Ferroptotic tumor cells reprogram tumor-associated macrophage antigen presentation to enhance the efficacy of immune checkpoint blockade

Summary

Although evidence links ferroptosis to tumor immunity, the rationale and translational potential of ferroptosis-based therapy remain unresolved. Here, we show that inducing tumor-cell ferroptosis enhances anti-tumor immunity by potentiating major histocompatibility complex II (MHC-II)-dependent antigen presentation in tumor-infiltrating macrophages. Multi-omics analyses reveal that all-trans retinoic acid (ATRA) released from ferroptotic tumor cells directly targets CD38 through the transcriptional factor retinoic acid receptor alpha (RARα) and activates transcription factor EB (TFEB) to control MHC-II expression in macrophage by inducing autophagy. Clinically, a ferroptosis signature correlates with improved immunotherapy response. We also developed a drug-free nano-redox lever that selectively targets and disrupts glutathione metabolism in hypoxic tumor regions by accepting electrons, thereby potentiating ferroptosis-mediated immune stimulation. This creates a positive feedback loop wherein activated macrophages further promote immune-driven tumor ferroptosis, synergizing with anti-PD-1 (programmed cell death protein 1) therapy across preclinical models. Together, our study identifies an uncovered role for ferroptosis in tumor immunity and provides a clinically translatable approach to enhance immunotherapy efficacy.

Graphical abstract

Highlights

• •Ferroptosis triggers anti-tumor immunity by activating macrophage antigen presentation
• •Ferroptotic tumor cell-derived ATRA elevated MHC-II expression on macrophages
• •RARα-CD38-TFEB axis induces autophagy to control MHC-II expression
• •A tumor-selective DFNRL enhances ferroptosis and synergizes with anti-PD-1 therapy

Sun et al. demonstrate that inducing tumor ferroptosis activates anti-tumor immunity by potentiating MHC-II-dependent macrophage antigen presentation. A drug-free nano-redox lever selectively targets hypoxic tumor regions to stimulate ferroptosis-mediated immune activation and synergizes with anti-PD-1 therapy, providing a clinically translatable approach to enhance immunotherapy efficacy.

Introduction

Immune checkpoint blockade has revolutionized cancer therapy over the past decade.¹ However, response rates remain limited among most solid tumors, such as colorectal cancer (CRC) and hepatocellular carcinoma (HCC).^2,3 Lack of T cell infiltration, defective antigen presentation, and immunosuppressive signals contribute to poor immunotherapy response.⁴ Recent works highlight the indispensable role of antigen presentation in anti-tumor immunity, as it initiates adaptive immune responses against cancer.^5,6 Expression of major histocompatibility complex (MHC), the critical mediator of antigen presentation, correlates with favorable immune checkpoint therapy in tumors.⁷ Enhancing MHC expression to augment antigen presentation holds promise for improving immunotherapy effectiveness. Nonetheless, clinically translatable approaches targeting antigen presentation remain elusive.

Innate immunocytes, including dendritic cells (DCs) and macrophages, are essential components of the tumor immune microenvironment (TIME) that express MHC molecules and act as antigen-presenting cells (APCs) critical for activating T cell-mediated adaptive immunity against tumors.⁸ While DCs strongly prime T cell responses, they are scarce in tumors and exhibit dampened antigen presentation capacity.⁹ In contrast, tumor-associated macrophages (TAMs) represent one of the most abundant immune populations within the TIME and are classically categorized into M1 anti-tumor and M2 pro-tumor subtypes.¹⁰ Dynamic crosstalk between tumor cells and TAMs orchestrates the functional states of TAMs.¹¹ Although TAMs are often considered immunosuppressive, re-education of TAMs toward a tumor-suppressive phenotype and restoration of their antigen presentation capacity have emerged as promising anti-cancer strategies.¹² Despite extensive studies on TAM polarization, the regulation of antigen presentation in these key APCs remains unclear.

Ferroptosis is an iron-dependent form of regulated cell death driven by metabolic dysfunction, leading to lipid peroxidation accumulation and oxidative damage.¹³ Recent evidence supports ferroptosis as an immunogenic cell death modality. Immunomodulating molecules, including ATP, high-mobility group box 1 (HMGB1), and certain lipid metabolites released from dying ferroptotic cells, can act as damage-associated molecular patterns to activate the innate immune response.^14,15 Conversely, interferon-γ (IFN-γ) secreted by activated CD8⁺ T cells can trigger cancer cell ferroptosis by reducing system Xc⁻ expression.^16,17 This positive feedback loop provides a rationale for ferroptosis-based strategies to improve immunotherapy. However, ferroptosis does not always stimulate the TIME, underscoring context-dependent effects and adding complexity to maximizing anti-tumor immunity. For instance, ferroptosis can also lead to the recruitment of tumor-associated polymorphonuclear myeloid-derived suppressor cells, thereby attenuating responses to anti-PD-1 (programmed cell death protein 1) therapy, which emphasizes that mechanisms influencing the effects of ferroptosis on immunity require further clarification.¹⁸

Among the various identified pathways defending against ferroptosis, the glutathione (GSH)/glutathione peroxidase 4 (GPX4) system plays the most central role.¹⁹ Depletion of GSH or inactivation of GPX4 induces robust ferroptosis; however, ferroptosis inducers such as erastin and RSL3 lack Food and Drug Administration (FDA) approval and tumor-targeting specificity, raising concerns about their clinical applicability.²⁰ Nicotinamide adenine dinucleotide phosphate (NADPH) is a key cofactor that sustains cellular redox balance by regenerating reduced GSH from oxidized glutathione (GSSG). Hypoxic tumor cells frequently express elevated levels of nitroreductases, which utilize NADPH to catalyze the reduction of nitroimidazole compounds.²¹ Depletion of NADPH through nitroimidazole metabolism results in GSSG accumulation and a significant decrease in GSH levels, ultimately leading to impairment of the GSH defense system and amplification of lipid peroxidation.^22,23 This biochemical vulnerability provides a potential strategy for tumor-selective induction of ferroptosis.

Herein, we demonstrate that ferroptosis stimulates anti-tumor immunity through macrophage antigen presentation, where ferroptotic tumor cell-derived all-trans retinoic acid (ATRA) activates the retinoic acid receptor (RAR) alpha (RARα)-CD38-TFEB (transcription factor EB) pathway to upregulate MHC-II expression in macrophages. This creates a positive feedback loop where activated macrophages further promote immune-mediated tumor ferroptosis, which we leveraged by developing a drug-free nano-redox lever (DFNRL) that selectively depletes GSH to enhance ferroptosis and improve anti-PD-1 therapy across preclinical models.

Results

Triggering ferroptosis activates tumor immune microenvironment

Compared with the known cancer cell-intrinsic function, the influence of tumor ferroptosis on the TIME remains poorly understood. To investigate how tumor ferroptosis acts on TIME, we elicited ferroptosis using two small molecules that disrupt GSH-GPX4 antioxidant system in vivo, RSL3 and erastin, respectively (Figure 1A). Treatment with RSL3 and erastin led to the accumulation of the lipid peroxidation by-products 4-hydroxynonenal and malondialdehyde, as well as downregulation of GPX4 in tumor lesions, indicating successful induction of tumor ferroptosis (Figure S1A). MC38 tumor-bearing mice treated with RSL3 showed reduced tumor volumes and weights compared with vehicle controls (Figures 1B and 1C). Fluorescence-activated cell sorting (FACS) analysis revealed increased tumor infiltration of CD4⁺ T helper and CD8⁺ cytotoxic T lymphocytes (CTLs) following RSL3 treatment (Figure 1D). Specifically, the proportion of exhausted T cells expressing the checkpoint molecules PD-1 and T cell immunoglobulin and mucin domain-containing protein 3 (TIM-3) decreased, while the proportion of IFN-γ⁺CD8⁺ T cells increased following RSL3 treatment (Figure 1D). The cytotoxic cytokine granzyme B secreted by activated CTLs was also highly upregulated in the tumor microenvironment by RSL3 (Figure S1B). Similar findings were also observed in erastin-treated Hepa1-6 tumor-bearing mice (Figures 1E–1H and S1B). These findings indicate that ferroptosis induction elicits anti-tumor immune responses.

Figure 1.

Induction of ferroptosis triggers anti-cancer immune activation

(A) Schematic representation of therapeutic schedule for RSL3.

(B and C) Tumor growth curves (B) and tumor weights (C) of MC38-bearing mice receiving 5 mg/kg RSL3 or vehicle control (n = 6 mice per group). Data are presented as mean ± SD. ∗p < 0.05, ∗∗∗p < 0.001; (B) two-way ANOVA with Tukey’s multiple comparison test, (C) two-sided unpaired Student’s t test.

(D) Representative flow images and quantification of tumor-infiltrating CD3⁺CD4⁺ helper T cells, CD3⁺CD8⁺ cytotoxic T cells, and PD-1⁺ , TIM-3⁺, and IFNγ⁺ CD8⁺ T cells measured by flow cytometry for the indicated groups (n = 6 mice per group for CD3⁺CD4⁺, CD3⁺CD8⁺, CD8⁺PD-1⁺, and CD8⁺TIM-3⁺; n = 4 mice per group for CD8⁺IFNγ⁺). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(E) Schematic representation of therapeutic schedule for erastin.

(F and G) Tumor growth curves (F) and tumor weights (G) of Hepa1-6-bearing mice receiving 15 mg/kg erastin or vehicle control (n = 6 mice per group). Data are presented as mean ± SD. ∗∗p < 0.01, ∗∗∗p < 0.001; (F) two-way ANOVA with Tukey’s multiple comparison test, (G) two-sided unpaired Student’s t test.

