Ferroptosis Signature Shapes the Immune Profiles to Enhance the Response to Immune Checkpoint Inhibitors in Head and Neck Cancer

Abstract As a type of immunogenic cell death, ferroptosis participates in the creation of immunoactive tumor microenvironments. However, knowledge of spatial location of tumor cells with ferroptosis signature in tumor environments and the role of ferroptotic stress in inducing the expression of immune‐related molecules in cancer cells is limited. Here the spatial association of the transcriptomic signatures is demonstrated for ferroptosis and inflammation/immune activation located in the invasive front of head and neck squamous cell carcinoma (HNSCC). The association between ferroptosis signature and inflammation/immune activation is more prominent in HPV‐negative HNSCC compared to HPV‐positive ones. Ferroptotic stress induces PD‐L1 expression through reactive oxygen species (ROS)‐elicited NF‐κB signaling pathway and calcium influx. Priming murine HNSCC with the ferroptosis inducer sensitizes tumors to anti‐PD‐L1 antibody treatment. A positive correlation between the ferroptosis signature and the active immune cell profile is shown in the HNSCC samples. This study reveals a subgroup of ferroptotic HNSCC with immune‐active signatures and indicates the potential of priming HNSCC with ferroptosis inducers to increase the antitumor efficacy of immune checkpoint inhibitors.


Introduction
Recent advances in understanding the new form of cell death, ferroptosis, which is different from other well-known forms of cell death, [1] have attracted the attention of cancer researchers due to the high potential of ferroptosis in cancer therapies. Ferroptosis is an iron-dependent nonapoptotic cell death driven by the accumulation of lipid reactive oxygen species (ROS). Glutathione (GSH)/glutathione peroxidase 4 (GPX4) plays a pivotal role in protecting cells from ferroptosis by hydrolyzing lipid hydroperoxides. [1] Mounting evidence supports the crucial role of ferroptosis in tumor suppression. For example, a strong dependence of cancer cells on GPX4 and system Xc-has been noted in multiple types of cancer, [2] and the inhibition of the system Xc-coding gene SLC7A11 induces ferroptotic cell death and enhances cytotoxicity in therapy-resistant cancer cells. [3] Importantly, cancer stem cells (CSCs) or mesenchymal-like cancer cells have been found to be especially vulnerable to ferroptotic cell death. [4,5] This finding highlights the potential of ferroptosis induction as a new anticancer therapy, particularly targeting drug-resistant CSCs. [6,7] However, the available strategy for ferroptosis induction as an adjunct to cancer therapy in clinical oncology is still limited.
In addition to the vulnerability of CSCs to ferroptotic cell death, another attracting point of ferroptosis in cancer treatment is that ferroptosis is closely related to the inflammatory response of the tumor microenvironment (TME). Ferroptosis impacts immune microenvironments in two different ways. [8] First, ferroptosis affects the number and function of immune cells themselves. Ferroptosis influences macrophage polarization to modulate immune environments. [9] Lipid peroxidation and ferroptosis regulate the viability and activity of CD8 + cytotoxic T cells and CD4 + helper T cells. [10,11] Second, ferroptotic cells themselves release damage-associated molecular patterns (DAMP) which can be recognized by immune cells to trigger subsequent inflammatory responses. [12,13] Ferroptotic cells also produce abundant pro-inflammatory cytokines and chemokines to create an inflamed microenvironment. [14,15] However, the clinical relevance and therapeutic impact of the signature created by endogenous ferroptotic stress in tumor cells, the "ferroptotic signature," remains elusive.
