC, supporting its potential as a therapeutic target in pancreatic cancer.

Development and characterization of a high-affinity humanized anti-CXCR2 antibody

Using a peptide-based immunization strategy followed by hybridoma screening, we generated a murine monoclonal antibody, designated NICR2-5, that specifically recognizes a conserved peptide epitope of human CXCR2. Enzyme-linked immunosorbent assay (ELISA) demonstrated that NICR2-5 exhibited strong and selective binding to the CXCR2 antigenic peptide, while showing no detectable reactivity toward the closely related CXCR1 peptide or irrelevant control peptides (Fig. 2A).

Fig. 2.

Development and characterization of a high-affinity humanized anti-CXCR2 antibody. A Enzyme-linked immunosorbent assay (ELISA) screening of mouse anti-human CXCR2 monoclonal antibody clones generated by peptide-based immunization. The lead clone, NICR2-5, showed strong binding to the CXCR2 antigenic peptide, with no detectable cross-reactivity to CXCR1 or irrelevant control peptides. B Humanization of NICR2-5 was performed based on next-generation sequencing (NGS)-identified complementarity-determining regions (CDRs) of the heavy and light chains. The resulting humanized antibody (huCXCR2-Ab) retained high binding specificity to CXCR2 as confirmed by ELISA. C Surface plasmon resonance (SPR) analysis demonstrated that huCXCR2-Ab exhibited high-affinity binding to recombinant human CXCR2 protein, with an equilibrium dissociation constant (KD) of 0.3 nM. D Predicted structural model of huCXCR2-Ab bound to the extracellular domain of CXCR2. The antibody heavy chain is shown in blue, light chain in yellow, and complementarity-determining regions (CDRs) in orange; the CXCR2 seven-transmembrane domain is displayed in magenta. E Flow cytometric analysis showing specific recognition of endogenous CXCR2 on human pancreatic cancer cell lines (PANC-1 and MiaPaCa-2) and leukemia cell lines by huCXCR2-Ab, confirming its functional specificity and potential for therapeutic application across CXCR2-expressing tumors

To enable therapeutic development, next-generation sequencing (NGS) was performed on the NICR2-5 hybridoma to determine the full-length antibody sequences and precisely define the complementarity-determining regions (CDRs) of both heavy and light chains. Based on these sequences, antibody humanization was carried out to generate a humanized anti-CXCR2 antibody (huCXCR2-Ab). Importantly, the humanized antibody retained robust and specific binding to the CXCR2 peptide, as confirmed by ELISA (Fig. 2B), indicating successful preservation of antigen recognition following humanization. The binding affinity of huCXCR2-Ab to recombinant human CXCR2 protein was next evaluated using surface plasmon resonance (SPR) analysis. Biacore measurements revealed that huCXCR2-Ab bound CXCR2 with high affinity, exhibiting an equilibrium dissociation constant (Kᴅ) of approximately 0.3 nM (Fig. 2C), consistent with a strong and stable antibody–antigen interaction. To gain structural insight into the molecular basis of CXCR2 recognition, a docking model of the huCXCR2-Ab–CXCR2 complex was generated using AlphaFold2-based structure prediction and ClusPro docking analysis. The model suggested that huCXCR2-Ab engages the extracellular domain of CXCR2 primarily through interactions mediated by the heavy-chain CDRs, forming a stable contact interface with the receptor’s extracellular loops (Fig. 2D). This predicted binding mode provides a structural rationale for the high specificity and affinity observed experimentally. To confirm direct recognition of CXCR2 on the cell surface, flow cytometric analysis was performed using human cancer cell lines expressing endogenous CXCR2. huCXCR2-Ab robustly bound to CXCR2-positive pancreatic cancer cell lines, including PANC-1 and MiaPaCa-2, as well as the leukemia cell line HMC-1, which served as a positive control (Fig. 2E). These results demonstrate that huCXCR2-Ab effectively recognizes native CXCR2 in a cellular context.

huCXCR2-Ab suppresses cancer cell proliferation in vitro and inhibits CXCR2-driven signaling

To investigate the functional impact of huCXCR2-Ab on CXCR2 signaling inhibition and tumor cell growth, we first examined its effects on key downstream signaling pathways activated upon receptor engagement. IL-8 stimulation is known to promote β-arrestin recruitment to CXCR2 and activate Gi-coupled signaling cascades. As shown in Fig. 3A, huCXCR2-Ab effectively blocked IL-8–induced β-arrestin recruitment in a dose-dependent manner, indicating efficient disruption of ligand–receptor interaction and receptor activation. Consistent with this observation, huCXCR2-Ab markedly reversed IL-8–mediated inhibitory guanine nucleotide-binding protein (Gi) activation and restored intracellular cAMP levels in both HMC-1 and MiaPaCa-II cells (Fig. 3B–C), showing functional inhibition of CXCR2-dependent Gi signaling in distinct CXCR2-expressing tumor cell types.