(H) Representative flow images and quantification of tumor-infiltrating CD3⁺CD4⁺ helper T cells, CD3⁺CD8⁺ cytotoxic T cells, and PD-1⁺ , TIM-3⁺, and IFNγ⁺ CD8⁺ T cells measured by flow cytometry for the indicated groups (n = 6 mice per group for CD3⁺CD4⁺, CD3⁺CD8⁺, CD8⁺PD-1⁺, and CD8⁺TIM-3⁺; n = 4 mice per group for CD8⁺IFNγ⁺). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

Antigen presentation of macrophages via MHC-II is involved in ferroptosis-induced tumor immune activation

We collected tumor tissues from mice receiving RSL3 or vehicle treatment, and RNA sequencing (RNA-seq) transcriptomic analysis consistently supported the anti-tumor immune effects of RSL3 in vivo. Gene Ontology (GO) enrichment pinpointed that modulation of biological processes involved in anti-tumor immunity was among the top changes (Figure 2A). Among these top 10 enrichments, we noticed cellular response to IFN, immune system process, and innate immune response, indicating a potential involvement of innate immunity in ferroptosis-induced immune activation. Further exploration of the sequencing data using gene set enrichment analysis (GSEA) revealed macrophage activation and antigen presentation pathway as prominently enriched pathways (p < 0.05, false discovery rate <0.25; Figure 2B). These combined results suggest that ferroptosis induction elicits anti-tumor immune responses, with activation of macrophage antigen presentation as a potential key contributor. We depleted macrophages using Clodrosome (clodronate-encapsulated liposomes) and evaluated the anti-tumor effect of RSL3 in immune-competent MC38-bearing mice (Figure 2C). Macrophage depletion alone had no effect on tumor growth^24,25 but largely reversed the anti-tumor efficacy of RSL3, supporting a functional contribution of tumor-infiltrating macrophages to ferroptosis-mediated immune activation (Figure 2D).

Figure 2.

Elevated macrophage MHC-II contributes to ferroptosis-induced immune activation

(A) GO analysis shows altered pathways after 5 mg/kg RSL3 treatment.

(B) GSEA analysis shows enrichment of pathways involved in antigen processing and presentation and macrophage activation in tumors with 5 mg/kg RSL3 treatment. NES, normalized enrichment score.

(C) Schematic representation of therapeutic schedule for 5 mg/kg RSL3 and 200 μg clodronate liposomes.

(D) Growth curves of MC38-bearing mice receiving indicated treatment (n = 6 mice per group). Data are presented as mean ± SD. NS, not significant, ∗∗∗p < 0.001; two-way ANOVA with Tukey’s multiple comparison test.

(E) Representative flow images and quantification of MHC-II⁺ CD11b⁺F4/80⁺ macrophages from MC38 tumors receiving RSL3 or vehicle treatment by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05; two-sided unpaired Student’s t test.

(F) Representative flow images and quantification of MHC-II expression in BMDMs cultured in RSL3 or DMSO-treated conditioned medium by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(G) Model for OVA^323-339-specific antigen presentation analysis.

(H) Proliferation rate of OT-II CD4⁺T cells co-cultured with OVA^323-339-loaded wild-type or I-A/I-E^−/− BMDMs with indicated treatment (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

To elucidate the underlying molecular mechanisms by which macrophages regulate immune activation, we first assessed hallmark genes expressed by tumor-suppressing (M1) and tumor-promoting (M2) macrophages in our transcriptome data, finding comparable gene profiling between groups (Figures S1C and S1D). Instead, FACS analysis confirmed heightened surface MHC-II protein on tumor-infiltrating macrophages from mice receiving RSL3 treatment (Figure 2E), while MHC-I remained unchanged (Figure S1E). Considering the critical role of DCs in antigen presentation, we also examined the MHC molecules on DCs. DC MHCs-I and -II did not differ between groups (Figures S1F and S1G). Together, these results suggest that elevated macrophage MHC-II-related antigen presentation shall be essential for ferroptosis-mediated immune activation.

To further dissect how ferroptosis regulated MHC-II expression in macrophages, we first treated bone marrow-derived macrophages (BMDMs) with RSL3 but found that RSL3 failed to regulate macrophage MHC-II in vitro directly (Figure S1H). We, therefore, speculated whether factors released from ferroptotic tumor cells might be involved. Although conditioned medium from MC38 cells lethally treated with RSL3 did not alter the expression of genes involved in MHC-related pathways, we found that this conditioned medium increased surface MHC-II expression on BMDMs (Figures 2F and S1I). In contrast, conditioned medium from MC38 cells lethally treated with the apoptosis inducer carbonyl cyanide m-chlorophenyl hydrazone (CCCP) failed to increase surface MHC-II expression on BMDMs (Figure S1J). Since binding of antigen-specific MHC-II provides critical “first signal” activation of cognate CD4⁺ T cells in initiating anti-tumor immunity to support CD8⁺T cell activation,²⁶ we, therefore, tested if supernatants from RSL3-treated tumor cells could potentiate MHC-II-dependent activation of antigen-specific CD4⁺ T cells. BMDMs loaded with chicken ovalbumin peptide ISQAVHAAHAEINEAGR (OVA^323-339) and cultured in conditioned media from RSL3-treated cells stimulated greater OT-II transgenic CD4⁺ T cell proliferation than vehicle media, an effect wholly reversed upon BMDM MHC-II depletion (Figures 2G and 2H). These data indicate that ferroptotic tumor cells can indirectly upregulate macrophage MHC-II expression, illuminating a key role for enhanced macrophage antigen presentation in ferroptosis-induced anti-tumor immunity.

Ferroptotic tumor cell-derived ATRA drives macrophage MHC-II upregulation by induction of autophagy

Ferroptosis results from aberrant metabolic pathways. Cell metabolism, including iron, lipid, and redox metabolism, plays a unique role in ferroptosis compared with other cell death modalities. Ferroptosis induction substantially alters cancer cell metabolism, leading to the accumulation of oxidative and reductive metabolites that may be released into the tumor microenvironment.^27,28 We, therefore, focused on the secreted metabolome for an in-depth investigation. Targeted metabolomics were used to identify potential bioactive metabolites enriched following RSL3 treatment that could upregulate macrophage MHC-II. Five cell-intrinsic metabolites were found to increase in the supernatant from RSL3-treated MC38 cells (p < 0.01, fold change >2. Figures 3A and S2A). Screening differential metabolites revealed only ATRA, the biologically active vitamin A derivative, increased surface MHC-II in BMDMs (Figures 3B, 3C, and S2B). ATRA is synthesized via the irreversible oxidation of retinol by retinaldehyde dehydrogenases (RALDHs), including ALDH1A1, ALDH1A2, and ALDH1A3 (Figure 3D).²⁹ We found that the precursor of ATRA, all-trans retinol (vitamin A), was dramatically consumed following RSL3 but not CCCP treatment, supporting an increase in ATRA biosynthesis under ferroptosis stress (Figure S2C). We also found that ALDH1A1, but not the other two RALDHs, was highly expressed in MC38 and Hepa1-6 cells, suggesting that it might be responsible for catalyzing ATRA production in these cells (Figures 3E and S2D–S2F). These findings are consistent with previous reports indicating cell-type-specific expression of ALDH1A1 in HCC and CRC cells.^30,31 In contrast, CCCP did not increase ALDH1A1 expression (Figure S2G). To further identify the major role of ALDH1A1 in ATRA-mediated MHC-II regulation, we knocked down Aldh1a1 in both MC38 and Hepa1-6 cells and showed that conditioned medium from cells with Aldh1a1 knockdown failed to upregulate MHC-II expression in BMDMs (Figures 3F, S2H, and S2I). Similar findings were also observed in vivo (Figure S2J). These results supported the idea that ferroptotic tumor cell-derived ATRA regulates MHC-II expression in macrophages.

Figure 3.

Ferroptotic tumor cell-derived ATRA controls MHC-II expression in macrophages

(A) Volcano plots show top altered metabolites in supernatant from RSL3 or DMSO-treated MC38 cells. FC, fold change.

(B) Relative fold change of ATRA in supernatant from RSL3- or DMSO-treated MC38 cells (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗p < 0.01; two-sided unpaired Student’s t test.

(C) Representative images of flow cytometry analysis of MHC-II in BMDMs treated with 200 nM ATRA or DMSO, and quantification was evaluated by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(D) Metabolism of vitamin A.

(E) Relative mRNA expression of Aldh1a1, Aldh1a2, and Aldh1a3 of MC38 cells treated with or without 5 μM RSL3 measured by RT-qPCR (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(F) Representative flow images and quantification of MHC-II in BMDMs cultured in conditioned medium of RSL3-treated MC38 cells or MC38 cells with Aldh1a1 knockdown by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(G) Volcano plots of TT-seq data show newly born differentially expressed genes between ATRA and DSMO treatment groups.

(H) Gene Ontology and KEGG analysis of upregulated differentially expressed genes from TT-seq analysis.

(I) Heatmap of Gene Ontology enrichment analysis of genes upregulated at the transcriptional level after 200 nM ATRA treatment.

(J) UCSC tracks of TT-seq signals of autophagy-related genes in DMSO- and ATRA-treated BMDMs. Relative mRNA expression of Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1 of MC38 cells treated with 200 nM ATRA or DMSO measured by RT-qPCR (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(K) Representative immunoblot of LC-3A/B in BMDMs with indicated treatment, and the ratio of LC-3A/B-II and LC-3A/B-I was quantified (n = 3 independent biological experiments). β-actin serves as an internal control. Data are presented as mean ± SD. ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(L) Representative flow images and quantification of MHC-II in BMDMs after indicated treatment by flow cytometry. ATRA, 200 nM. CQ, 10 μM (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

To better understand how ATRA regulates MHC-II expression in macrophages, we performed both bulk RNA-seq and transient transcriptome sequencing (TT-seq) to identify newly born differentially expressed genes in BMDMs treated with ATRA or DMSO. Compared with the control group, we identified a total of 583 upregulated genes and 700 downregulated genes in the ATRA-treated group according to RNA-seq analysis (Figure S2K) and a total of 1,173 upregulated genes and 1,454 downregulated genes in the ATRA-treated group according to TT-seq analysis (Figure 3G). GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis of data from RNA-seq and TT-seq indicated that autophagy and lysosome pathway were highly enriched (Figures 3H and S2L). The key step in autophagy is the fusion of autophagosomes with lysosomes.³² Since autophagy has been reported to be involved in MHC-II regulation and antigen presentation in macrophages,^33,34 we, therefore, hypothesized that ATRA might regulate MHC-II through the autophagy pathway. Autophagy-related genes were commonly upregulated in ATRA-treated BMDMs (Figure 3I). UCSC tracks also showed that signals of key autophagy regulators, Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1, were elevated after ATRA treatment and validated by quantitative reverse-transcription PCR (RT-qPCR) (Figure 3J). By monitoring the conversion of LC-3A/B from a cytosolic form (LC-3A/B-I) to its autophagosome-associated form (LC-3A/B-II), we confirmed that ATRA treatment induced autophagy in BMDMs (Figure 3K). Previous study has pointed out the role of autophagy in MHC-II regulation. As expected, chloroquine (CQ), an inhibitor of autophagy, was able to decrease MHC-II expression in BMDMs (Figure S2M). Importantly, CQ completely blocked ATRA-induced MHC-II upregulation in BMDMs (Figure 3L). Together, these data suggested that ferroptotic tumor cell-derived ATRA controls MHC-II expression in macrophages, probably via induction of autophagy.