The clinical characteristics of head and neck squamous cell carcinoma (HNSCC) are distinct from those of cancers originating from other tissues/organs: advanced HNSCC is often associated with severe destruction of surrounding tissues and neck lymphadenopathy, and local-regional recurrence is the main pattern of treatment failure. [16] In recent years, the great success of anticancer immunotherapy, especially immune checkpoint inhibitors (ICIs), has led to a paradigm shift in cancer treatments. [17] In HNSCC, ICIs showed promising results in recurrent/metastatic (R/M) HNSCC. [18][19][20] The programmed death ligand-1 (PD-L1) is expressed on the surface of multiple types of cells in TME including cancer cells and is considered a major indicator for ICI treatments. [21,22] However, clinical strategies to enhance PD-L1 expression and remodel the immune environment to increase ICI efficacy are limited. Identifying the ICIsusceptible subgroup of HNSCC and developing a strategy to sensitize HNSCC to ICI treatment are urgent and unmet medical needs.
In this study, we demonstrate that ferroptotic stress induces the inflammation signature and PD-L1 expression in HNSCC. We also reveal that the ferroptosis signature of HNSCC shapes TME to an active immune state. This finding provides a potential strategy to enhance the efficacy of ICIs by identifying the optimal group of patients for treatment and priming HNSCC with potential ferroptosis inducers.

The Ferroptosis Signature Correlates with Inflammation in HNSCC Samples
To systemically approach the heterogeneous characteristics of HNSCC representing different steps in the metastatic journey, we collected matched pairs of primary tumors (including the inner core of the tumor, IC; and the invasive front, IF), metastatic tumors (M), and adjacent normal tissues from 21 patients with HNSCC receiving treatments at Taipei Veterans General Hospital (TVGH). Bulk RNA sequencing (RNA-seq) was performed from the 65 collected samples ( Figure S1A, Table S1, Supporting Information). Weighted correlation network analysis (WGCNA) was implemented for 18428 protein-encoding genes from the samples [23] ( Figure S1B, Table S2, Supporting Information). Differential patterns of signaling networks, especially inflammation/immune-related signatures and ferroptosis signals, were observed among the three parts of the tumors. The epithelial-mesenchymal transition (EMT) signature was also observed to validate the spatial gene expression relationship in the collected samples ( Figure 1A). Therefore, we annotated the modules with a hypergeometric test in the ferroptosis signature (with reference to WP_FERROPTOSIS), [24] microenvironmental inflammation (with reference to BIO-CARTA_INFLAM_PATHWAY), tumor inflammation, [25] EMT, [26] and interferon-stimulated gene (ISG) [27] signature sets (Table S3, Supporting Information). For EMT, an enriched signature from IC to IF to M was observed by network analysis with overlay of the log 2 T/N values between tumors (T) and corresponding normal tissues (N) ( Figure S1C,E, Supporting Information), which validated the spatial gene expression pattern of the samples. Regarding the subnetworks centered on the ferroptosis and immune-related signature modules, a gradual increase in the expression of microenvironmental inflammation signature ( Figure S1D,E, Supporting Information), tumor inflammation ( Figure 1B; Figure S1E, Supporting Information), ISG ( Figure 1C; Figure S1E, Supporting Information), and ferroptosis ( Figure 1D; Figure S1E, Supporting Information) from IC to IF to M was revealed. Data indicate that tumor aggressiveness was associated with EMT and inflammation as expected, and an increased ferroptosis signature was also revealed together with increased aggressiveness.
Next, we applied single-cell RNA sequencing (scRNA-seq) to three primary HNSCC samples to investigate the correlation between tumor inflammation and ferroptosis. The characteristics of the patient are shown in Figure S1A and Table S1, Supporting Information. The unsupervised clustering of 3369 cells generated 11 clusters of cells ( Figure 1E). We identified tumor cells or non-tumor cells population from scRNA seq data according to an epithelial score [28] (Figure S1F, Supporting Information). We analyzed the signature of microenvironment cells, including T cells, B/plasma cells, dendritic cells, macrophages, fibroblast, myocytes mast cells, and endothelial cells ( Figure S1G, Supporting Information). Next, we focused on the analysis of tumor cell clusters (comprising clusters 3 and 11 in the dotted square box of Figure 1E). A re-clustering of tumor cells generated five clusters distributed by UMAP ( Figure 1F; Table S4, Supporting Information). The correlation of tumor inflammation and ferroptosis signature was demonstrated between tumor cell groups ( Figure 1G; Table S5, Supporting Information). MSigDB hallmark analysis was used to compare the expression of the top 50 genes identified by scRNA-seq in ferroptosis_high versus ferroptosis_low cells. The results showed that the EMT and inflammation response were significantly enriched signatures in ferroptosis_high cells ( Figure 1H). We also found that ferroptosis_high cells expressed a higher tumor inflammation signature compared to ferropto-sis_low cells ( Figure 1I). Taken together, these results indicate a significant association of ferroptosis signature and inflammation in HNSCC samples.