Fig. 3.

huCXCR2-Ab suppresses CXCR2 signaling and inhibits tumor cell proliferation in vitro. A β-arrestin recruitment assay showing that huCXCR2-Ab effectively inhibits IL-8–induced β-arrestin recruitment to CXCR2 in a dose-dependent manner, indicating blockade of receptor–ligand interaction. B–C Measurement of Gi protein activation and intracellular cAMP levels in HMC-1 and MiaPaCa-II cells following IL-8 stimulation. huCXCR2-Ab reversed IL-8–induced Gi activation and restored intracellular cAMP concentration, confirming disruption of downstream G-protein signaling. D Western blot analysis of ERK1/2 phosphorylation showing that huCXCR2-Ab suppresses ligand-induced ERK activation, indicating inhibition of CXCR2-mediated mitogenic signaling. E In vitro proliferation assays demonstrating that huCXCR2-Ab significantly inhibits the growth of CXCR2-expressing pancreatic (MiaPaCa-II) and leukemia (HMC-1) cells in a dose-dependent manner. Data represent mean ± SD from three independent experiments. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test

In addition to Gi-coupled signaling cascades, CXCR2 activation is known to promote mitogenic responses through the MAPK/ERK pathway. Western blot analysis revealed that huCXCR2-Ab significantly suppressed ERK1/2 phosphorylation induced by multiple CXCR2 ligands including C-X-C motif chemokine ligand 2 (CXCL2), CXCL5 and CXCL8 (also named as IL-8), while total ERK levels remained unchanged, confirming effective blockade of CXCR2 downstream proliferative signaling (Fig. 3D). Finally, in vitro proliferation assays demonstrated that huCXCR2-Ab markedly inhibited the growth of CXCR2-expressing pancreatic ductal adenocarcinoma (MiaPaCa-II) and leukemia (HMC-1) cells in a dose-dependent manner (Fig. 3E), supporting a direct functional consequence of CXCR2 signaling inhibition on tumor cell proliferation.

ARID1A deficiency remodels the tumor immune microenvironment toward a CXCR2-high, myeloid-dominant state

Identification of specific biomarker for patient selection is an important issue for precision cancer therapy. We recently established a genetically engineered Pdx1-CreKras^G12DArid1a^L/L (KAR) mice to address the role of Arid1a in pancreatic tumorigenesis [20]. This PDAC model is driven by constitutively active oncogenic Kras^G12D and pancreas-specific Arid1a deletion. To characterize how ARID1A deficiency reshapes the TME in PDAC, we first examined the expression of immune regulatory molecules at the tissue level. IHC analysis revealed markedly increased CXCR2 expression in ARID1A-deficient (KAR) tumors compared with KPC tumors, accompanied by a concomitant upregulation of PD-L1 (Fig. 4A). Quantitative assessment confirmed significantly higher CXCR2- and PD-L1–positive areas in KAR tumors, indicating establishment of an immune suppressive microenvironment associated with ARID1A loss (Fig. 4B).

Fig. 4.

ARID1A deficiency remodels the tumor immune microenvironment toward a CXCR2-high, myeloid-dominant state. A Representative immunohistochemical (IHC) staining of CXCR2 and PD-L1 in pancreatic tumors from KPC and ARID1A-deficient (KAR) mice. Scale bars, 100 μm. B Quantification of CXCR2-positive and PD-L1-positive areas in KPC and KAR tumors. Data are presented as mean ± SD. Statistical significance was determined using unpaired two-tailed Student’s t test and P values are indicated. C UMAP visualization of single-cell RNA sequencing (scRNA-seq) data showing major immune and stromal cell populations identified in KPC and KAR tumors, including T cells, B cells, neutrophils, macrophages, monocytes, dendritic cells, myeloid-derived suppressor cells (MDSCs), and cancer-associated fibroblasts (CAFs). Single-cell suspensions from three mice per group were pooled to generate one scRNA-seq library. D Dot plot showing representative marker gene expression used for annotation of major cell populations identified by scRNA-seq. Dot size represents the percentage of cells expressing each gene, and color intensity indicates the average expression level. E Proportional analysis of major immune and stromal cell populations in KPC and KAR tumors. F Box plots showing the proportion of individual cell types as a percentage of total cells in KPC and KAR tumors. G Feature plots illustrating the expression of Cxcr2 and its ligands (Cxcl1, Cxcl2, and Cxcl5) in KPC and KAR tumors at single-cell resolution