ATRA transcriptionally regulates CD38 expression through RARα to induce autophagy

As the primary ligand for the transcriptional factor RAR, ATRA binds to retinoic acid response elements (RAREs) in the promoter regions of target genes to control their expression. RARE-RFP reporter assay showed that conditioned medium from RSL3-treated control MC38 cells, but not shAldh1a1 MC38 cells, activated RAR signaling, indicating that ferroptotic tumor cell-derived ATRA activated RAR signaling in BMDMs (Figure S3A). However, we did not find the core RARE motif in the promoter region of Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1, suggesting that these autophagy-related genes are indirectly regulated by ATRA-RAR signaling (Figure S3B).

TFEB is a master regulator of autophagy and lysosomal biogenesis that controls the expression of Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1.³⁵ We re-analyzed our bulk RNA-seq and TT-seq data. We found that CD38, an ectoenzyme that has previously been shown to be required for TFEB activation and TFEB-mediated autophagy,^36,37 was among the top differences from both bulk RNA-seq and TT-seq (Figure 4A), suggesting that CD38 might be a downstream factor of ATRA-RAR signaling to regulate TFEB-mediated autophagy. RT-qPCR showed that both ATRA and conditioned medium from RSL3-treated MC38 cells were able to increase Cd38 mRNA expression in BMDMs (Figures S3C and S3D). Functionally, loss of Cd38 in BMDMs attenuated the mRNA expression of Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1 (Figure 4B). To investigate whether CD38-TFEB contributes to ATRA-induced autophagy and MHC-II expression, we silenced Cd38 in BMDMs together with ATRA administration. We found that ATRA treatment induced nuclear translocation of TFEB; however, knockdown of Cd38 blocked this process (Figure 4C). Loss of Cd38 also reduced MHC-II expression and prevented ATRA-induced MHC-II upregulation in BMDMs (Figure 4D). All these results strongly supported that ATRA regulated CD38 to mediate TFEB activation and MHC-II expression.

Figure 4.

ATRA-RARα signaling directly regulates CD38 transcription to induce autophagy and MHC-II expression

(A) Venn diagram of shared differential genes from overlapping RNA-seq and TT-seq transcriptome.

(B) Relative mRNA expression of Ctsa, Ctsb, Ctsd, Lamp1, and Atp6v0d1 of siCtrl or siCd38 BMDMs measured by RT-qPCR (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(C) Representative immunoblot of cytoplasmic and nuclear TFEB in BMDMs with indicated treatment. Vinculin serves as an internal control of cytoplasmic fraction, and Lamin B1 serves as an internal control of nuclear fraction (n = 3 independent biological experiments).

(D) Representative flow images and quantification of MHC-II in siCtrl or siCd38 BMDMs after indicated treatment by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. NS, not significant, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(E) (Top) UCSC tracks showing the TT-seq signals of Cd38 in DMSO- and ATRA-treated BMDM cells. (Bottom) ATAC-seq of DMSO- or ATRA-treated BMDM cells.

(F) CUT&TAG signals for H3K27ac, H3K4me3, and Rpb1in the promoter and enhancer regions of Cd38 after ATRA treatment.

(G) The upstream enhancer, promoter, and downstream enhancer regions of Cd38, and the mutation strategy for plasmid construction for dual-luciferase reporter assay.

(H) The luciferase activity (Firefly/Renilla) of wild or upstream enhancer, promoter, and downstream enhancer regions of mutated Cd38 after indicated treatment. Data are presented as mean ± SD. NS, not significant, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

We next asked whether ATRA directly regulated Cd38 expression. By mining publicly available chromatin immunopreciptation sequencing data, we observed that RARα, a major subtype of RAR, could be enriched at the promoter regions of CD38 (Figure S3E). Depletion of Rara, which encodes RARα, blocked ATRA-induced Cd38 expression in BMDMs (Figure S3F), whereas loss of RARβ and RARγ did not affect ATRA-induced Cd38 and MHC-II expression in BMDMs (Figures S3G and S3H). These data suggested that Cd38 might directly respond to the ATRA stimulus. Consistent with TT-seq data, ATAC-seq showed that the chromatin accessibility became more open in the promoter and enhancer regions of Cd38 after ATRA treatment (Figure 4E). CUT&TAG assay also found increased signals of active epigenetic markers H3K27ac and H3K4me3 as well as the active transcription marker Rpb1, the largest subunit of RNA polymerase II, in both the promoter and enhancer regions of Cd38 after ATRA treatment (Figure 4F). Meanwhile, we observed an increase in the genome-wide occupancy of Rpb1, indicating an overall transcriptional activation following ATRA treatment (Figure S3I). To confirm our supposal further, we carried out dual-luciferase reporter assay. We noted the presence of core RARα-binding RARE motif, 5′-AGGTCA-3′ or 5′-GTGTCA-3′, in both the promoter and enhancer regions of Cd38 (Figure 4G). The results showed that ATRA treatment could increase the activity of luciferase reporter containing the upstream and downstream enhancers in CD38, while RARE motif-mutated enhancers in CD38 failed to respond to ATRA stimulus completely (Figure 4H). Collectively, these data provided strong evidence that CD38 is a direct target of transcription factor RARα and is responsible for ATRA-mediated autophagy and MHC-II regulation.

Elevated ferroptosis signature associates with better immunotherapy outcomes

To better evaluate the clinical relevance of ferroptosis and the MHC-II antigen presentation pathway, we derived a ferroptosis signature according to our RNA-seq data. Among the top 200 differential genes, TFRC, ACSL4, ENO1, and IDO1 are four reported putative ferroptosis-related genes (Figure 5A). We used these genes as signature genes and calculated ferroptosis signature scores in each patient by averaging Z scores of TFRC, ACSL4, ENO1, and IDO1. We then used these scores to analyze the correlation of ferroptosis with CTL infiltration and CD38 expression using The Cancer Genome Atlas (TCGA) data. The ferroptosis signature positively correlated with CTL infiltration and CD38 expression in the majority of TCGA malignancies (Figure 5B). Analysis of published immunotherapy clinical trial transcriptomic data^{38,39,40,41,42,43} also validated coordinated correlation of our 4-gene ferroptosis signature with CTL infiltration and CD38 expression across solid tumor malignancies (Figure 5C). Notably, higher ferroptosis signature scores were observed in responders, when compared with non-responders (Figures 5D and 5E). Also, a higher ferroptosis signature score was indicative of superior clinical response to anti-PD-1 and prolonged disease-free survival in a melanoma cohort (Figure 5F).

Figure 5.

Ferroptosis signature associates with better responses to immunotherapy

(A) Volcano plot analysis of differentially expressed genes from RNA-seq data of RSL3- and vehicle-treated MC38-bearing mice.

(B) Correlation between ferroptosis signature score and cytotoxic T cell (CTL) signature score and CD38 expression in patients across solid malignancies. The Spearman correlation coefficient and associated probability were determined. All the data were obtained from TCGA.

(C) Ferroptosis signature is positively correlated with CTL signatures and CD38 expression in most clinical trials for immunotherapy. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; Spearman correlation.

(D) Ferroptosis signature score is higher in patients from the Riaz 2017 cohort evaluated as responders than in those evaluated as nonresponders. Data are presented as mean ± SD. ∗p < 0.05; two-sided unpaired Student’s t test.

(E) Ferroptosis signature score is higher in patients from Liu 2019 cohort evaluated as responders than in those evaluated as nonresponders. Data are presented as mean ± SD. ∗p < 0.05; two-sided unpaired Student’s t test.

(F) Patients with higher ferroptosis signature score have better progress-free survival in patients from the Liu 2019 cohort treated with anti-PD-1; log rank (Mantel-Cox) test.

(G and H) Representative multiplex immunofluorescence images of TFRC, IDO1, ENO1, and ACSL4 (G) were shown from biopsy samples of responders (n = 6) or non-responders (n = 6) to immunotherapy in CRC, and the fluorescence signals were quantified (H). Scale bar: 50 μM. Data are presented as mean ± SD. ∗∗∗p < 0.001; two-sided unpaired Student’s t test.

(I) Representative multiplex immunofluorescence images of CD68 and MHC-II co-location from biopsy samples of responders (n = 6) or non-responders (n = 6) to immunotherapy in CRC. Scale bar: 50 μM.

(J) Correlation between the number of MHC-II⁺CD68⁺ macrophages and ferroptosis signature score. Spearman correlation.

(K) Representative positron emission tomography-computed tomography (PET-CT) of patients receiving anti-PD-1 antibody treatment.

(L) Representative CT of patients receiving anti-PD-1 antibody treatment.

We additionally retrospectively analyzed 12 unresectable colon cancer patients who received anti-PD-1 after failing standard therapies in our hospital (6 responders and 6 non-responders, Table S1). All patients exhibited disease progression after multiple cycles of standard systemic therapy prior to receiving anti-PD-1 immunotherapy. Multiplex immunofluorescence staining of biopsy tissues obtained before anti-PD-1 therapy showed higher protein expression of the 4-gene ferroptosis signature in responding compared with non-responding patients (Figures 5G and 5H). MHC-II⁺CD68⁺ macrophages also infiltrated more in responders than in non-responders (Figure 5I). The number of MHC-II⁺CD68⁺ macrophages was positively correlated with ferroptosis signature score (Figure 5J). Representative positron emission tomography-computed tomography (CT) and CT abdomen/lung images before and after anti-PD-1 therapy were also shown for the responder (Figure 5K) and non-responder (Figure 5L) patients. These data support ferroptosis activation associated with heightened tumor MHC-II antigen presentation and responsiveness to immune checkpoint inhibitors across cancers.

