Nrf2 deficiency converts the ESCC microenvironment into an immunologically active state via the GPX2-ICD-DC signaling path

Abstract

Background

Radiotherapy (RT) resistance remains a significant challenge in esophageal squamous cell carcinoma (ESCC). While Nrf2 is known to mediate antioxidant defense, its role in modulating the immunogenicity of radiotherapy-induced cell death is poorly understood.

Methods

We established a spontaneous esophageal cancer model using genetically engineered mice with conditional knockout of Nrf2, and analyzed the tumor immune microenvironment by single-cell RNA sequencing. In vitro, we performed co-culture experiments using CRISPR/Cas9-mediated Nrf2-knockout esophageal cancer cell lines with dendritic cells and T cells to validate immune activation. In addition, in vivo validation was conducted using a mouse subcutaneous tumor model.

Results

ScRNA-seq revealed that Nrf2 deficiency significantly remodeled the myeloid compartment, characterized by a population shift from Folr2 + to Mrc1 + macrophages and an expansion of effector T cells. Mechanistically, Nrf2 deletion downregulated the expression of Gpx2, impairing antioxidant defenses. This sensitized ESCC cells to RT, triggering the release of immunogenic cell death (ICD) markers, including ATP, HMGB1, and surface calreticulin (CRT). In tumor-DC-T cell co-culture systems, Nrf2-deficient cells stimulated dendritic cells (DCs) to secrete IP-10 (CXCL10), which was indispensable for the recruitment and activation of CD8 + T cells. Finally, in vivo experiments confirmed that Nrf2 deficiency enhanced radiosensitivity and promoted CD8 + T cell infiltration.

Conclusion

Our findings identify the Nrf2-Gpx2 axis as a master regulator of immunogenicity in ESCC. Targeting this axis represents a promising strategy to convert “cold” tumors into “hot” environments, thereby improving the efficacy of radiotherapy and immunotherapy.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08323-7.

Introduction

Esophageal cancer is the eleventh most common cancer worldwide and the seventh leading cause of cancer-related death, accounting for 4.6% of all cancer deaths [1]. Esophageal cancer is mainly classified into two types: esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma, with ESCC being the predominant type, accounting for approximately 90% of all cases [2]. For locally advanced esophageal cancer, local recurrence is the main cause of failure of radiotherapy-based multimodal treatments, such as neoadjuvant chemoradiotherapy or definitive chemoradiotherapy [3]. ESCC is typically associated with an immunosuppressive tumor microenvironment, characterized in particular by exhaustion of CD8⁺ T cells and infiltration of regulatory T cells [4]. The combination of chemoradiotherapy and immunotherapy can significantly improve clinical response rates in patients with ESCC [5]. How radiotherapy reshapes the immune microenvironment and thereby enhances the benefit of immunotherapy represents an important area of research.

Nuclear factor erythroid 2-related factor 2 (NRF2) is a critical transcription factor that governs cellular defense mechanisms against oxidative stress and xenobiotics, modulating the expression of over 200 genes involved in antioxidant response, drug metabolism, and cellular resilience [6]. Under normal physiological conditions, Nrf2 levels remain low in healthy cells; however, its activation protects against oxidative stress and inflammation [6, 7]. Nevertheless, constitutive activation of Nrf2 is frequently observed in various cancers, contributing to oncogenesis, immune evasion, and therapeutic resistance [6]. Our previous studies found that esophageal cancer patients in Asian populations have a higher frequency of NRF2 mutations [8]. These mutations can lead to either enhanced or reduced NRF2 function. NRF2 overexpression is associated with poor patient prognosis and heightened resistance to chemoradiation therapy, whereas its negative expression is linked to improved therapeutic efficacy [9–11]. Given this dichotomous role, understanding the precise mechanisms by which NRF2 influences tumor immunobiology, particularly its interplay with specific downstream effectors like GPX2, is crucial for developing targeted therapeutic strategies in esophageal cancer [12]. This investigation specifically aims to elucidate how Nrf2-mediated downregulation of GPX2 contributes to immunogenic cell death and subsequent anti-tumor immunity within the esophageal cancer context, providing a potential avenue for therapeutic intervention.

Immunogenic cell death (ICD) represents a unique modality of regulated cell death that triggers a robust adaptive immune response [13]. The onset of ICD is fundamentally driven by ER stress and the accumulation of reactive oxygen species (ROS), leading to the characteristic release of damage-associated molecular patterns (DAMPs) that recruit and activate immune effectors [14]. Despite its therapeutic potential, effective clinical modalities for inducing ICD in the tumor microenvironment remain scarce. Consequently, developing strategies to augment ICD and enhance the clinical benefits of immunotherapy by modulating the immune activation status within tumors is a critical research priority.

In this study, we utilized a spontaneous ESCC murine model using genetically engineered mice combined with CD45 + immune cells enriched single-cell RNA sequencing (scRNA-seq) to map the dynamic immune landscape following Nrf2 deletion. We demonstrated that epithelial Nrf2 deficiency fundamentally remodels the myeloid compartment, characterized by a significant shift in macrophage subsets and an expansion of effector T cell populations. We demonstrated that molecular pathways through which Nrf2-mediated GPX2 suppression elicits robust anti-tumor immune responses, potentially transforming current therapeutic paradigms for esophageal cancer. Specifically, we hypothesize that inhibiting Nrf2 will lead to a decrease in GPX2, thereby promoting immunogenic cell death and enhancing anti-tumor immunity in esophageal cancer.

Method

Cell culture

The human cell lines KYSE150, KYSE30, and THP-1 were purchased from the National Collection of Authenticated Cell Culture (Shanghai, China). The MEC25 cell line was acquired from Shenzhen Wenhua Times Technology Co., Ltd. All cells were cultured in RPMI 1640 medium or Dulbecco’s Modified Eagle’s Medium (DMEM; Gibco), supplemented with 10% fetal bovine serum (FBS; Gibco). The cells were maintained in a humidified incubator with 5% CO₂ at 37 °C. All cell lines were tested and confirmed negative for mycoplasma.

ATP release assay

KYSE30 and KYSE150 cells were seeded in six-well plates and subjected to radiotherapy for 24–48 h at doses of 3, 6, or 9 Gy. Extracellular ATP release was measured using an ATP Detection Kit (MCE HY-K0314) according to the manufacturer’s instructions. The luminescence (RLU) values of ATP standards were plotted against their respective concentrations to generate a standard curve, from which the extracellular ATP levels of the samples were determined.