Spatial Association between the Ferroptosis Signature and Tumor Inflammation in HNSCC
To track the ferroptosis signature throughout the metastatic journey of HNSCC cells, we applied the 10x Genomics Visium spatial transcriptomic platform for analyzing two independent HNSCC samples. The characteristics of the patients are shown in Table  S1, Supporting Information. The workflow of 10x Genomics Visium spatial transcriptomic analysis is illustrated in Figure 2A. The samples were harvested immediately after surgery and then divided the samples from the outer part of the tumor to the inner tumor part by cryosection and analyzed using 10x Genomics Visium. For patient no. 1, unsupervised clustering generated 12 clusters in four analyzed slides ( Figure  Here, we identified the tumor invasive fronts region and the inner tumor region in one of the section tumors ( Figure 2B). A similar spatial distribution pattern of tumor inflammation and ferroptosis signatures was observed in the samples analyzed by 10x Genomics Visium ( Figure 2C,D). Next, we focused on tumor cell clusters (clusters T1-T6). The ferroptosis_high group revealed a higher expression level of the tumor inflammation signature ( Figure 2E). An increased expression of the inflammation/immune-related genes HLA-DQA1, HLA-DRB1, CCL5, and CXCL9 was shown in the ferroptosis_high group ( Figure 2F). The positive correlation between tumor inflammation and ferroptosis signatures was validated in patient no. 2 ( Figure S3, Supporting Information). Together, spatial transcriptomic studies further support the colocalization of tumor inflam-mation and ferroptosis signatures in the same histological regions of HNSCC.

Ferroptosis Induces the Immunogenic Signature and PD-L1 Expression in HNSCC
Our observation of the clinical sample suggested that the tumor inflammation signature was correlated with the ferroptosis signature of HNSCC (Figures 1,2). Therefore, we sought the group of major genes influenced by ferroptotic stress in HNSCC. To examine the impact of non-lethal ferroptotic stress on HNSCC cells, we applied two ferroptosis inducers, (1S,3R)-RSL3 (RSL3) and FIN56 [29] to treat HNSCC. The mechanism of action of these two drugs is summarized in Figure S4A, Supporting Information. A panel of HNSCC cell lines and a primary HNSCC culture with different spectrums of EMT phenotype ( Figure S4B, Supporting Information) were subjected to the lipid ROS analysis and the WST-1 assay under treatment with ferroptosis inducers. Epithelial-type FaDu and SAS cells were more resistant to the ferroptosis inducers RSL3 and FIN56 and generated fewer lipid ROS after treatment, and mesenchymal-type OECM-1 cells were the most susceptible to inhibition of GPX4 among these cell lines and generated the most lipid ROS after treatment ( Figure S4C-E, Supporting Information). We applied sublethal doses of ferroptosis inducers to treat HNSCC cells and examined the ferroptotic stress-induced transcriptomic change in subsequent experiments ( Figure S4F, Supporting Information). The experimental schema is illustrated in Figure 3A. We intersected the genes up-regulated in HNSCC samples analyzed by two platforms: the top 1500 genes up-regulated in ferroptosis_high group by spatial transcriptomics (Table S6A, Supporting Information) and the genes up-regulated in the RNA sequencing results of HSC-3 cells treated with a sublethal concentration of FIN56 versus control (n = 539) (FPKM fold change ≥ 1.5) (Table S6B, Supporting Information). 