Development of drug-free nano-redox lever that disrupts tumor cell redox homeostasis to sensitize to ferroptosis

Given the potential to enhance immunotherapy, we aimed to develop an approach to induce tumor ferroptosis. Unfortunately, most experimental inducers show anti-tumor effects in vitro but lack pharmacokinetics or specificity and can have organ toxicities. To overcome these limitations, we synthesized an amphiphilic polymer, methoxy (polyethylene glycol)-polylysine-nitroimidazole (mPEG-pLys-NI) copolymer, that self-assembles into hypoxia-targeted levers to disrupt the GSH/GPX4 axis (Figure S4A). Transmission electron microscopy revealed spherical levers with size range of 44.62 ± 5.35 nm (Figure S4B). Dynamic light scattering confirmed uniform size distribution (Figure S4C) and negative surface charge with −3.52 ± 0.22 mV (Figure S4D).

The DFNRL contained a hydrophilic PEG segment and a hydrophobic NI-conjugated lysine segment. NI served as a substrate for nitroreductases (including NQO1, CYPOR, and XDH) selectively expressed in hypoxic tumor regions. Published single-cell data confirmed predominant nitroreductase expression within cancer cells (Figure S4E). Tumor tissues also expressed high levels of HIF1α, a key marker of hypoxia (Figure S4F). Nitroreduction triggers stepwise electron transfer from NADPH, successively reducing NI to hydroxylamine and amine intermediates (Figure S4G). Because the unstable intermediates in this chemical process were technically difficult to isolate, we instead detected the chemical shifts of hydrogen atoms during the reaction process by ¹H-NMR to indirectly confirm the electron transfer mechanism (Figure S4H). Accordingly, DFNRL administration extremely impeded NADPH/NADP⁺ and GSH/GSSH ratios in both MC38 and Hepa1-6 cells (Figure S4I).

We next examined in vitro cellular uptake and in vivo distribution of DFNRL. Dose-dependent internalization of coumarin-6-labeled DFNRL occurred in MC38 and Hepa1-6 cells (Figure S4J). The in vivo distribution of DFNRL primarily involved two steps. First, particles smaller than 200 nm passively accumulate in tumors through the enhanced permeation and retention (EPR) effect. Subsequently, upon penetrating hypoxic regions, DFNRL is specifically bioactivated by the overexpressed nitroreductase, triggering its redox-disrupting effect. This enzyme-responsive bioactivation enhances nanoparticle retention at the tumor site, thereby achieving tumor microenvironment-responsive selective delivery. To clarify DFNRL biodistribution, IR780-labeled DFNRL was intravenously injected into tumor-bearing mice. Whole-body imaging revealed IR780-DFNRL, but not free IR780, specifically localized to tumors over 24 h of detection (Figure S4K). Ex vivo organ fluorescence quantification also supported tumor-selective accumulation (Figure S4L).

DFNRL potentiates pharmacologic ferroptosis activation and downstream immunogenic signaling

Given the pro-oxidative capacity of DFNRL, we next asked whether DFNRL could synergize with pharmacological ferroptosis induction. We combined DFNRL with ferroptosis inducers, including RSL3 or arachidonic acid (AA) plus IFN-γ, an immune-mediated ferroptosis-mimetic model in vitro¹⁷ (Figure 6A). IFN-γ suppresses the expression of system Xc⁻, stimulates ACSL4, and alters tumor cell lipid composition, thereby increasing the incorporation of AA into phospholipids containing C16 and C18 acyl chains to induce ferroptosis. Morphologically, DFNRL enhanced cell membrane blebbing and rounding up of cells triggered by RSL3 or AA + IFN-γ, indicating boosted ferroptotic death (Figure S5A). DFNRL also induced shrunken mitochondria with heavily condensed membranes, suggesting a severe ferroptosis phenotype (Figure S5B). This translated to reduced viability that was completely rescued by the ferroptosis inhibitors ferrostatin-1 (Fer-1) and deferasirox, but not apoptosis inhibitor Z-VAD-FMK (ZVAD) and necrosis inhibitor necrosulfonamide (Necro; Figures 6B, 6C, S5C, and S5D). In support, DFNRL drove further GSH consumption and lipid reactive oxygen species accumulation in the presence of RSL3 or AA + IFN-γ (Figures 6D, 6E, S5E, and S5F). Conditioned medium from cells treated with DFNRL plus RSL3 or AA plus IFN-γ strongly upregulated macrophage MHC-II and CD38 expression (Figures 6F and S5G). In vivo, DFNRL enhanced RSL3-induced ferroptosis in both MC38 and Hepa1-6 xenograft models (Figures 6G and S5H). Both DFNRL and RSL3 increased MHC-II expression in tumor-infiltrating macrophages, and the combination of DFNRL and RSL3 achieved maximum effect (Figures 6H and S5I). Collectively, our results support that DFNRL synergistically enhances pharmacologic ferroptosis activation and downstream immunomodulation.

Figure 6.

DFNRL potentiates pharmacologic ferroptosis activation

(A) Cartoon model for ferroptosis induced by RSL3 or CD8⁺ T cells and the anti-ferroptosis mechanism of DFNRL.

(B and C) Cell viability of MC38 cells with indicated treatment under hypoxia assessed by CCK-8 analysis. Fer-1, ferrostatin; DFX, deferasirox; ZVAD, ZVAD-FMK; Necro, necrosulfonamide (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(D) GSH content of MC38 cells with indicated treatment under hypoxia (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(E) Representative flow images and quantification of BODIPY C11 fluorescence of MC38 cells with indicated treatment under hypoxia by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(F) Quantification of MHC-II and CD38 in BMDMs cultured in conditioned medium from MC38 cells with indicated treatment under hypoxia, evaluated by flow cytometry (n = 3 independent biological experiments). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(G) Growth curves of MC38-bearing mice receiving indicated treatment (n = 6 mice per group). Data are presented as mean ± SD. ∗∗∗p < 0.001; two-way ANOVA with Tukey’s multiple comparison test.

(H) Representative flow images and quantification of MHC-II in tumor-infiltrating macrophages by flow cytometry from MC38-bearing mice receiving indicated treatment (n = 3 mice per group). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

DFNRL synergistically enhances anti-PD-1 therapy efficacy by escalating tumor ferroptosis and immune activation in vivo

Anti-PD-1 immunotherapy efficacy involves T cell-derived IFN-γ inhibiting cystine uptake via system Xc⁻, evoking ferroptosis.¹⁶ Ferroptotic cells then stimulate antigen presentation to propagate anti-tumor immunity. We hypothesized that combining DFNRL with anti-PD-1 could synergistically boost this cycle. MC38 and Hepa1-6 xenograft models showed anti-tumor effects of DFNRL or anti-PD-1 alone, but combination therapy gave superior repression (Figures 7A, 7B, S6A, and S6B).

Figure 7.

DFNRL synergizes with anti-PD-1 in vivo and ex vivo

(A) Representative tumor images from MC38-bearing mice with DFNRL, anti-PD-1, or combination treatment.

(B) Tumor growth curves of MC38-bearing mice receiving 100 μg anti-PD-1, 100 μg DFNRL, a combination of anti-PD-1 and DFNRL, or vehicle plus IgG (n = 6 mice per group). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; two-way ANOVA with Tukey’s multiple comparison test.

(C and D) Representative flow images and quantification of CD11b⁺F4/80⁺MHC-II⁺ macrophages (C) and CD4⁺ T cells and CD8⁺ T cells (D) in MC38 tumors receiving 100 μg anti-PD-1, 100 μg DFNRL, a combination of anti-PD-1 and DFNRL, or vehicle plus IgG (n = 6 mice per group). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(E) Schematic representation for patient-derived organotypic tumor spheroid (PDOT) establishment and following treatments.

(F) Representative bright-field images and Ki-67 immunofluorescence images of PDOTs receiving 10 μg/mL anti-PD-1, 100 μg DFNRL, or the combination under hypoxia (n = 15 fields per group). Scale bar: 20 μm (up) and 50 μm (down). Data are presented as mean ± SD. ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

(G and H) Representative flow images and quantification of CD11b⁺F4/80⁺MHC-II⁺ macrophages (G) and CD4⁺ T cells and CD8⁺ T cells (H) in PDOTs receiving 10 μg/mL anti-PD-1, 100 μg DFNRL, or the combination under hypoxia (n = 5 PDOTs per group). Data are presented as mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001; one-way ANOVA with Tukey’s multiple comparison test.

Combination treatment caused severe intratumoral ferroptosis by immunohistochemistry and the strongest GSH and NADPH depletion (Figures S6C–S6F). Mirroring microenvironmental changes, combination therapy gave the strongest macrophage MHC-II upregulation (Figures 7C and S6G). Also, this combination therapy-induced increase in macrophage MHC-II expression aligned with heightened intratumoral infiltration of both CD4⁺ and CD8⁺ T cells (Figures 7D and S6H). No histological organ damage, weight changes, or biochemical liver/kidney dysfunction occurred, supporting biosafety (Figure S7).

Patient-derived organoids (PDOs) of CRC and matched peripheral blood monocyte co-culture models were also utilized to evaluate a potential clinical translation (Figure 7E). PDO growth was inhibited by around 40% by DFNRL or anti-PD-1 alone. However, the combination treatment profoundly suppressed expansion by over 90% (Figure 7F). This mirrored decreases in staining of proliferation marker Ki-67 (Figure 7F). Combination treatment elicited the greatest MHC-II induction on co-cultured macrophages (Figure 7G) alongside heightened CD4⁺ and CD8⁺ T cells (Figure 7H). Together, these in vivo and ex vivo analyses strongly demonstrate anti-tumor synergy and on-target effects of DFNRL with anti-PD-1 therapy.

Discussion

Immunotherapy clinical responses remain unsatisfactory for numerous malignancies. In the present study, we demonstrate that induction of tumor ferroptosis provokes anti-tumor immunity by enhancing MHC-II-dependent macrophage antigen presentation. Since ferroptosis engages cancer-immunity interactions, enlarging tumor ferroptosis establishes a positive feedback loop potentiating immune checkpoint blockade therapeutic efficacy. As such, a DFNRL eliciting tumor-specific NADPH and GSH depletion and ferroptosis was designed to synergize with anti-PD-1 therapy.