Co-culture of DC and T cells

THP-1 cells were seeded in six-well plates at a density of 1*10⁶ cells per well in RPMI 1640 complete medium. To induce differentiation into dendritic cells (DCs), the cells were treated with PMA (50 ng/ml) and IL-4 (40 ng/ml) on Day 0. On Day 3, the medium was replaced with fresh medium containing IL-4 (40 ng/ml). On Day 5, the medium was replaced again with the addition of LPS (20 ng/ml) and IL-6 (20 ng/ml). Human peripheral blood–derived T cells were isolated and used without further subset purification. For T cell activation, T cells were pre-stimulated with CD28 and CD3 antibodies for 48 h before being co-cultured with the irradiated tumor cells.

Flow cytometry for CRT and macrophage proportions

Flow cytometry was employed to precisely detect the cell surface expression of calreticulin (CRT), a hallmark of immunogenic cell death (ICD). Irradiated KYSE150 and KYSE30 cells were collected, washed twice, and incubated with the CRT antibody. The surface expression of CRT was then analyzed using a flow cytometer. FlowJo software was used for the analysis of the final data.

Western blotting for DAMPs analysis

For Western blot (WB) analysis, irradiated cells were harvested and lysed using RIPA lysis buffer. Protein concentrations were determined using a BCA Protein Assay Kit (Epizyme ZJ101). Protein samples (10 mg) were separated by SDS-PAGE and transferred onto PVDF membranes. After blocking, the membranes were incubated with primary antibodies overnight at 4 °C. The following antibodies were used: Beta-Actin (CST 4970), CRT (HUABIO ET1608), GPX2 (AbClonal A15999), Nrf2 (CST D1Z9C), and Nqo1 (Abcam EPR3309). Signals were visualized using an imaging system to evaluate the expression levels of Nrf2, GPX2, and CRT. Signals were visualized using an imaging system. Band intensities were quantified using ImageJ software by densitometric analysis and normalized to β-Actin, which was used as a loading control.

Mouse model and treatments

Genetically engineered mice Nrf2^flox+/+ K14^cre+/+were purchased from Shanghai Model Organisms Center, Inc, To generate epithelial-specific Nrf2 knockout mice, six-week-old mice were administered TAMOXIFEN (sigma); 20 mg/ml, via intraperitoneal injection at a dose of 100 mg/kg body weight for three consecutive days. Subsequently, the mice were given drinking water containing 0.1 mg/ml 4-nitroquinoline-1-oxide (4NQO), a carcinogen targeting the upper aerodigestive tract. 4NQO was first dissolved in DMSO (10 mg/ml) and then diluted in drinking water. This treatment lasted for 16 weeks, followed by 10 weeks of normal drinking water. Mice were then euthanized to collect esophageal tumor tissues.

For the subcutaneous tumor model using wild-type recipient mice, five mice were included in each group, unless otherwise specified. Mice were subcutaneously injected with 6 × 10⁶ Nrf2-knockout mEC25 cells (Nrf2-knockout group, n = 5) or an equal number of Nrf2 wild-type mEC25 cells (control group, n = 5). Approximately day 10 post-injection, both groups received localized radiotherapy (RT). RT was delivered using an electron beam in two fractions of 3 Gy (total dose of 6 Gy) on days 10 and 12 [15]. During irradiation, mice were anesthetized and protected by a lead shield featuring a 1 cm² aperture to specifically expose the tumor. Mice were sacrificed two weeks later, and the tumor volume and weight were measured. The in vivo experiments were independently repeated to ensure reproducibility.

Immunofluorescence and immunohistochemistry

For immunofluorescence (IF), cells seeded on coverslips in 24-well plates were treated with 6 Gy of radiation for 24 h. Cells were fixed with 4% paraformaldehyde, permeabilized with 0.4% Triton X-100 for 30 min, and blocked with goat serum for 1 h. Coverslips were incubated with primary antibodies against CRT and Nrf2 overnight at 4 °C, followed by incubation with CoraLite594/488-conjugated secondary antibodies (Elabscience) for 1 h. Nuclei were stained with DAPI, and images were captured using a confocal microscope. For immunohistochemistry (IHC), tumor tissues were embedded and sectioned. Sections were deparaffinized in xylene and dehydrated through a series of ethanol gradients. After rehydration, antigen retrieval was performed in boiling EDTA buffer for 15 min. Sections were permeabilized with 0.4% Triton X-100 and blocked with goat serum. Primary antibodies against CD8 (Abcam, Ab33786), CD68 (Abcam, Ab283654), and CD80 (Ab254579) were incubated overnight at 4 °C. For IHC staining, samples were incubated with HRP-conjugated secondary antibodies and developed using DAB. For tissue IF, sections were incubated with CoraLite594/488-conjugated secondary antibodies and DAPI, and visualized using a fluorescence microscope (IX73, Olympus). For each group, n = 3 independent biological replicates (slides) were used. Three non-overlapping fields of view were randomly selected from each slide for image acquisition and analysis. Quantitative data are presented as mean ± standard deviation (SD) or standard error of the mean (SEM). Statistical analysis was performed using appropriate tests, such as one-way analysis of variance (ANOVA).

Bulk RNA-seq and scRNA-seq of mouse esophageal samples

Dysplastic or malignant lesions were harvested from the spontaneous esophageal cancer model. Tissues were minced and digested in RPMI-1640 (Invitrogen) containing Collagenase IV (Gibco) and Hyaluronidase (Sigma-Aldrich). Cells were labeled with CD45⁻FITC antibody (553080, BD Biosciences; 1:20) and sorted via fluorescence-activated cell sorting (FACS) using a FACSAria (BD Biosciences). CD45⁺ and CD45⁻ cells were re-mixed at a 2:1 ratio to ensure sufficient immune cell representation for downstream immune landscape analysis.