42 genes were found to be co-upregulated in both platforms ( Figure 3B; Table S6C, Supporting Information), and these genes were subjected to GO ontology analysis. Interestingly, CD274 acts as the signal hub in these upregulated genes ( Figure S5A, Supporting Information and a zoomed-in illustration in Figure 3C; Table S7, Supporting Information), and upregulation was validated by RT-qPCR in HSC-3 cells treated with FIN56 ( Figure 3D). In subsequent experiments, we focused on ferroptosis-induced CD274 due to the high therapeutic impact of up-regulation of PD-L1 in cancer cells.   Next, we performed a subgroup analysis to explore the correlation between the expression of CD274 and ferroptosis-related genes in HPV-positive versus negative HNSCC because HPV positivity represents distinct clinical characteristics of HNSCC patients. A prominent correlation was shown between the expression of CD274 and ferroptosis-related genes in the HPV-negative group, that is, a negative correlation of CD274 and ferroptosisrelated genes (GPX4, GSS, GCLC, SLC3A2) and a positive correlation of CD274 and ACSL4 ( Figure S6A, Supporting Information). However, the correlation between CD274 and ferroptosisrelated genes was not as evident in HPV-positive cases, with the exception of the positive correlation between CD274 and ACSL4 ( Figure S6B, Supporting Information). Furthermore, we examined the expression level of the PD-L1 and GPX4 proteins in the TVGH HNSCC tumor samples, and a significant negative correlation between PD-L1 and GPX4 was revealed ( Figure 3I,J). The above data indicate a negative correlation between the expression of PD-L1 and ferroptosis-related genes in HNSCC samples, especially those negative for HPV, and a sublethal stress of ferroptosis up-regulates PD-L1 in HNSCC cells.

Ferroptosis-Induced NF-B Activation and Calcium Influx Contribute to Up-Regulation of PD-L1
Next, we sought the pathway(s) that mediate ferroptosis-induced up-regulation of PD-L1. The cellular ROS level was significantly increased after RSL3 treatment, as expected ( Figure 4A). The ROS scavenger N-acetyl cysteine (NAC) attenuated the RSL3 treatment-induced protein and mRNA expression of PD-L1 in CAL-27 cells ( Figure 4B,C). The transcription inhibitor actinomycin D repressed ferroptosis-induced upregulation of PD-L1 ( Figure 4D), indicating the potential for ferroptosis-induced expression of PD-L1 through ROS-mediated transcriptional activation of CD274. Next, we evaluated the possible ferroptotic ROSactivated pathways in HNSCC cells by co-treatment with FIN56 and different inhibitors for inflammatory signal pathways, including a STAT3 inhibitor, a STAT1 inhibitor, and an NF-B inhibitor. Suppression of NF-B but not STAT1 or STAT3 attenuated ferroptosis-induced PD-L1 expression ( Figure 4E). Analysis of the RNA sequencing data of HSC-3 cells treated with/without FIN56 showed that NF-B signal pathway was one of the major pathways in the analysis of the MSigDB hallmark pathways and the KEGG pathways ( Figure S7A,B, Supporting Information). Next, we examined whether ferroptosis activated NF-B signaling. Increased phosphorylation and nuclear translocation of p65 was observed in SAS cells under FIN56 treatment ( Figure 4F,G). Direct activation of NF-B signal by phorbol myristate acetate (PMA) upregulated PD-L1 in SAS cells ( Figure S7C,D, Supporting Information). Activation of NF-B by ferroptosis inducers was validated by EMSA ( Figure 4H) and ELISA ( Figure 4I) in HNSCC cell lines. Inhibition of NF-B by parthenolide downregulated FIN56-induced PD-L1 expression ( Figure 4J). The above data indicate that ferroptosis-induced ROS activates NF-B to upregulate PD-L1 in HNSCC.