To systematically define the cellular composition underlying these changes, we performed single-cell RNA sequencing of KPC and KAR tumors. Uniform manifold approximation and projection (UMAP) analysis identified major immune and stromal populations, including T cells, B cells, neutrophils, macrophages, monocytes, dendritic cells, myeloid-derived suppressor cells (MDSCs), and cancer-associated fibroblasts (CAFs) (Fig. 4C). Canonical marker gene expression confirmed accurate annotation of each cell type (Fig. 4D). Comparative analysis of cellular proportions revealed profound remodeling of the immune landscape in KAR tumors. ARID1A-deficient tumors exhibited a marked increase in myeloid populations, particularly neutrophils, MDSCs, and macrophages, whereas adaptive immune populations such as B cells and dendritic cells were significantly reduced (Fig. 4E). T-cell proportions showed no substantial compensatory increase, suggesting that ARID1A deficiency favors accumulation of innate immune cells rather than effective antitumor lymphocyte infiltration. Given the prominent expansion of myeloid cells, we next examined expression of the CXCR2 chemokine axis at single-cell resolution. Feature plots demonstrated robust upregulation of Cxcr2 expression in KAR tumors, predominantly within neutrophils and MDSCs, while minimal expression was observed in KPC tumors (Fig. 4F). In parallel, CXCR2 ligands, including Cxcl1, Cxcl2, and Cxcl5, were selectively enriched in KAR tumors, indicating activation of a CXCR2-dependent chemokine network that may promote recruitment and retention of suppressive myeloid populations (Fig. 4F). To further investigate whether the elevated expression of CXCR2 and its ligands translated into functional intercellular communication, we performed CellChat analysis to infer cell–cell communication networks within the tumor immune microenvironment. Compared with KPC tumors, KAR tumors exhibited increased overall interaction strength and a greater number of inferred cell–cell interactions, accompanied by distinct outgoing and incoming signaling patterns (Supplementary Fig. S1A–D). Ligand–receptor analysis further identified the Cxcl2–Cxcr2 interaction as a major communication axis within the CXCL signaling pathway, predominantly mediating crosstalk among neutrophils and MDSCs in KAR tumors (Supplementary Fig. S1E). These findings support the notion that ARID1A deficiency promotes a CXCR2-centered myeloid communication network that may facilitate the recruitment and maintenance of immunosuppressive myeloid populations. Collectively, these data demonstrate that ARID1A deficiency drives extensive immune remodeling in PDAC, characterized by enhanced CXCR2 and PD-L1 expression and preferential accumulation of CXCR2⁺ myeloid cells. This CXCR2-high, myeloid-dominant tumor microenvironment provides a mechanistic foundation for immune suppression and suggests that ARID1A-deficient tumors may be particularly dependent on CXCR2-mediated signaling for maintaining immune evasion. Given the established role of CXCR2⁺ myeloid populations in suppressing antitumor immunity [22, 23], we next sought to determine whether this CXCR2-driven myeloid remodeling functionally impacts T-cell states and contributes to immune dysfunction in ARID1A-deficient tumors.

CXCR2-associated myeloid programs reshape T-cell states in ARID1A-deficient pancreatic tumors

To investigate the functional consequences of CXCR2-driven myeloid remodeling, we first examined T-cell states in KPC and KAR tumors using single-cell RNA sequencing. Unexpectedly, T cells in KAR tumors exhibited significantly lower regulatory T-cell (Treg) and exhaustion scores compared with those in KPC tumors, indicating a relative reduction of canonical Treg and exhausted phenotypes despite the presence of a highly immunosuppressive tumor microenvironment (Fig. 5A). We next evaluated whether distinct myeloid populations in KAR tumors contribute to this altered T-cell state. By integrating gene signatures associated with T-cell attraction and suppression, we mapped major myeloid clusters according to their functional profiles. Myeloid populations from KAR tumors—including neutrophils, macrophages, and MDSCs—were preferentially positioned toward a T cell–suppressive quadrant, whereas corresponding populations from KPC tumors displayed a more balanced or T-cell–attractive profile (Fig. 5B). These results suggest that ARID1A deficiency skews myeloid cells toward a suppressive functional state capable of restraining effective T-cell responses independently of classical Treg expansion or exhaustion programs.