Single cells were captured into nanoliter droplets using the Chromium system (10× Genomics). Single-cell RNA sequencing (scRNA-seq) libraries were prepared with the Chromium Single Cell 5′ Reagent Kit and sequenced on an Illumina HiSeq x10 system. Raw gene expression matrices were generated using CellRanger (v2.1.0) and integrated via the Seurat R package (v2.3.4). Quality control (QC) involved filtering genes detected in < 0.1% of total cells and excluding cells with < 500 gene counts or > 10% mitochondrial gene expression. A total of 35,483 cells were retained. Highly variable genes (HVGs) were defined as those with standardized expression between 0.0125 and 3 and a dispersion > 0.5. Dimensionality reduction was performed via Principal Component Analysis (PCA), followed by t-SNE analysis using the RunTSNE function. Cell types were annotated based on canonical marker genes.

Clustering and marker gene identification

Epithelial and stromal cells were subjected to independent re-analysis. DoubletFinder was used to remove doublets, and cells with hybrid features (e.g., co-expressing Cd3d and Cd19) were manually excluded. Batch effects were corrected using the Harmony package. Clustering was performed using the FindClusters function in Seurat, and cluster-specific markers were identified using the FindAllMarkers function.

Gen set variation analysis (GSVA)

Pathway analysis was performed using KEGG and GO databases via the MSigDB database (v6.2). To estimate pathway activity at the single-cell level, we employed Gene Set Variation Analysis (GSVA) using the GSVA package (v1.30.0). Differential activity between clusters was analyzed using the Limma package (v3.38.3). Significant pathways (P < 0.05) were visualized via heatmaps based on t-values and average activity scores.

Statistical analyses

Data are presented as the mean ± standard deviation (SD) of at least three independent experiments. Statistical analyses were performed using GraphPad Prism 9 software. Statistical significance was assessed using a two-tailed Student’s t-test, and a P-value < 0.05 was considered statistically significant.

Human samples

The PBMCs used for the in vitro co-culture experiments were derived from healthy donors.

Results

ScRNA-seq reveals altered immune cell composition following epithelial Nrf2 deletion in esophageal cancer

We established a spontaneous murine model of esophageal cancer and validated the knockout efficiency through immunofluorescence. Eight-week-old genetically engineered mice with tamoxifen-inducible Nrf2-deficient esophageal epithelium (designated as Mod, for knock down) and their control littermates with the same genotype but without tamoxifen treatment (designated as Nc, for negative control) were treated with 4-nitroquinoline-1-oxide (4NQO) for 16 weeks to induce spontaneous tumorigenesis. (Fig. S1A-B). To better characterize the dynamics of immune cells within the tumor and facilitate robust analysis in the single-cell dataset, CD45+ immune cells and CD45- cells were mixed at a 2:1 ratio via fluorescence-activated cell sorting (FACS) prior to single-cell RNA sequencing (Fig. 1A). A total of 39,516 cells were ultimately captured, comprising 25,501 CD45+ immune cells and 14,015 CD45- cells. These cells were classified into seven major subsets, including epithelial cells, fibroblasts, endocrine cells, B cells, T/NK cells, neutrophils, and monocytes (Fig. 1D). The relative proportions of each cell type across the two mouse models were visualized using bar charts (Fig. 1B). Comparative analysis revealed that the proportion of fibroblasts in Nrf2-knockout esophageal tumors was significantly lower than that in wild-type mice. Conversely, the proportions of T/NK cells, neutrophils, and monocytes were significantly higher in the Nrf2-deficient group (Fig. 1C). Furthermore, we stratified the samples based on Nrf2 expression levels and compared their immune infiltration scores. These findings collectively reveal substantial remodeling of the esophageal immune microenvironment following Nrf2 deficiency, characterized by a significant increase in immune cell infiltration (Fig. 1E).

Fig. 1.

(A) Schematic illustration of the experimental workflow. Spontaneous ESCC was induced by 4NQO in K14Cre+/+ Nrf2flox+/+ mice. CD45+ leukocytes were enriched prior to scRNA-seq. (B) UMAP visualizing the seven major cellular ecosystems identified in the TME, color-coded by cell type. (C) Quantitative analysis of the fractional abundance of major cell lineages in Mod (knock down, Nrf2-deficient) versus Nc (negative control) tumors. (D) Dot plot visualizing the expression of signature marker genes for distinct major subsets. (E) Heatmap displaying the differential enrichment scores of immune-related signatures across Nrf2 High and Nrf2 Low. (F) Uniform Manifold Approximation and Projection (UMAP) visualization of the subclustered cell populations, revealing three distinct subsets: Dendritic cells (DC), M1-like macrophages, and M2-like macrophages. (G) Bar chart comparing the relative proportions of the DC, M1-like, and M2-like subclusters between the Mod and Nc groups. (H) Dot plot illustrating the expression patterns of canonical marker genes across the identified cell subsets. The size of the dot represents the percentage of cells expressing a specific gene within the corresponding cluster, while the color gradient indicates the scaled mean expression level

Nrf2 loss drives macrophage phenotypic shifts within the TME

Given the prominent alterations observed in immune infiltration, we next focused on the myeloid compartment. Based on canonical marker expression, myeloid cells were annotated into dendritic cells (DCs) and macrophage populations (Fig. 1F). DCs were identified by high expression of Cd209a, Ccl17, and Cd74, whereas macrophages were further classified into M1-like and M2-like subsets according to established transcriptional signatures (Fig. 1H).

Quantitative analysis revealed that both M1-like and M2-like macrophage populations were expanded in Nrf2-deficient tumors compared with controls, whereas DCs were present at relatively low abundance, limiting further subtype resolution (Fig. 1G). Additionally, the M1/M2 functional scores were higher in the Mod group than in the Nc group. (Fig. S1 C). These findings suggest that epithelial Nrf2 loss is associated with a preferential expansion of macrophage populations within the myeloid compartment, rather than a uniform alteration across all myeloid cell types.

To further resolve myeloid cell heterogeneity, we performed subclustering analysis and identified six transcriptionally distinct myeloid subsets, comprising five macrophage clusters and one dendritic cell (DC) cluster (Fig. 2A–C). Based on marker gene expression and M1/M2 polarization scores (Fig. S1D), these subsets were identified as DC cluster (characterized by Cd209a and Ccl17), an inflammatory chemokine-enriched cluster (Marco_Cxcl3 defined by high expression of Cxcl3/9/10), a Vcam1-high cluster (Marco_Vcam1), and a stress-response cluster expressing heat shock protein Hspa1a and Hspa1b (Marco_Hspa1b), Additionally two M2-like populations were identified: a Folr2⁺ tissue-resident cluster (Marco_Folr2) and an Mrc1⁺ macrophage cluster (Marco_Mrc1) displaying high levels of Mrc1 (CD206), Cd163, and Msr1.