The context-dependent consequences of ferroptosis on anti-cancer immunity likely reflect cell type-specific differences. Lipid peroxidation reportedly hinders cytotoxic CD8⁺ T cell function, and CD8⁺ T cells are sensitive to the GPX4 inhibitor RSL3.⁴⁴ However, the cystine-depleting agent cyst(e)inase spurs tumor ferroptosis without affecting CD8⁺ T cell activity, similar to our DFNRL.¹⁶ Thus, disrupting system Xc⁻, NADPH, and GSH pathways may elicit distinct outcomes compared with direct GPX4 inhibition. Further elucidation of the mechanisms governing differential immunocyte susceptibilities is warranted. Moreover, ferroptotic cancer cells exhibit multifaceted immunomodulation. Some evidence indicated that ferroptosis inhibits DCs and adaptive immunity.⁴⁵ Yet another group suggested that ferroptotic cells promote DC maturation, underscoring context specificity.⁴⁶ Comprehensively assessing effects across the tumor microenvironment is essential.

Beyond DCs and B cells, abundant TAMs represent predominant antigen-presenting phagocytes across malignancies.⁴⁷ TAMs are traditionally viewed as immunosuppressive with dampened antigen presentation capacity and heightened expression of inhibitory factors. Strategies targeting TAM depletion or reprogramming are actively explored, although multiple CSF1R inhibitor trials failed. Our data put forth an alternative approach by reinvigorating MHC II-mediated macrophage antigen presentation. While DCs exhibit robust antigen cross-presentation, scarce tumor-resident DCs with impaired function highlight TAM antigen presentation as potentially impactful.⁴⁸ This spotlights macrophage antigen presentation as integral for adaptive response induction after ferroptotic cell death, necessitating clarification of poorly understood mechanisms enabling macrophage antigen cross-presentation in tumors.

Current experimental ferroptosis-inducing agents, including erastin and RSL3, lack suitable pharmacological properties for clinical translation. Most reagents exhibit unfavorable hydrophobicity, restricting delivery alongside absent tumor targeting, resulting in uncontrolled organ toxicity.⁴⁹ To address this limitation, we fabricated a hypoxia-responsive micelle formulated using three biocompatible and FDA-approved compounds (i.e., PEG, lysine, and nitroimidazole). It is well established that nanoparticles under 200 nm permeate tumors through the EPR effect.⁵⁰ This passive accumulation phenomenon enables tumor-selective delivery across nanoparticulates, including our micelle. To date, most nanotechnology-based ferroptosis studies focus on the delivery of ferroptosis inducers. Although the amphiphilic block copolymer developed in this study is also capable of encapsulating such agents, we deliberately adopted a “drug-free” design. Drug-free nanoparticles offer simplified and more stable manufacturing processes, avoiding the stringent quality control challenges associated with drug-loaded nano-formulations. These advantages may accelerate clinical translation. Furthermore, the separate administration of DFNRL and ferroptosis inducers affords clinicians greater flexibility in dose optimization. Such modularity cannot be achieved with conventional drug-loaded nanoparticles and represents a resilient, platform-based strategy that preserves the intrinsic therapeutic function of the nanomaterial, while facilitating flexible integration with diverse treatment modalities.

In summary, our work delineates ferroptosis as a critical driver of anti-tumor immunity, revealing unidentified immunomodulatory roles. To this end, we developed a DFNRL that selectively targets the NADPH/GSH pathway to induce tumor-specific ferroptosis enhancement safely across preclinical models. This synergizes with immune checkpoint blockade, offering a translational nanotherapeutic strategy to overcome immunotherapy resistance by simultaneously escalating cancer cell death and immune activation.

Limitations of the study

While our data demonstrate that ferroptotic cells differentially regulate MHC-II expression but not MHC-I expression, we acknowledge that the mechanistic basis underlying this selectivity remains incompletely understood. Further investigation is warranted to fully elucidate how metabolites derived from ferroptosis, as compared to those from other cell death modalities, specifically regulate anti-tumor immune responses.

Resource availability

Lead contact

Reasonable requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jimin Zhu (zhu.jimin@zs-hospital.sh.cn).

Materials availability

Plasmids generated in this study will be made available on request, but we may require a payment and/or a completed materials transfer agreement if there is potential for commercial application.

Data and code availability

• •TT-seq, RNA-seq, ATAC-seq, and CUT&TAG data generated in this study have been deposited in NCBI Gene Expression Omnibus database with accession number GEO: GSE318119: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE318119.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (nos. 32270588, 82103252, 82173122, 82473853, 82273027, and 82303161), the Chinese Academy of Medical Sciences Innovation Fund for Medical Sciences (2024-I2M-3-011 and 2023-I2M-2-005), Shanghai Science and Technology Committee (nos. 20Y11908100, 21ZR1413700, and 22ZR1411800), and Shanghai Sailing Program (nos. 21YF1407400 and 23YF1405900).