Fig. 5.

CXCR2-associated myeloid programs reshape T-cell states in ARID1A-deficient pancreatic tumors. A Violin plots showing regulatory T-cell (Treg) score and T-cell exhaustion score in KPC and ARID1A-deficient (KAR) tumors based on single-cell RNA sequencing analysis. B Two-dimensional projection of myeloid clusters based on mean T-cell attraction score (x-axis) and mean T-cell suppression score (y-axis). Dot size reflects the number of cells per cluster. C Dot plot illustrating expression of Cxcr2 and its ligands (Cxcl1, Cxcl2, and Cxcl5) across major immune cell populations in KPC and KAR tumors. D Gene Ontology (GO) overrepresentation analysis of CXCR2⁺ neutrophils from KPC and KAR tumors. Dot size represents gene count, and color intensity indicates adjusted P values. E Feature plots showing single-cell module scores of representative GO programs (GO1–GO4) across the tumor microenvironment, visualized on UMAP projections. (GO1: myeloid leukocyte migration, GO2: chemotaxis, GO3: immune response–regulating signaling pathway, GO4: positive regulation of leukocyte activation). F Box plots comparing GO module scores (GO1–GO4) between KPC and KAR tumors. Statistical significance is indicated for each comparison. G Correlation analyses between GO module scores and expression levels of Cxcr2 and representative CXCR2-associated genes across myeloid populations in KPC and KAR tumors. H Flow cytometric analysis of tumor-infiltrating immune cells from KPC and KAR tumors. Quantification of IFNγ⁺ Th1 cells among CD4⁺ T cells, CD11b⁺F4/80⁺ macrophages and Foxp3⁺ regulatory T cells (Treg) among CD4⁺ T cells. Each dot represents one mouse. Data are presented as mean ± SEM. Statistical significance was determined using unpaired two-tailed Student’s t test. *p < 0.05, **p < 0.01, ***p < 0.001.(I) Immunohistochemical staining for myeloperoxidase (MPO) in KPC and KAR tumors. Areas of positive MPO were quantified by ImageJ software. Statistical significance was determined using unpaired two-tailed Student’s t test. Scale bars, 50 μm

Consistent with this observation, expression of CXCR2 and its cognate ligands (Cxcl1, Cxcl2, and Cxcl5) was highly enriched in myeloid populations, particularly neutrophils and MDSCs, in KAR tumors (Fig. 5C), implicating activation of the CXCR2 chemokine axis as a key driver of myeloid-mediated immune suppression. To further characterize the transcriptional programs associated with CXCR2⁺ myeloid cells, we performed Gene Ontology (GO) overrepresentation analysis. CXCR2⁺ myeloid populations in KAR tumors were significantly enriched for pathways related to leukocyte chemotaxis, myeloid leukocyte migration, and immune response-regulating signaling pathways (Fig. 5D). Visualization of representative GO module scores across the tumor microenvironment revealed broad activation of these chemotactic and immune regulatory programs in KAR tumors (Fig. 5E). Quantitative comparison confirmed that GO module scores associated with chemotaxis and immune regulation were significantly higher in KAR tumors than in KPC tumors (Fig. 5F). Notably, CXCR2 expression levels showed strong positive correlations with these GO module scores across myeloid populations, supporting a direct link between CXCR2 signaling and activation of immunosuppressive transcriptional programs (Fig. 5G). Flow cytometric analysis demonstrated increased frequencies of IFN-γ-expressing Th1 cells and CD11b⁺ myeloid cells, accompanied by reduced frequencies of Foxp3⁺ regulatory T cells in KAR tumors compared with KPC tumors, consisting with the scRNA-seq findings (Fig. 5H). IHC staining for myeloperoxidase (MPO) further confirmed elevated neutrophil infiltration in KAR tumors (Fig. 5I). These data demonstrate that ARID1A deficiency establishes a CXCR2-driven, myeloid-dominant immune microenvironment that reshapes T-cell states and enforces immune suppression primarily through myeloid cell-mediated chemotactic and regulatory programs rather than through expansion of classical Treg or exhausted T-cell compartments.