Fig. 2.

Nrf2 deficiency drives phenotypic remodeling of myeloid subsets. (A) UMAP plot showing the sub-clustering of myeloid cell subsets. (B) Stacked bar chart displaying the proportional distribution of each myeloid subset in the Mod and Nc groups. (C) Dot plot visualizing the expression of signature marker genes for distinct myeloid subsets. (D) Expression of specific functional genes in macrophage subsets. (E) Comparison of MDSC function scores (in Neutrophils) between Mod and Nc groups, based on S100a8, S100a9, Arg1, Mmp9, and Itgam. (F) Expression of S100a8 in neutrophils. (G) Comparison of immunosuppressive gene expression in mast cells between Mod and Nc groups. (H) Relative proportion of MDSCs in Mod and Nc groups. (I) Expression of MDSC-associated genes in Mod and Nc groups

Notably, Nrf2 deficiency induced a selective phenotypic shift from Folr2+ to Mrc1+ macrophages (Fig. S1E, F). This shift does not represent a simple transition between classical M2 states, but rather suggests a remodeling of macrophage functional states. Crucially, further analysis of functional gene expression in the relevant macrophage subclusters showed that CXCL1, CXCL2, and CXCL3 were markedly upregulated in the Marco_Cxcl3 subset (Fig. 2D). Previous studies have reported that the CXCL1/2/3 chemokine axis is associated with the recruitment of myeloid cells, including tumor-associated macrophages and MDSCs [16].

Consistent with the reduction of this chemokine-enriched subset, which has been associated with the recruitment of immunosuppressive myeloid cells, the MDSC functional score (based on S100a8, S100a9, Arg1, Mmp9, and Itgam) within the neutrophil compartment was significantly lower in the Mod group (Fig. 2E, F). While the expression of the immune checkpoint molecule CD274 (PD-L1) was markedly increased in mast cells (Fig. 2G). These findings suggest that Nrf2 deficiency may influence the immune regulatory state of multiple myeloid cell populations in the tumor microenvironment. Further single-cell proportion analysis revealed that the frequency of monocytic MDSCs (M-MDSCs) in the tumor microenvironment was increased, rising from 0.39% in the control group to 0.93% in the Mod group (Fig. 2H). In contrast, polymorphonuclear MDSCs (PMN-MDSCs) remained at a very low level in both groups (approximately 0.03%). However, detailed analysis of the gene expression profiles of MDSC subsets (Fig. 2I) showed that there were no significant differences between the Mod and Nc groups in the expression proportion (dot size) or average expression level of classical immunosuppressive genes, including Arg1, Nos2, S100a8, S100a9, Tgfb1, and Il10. These findings suggest that although NRF2 deficiency is associated with a modest expansion of the M-MDSC population, there is no clear evidence of enhanced immunosuppressive transcriptional activity. Therefore, MDSCs are unlikely to represent the dominant source of immunosuppression in Nrf2-deficient tumors.

Taken together, the selective shift from Folr2⁺ tissue-resident macrophages toward Mrc1⁺ macrophages, together with the reduced representation of chemokine-driven immunosuppressive subsets and the limited functional contribution of MDSCs, indicates a substantial reprogramming of the myeloid landscape toward a less suppressive immune context, potentially supporting enhanced CD8⁺ T cell infiltration.

NRF2 loss reshapes the T cell compartment and drives progenitor exhausted CD8⁺ T cell accumulation

T cells play a critical role as cytotoxic effector cells in tumor immunity. To evaluate the impact of NRF2 deficiency on the T cell compartment within the tumor immune microenvironment (TIME), we performed subclustering analysis of T/natural killer (NK) cells. Based on the expression of signature genes, eight distinct subpopulations were identified: CD4_Naive,( defined by Cd4, Ccr7, Il7r, and Sell) CD4+T_Il18r1 (Il18r1), CD4+T_Tnfsf11 (Tnfsf11), CD4_Treg (Foxp3, Pdcd1, and Tigit), CD8_Naive (Cd8a, Ccr7, and Sell), CD8_Gzmk (Cd8a, Gzmk, and Nkg7), γδT (Trdc and Trgc1), and NK cells (Ncr1 and Klrd1) (Fig. 3A, B, E).Comparison of subpopulation proportions revealed markedly increased CD8⁺ T cell infiltration in the TIME of NRF2‑deficient tumors (Fig. S1G). Specifically, the proportion of the CD8_GZMK subset increased from 4.61% to 6.46%, while the CD8_Naive subset rose from 1.58% to 7.81%. Concurrently, the immunosuppressive CD4_Treg population decreased from 27.23% to 24.10%, and γδT cells decreased from 26.47% to 19.13% (Fig. 3C, Fig. S1H). Immunohistochemical analysis further confirmed a significant increase in CD8⁺ T cell infiltration in Nrf2‑deficient tumors (Fig. 3D).

Fig. 3.

Nrf2 deficiency remodels the T/NK cell landscape in the tumor immune microenvironment. (A) UMAP clustering of T cell and NK cell subsets. (B) Stacked bar chart displaying the proportional distribution of each T and NK subset in the Mod and Nc groups. (C) Proportional distribution of T cell subsets between the two groups. (D) Representative immunohistochemical (IHC) staining of CD8+ T cells in the esophageal epithelium, showing early immune infiltration patterns. (E) Dot plot showing the expression of canonical marker genes for T cell subsets (e.g., Gzmk, Foxp3, Pdcd1). (F) CD8+T cell functional marker. (G) Heatmap displaying the functional signature scores of different T cell subsets. (H–J) Dot plots comparing the expression of key functional genes in CD4+ T cells (H), myeloid cells (I), and DC cells (J)

We next evaluated the functional status of CD8⁺ T cells. Comparative analysis of total CD8⁺ T cell function‑associated genes showed that, compared with the Nc group, the Mod group exhibited significantly reduced expression of the classical cytotoxic effector molecules Prf1, Ifng, and Fasl, while expression of immunosuppressive and exhaustion‑related molecules Ctla4, Pdcd1, and Tox was increased. Moreover, Tcf7, a transcription factor associated with stem‑like/progenitor status, was markedly upregulated in the Mod group (Fig. 3F). These results suggest that although NRF2 deficiency promotes CD8⁺ T cell infiltration, these T cells do not display enhanced cytotoxic function and are more likely to exist in a progenitor exhausted state rather than a fully activated cytotoxic effector state.