Author contributions

J.-M.Z., T.S., and S.-B.H. conceived the study and designed the experiments. J.-L.S. and F.W. carried out in vitro and in vivo assays. Y.-C.C. designed and synthesized micelles. D.-D.Y. and S.-B.H. performed transcriptome and data analyses. J.-R.L., R.-C.X., H.-H.L., Z.-R.Q., X.S., and X.-N.Y. helped in in vitro and in vivo experiments. Y.-K.Y. provided mice and provided scientific suggestions. T.-T.L., S.-Q.W., and L.D. provided clinical samples and follow-up data. J.-L.S., X.-Z.S., S.-B.H., T.S., and J.-M.Z. provided funding for the study. J.-L.S., Y.-C.C., F.W., and D.-D.Y. drafted the manuscript. S.-B.H., T.S., and J.-M.Z. revised the manuscript. All authors read and approved the final version of the manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Rabbit monoclonal anti-human/mouse TFRC antibody	Abcam	Cat# ab214039; RRID: AB_2904534
Rabbit polyclonal anti-4 Hydroxynonenal antibody	Abcam	Cat# ab46545; RRID: AB_722490
Mouse monoclonal Anti-Malondialdehyde antibody	Abcam	Cat# ab243066; RRID: AB_3662056
Rabbit polyclonal anti-human Indoleamine 2, 3-dioxygenase antibody	Abcam	Cat# ab211017; RRID: AB_2936946
Rabbit monoclonal anti-human CD68 antibody	Abcam	Cat# ab213363; RRID: AB_722490
Mouse monoclonal Anti- human MHC Class II antibody	Abcam	Cat# ab55152; RRID: AB_ 944199
Rabbit monoclonal anti-human/mouse ACSL4 antibody	ABclonal	Cat# A20414; RRID: AB_2909505
Rabbit polyclonal anti-human/mouse ENO1 antibody	ABclonal	Cat# A1033; RRID: AB_2757874
Rabbit polyclonal anti-human/mouse HIF1α antibody	ABclonal	Cat# A11945; RRID: AB_2940796
Rabbit monoclonal anti-human/mouse ALDH1A1 antibody	ABclonal	Cat# A0157; RRID: AB_ 2861455
Rabbit monoclonal anti-human/mouse ALDH1A antibody	ABclonal	Cat# A9123; RRID: AB_2863659
Rabbit monoclonal anti-human/mouse ALDH1A3 antibody	ABclonal	Cat# A15230; RRID: AB_2762125
HRP-conjugated Goat anti-Rabbit IgG (H + L)	ABclonal	Cat# AS014; RRID: AB_2769854
HRP-conjugated Goat anti-Mouse IgG (H + L)	ABclonal	Cat# AS003; RRID: AB_2769851
Rabbit monoclonal anti-human/mouse LC3A/B antibody	Cell Signaling Technology	Cat# 12741; RRID: AB_2617131
Rabbit monoclonal anti-human GPX4 Antibody	Cell Signaling Technology	Cat# 59735; RRID: AB_2940796
Rabbit monoclonal anti-human/mouse CD38 antibody	Cell Signaling Technology	Cat# 68336; RRID: AB_3674060
Rabbit monoclonal anti-human/mouse TFEB antibody	Cell Signaling Technology	Cat# 83010; RRID: AB_3105800
Rabbit monoclonal anti-human/mouse Vinculin antibody	Cell Signaling Technology	Cat# 13901; RRID: AB_2728768
Rabbit monoclonal anti-human/mouse Lamin B1 antibody	Cell Signaling Technology	Cat# 13435; RRID: AB_2737428
Rabbit monoclonal anti- human/mouse Ki-67 antibody	Cell Signaling Technology	Cat# 9449; RRID: AB_ 2797703
Rat Anti-mouse CD45 BD Horizon™ BUV395 (30-F11)	BD Biosciences	Cat# 565967; RRID: AB_2739420
Armenian hamster monoclonal anti-mouse CD3ε PE/Cyanine7 (145-2C11)	BioLegend	Cat# 100320; RRID: AB_312685
Rat monoclonal anti-mouse CD4 eFluor™ 450 (RM4-5)	Thermo Fisher Scientific	Cat# 48-0042-82, RRID; AB_1272194
Rat monoclonal anti-mouse CD8a PerCP-Cyanine5.5 (53–6.7)	Thermo Fisher Scientific	Cat# 45-0081-82; RRID: AB_1107004
Rat monoclonal anti-mouse IFN gamma PE (XMG1.2)	Thermo Fisher Scientific	Cat# 12-7311-82; RRID: AB_466193
Rat monoclonal anti- human/mouse CD11b PerCP-Cyanine5.5 (M1/70)	BioLegend	Cat# 101228; RRID: AB_893232
Rat monoclonal anti-mouse F4/80 PE (BM8)	BioLegend	Cat# 123110; RRID: AB_893486
Mouse monoclonal anti-human CD45 PE (HI30)	BioLegend	Cat# 982322; RRID: AB_2936522
Mouse monoclonal anti- human CD3 PE/Cyanine7 (OKT3)	BioLegend	Cat# 317333; RRID: AB_2561451
Mouse monoclonal anti- human CD4 Brilliant Violet 421™ (OKT4)	BioLegend	Cat# 317433; RRID: AB_11150413
Mouse monoclonal anti- human CD8 PerCP-Cyanine5.5 (SK1)	BioLegend	Cat# 344709; RRID: AB_2044009
Mouse monoclonal anti-human IFN gamma APC/Cyanine7 (4S.B3)	BioLegend	Cat# 502529; RRID: AB_10663411
Rat monoclonal anti-mouse I-A/I-E FITC (M5/114.15.2)	BioLegend	Cat# 107606; RRID: AB_313321
Mouse monoclonal anti-mouse H-2Kd/H-2Dd APC (29E.2A3)	BioLegend	Cat# 114713; RRID: AB_2734173
Rat monoclonal anti-mouse CD38 Antibody PE/Cyanine7	BioLegend	Cat# 102717; RRID: AB_ 2072892
Armenian hamster monoclonal anti-mouse CD11c PE/Cyanine7 (N418)	BioLegend	Cat# 117317; RRID: AB_ 93569
GoInVivo™ Purified anti-mouse CD279 (PD-1) Antibody	BioLegend	Cat# 135236; RRID: AB_ 2616837
GoInVivo™ Purified anti-human CD279 (PD-1) Antibody	BioLegend	Cat# 329946; RRID: AB_ 2566290
GoInVivo™ Purified Rat IgG2a, κ Isotype Ctrl Antibody	BioLegend	Cat# 400559; RRID: AB_ 2819062
GoInVivo™ Purified Mouse IgG1, κ Isotype Ctrl Antibody	BioLegend	Cat# 400189; RRID: AB_ 3097008
anti-human/monkey CD28	InVivoMAb	Cat# BE0291; RRID: AB_ 2687814
anti-human CD3	InVivoMAb	Cat# BE0001-2; RRID: AB_ 1107632
anti-mouse CD28	InVivoMAb	Cat# BE0015-1; RRID: AB_ 1107624
anti-mouse CD3ε	InVivoMAb	Cat# BE0001-1; RRID: AB_ 1107634
Chemicals, peptides, and recombinant proteins
Human IL-2	Novoprotein	Cat# C013
Human M-CSF	Novoprotein	Cat# C417
Mouse M-CSF	Novoprotein	Cat# CB34
Mouse IFN gamma	Novoprotein	Cat# C746
FerroOrange	DOJINDO	Cat# F374
CCK-8	DOJINDO	Cat# CK04
BODIPY™ 581/591 C11	Thermo Fisher Scientific	Cat# D3861
DMSO	Thermo Fisher Scientific	Cat# D12345
Sodium pyruvate	Thermo Fisher Scientific	Cat# 11360070
Fixable Viability Dye eFluor™ 506	Thermo Fisher Scientific	Cat# 65-0866-14
Lipofectamine™ LTX	Thermo Fisher Scientific	Cat# 15338100
Lipofectamine™ RNAiMAX	Thermo Fisher Scientific	Cat# 13778100
DNase I, RNase free(1U/L)	Thermo Fisher Scientific	Cat# EN0521
Foxp3/Transcription Factor Staining Buffer Set	Thermo Fisher Scientific	Cat# 00-5523-00
Perm/Wash Buffer	Thermo Fisher Scientific	Cat# 554723
Collagenase from Clostridium histolyticum	Sigma	Cat# C5138
ATRA	Sigma	Cat# R2625
ATP	Sigma	Cat# A1852
αKG	Sigma	Cat# K1128
GABA	Sigma	Cat# A2129
Arachidonic acid	Sigma	Cat# 10931
Thiamine	Sigma	Cat# T1270
GolgiStop™ Protein Transport Inhibitor (Containing Monensin)	BD Biosciences	Cat# 554724
Fixation and Permeabilization Solution	BD Biosciences	Cat# 554722
Cell Activation Cocktail (without Brefeldin A)	BioLegend	Cat# 423303
Ferrostatin-1	Selleck	Cat# S7243
RSL3	Selleck	Cat# S8155
Chloroquine diphosphate	Selleck	Cat# S4157
Deferoxamine	Selleck	Cat# S5742
Erastin	APExBIO	Cat# B1524
Clophosome and control liposomes	FormuMax	Cat# F70101C-AC
Necrostatin-1	MedChemExpress	Cat# HY-15760
Z-VAD-FMK	MedChemExpress	Cat# HY-16658B
OVA (323–339)	MedChemExpress	Cat# HY-P0286F
PMSF	Yeasen Biotechnology	Cat# 20104ES03
Phosphatase Inhibitor Cocktail	Yeasen Biotechnology	Cat# 20109ES05
Protease Cocktail	Yeasen Biotechnology	Cat# 20123ES10
RIPA lysis buffer	Yeasen Biotechnology	Cat# 20101ES60
DMEM	Yeasen Biotechnology	41401ES76
Percoll	GE Healthcare	Cat# 17-0891-09
TB Green® Premix Ex Taq™	Takara	Cat# RR420
RNAiso Plus	Takara	Cat# 9109
Human Colorectal Cancer Organoid Culture Medium	D1 MEDICAL TECHNOLOGY	Cat# K211M01
Critical commercial assays
RT reagent Kit with gDNA Eraser	Takara	Cat# RR047
GSH/GSSG assay kit	Beyotime	Cat# S0053
NADP+/NADPH assay kit	Beyotime	Cat# S0179
Mouse CD4 T cell Isolation Kit	BioLegend	Cat# 480005
Vitamin A ELISA Kit	Elabsciencee	Cat# E-EL-0135
Deposited data
Raw data	This paper	Database: GEO, Accession number: GEO: GSE318119: https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE318119
Experimental models: Cell lines
293T	ATCC	Cat# CRL-3216
MC38	cell bank of the Chinese Academy of Science	Cat# SCSP-5431
Hepa1-6	cell bank of the Chinese Academy of Science	Cat# SCSP-512
Experimental models: Organisms/strains
NOD.Cg-PrkdcscidIl2rgtm1SugB2mem1TacH2-Ab1tm1Doi/JicCrl	Charles River	Cat# 411
OT-II mice	Dr. Yao Yikun	N/A
Oligonucleotides
CD274 forward 5′ GGTGCCGACTACAAGCGAAT	This paper	N/A
CD274 reverse 5′ AGCCCTCAGCCTGACATGTC	This paper	N/A
Atp6v0d2 forward 5′ CTGGTTCGAGGATGCAAAGC	This paper	N/A
Atp6v0d2 reverse 5′ GTTGCCATAGTCCGTGGTCTG	This paper	N/A
Lamp1 forward 5′ TGCTCCGGGATGCCACTAT	This paper	N/A
Lamp1 reverse 5′ TGTTGTCCTTTTTCAGGTAGGTG	This paper	N/A
Ctsb forward 5′ CAGGCTGGACGCAACTTCTAC	This paper	N/A
Ctsb reverse 5′ TCACCGAACGCAACCCTTC	This paper	N/A
Ctsa forward 5′ CAGCCCTCTTTCCGGCAATA	This paper	N/A
Ctsa reverse 5′ TTTGGGTCGTTCTGCGACTC	This paper	N/A
Ctsd forward 5′ CGATTATCAGAATCCCTCTGCG	This paper	N/A
Ctsd reverse 5′ GGTCTTAGGCGATGACTGCAT	This paper	N/A
Cd38 forward 5′ TCCCTCCGTGAGCCATTTTAC	This paper	N/A
Cd38 reverse 5′ CGATGTCGTGCATCACCCA	This paper	N/A
Ciita forward 5′ GGAGGAGATCGAACTCAGCTC	This paper	N/A
Ciita reverse 5′ GTTCCGCAATGTTGGCATAGG	This paper	N/A
H2-k1 forward 5′ ACCAGCAGTACGCCTACGA	This paper	N/A
H2-k1 reverse 5′ AACCAGAACAGCAACGGTCG	This paper	N/A
H2-d1 forward 5′ TGGTGCTGCAGAGCATTACA	This paper	N/A
H2-d1 reverse 5′ TGTGCCTTTGGGGAATCTGT	This paper	N/A
H2-l forward 5′ GATGCAGAGCATTACAGGGC	This paper	N/A
H2-l reverse 5′ GCCAGGTCAGGGCAATGTC	This paper	N/A
Tap1 forward 5′ GGACTTGCCTTGTTCCGAGAG	This paper	N/A
Tap1 reverse 5′ CAGCATCCGACACAGCATGT	This paper	N/A
H2-Ab1 forward 5′ AGCCCCATCACTGTGGAGT	This paper	N/A
H2-Ab1 reverse 5′ GATGCCGCTCAACATCTTGC	This paper	N/A
H2-Eb1 forward 5′ GCGGAGAGTTGAGCCTACG	This paper	N/A
H2-Eb1 reverse 5′ CCAGGAGGTTGTGGTGTTCC	This paper	N/A
H2-Aa forward 5′ TCAGTCGCAGACGGTGTTTAT	This paper	N/A
H2-Aa reverse 5′ GGGGGCTGGAATCTCAGGT	This paper	N/A
Aldh1a1 forward 5′ ATACTTGTCGGATTTAGGAGGCT	This paper	N/A
Aldh1a1 reverse 5′ GGGCCTATCTTCCAAATGAACA	This paper	N/A
Aldh1a2 forward 5′ CAGAGAGTGGGAGAGTGTTCC	This paper	N/A
Aldh1a2 reverse 5′ CACACAGAACCAAGAGAGAAGG	This paper	N/A
Aldh1a3 forward 5′ GGGTCACACTGGAGCTAGGA	This paper	N/A
Aldh1a3 reverse 5′ CTGGCCTCTTCTTGGCGAA	This paper	N/A
Actin forward 5′ GGCTGTATTCCCCTCCATCG	This paper	N/A
Actin reverse 5′ CCAGTTGGTAACAATGCCATGT	This paper	N/A
Mouse Aldh1a1 siRNA#1 sense 5′ CCCAGUUCUUAUCCAAGAAUAdTdT	This paper	N/A
Mouse Aldh1a1 siRNA#1 antisense 5′ UAUUCUUGGAUAAGAACUGGGdTdT	This paper	N/A
Mouse Aldh1a1 siRNA#2 sense 5′ GCUCAUGUUCAUUUGGAAGAUdTdT	This paper	N/A
Mouse Aldh1a1 siRNA#2 antisense 5′ AUCUUCCAAAUGAACAUGAGCdTdT	This paper	N/A
Mouse Aldh1a1 siRNA#3 sense 5′ CGCAAUGAAGAUAUCUCAGAAdTdT	This paper	N/A
Mouse Aldh1a1 siRNA#3 antisense 5′ UUCUGAGAUAUCUUCAUUGCGdTdT	This paper	N/A
Software and algorithms
GraphPad Prism	N/A	https://www.graphpad.com; RRID:SCR_002798
Fiji	N/A	https://ImageJ.net; RRID:SCR_008520
FlowJo_V10 software	N/A	https://docs.flowjo.com/flowjo/; RRID:SCR_002285
STAR v2.7.9a	Dobin et al.51	https://github.com/alexdobin/STAR; RRID:SCR_004463
SAMtools v1.11	Li et al.52	http://samtools.sourceforge.net;
deeptools v2.5.7	Ramirez et al.53	https://deeptools.readthedocs.io/en/develop/; RRID:SCR_016366
Rsubread v2.0.3	Liao et al.54	https://bioconductor.org/packages/release/bioc/html/Rsubread.html; RRID:SCR_016945
DESeq2 v1.38.3	Love et al.55	https://bioconductor.org/packages/release/bioc/html/DESeq2.html; RRID:SCR_015687
Bowtie 2 v2.3.5.1	Langmead and Salzberg56	http://bowtie-bio.sourceforge.net/bowtie2/
macs2 v2.1.1	Zhang et al.57	https://pypi.org/project/MACS2/; RRID:SCR_021317

Experimental model and study participant details

Patients and samples

The study retrospectively allocated and analyzed tumor samples from 12 patients with colorectal cancer who received anti-PD-1 immunotherapy after disease progression on standard chemotherapy. Pre-treatment biopsy samples were stained by multiplex immunohistochemistry for the ferroptosis markers (i.e., TFRC, ACSL4, ENO1, and IDO1). Tumor responses were evaluated using modified Response Evaluation Criteria in Solid Tumors (mRECIST). Patients who achieved complete or partial responses were classified as anti-PD-1 sensitive (n = 6), while those with stable and progressive disease were termed resistant (n = 6). Specimens were obtained under informed consent approved by the Ethics Committee of Zhongshan Hospital, Fudan University (B2024-223).