Combined CXCR2 and PD-L1 blockade reshapes the tumor immune microenvironment and synergistically inhibits tumor growth

Given that ARID1A deficiency establishes a CXCR2-driven, myeloid-dominant immunosuppressive microenvironment that restrains effective T-cell responses independently of classical Treg expansion or exhaustion, we next investigated whether therapeutic blockade of CXCR2 could restore antitumor immunity and enhance the efficacy of immune checkpoint inhibition. To this end, we developed a mouse anti-CXCR2 monoclonal antibody (muCXCR2-Ab) which targets the similar epitope identified by huCXCR2-Ab in mouse CXCR2 protein.

KAR mice bearing established pancreatic tumors were treated with anti-PD-L1 antibody, muCXCR2-Ab, or the combination of both agents according to the indicated treatment schedule (Fig. 6A). While monotherapy with either anti-PD-L1 or anti-CXCR2 resulted in a modest suppression of tumor progression, combined CXCR2 and PD-L1 blockade led to a pronounced reduction in tumor burden (Fig. 6B). Quantitative analyses revealed that tumors from mice receiving the combination therapy exhibited significantly lower tumor weights and smaller tumor sizes compared with those from vehicle- or monotherapy-treated groups (Fig. 6C), indicating a synergistic antitumor effect. IHC analysis of tumor tissues revealed a marked reduction in Ki67-positive proliferating cells and a corresponding increase in cleaved Caspase-3 staining in the combination group, indicating suppressed tumor cell proliferation and enhanced apoptosis (Fig. 6D). In addition, combination therapy significantly increased infiltration of CD8⁺ T cells and expression of IFN-γ, while reducing the abundance of F4/80⁺ tumor-associated macrophages (Fig. 6E). Flow cytometric analysis further demonstrated that combined PD-L1 and anti-CXCR2 antibody treatment markedly increased the frequency of intratumoral CD8⁺ T cells compared with vehicle or either monotherapy alone (Fig. 6F). These findings indicate that combined CXCR2 and PD-L1 blockade is associated with enhanced CD8⁺ T-cell infiltration in ARID1A-deficient tumors. Moreover, treatment with anti-CXCR2 antibody, either alone or in combination with PD-L1 blockade, significantly reduced the frequency of intratumoral CD11b^high^Ly6G^high^ myeloid cells compared with vehicle-treated controls (Fig. 6G). These observations are consistent with reduced CXCR2-associated recruitment of granulocytic myeloid cells and may contribute to the increased CD8⁺ T-cell infiltration observed following combination therapy. However, the present study does not directly establish that CD8⁺ T cells are the primary mediators of the therapeutic efficacy.

Fig. 6.

Combination therapy targeting CXCR2 and PD-L1 remodels the tumor microenvironment and induces tumor regression in ARID1A-deficient pancreatic cancer. A Schematic illustration of the treatment regimen. Eight-week-old ARID1A-deficient (KAR) mice bearing established pancreatic tumors (n = 5 per group) were treated with anti-PD-L1 antibody, anti-CXCR2 antibody, or the combination of both agents according to the indicated schedule. Mice were sacrificed at the experimental endpoint for downstream analyses. B Representative images of pancreatic tumors harvested from mice treated with vehicle (DMSO), anti-PD-L1, anti-CXCR2, or the combination of anti-PD-L1 and anti-CXCR2. C Tumor weight and tumor size were measured at the endpoint and are shown as mean ± SEM. While treatment with either anti-PD-L1 or anti-CXCR2 alone resulted in modest tumor growth inhibition, combination treatment significantly reduced tumor weight and tumor size compared with vehicle or monotherapy groups. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test. D Representative immunohistochemical (IHC) staining of Ki67, cleaved caspase-3, and CXCR2 in tumor tissues from each treatment group. E Immunohistochemical analysis of immune cell infiltration and effector markers in tumor tissues following treatment. Tumor sections were stained for CD8, F4/80, and IFN-γ to evaluate cytotoxic T-cell infiltration, macrophage abundance, and effector activity, respectively. F Flow cytometric analysis of tumor-infiltrating immune cells. Single-cell suspensions from tumor tissues were analyzed for CD4⁺ and CD8⁺ T cells and F4/80⁺ myeloid cells. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test. G Flow cytometric analysis of intratumoral granulocytic myeloid cells. Single-cell suspensions from tumor tissues were analyzed for CD11b^high Ly6G.^high populations. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple comparisons test. Data are presented as mean ± SEM (n = 5 biologically independent mice per group)