Further functional characterization at the subpopulation level revealed that CD4⁺ and CD8⁺ naive T cells were primarily enriched in naive‑related pathways, whereas CD8_GZMK, γδT, and NK cells exhibited higher activity in effector‑related pathways such as cytotoxic program, IFN response, and cytokine/chemokine signaling, indicating that these cells possess more pronounced effector or inflammatory transcriptional features. Notably, CD8_GZMK cells showed relatively higher cytotoxicity-related pathway activity compared with other T cell subsets (Fig. 3G), suggesting that this population retains certain effector-like transcriptional features. However, consistent with the overall CD8⁺ T cell analysis, these infiltrating CD8⁺ T cells in the NRF2-deficient group did not exhibit enhanced cytotoxic gene expression, supporting the notion that they represent a progenitor-exhausted population rather than fully functional cytotoxic T cells.

To explore the mechanisms underlying the increased infiltration of CD8⁺ T cells, we further analyzed changes in chemokine signaling. Analysis of key functional genes in CD4⁺ T cells revealed that CD4⁺ T cells in the NRF2‑knockout group expressed significantly higher levels of chemokines such as Cxcl9, Ccl3, and Ccl4 (Fig. 3H), which have been shown to promote effector T cell recruitment into tumor tissues. Additionally, marked upregulation of Cxcl9 and Cxcl10 was observed in myeloid cells and dendritic cells (DCs), and DCs also showed high expression of the maturation markers CD83 and CD80 (Fig. 3I, J). Collectively, these results suggest that NRF2 deficiency may enhance T cell recruitment into the tumor microenvironment by promoting chemokine secretion and facilitating DC maturation, ultimately leading to increased CD8⁺ T cell infiltration. These transcriptional features suggest that CD8⁺ T cells in Nrf2-deficient tumors may shift toward a progenitor exhausted state rather than a fully cytotoxic phenotype.

Together, these findings suggest that epithelial Nrf2 loss reshapes the tumor immune microenvironment into a stat characterized by enhanced T cell recruitment but impaired cytotoxic function, reflecting a “T cell–inflamed yet dysfunctional” immune landscape.

Epithelial Nrf2 deficiency leads to the collapse of antioxidant defense homeostasis via the Gpx2 axis

As Nrf2 deletion was restricted to esophageal epithelial cells in this model, we next examined transcriptional alterations within the epithelial compartment to explore the mechanism of TME. Differential gene expression and gene set enrichment analyses revealed marked changes in metabolic and stress-response pathways. Pathways associated with oxidative stress responses were significantly enriched in control tumors, whereas oxidative phosphorylation and translation-related pathways were relatively activated in Nrf2-deficient epithelial cells (Fig. 4A, B). In contrast, pathways related to cell motility and epidermal development were suppressed. These transcriptional changes suggest a substantial disruption of redox-associated programs in Nrf2-deficient epithelial cells.

Fig. 4.

Nrf2 deficiency impairs antioxidant defense and induces transcriptional reprogramming in esophageal epithelial cells.(A) GSEA showing the Normalized Enrichment Scores (NES) of significantly enriched pathways. Note the enrichment of metabolic and stress-related pathways (e.g., Oxidative phosphorylation, Response to oxidative stress) and the suppression of developmental pathways. (B) GSEA enrichment plot for the “Response to Oxidative Stress” pathway. (C) Volcano plot visualizing differentially expressed genes (DEGs) in Nc versus Mod epithelial cells. Blue dots represent downregulated genes, highlighting the suppression of key Nrf2 downstream targets involved in antioxidant defense (Gpx2, Mgst1 and Ftl1). Red dots indicate upregulated genes. (D) Immunohistochemical validation of GPX2 gene expression in spontaneously tumor-bearing mice.

Volcano plot analysis demonstrated broad transcriptional remodeling following Nrf2 loss, with significant downregulation of genes involved in antioxidant defense and metabolic regulation. Multiple Nrf2-regulated antioxidant and metabolic genes were downregulated (e.g., Rplp0, Mgst1, Ftl1, Adh7) (Fig. 4C). Among these, Gpx2 showed a marked decrease and represents a well-established NRF2 downstream antioxidant enzyme involved in cellular redox homeostasis. Considering that radiotherapy-induced immunogenic cell death (ICD) critically depends on the accumulation of reactive oxygen species (ROS), genes involved in ROS detoxification are likely to influence this process. Among these, Gpx2—a key antioxidant enzyme and known Nrf2 target—was prominently reduced, a finding further validated by immunohistochemical analysis (Fig. 4D). These results suggest a substantial disruption of redox-associated programs in Nrf2-deficient epithelial cells. Given the marked downregulation of GPX2, a key antioxidant target of Nrf2, we next examined whether Nrf2 deficiency is associated with enhanced susceptibility to immunogenic cell death following radiotherapy.

Disruption of the Nrf2/Gpx2 axis sensitizes ESCC cells to radiation-induced immunogenic cell death

To further elucidate how radiotherapy-treated Nrf2-deficient tumor cells influence the paracrine characteristics of dendritic cells (DCs), we established an in vitro co-culture system and performed a systematic analysis of the cytokine profiles in the culture supernatants using multiplex bead-based assays (Fig. S2A). Our results demonstrated that Nrf2 deficiency significantly remodeled the cytokine landscape generated by tumor–DC interactions compared to the control group (Fig. 5A).

Fig. 5.