Two patient-derived colon tissues were obtained from patients undergoing primary surgery at Zhongshan Hospital of Fudan University. The patients provided informed consent under Institutional Review Board-approved protocols.

Cell lines and cell culture

The authenticated murine HCC cell line Hepa1-6 and the CRC cell line MC38 were obtained from the cell bank of the Chinese Academy of Science (Shanghai). These cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Hyclone, USA) supplemented with 10% fetal bovine serum and maintained at 37°C in a humidified incubator with 5% CO₂ (Thermo Fisher Scientific). All cells were tested and confirmed to be free from mycoplasma contamination.

Animal models

Male 6-week-old C57BL/6 wild-type mice were purchased from Charles River Laboratories (Shanghai, China). Male 6-week-old NOG-MHC I/II-2 KO mice (NOD.Cg-Prkdc^scidIl2rg^tm1SugB2m^em1TacH2-Ab1^tm1Doi/JicCrl) were obtained from Shanghai Model Organisms Center, Inc. Male 6-week-old OT-II transgenic mice were kindly provided by Prof. Yi-Kun Yao at the Chinese Academy of Science, Shanghai, China. All animal care and experimental protocols were approved by the Institutional Animal Care and Use Committee of Zhongshan Hospital, Fudan University (Approval No: 2024–181).

To establish the MC38 and Hepa1-6 xenograft model, approximately 10⁶ cells were collected from 10 cm culture dishes, resuspended in 200 μL PBS, and injected into the right flank of male 6-week-old C57BL/6 mice. For RSL3 treatment, RSL3 was administered i.p. at a dose of 5 mg/kg three times per week. For DFNRL and anti-PD-1 combination therapy, DFNRL (1 mg/mL) was administered i.v. at a volume of 150 μL, and anti-PD-1 antibody (2 mg/mL, Biolegend, USA) was administered i.p. at a volume of 50 μL, twice per week, either alone or in combination. DMSO, PBS, and isotype IgG₂ served as vehicle controls. For in vivo macrophage depletion, Clophosome or control liposomes (200 μL, FormuMax, USA) were first injected i.v. 24 h before tumor cell implantation and then i.v. 3 times per week. Detailed treatment schedules were presented in the corresponding figures.

Tumor volumes were measured every 3 days using the formula: volume = length × width² × Π/6. Mice were sacrificed for further experiments when the tumor size exceeded 1500 mm³. Otherwise, tumors were allowed to grow for 15 days. Upon sacrifice, tumors were dissected immediately, and tumor weights were measured.

Method details

Preparation of mouse bone marrow-derived macrophages (BMDMs)

To obtain BMDMs, bone marrow cells were isolated from the tibia and femur of male 6-week-old wild-type or I-A/I-E-KO C57BL/6 mice. After lysing the red blood cells, the remaining cells were cultured in 10 cm dishes with DMEM supplemented with 50 ng/mL recombinant mouse macrophage colony stimulating factor (M-CSF; Novoprotein, China). Forty-eight hours later, the culture medium was replaced with fresh DMEM containing 50 ng/mL M-CSF. The adherent cells were allowed to grow for an additional 5 days. On day 7, the adherent macrophages were collected for further experiments.

Flow cytometry (FACS)

For in vitro experiments, cells were harvested by trypsinization, resuspended in cold PBS, and stained with indicated antibodies on ice.

For in vivo experiments, fresh tumor tissues were dissected and digested with 1 mg/mL collagenase IV (Sigma) and 10⁻³ U/L DNase I (Invitrogen) for 1 h. Tumor-infiltrating lymphocytes were isolated after density gradient centrifugation using Percoll (Pharmacia) according to the manufacturer’s instructions and stained with the indicated antibodies. Tumor-associated macrophages (CD45⁺CD11b⁺F4/80⁺) were sorted by FACS, followed by quantitative reverse transcription-polymerase chain reaction (qRT-PCR) analysis.

Cells were stained with the following antibodies (BioLegend, USA): BV606 anti-mouse CD45 antibody, PE-Cy7 anti-mouse CD3 antibody, Fluor450 CD4 antibody, PerCP cy5.5 anti-mouse CD8 antibody, PerCP cy5.5 anti-mouse CD11b antibody, PE anti-mouse CD27 antibody, PE anti-mouse F4/80 antibody, FITC anti-mouse PD-1 antibody, FITC anti-mouse MHC-II (I-A/I-E) antibody, APC anti-mouse MHC-I (H-2Kd/H-2Dd) antibody, and APC anti-mouse TIM-3 antibody.

Patient-derived organoid

Tumor spheroids were isolated and cultured following previously reported protocols with some modifications.⁵⁸ Fresh tumor tissues from primary CRC patients were collected and immersed immediately in cold DMEM after resection. The tissues were minced and incubated with 1 mg/mL collagenase IV (Sigma), 0.05% hyaluronidase (Sigma), and 0.01% deoxyribonuclease (Sigma) for 15 min. The cell suspension was filtered through a 70 μm filter and incubated in red blood cell lysis buffer for 5 min. Paired peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll-Paque density gradient centrifugation. Isolated PBMCs were cultured with anti-human CD3 (2.5 μg/mL) and CD28 (2.5 μg/mL) antibodies (BioXcell, USA) and 50 ng/mL recombinant human M-CSF (Novoprotein, China) for 5 days prior to experiments. After washing 3 times with cold PBS, tumor spheroids and activated PBMCs were resuspended in 50 μL Matrigel (Corning) at a 1:10 ratio and cultured in classic human CRC organoid medium (D1 MEDICAL TECHNOLOGY) for 3 days.

Quantitative reverse transcription PCR (qPCR)

Cells were lysed in RNAiso Plus reagent (Takara), and total RNA was extracted. cDNA was synthesized using an RT reagent Kit with gDNA eraser (Takara) according to the manufacturer’s instructions. qPCR was performed on an Applied Biosystems 7500 Sequence Detection System (Applied Biosystems, USA) using TB Green premix (Takara). A total of 1 μg cDNA was used per 20 μL reaction with gene-specific primers. mRNA expression was quantified using the 2^−ΔΔCT method relative to Actb. Data are presented as fold changes in expression levels compared to the control group.

Small interfering RNA (siRNA) and transfection

For Aldh1a1 knockdown, Hepa1-6 and MC38 cells were reverse-transfected with 100 pmol of three independent murine Aldh1a1-targeting or control siRNAs using Lipofectamine RNAiMAX (Thermo Fisher) according to the manufacturer’s protocol. After 48 h, cells were harvested for qPCR and Western blot to assess Aldh1a1 knockdown efficiency.

Cell viability

Cells (1 × 10⁴) were seeded in 96-well plates and allowed to adhere for 24 h before treatment with the indicated reagents in the culture medium. At the endpoint, cells were washed with PBS, and cell viability was quantified using the CCK-8 assay (Dojindo) according to the manufacturer’s instructions.

GSH/GSSG and NADPH/NADP⁺ determination

Intracellular GSH/GSSG and NADPH/NADP⁺ ratios were determined using commercial assay kits (Beyotime) as previously described. Briefly, 1 × 10⁶ cells were lysed, and the GSH Detection Mixture was added to convert GSSG into GSH by glutathione reductase for spectrophotometric 5,5′-dithiobis (2-nitrobenzoic acid) (DTNB) reduction assay at 420 nm. Standard curves were used to quantify GSH levels. For NADP⁺/NADPH, half of the lysate was heat-inactivated at 60°C to deplete NADP⁺. Unheated samples were kept on ice containing both NADP⁺ and NADPH. NADP⁺ reacted with the working buffer and was reduced into NADPH. At the same time, the newly formed NADPH reduced 2-(4-iodophenyl)-3-(4-nitrophenyl)-5-(2,4-disulfophenyl)-2H-tetrazolium (WST-8) to formazan, which could be measured spectrophotometrically at 450 nm. NADPH/NADP⁺ ratios were calculated using the following formula: (intensity of heated sample)/(intensity of unheated sample – intensity of heated sample).

Detection of intracellular Fe²⁺

After indicated treatments, cells (1 × 10⁵/mL in PBS) were incubated with the Fe²⁺ probe FerroOrange (Dojindo) at a dilution of 1:1000 on ice for 15 min. Fluorescence was then detected by FACS.

Immunohistochemistry (IHC)

Formalin-fixed, paraffin-embedded tumor tissues were sectioned, deparaffinized, and rehydrated. To quench endogenous peroxidase activity, slides were incubated with 3% H₂O₂ in methanol for 10 min at room temperature. Antigen retrieval was performed by heat-induced epitope retrieval, where the sections were boiled in 10 mM sodium citrate buffer (pH 6.0) for 10 min. The sections were then blocked with 10% normal goat serum and incubated overnight at 4°C with primary antibodies. The next day, slides were incubated with HRP-conjugated secondary antibodies for 1 h at room temperature, followed by development with DAB substrate. Hematoxylin was used for counterstaining, and the slides were dehydrated and mounted. Mouse or rabbit IgG served as negative controls.