Discussion

PDAC remains one of the most lethal cancers, primarily due to its profoundly immunosuppressive TME and poor response to ICB. Although anti-PD-1/PD-L1 therapies have transformed treatment outcomes for melanoma and non-small cell lung cancer, similar success has not been achieved in PDAC[24]. One prominent barrier is the dense infiltration of immunosuppressive myeloid subsets—including tumor-associated neutrophils (TANs), myeloid-derived suppressor cells (MDSCs), and tumor-associated macrophages (TAMs)—which collectively suppress T-cell priming and cytotoxicity. Notably, many of these populations are recruited via CXCR2-dependent chemokine gradients [25].

Our results provide clear preclinical evidence that therapeutic blockade of CXCR2 effectively dismantles this suppressive network and potentiates the antitumor efficacy of PD-L1 inhibition. In ARID1A-deficient PDAC mice, we observed marked elevation of CXCR2 expression in both tumor cells and infiltrating neutrophils alongside high PD-L1 levels. Dual inhibition with anti-CXCR2 plus anti-PD-L1 significantly reduced tumor burden, induced apoptosis, increased CD8⁺ T-cell infiltration, and concomitantly depleted neutrophils in the TME. These findings suggest cooperative therapeutic activity between CXCR2 blockade and PD-L1 inhibition and are consistent with a transition toward a more immune-responsive tumor microenvironment. This aligns with a previous study that demonstrated CXCR2 signaling drives neutrophil extracellular trap (NET) formation, accelerates PDAC progression, and correlates with poor survival[26]. Furthermore, they showed that gemcitabine enhances neutrophil recruitment and NET formation, suggesting chemotherapy may inadvertently promote tumor progression through a CXCR2-dependent immunosuppressive axis.

Our conclusions are well-supported by previous preclinical reports. Steele et al. showed that CXCR2 inhibition facilitates T-cell infiltration and sensitizes PDAC tumors to PD-1 blockade [3], while Piro et al. reported that the CXCR1/2 inhibitor ladarixin improves ICB responsiveness in resistant PDAC models [27]. Similar effects have been observed beyond PDAC. CXCR2 suppression reprogrammed TANs and synergized with anti-PD-1 therapy in non-alcoholic steatohepatitis and hepatocellular carcinoma[28], and enhanced PD-L1 blockade efficacy in neutrophil-rich triple-negative breast cancer[29]. Together, these studies suggest that the CXCR2/PD-L1 axis is a broadly targetable immune-modulatory circuit across cancer types.

Recent advances in myeloid-targeted immunotherapy have highlighted the central role of tumor-associated myeloid populations, including macrophages, neutrophils, and myeloid-derived suppressor cells, in shaping immunosuppressive TMEs and limiting immune checkpoint blockade efficacy [30–32]. Emerging studies have further shifted the field beyond simple depletion of suppressive myeloid populations toward strategies that coordinately remodel both myeloid and lymphoid immune compartments. For example, a recent study demonstrated that TREM2-targeted immunocytokines incorporating protease-activatable masked IL-2 simultaneously antagonized immunosuppressive tumor-associated macrophages (TAMs) while enhancing T-cell and NK-cell activity, thereby generating potent multi-axis antitumor immunity with reduced systemic toxicity [33]. In parallel, targeted degradation of the myeloid checkpoint VISTA significantly enhanced antitumor immunity and improved responsiveness to immunotherapy, further emphasizing the therapeutic importance of remodeling suppressive myeloid ecosystems in cancer [34]. Similarly, IDH1-mutant cholangiocarcinoma was recently shown to establish a CCL2-dependent TAM-enriched microenvironment, and CCL2 neutralization reduced TAM infiltration and suppressed tumor progression, supporting the concept that tumor-specific genetic alterations actively shape immunosuppressive myeloid niches [35]. Emerging strategies now extend beyond systemic antibody blockade to include localized immune remodeling approaches, such as implant-mediated slow-release systems designed to reprogram suppressive myeloid populations within highly immunosuppressive tumors[36]. In addition, recent computational and single-cell interactome studies demonstrated that the efficacy of myeloid-targeted therapies is highly dependent on therapeutic timing, microenvironmental context, and restoration of coordinated myeloid–T-cell communication networks [37, 38]. Together, these findings are consistent with the concept that CXCR2 blockade is associated with remodeling of suppressive myeloid ecosystems, thereby potentially enhancing responsiveness to immune checkpoint blockade.