NRF2 knockdown potentiates radiation-induced immunogenic cell death (ICD) in esophageal squamous cell carcinoma cells.(A, D) Western blot analysis of NRF2, Calreticulin (CRT), and GPX2 protein expression in KYSE150 (A) and KYSE30 (D) cells following irradiation (0, 6, 9 Gy). (B, E) Quantification of extracellular ATP secretion in the culture supernatants of KYSE150 (B) and KYSE30 (E) cells at 24 and 48 hours post-irradiation.(C, F) ELISA analysis of HMGB1 release in the supernatants of KYSE150 (C) and KYSE30 (F) cells 24 hours after irradiation with the indicated doses.(G, H) Flow cytometric analysis of cell surface CRT exposure in KYSE150 (G) and KYSE30 (H) cells. (I–L) Representative immunofluorescence images and quantitative analysis of NRF2 (red) and CRT (green) expression in KYSE30 (I, J) and KYSE150 (K, L) cells. Nuclei were counterstained with DAPI (blue). Scale bars = 20 μm. For each group, n = 3 independent biological replicates (slides) were used. Three non-overlapping fields of view were randomly selected from each slide for image acquisition and analysis. Quantitative data are presented as mean ± standard deviation (SD) or standard error of the mean (SEM).

Within the co-culture supernatants, the secretion of IP-10 (CXCL10) was markedly elevated in the Nrf2-knockout group. As a prototypical Th1-type chemokine essential for the recruitment of CD8⁺ T cells, the elevation of IP-10 suggests a robust enhancement of chemotactic signaling within this co-culture system. Concurrently, MCP-1 (CCL2) levels were also significantly higher in the Nrf2-deficient group than in the control. Given the established role of MCP-1 in recruiting monocytes and macrophages, this finding aligns with the increased proportions of myeloid cells observed in our previous single-cell RNA sequencing (scRNA-seq) analysis.

Furthermore, compared to the control, the Nrf2-deficient co-culture system exhibited varying degrees of elevation in the pro-inflammatory cytokine IL-6, the hematopoietic factor G-CSF, and the immunomodulatory factor IL-10, whereas levels of the pro-angiogenic factor VEGF were significantly diminished. Collectively, these cytokine alterations indicate that Nrf2 deficiency substantially reshapes the cytokine landscape associated with tumor–DC interactions, characterized by the upregulation of multiple chemotactic and inflammatory mediators alongside a reduction in VEGF signaling.

Given the observed transcriptional alterations in redox regulation, we next examined whether NRF2 loss influences tumor cell responses to radiotherapy-induced immunogenic cell death (ICD). Using CRISPR/Cas9-mediated knockdown of NRF2 in KYSE30 and KYSE150 cell lines, quantitative analysis of western blot band intensities further confirmed the significant reduction of GPX2 and NQO1 protein levels following NRF2 silencing (Fig. S1I, J). Consistently, GPX2 protein expression was reduced following NRF2 silencing, along with decreased expression of NQO1, a canonical NRF2 downstream target involved in cellular antioxidant defense, thereby confirming effective inhibition of NRF2 signaling (Fig. 5A, D). To assess ICD-associated molecular features, cells were exposed to varying doses of ionizing radiation (3, 6, and 9 Gy). Radiotherapy induced increased release of HMGB1 in both cell lines (Fig. 5C, F). ATP release was further evaluated at 24 and 48 hours following irradiation. While temporal and dose-dependent effects varied between cell lines, NRF2-deficient cells consistently exhibited higher extracellular ATP levels compared with controls (Fig. 5B, E). Based on these results, a 6 Gy dose was selected for subsequent experiments as it elicited consistent ICD-associated responses across both cell lines. Surface exposure of calreticulin (CRT), another hallmark of ICD, was assessed by immunofluorescence and flow cytometry. Both assays demonstrated increased CRT exposure in NRF2-deficient KYSE30 and KYSE150 cells following radiotherapy (Fig. 5G–L). The rightward shift observed in flow cytometry histograms reflects an increase in CRT surface expression intensity, which does not necessarily translate into a proportional increase in the percentage of CRT-positive cells.Notably, CRT exposure detected by flow cytometry and immunofluorescence reflects cell-surface translocation rather than changes in total protein abundance, which may not necessarily be captured by whole-cell lysate Western blot analysis. Collectively, these data indicate that NRF2-deficient tumor cells exhibit an enhanced ICD-associated molecular signature in response to radiotherapy.

NRF2/GPX2 axis remodel the tumor-DC cytokine milieu and enhance recruitment-related signaling

To further elucidate how radiotherapy-treated NRF2-deficient tumor cells influence the paracrine characteristics of dendritic cells (DCs), we established an in vitro co-culture system and performed a systematic analysis of the cytokine profiles in the culture supernatants using multiplex bead-based assays (Fig. S2A). Our results demonstrated that NRF2 deficiency significantly remodeled the cytokine landscape generated by tumor–DC interactions compared to the control group.

Within the co-culture supernatants, the secretion of IP-10 (CXCL10) was markedly elevated in the NRF2-knockout group. As a prototypical Th1-type chemokine essential for the recruitment of CD8⁺ T cells, the elevation of IP-10 suggests a robust enhancement of chemotactic signaling within this co-culture system. Concurrently, MCP-1 (CCL2) levels were also significantly higher in the NRF2-deficient group than in the control (Fig. 6A). Given the established role of MCP-1 in recruiting monocytes and macrophages, this finding aligns with the increased proportions of myeloid cells observed in our previous single-cell RNA sequencing (scRNA-seq) analysis.

Fig. 6.

NRF2-deficient tumor cells reshape the cytokine milieu and enhance CD8+ T cell effector function in a dendritic cell-dependent manner. (A) Violin plots displaying the multiplex cytokine analysis of supernatants from tumor-DC co-cultures. (B) Schematic of the co-culture experimental design (Triple vs. Double systems). (C) Flow cytometric quantification of TNF-α, Granzyme B (GZMB), and IFN-γin CD8+ T cells from the Double co-culture system (no DCs). (D) Comparative analysis of T cell effector function in NRF2-knockout groups in the presence or absence of DCs. (E) T cell activation in Triple co-culture (with DCs). *P < 0.05, **P < 0.01, ***P < 0.001.

Furthermore, compared to the control, the NRF2-deficient co-culture system exhibited varying degrees of elevation in the pro-inflammatory cytokine IL-6, the hematopoietic factor G-CSF, and the immunomodulatory factor IL-10, whereas levels of the pro-angiogenic factor VEGF were significantly diminished. Collectively, these cytokine alterations indicate that NRF2 deficiency substantially reshapes the cytokine landscape associated with tumor–DC interactions, characterized by the upregulation of multiple chemotactic and inflammatory mediators alongside a reduction in VEGF signaling.