Co-culture of OT-II T cells and BMDMs

OVA^323-339-specific CD4⁺ T cells were isolated from splenocytes of OT-II transgenic mice by magnetic separation (BioLegend, USA). The isolated OT-II CD4⁺ T cells were cultured with anti-CD3 (2.5 μg/mL) and anti-CD28 (2.5 μg/mL) antibodies (BioXcell, USA). BMDMs were pulsed with OVA^323-339 peptide (1 μg/mL) for 48 h, then co-cultured with OT-II CD4⁺ T cells at a 1:1 ratio for 72 h. The proliferation rates of OT-II CD4⁺ T cells were assessed by monitoring the carboxyfluorescein diacetate succinimidyl ester (CFSE) signal using FACS.

RNA-sequencing and analysis

RNA sequencing for whole transcriptome analysis was performed using the Illumina NovaSeq 6000 platform. Every experimental group had 3 biological replicates, and all RNA samples passed quality control with 5–8 Gb and a Q20 ≥ 90%. Transcript abundances (including mRNAs, pseudogenes, noncoding RNAs, and other predicted RNAs) were calculated in FPKM (fragments per kilobase of transcript per million mapped reads) and used for downstream Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway, and Gene Set Enrichment Analysis (GSEA) analysis. For generating heatmaps, gene expression levels were normalized to the control group.

Bioinformatics

Gene expression data from TCGA and published cohort RNA-seq were normalized to z-scores. Signature scores were calculated by averaging z-scores of the following gene sets in each sample: Ferroptosis (including TFRC, IDO1, ENO1, ACSL4) and cytotoxic T lymphocyte (including CD8A, CD8B, GZMA, GZMB, PRF1). The Tumor Immune Single-Cell Hub 2 (TISCH2) scRNA-seq database (http://tisch.comp-genomics.org) was analyzed to examine cancer cell subtype-specific expression of the nitroreductases, i.e., NQO1, CYPOR, and XDH, within the tumor microenvironment.

Transient transcriptome sequencing (TT-seq)

TT-seq was carried out as described previously.⁵⁹ BMDM cells were grown in 15 cm dishes, and the RNA was labeled with 4-thiouridine (4sU) (Cool Chemistry) in vivo for 15 min. TRIzol (Invitrogen) was added to stop the reaction, and total RNA was extracted according to instructions. As a control, 4sU-labeled DLD-1 RNA was mixed as the spike-in. The mix was fragmented with 0.2 M NaOH for 14 min, and Tris-HCl (pH 6.8) was added to stop the fragmentation. Biotinylation of labeled RNA was carried out in 150 μL of biotinylation mix (100 μg fragmented total RNA, 10 mM HEPES pH 7.5, 1 mM EDTA, 0.167 mg/mL MTSEA-biotin) for 30 min and purified with Streptavidin Magnetic Beads (NEB). The biotinylated RNA was eluted with 100 mM DTT and purified with RNA Clean beads (Vazyme, N412). The RNA libraries were constructed from the 100 μg purified RNA sample using the Stranded mRNA-seq Lib Prep Module for Illumina (ABclonal, RK20349) following the manual and sequenced with the NovaSeq 6000 platform (150 bp, pair-end; Annoroad Gene Technology, Beijing, China).

Raw reads were trimmed with Trim Galore v0.6.6 (https://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) to remove adaptors and low-quality reads with the parameter “-q 25”, followed by mapping to the mm10 and hg19 genome using STAR v2.7.9a with parameters “-outSAMtype BAM SortedByCoordinate --outFilterMultimapNmax 1 --alignSJoverhangMin 8 --alignSJDBoverhangMin 1 --outFilterMismatchNoverReadLmax 0.04 --alignIntronMin 20 --alignIntronMax 1000000 --alignMatesGapMax 1000000”.⁵¹ The number of spike-in hg19 reads counted by SAMtools v1.11 was used to calculate the normalization factor alpha = 1e5/hg19_count.⁵² The TT-seq signal normalized by spike-in factor was generated with bamCoverage from deeptools v2.5.7.⁵³ Gene expression quantification was performed using featureCounts from Rsubread v2.0.3.⁵⁴ Differentially expressed genes were calculated by DESeq2 v1.38.3⁵⁵ with a cutoff as follows: Padj <0.05. Functional annotations were done using enrichGO from the clusterProfiler package.⁶⁰

Assay for transposase-accessible chromatin with high-throughput sequencing (ATAC-seq)

ATAC-seq assays were performed according to the manual of the Hyperactive ATAC-Seq Library Prep Kit for Illumina (Vazyme, TD711). Briefly, 5 × 10⁵ cells were lysed in ice-cold lysis buffer, followed by resuspending in fragmentation buffer and incubation at 37°C for 30 min. The fragmented DNA products were extracted using ATAC DNA Extract Beads. After amplification, DNA was purified using ATAC DNA Clean Beads through a two-step sorting protocol (0.55 ×/1.2 ×). Then, the libraries were sequenced on the NovaSeq 6000 platform in PE150 mode.

After trimming the adaptors and low-quality reads as described for TT-seq above, the remaining reads were aligned to the mm10 genome using Bowtie 2 v2.3.5.1 with parameters “-k 1 -X 2000”.⁵⁶ We used SAMtools v1.11 and Picard v2.23.3 (https://broadinstitute.github.io/picard/) to remove low mapping quality reads, mitochondrial DNA, and PCR duplicates. Reads counts adjusted to reads per kilobase per million mapped reads (RPKM) using deeptools v2.5.7, and peaks were called by macs2 v2.1.1.⁵⁷ Differential accessibility analysis between ATRA- and DMSO-treated cells on the ATAC-seq peaks was performed with the DiffBind R package v3.8.4.⁶¹

Cleavage under targets and tagmentation (CUT&Tag) sequencing

CUT&Tag assays were performed using the handbook of NovoNGS CUT&Tag 4.0 High-Sensitivity Kit (for Illumina, Novoprotein, H259-YH01). Briefly, 5 × 10⁵ ATRA- or DMSO-treated cells were washed twice with wash buffer and then incubated with activated concanavalin A beads at RT for 10 min. The beads binding cells were permeabilized with 0.05% digitonin and incubated with the primary antibody of H3K27ac antibody (1:100; Abclonal, A7253), H3K4me3 antibody (1:100; Abclonal, A2357), Rpb1 NTD (D8L4Y) antibody (1:100; CST, 14958) at room temperature for 2 h, followed by incubation with the rabbit IgG antibody at room temperature for 1 h at a dilution of 1:100. Then the cells were washed and incubated with ChiTag transposomes for 1 h at room temperature, and the incubation was stopped by adding 10 μL 10% SDS at 55°C for 10 min. DNA fragments were extracted, purified, and submitted to generate libraries. DNA libraries were sequenced using NovaSeq 6000 with PE150.

After trimming the adaptors and low-quality reads with Trim Galore v0.6.6, the remaining reads were aligned to the mm10 genome and Lambda DNA assembly using Bowtie 2 v2.3.5.1 with parameters “--local --very-sensitive --no-mixed --no-discordant --phred33 -I 10 -X 700”. The number of spike-in Lambda DNA reads was used to calculate the normalization factor alpha = 1e4/Lambda DNA_count. Normalized bigwig files were generated using deeptools v2.5.7, and peaks were called by macs2 v2.1.1 with parameters “-p 1e−5, -keep-dup all”.

Preparation and characterization of mPEG-pLys-NI DFNRL

The preparation of hypoxia-responsive DFNRL was performed by a film hydration protocol using methoxy (polyethylene glycol)-polylysine-nitroimidazole (mPEG-pLys-NI) copolymer. Briefly, 20 mg of mPEG-pLys-NI was dissolved in 2 mL of N, N-dimethylformamide (DMF). The DMF was then removed under rotary evaporation by an oil pump, and the obtained thin film was hydrated with 20 mL of PBS for 2 h. The hypoxia-responsive DFNRL was obtained after filtration through a 0.45 μm membrane. A detailed synthesis route can be found in the Supplementary File. For in vitro and in vivo imaging studies, Coumarin-6 and IR780 were encapsulated in the DFNRL using the same procedure, respectively. To visualize the morphology of hypoxia-responsive DFNRL, the DFNRL was diluted, negatively stained with 1% (w/v) uranyl acetate, air-dried, and the grid was observed under a transmission electron microscope (TecnaiTM G2 spirit BioTWIN, Hillsboro). The physical properties of hypoxia-responsive DFNRL, including particle diameter, polymer dispersity index, and zeta potential, were measured by the dynamic light scattering (DLS) technique (Zetasizer Nano-ZS).

Biodistribution of mPEG-pLys-NI DFNRL

MC38 and Hepa1-6 tumor-bearing mice were intravenously injected with hypoxia-responsive IR780-labeled DFNRL (IR780-DFNRL) 10 days after tumor implantation. Whole-body fluorescence was tracked at 4h, 12h, 18h, and 24h using an IVIS Spectrum imager (PerkinElmer) to assess DFNRL accumulation and retention in vivo. Free IR780 dye served as a control. After 24 h, ex vivo organ fluorescence was quantified by IVIS imaging following necropsy and tissue harvesting. Major organs, including the heart, lung, liver, kidney, spleen, intestine, and colon, were collected, washed in PBS, and imaged to compare DFNRL localization in tumor tissues versus healthy tissues.

¹H NMR monitoring of mPEG-pLys-NI DFNRL

The polymer mPEG-pLys-NI (10 mg) was dissolved in 0.5 mL of deuterated dimethyl sulfoxide (d6-DMSO). An excess of sodium dithionite (Na₂S₂O₄) was added as an oxygen scavenger. The mixture was centrifuged at 5000 rpm for 5 min to remove insoluble material. The supernatant was transferred to an NMR tube for analysis. ¹H NMR spectra were recorded at 0, 0.25, 1, and 4 h after sample preparation.

Quantification and statistical analysis

Data are presented as mean ± standard deviation (SD). Statistical analyses were performed using Prism 8.0 (GraphPad). Unpaired Student’s t-tests were used to compare quantitative variables between two groups. The association between two continuous variables was assessed using Pearson’s correlation coefficient. Two-sided p < 0.05 was considered statistically significant. All cellular experiments were performed with at least three independent biological replicates. The number of technical replicates is indicated in the figure legends.

Published: April 29, 2026