Compared with currently investigated CXCR2 small-molecule inhibitors, antibody-based therapeutics possess several pharmacological advantages. Owing to their high target specificity, monoclonal antibodies are less likely to interact with structurally related receptors, thereby reducing off-target effects. In addition, the long serum half-life of IgG antibodies enables sustained receptor occupancy and less frequent dosing, while Fc-mediated immune effector functions may further contribute to antitumor activity through mechanisms such as antibody-dependent cellular cytotoxicity (ADCC), which are not achievable with small-molecule antagonists [39–41]. In contrast, small-molecule inhibitors generally exhibit superior tissue penetration, intracellular accessibility, and the convenience of oral administration, which may facilitate drug distribution within dense tumor tissues[42, 43]. The recent clinical success of the oral RAS(ON) inhibitor daraxonrasib in metastatic pancreatic cancer further highlights the therapeutic potential of small-molecule targeted agents, particularly for intracellular oncogenic signaling molecules[44]. However, because CXCR2 is a cell-surface G protein-coupled receptor, antibody-based blockade may provide additional pharmacological advantages by achieving prolonged receptor occupancy, enhanced target selectivity, and potential Fc-mediated immune functions [45–48]. Nevertheless, antibody therapeutics also have inherent limitations, including higher manufacturing costs, the requirement for parenteral administration, and relatively restricted tissue penetration due to their large molecular size[49, 50]. Therefore, these two therapeutic modalities should be considered complementary rather than mutually exclusive, with antibody-based CXCR2 blockade representing a promising strategy for achieving sustained and selective inhibition of CXCR2 signaling while minimizing off-target toxicities.

However, translational challenges remain. A phase II trial evaluating the CXCR2 inhibitor navarixin with pembrolizumab demonstrated limited clinical responses [51], highlighting the necessity for biomarker-guided patient selection. CXCR2 ligand levels—including CXCL8 (IL-8)—may help identify patients who may benefit most from CXCR2-targeted therapy. Supporting this, IL-8 neutralization was reported to enhance response to PD-1 therapy in humanized PDAC models [26]. Yet, as CXCR2 binds at least seven ligands in humans, blockade of a single ligand is not sufficient due to compensatory overproduction of alternative chemokines such as CXCL1, CXCL2, or CXCL5, all elevated in PDAC. This supports our strategy of developing a direct anti-CXCR2 therapeutic antibody, which effectively prevents receptor activation by multiple ligands (CXCL2, CXCL5, IL-8) and suppresses proliferation and neutrophil recruitment (Fig. 3A–3D).

A second major contribution of our work is the identification of a potential responsive molecular subgroup. ARID1A-deficient and KRAS-mutant PDAC exhibited elevated CXCR2 expression and increased neutrophil infiltration, suggesting that tumors harboring these genetic alterations may be particularly sensitive to CXCR2 antibody + ICB co-treatment. Given that ARID1A mutations also occur in other immunotherapy-relevant cancers—such as high-TMB cervical cancer [52]—our findings support the possibility that ARID1A-mutant/CXCR2-high tumors represent a genomically sensitized cohort for immunotherapy.

In summary, our study shows that dual blockade of CXCR2 and PD-L1 is associated with remodeling of the tumor immune microenvironment and significant tumor regression in a preclinical model of ARID1A-deficient PDAC. These findings provide a strong preclinical rationale for further evaluation of CXCR2 antibody-based combination immunotherapy in PDAC and other immune checkpoint blockade–resistant malignancies. In addition, our results suggest that ARID1A deficiency warrants further investigation as a potential predictive biomarker for CXCR2-targeted immunotherapy.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

T.L.K. performed the experiments; T.L.K., Y.C.H., and W.C.H. analyzed the data; Y.C.H. and Y.S.S. provided critical samples/materials and reagents/ protocols; T.L.K, and W.C.H. wrote and edited the manuscript. Y.S.S., L.T.C., and W.C.H. conceived the project idea.

Funding

National Science and Technology Council, 110-2327-B-037-001, 114-3111-Y-043-001.

Data availability

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

Declarations

Conflict of interest

The authors have declared that no conflict of interest exists.