DC-mediated antigen cross-presentation is essential for T-cell activation triggered by NRF2 deficiency

To delineate the cellular mechanisms underlying T cell activation in NRF2-deficient tumors, we established tumor–PBMC co-culture systems in the presence or absence of dendritic cells under conditions of strong (anti-CD3/CD28) or suboptimal (anti-CD28 alone) T cell stimulation (Fig. 6B). Flow cytometric analysis of T cell effector molecules revealed that, in the absence of DCs, NRF2-deficient tumor cells did not induce significant changes in CD8⁺ T cell expression of GZMB or IFN-γ compared with control tumor cells, regardless of the strength of T cell activation (Fig. 6C).In contrast, the inclusion of DCs in three-cell co-culture systems resulted in pronounced differences. Under suboptimal T cell stimulation conditions, co-culture with NRF2-deficient tumor cells significantly increased CD8⁺ T cell expression of GZMB and IFN-γ (Fig. 6E). Similar enhancement was observed under strong T cell activation conditions (Fig. 6D), and these findings were consistently reproduced in an independent cell line, KYSE30 (Fig. S2B).These results indicate that the enhanced CD8⁺ T cell effector responses observed in NRF2-deficient settings are not mediated by direct tumor–T cell interactions but require the presence of dendritic cells.

NRF2 loss is associated with improved tumor control following radiotherapy in vivo

To evaluate the in vivo relevance of these findings, we established a subcutaneous tumor model using NRF2-deficient Mec25 cells and assessed tumor response to radiotherapy. Tumors derived from NRF2-deficient cells exhibited significantly reduced growth following irradiation compared with control tumors (n = 5 mice per group), as reflected by both tumor volume and weight measurements (Fig. 7A–B).Multiplex immunofluorescence analysis revealed increased infiltration of CD8⁺ T cells, CD80⁺ antigen-presenting cells, and CD68⁺ macrophages in irradiated NRF2-deficient tumors (Fig. 7C). Quantification was performed based on multiple randomly selected fields per slide, as described in the Methods section. We further validated these findings by immunohistochemistry (IHC) analysis of CD8 and CD80 expression. While CD8 staining showed a clear increase, CD80 expression was less pronounced by IHC, possibly due to its relatively low expression level and the limited sensitivity of chromogenic detection methods. (Fig. 7D).These observations indicate that epithelial NRF2 loss is associated with an immune-inflamed tumor microenvironment that coincides with improved tumor control following radiotherapy.

Fig. 7.

Nrf2 deficiency enhances radiosensitivity and promotes immune infiltration in vivo. (A) Representative images of subcutaneous tumors excised from Sg-Con and Sg-Nrf2 mice following radiotherapy. (B) Tumor growth curves measuring tumor volume over time. Data are presented as mean ± SD. (C) Representative multiplex immunofluorescence (mIF) images of tumor tissues stained for CD8 (T cells), CD80 (activated APCs), and CD68 (macrophages). Scale bars = 100 μm. Quantification of CD8⁺ T cells, CD80⁺ cells, and CD68⁺ macrophages based on mIF images. Data are presented as mean ± SD (n = 5 mice per group). *p < 0.05, **p < 0.01, ***p < 0.001 (D) Representative immunohistochemical (IHC) images of CD8 (T cell marker) and CD80 (activation marker for APCs) in tumor tissues from the subcutaneous xenograft model.

Discussion

Esophageal squamous cell carcinoma (ESCC) is a malignancy characterized by poor prognosis and high rates of local recurrence. Radiotherapy and definitive chemoradiotherapy are the primary treatments for patients with locally advanced ESCC or those ineligible for surgery [17]. Beyond its direct cytotoxic effects, radiotherapy can stimulate the release of tumor antigens and heat shock proteins, thereby activating anti-tumor immune responses [18]. However, radioresistance remains a formidable obstacle limiting its clinical efficacy. Previous studies have established that Nrf2 is frequently overexpressed in ESCC, correlating with resistance to chemoradiotherapy and poor prognosis; conversely, negative Nrf2 expression in clinical samples often predicts a superior therapeutic response [19]. Nevertheless, the precise mechanisms by which Nrf2 modulates the tumor immune microenvironment (TME) through its influence on tumor immunobiology remain poorly understood. Our study addresses this critical question by examining tumor-intrinsic redox programming. We demonstrate that epithelial-specific Nrf2 deficiency significantly enhances the immune response to radiotherapy in ESCC. This effect is not merely a result of increased radiosensitivity but is driven by a profound remodeling of the TME. Utilizing a 4NQO-induced spontaneous ESCC model and single-cell RNA sequencing (scRNA-seq) analysis, we observed a significant increase in immune cell infiltration within Nrf2-deficient tumors, characterized notably by the expansion of effector CD8⁺ T cells and the restructuring of the myeloid compartment. Notably, the remodeling of the myeloid compartment did not follow a simple pro- versus anti-inflammatory dichotomy. Although a shift from Folr2⁺ tissue-resident macrophages toward Mrc1⁺ macrophages was observed, this change may not necessarily indicate enhanced immunosuppression, as macrophage states in tumors exist along a functional spectrum rather than discrete M1/M2 categories. In parallel, the reduction of the CXCL1/2/3-enriched macrophage subset suggests diminished myeloid-driven recruitment signals, indicating a reorganization rather than a uniform activation of the myeloid compartment [16].

Importantly, our data also reveal a dissociation between immune cell recruitment and effector function. While CD8⁺ T cell infiltration was increased, these cells exhibited features of functional impairment and progenitor exhaustion. This discrepancy may be explained by enhanced antigen release and dendritic cell activation driven by immunogenic cell death, together with compensatory chemokine signaling (e.g., CXCL9/10), which promote T cell recruitment independently of classical CXCL1/2/3-mediated pathways. This shift was accompanied by a reduction in the proportion of fibroblasts and immunosuppressive Treg populations, indicating a transition from a relatively immunosuppressive TME to an “immuno-inflamed” state [20].

Nrf2 functions as a master transcription factor that governs cellular defense against oxidative damage and ionizing radiation by activating downstream target genes involved in reactive oxygen species (ROS) detoxification and DNA repair [21, 22]. Substantial evidence has demonstrated that Nrf2 activation in cancer cells promotes tumor progression and metastasis while conferring resistance to chemoradiotherapy [23]. Interestingly, in an immunological context, Nrf2 downregulation has been reported to impair CD8⁺ T cell functionality by altering glutathione (GSH) and ROS levels [24]. Furthermore, Nrf2 influences radiotherapy sensitivity through the regulation of redox homeostasis, which prevents high-Nrf2-expressing populations from benefiting effectively from treatment [25]. GPX2 is a crucial downstream mediator of Nrf2-driven antioxidant stress responses. In hepatocellular carcinoma, downregulation of GPX2 has been shown to sensitize tumor cells to immunogenic cell death (ICD), subsequently activating dendritic cells (DCs) [26].Consistent with these findings, we observed that Nrf2-deficient tumor cells exhibit markedly enhanced ICD hallmarks under radiotherapy, including increased ATP release, heightened HMGB1 secretion, and the surface exposure of calreticulin (CRT). These molecular events are fundamental markers of ICD, effectively promoting antigen presentation and activating both innate and adaptive immune responses [27]. Therefore, the Nrf2–GPX2 axis acts as a critical determinant of whether tumor cell death possesses immune-activating potential.

Previous research indicates that damage-associated molecular patterns (DAMPs) released during ICD must be captured, processed, and cross-presented by DCs to be translated into robust CD8⁺ T cell-mediated immune responses [28]. Our study further elucidates the indispensable role of DCs as mediators in this immune amplification process. In an in vitro co-culture system, Nrf2-deficient tumor cells did not directly enhance T cell effector function; rather, significant augmentation of CD8⁺ T cell cytotoxicity occurred only in the presence of DCs. This result further supports the premise that radiotherapy-induced immune responses depend on DC-mediated “signal conversion” rather than direct tumor-T cell interactions.

From a clinical translation perspective, our findings offer significant insights. While the combination of radiotherapy and immune checkpoint inhibitors (ICIs) has become a vital therapeutic direction for ESCC, clinical outcomes vary significantly between individuals [29, 30]. Based on our results, tumors with Nrf2 overexpression or constitutive activation may form an “immunologically cold” microenvironment—characterized by low ROS levels, low immunogenicity, and minimal DC activation—following radiotherapy, thereby limiting the efficacy of combined immunotherapy. Radiotherapy-induced ICD can promote antigen release and dendritic cell activation, thereby providing the necessary priming signals for T cell–mediated immunity [31]. However, the effectiveness of immune checkpoint inhibitors largely depends on the presence of a pre-existing inflamed tumor microenvironment. In this context, NRF2 inhibition, by enhancing ICD-associated signals and immune cell infiltration, may improve tumor immunogenicity and thereby potentiate the therapeutic efficacy of combined radiotherapy and immune checkpoint blockade.

Although this study utilized a genetically engineered NRF2-deficient model, the NRF2 pathway may also be pharmacologically targetable. Several small molecules capable of suppressing NRF2 signaling have been reported in preclinical studies. For example, ML385 has been shown to inhibit NRF2 transcriptional activity by disrupting its binding to antioxidant response elements, thereby sensitizing tumor cells to chemotherapy and radiotherapy [32, 33]. In addition, agents such as brusatol have been reported to transiently suppress Nrf2 signaling and enhance the efficacy of anti-cancer therapies [34–37]. However, these agents remain at the preclinical stage, and their specificity, pharmacokinetics, and safety profiles require further investigation. Therefore, while pharmacological targeting of NRF2 is promising, its clinical applicability warrants careful validation.

This study acknowledges several limitations. First, while GPX2 was identified as a key downstream effector linking Nrf2 to ICD susceptibility, the potential synergy of other Nrf2-regulated metabolic and stress-related pathways in immune remodeling warrants further investigation. Second, as this study was primarily conducted using murine models, the relationship between the Nrf2–GPX2 axis and radiotherapy-induced immune efficacy must be validated in more clinically relevant models and patient samples.

In conclusion, our study reveals that epithelial Nrf2 maintains redox homeostasis via GPX2, thereby restraining radiotherapy-induced ICD and the subsequent DC-dependent activation of T cell immunity. Targeting this regulatory axis represents a promising strategy to transform radiotherapy-induced cell death into a potent anti-tumor immune response, ultimately enhancing the overall efficacy of radiotherapy in ESCC.

Conclusion

In summary, this study identifies epithelial Nrf2 as a key regulator of redox balance, immune activation, and radiotherapy response in esophageal squamous cell carcinoma. Loss of Nrf2 disrupts the Nrf2–GPX2 antioxidant axis, increases tumor cell susceptibility to radiotherapy-induced immunogenic cell death, and promotes dendritic cell–dependent activation of cytotoxic CD8⁺ T cells (Fig. 7). This remodeling of the tumor immune microenvironment toward an immune-inflamed state significantly enhances tumor control after radiotherapy. Collectively, these findings highlight the Nrf2–GPX2 axis as a critical determinant of radiotherapy efficacy and suggest Nrf2 as a promising target for improving radiotherapy outcomes through immune-based combination strategies.

Fig. 8.

Loss of Nrf2 in tumor cells reduces the expression of GPX2, thereby weakening the antioxidant stress capacity of tumor cells. Radiotherapy effectively induces immunogenic cell death (ICD) in Nrf2-deficient tumor cells, leading to the release of CRT, HMGB1, and ATP, which subsequently activate T-cell–mediated cytotoxicity through antigen presentation by dendritic cells

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Author contributions

Yu Liu: Investigation and Writing—Original Draft. Qing Guo: Conceptualization and Project administration. Ming Liu: Data curation and Funding acquisition. Chong Li: Methodology and Validation. Dashan Ai and Qi Liu: Resources and Supervision. Yun Chen and Shengnan Hao: Software and Visualization. Kuaile Zhao and Fengtao Su: Writing—Review & Editing and Funding acquisition. All authors contributed to the article and approved the submitted version.

Funding

This study was supported by the National Natural Science Foundation of China (U22A20326 and 82172666, 82473577 to F. Su).

Data availability

The datasets analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethical approval

The animal experiments received approval from the Animal Ethics Committee at Fudan Cancer research Institute Technology Platform (FUSCC-IACUC-2024136). All animal procedures were performed in strict accordance with the International Guide for Animal Research Reporting (ARRIVE 2.0).

Consent for publication

Not applicable.

Competing interests

The authors declare no conflicts of interest.