Previous studies showed that calcium signal is a character of ferroptosis and a major signal in ROS induction. [30,31] We examined whether the calcium signal is also involved in the regulation of ferroptosis-induced PD-L1. Treatment of ferroptosis inducers did not induce significant calcium influx or change the pattern of calcium mobilization induced by a calcium ionophore A23187 within 10 min in HNSCC cells ( Figure S8A,B, Supporting Information). However, a gradual elevation of cytosolic calcium was detected after hours of FIN56 treatment ( Figure S8C, Supporting Information). Nevertheless, FIN56-induced calcium influx was abrogated in HNSCC cell cultivated with the calcium-free medium or intracellular calcium chelator BAPTA treatment (Figure S8D,E, Supporting Information). Prolonged treatment with FIN56 (24 h) significantly reduced A23187-induced calcium mobilization ( Figure S8F, Supporting Information). In contrast, prolonged FIN56 incubation did not affect the calcium mobilization induced by an ER calcium ATPase inhibitor thapsigargin ( Figure  S8G, Supporting Information). Treatment with calcium chelators EGTA-AM or BAPTA attenuated FIN56-induced PD-L1 expression ( Figure S8H, Supporting Information), which validates the role of ferroptosis-induced calcium influx in PD-L1 upregulation. The above results indicate that extracellular calcium is responsible for ferroptosis-induced calcium influx, and the contribution of ER-stored calcium in ferroptosis-induced calcium influx is yet to be determined.

Ferroptosis Inducers Suppress HNSCC Growth, Modulate Tumor Microenvironments, and Sensitize Murine HNSCC to Anti-PD-L1 Antibody Treatment
We validated the antitumor activity of the ferroptosis inducers in three xenografted HNSCC models. Both the human HNSCC cell line HSC-3-formed tumors and primary HNSCC cultureformed tumors were repressed by RSL3 treatment (Figure 5A,B). RSL3 also inhibited the growth of four patient-derived HNSCC xenografted tumors (PDX) (Figure 5C,D).
We next investigated the influence of the ferroptosis inducer treatment on tumor microenvironments. The murine syngeneic HNSCC model by inoculating the murine HNSCC cell line MTCQ1-2 [32] to the subcutaneous region of C57BL/6J mice was applied ( Figure 6A). Immunohistochemical staining of the harvested tumors demonstrated the increased infiltration of CD4 + cells, CD8 + cells, and granzyme B + cells in the FIN56-treated tumors ( Figure 6B). Flow cytometric analysis of the tumor-infiltrated immune cells revealed an increased CD4 + cells, CD4 + Pd1 + cells, a reduced CD11b + F4/80 + macrophages, and a trend of increased CD8 + cells, CD8 + Ifn + cells, CD8 + Pd1 + cells, CD4 + Ifn + cells, and B cells in FIN56-treated tumors   Mice were sacrificed on the 28th day and tumors were harvested. Left lower, the photo of the tumors harvested. Right lower, the tumor weight of the two groups. DMSO group n = 5, RSL3 group n = 4. The data is presented in mean ± S.D; *p < 0.05 (Student's t-test). B) Upper, a schema to illustrate the experiment. 1 × 10 6 of the primary HNSCC cells were inoculated into the subcutaneous region of the nude mice. Intratumoral injection of RSL-3 (100 mg kg −1 ) was given on 5th and 7th day. The mice were sacrificed on the 28th day and tumors were harvested. Left lower, the photo of harvested tumors. Right lower, the tumor weight of the two groups. DMSO group n = 5, RSL3 group n = 5. The data is presented in mean ± S.D (Student's t-test). C) Upper left, a schema for illustrating the experiment. Patient-derived xenograft tumors (PDX) were inoculated to the subcutaneous region of the nude mice. Intratumoral injection of RSL-3 (100 mg kg −1 ) or DMSO control was administered on day1. The mice were sacrificed on the day10 and the tumors were harvested. Lower left, photo of harvested tumors. Right, the tumor weight of the two groups. DMSO group n = 4, RSL3 group n = 4. The data is presented in mean ± S.D (Student's t-test). D) Left, the tumor volume at day1, day4, day7, and day10. Right, a waterfall plot to show the tumor volume change by quantifying the ratio of tumor volume at the day1 to the day10. DMSO group n = 4, RSL3 group n = 4. **p < 0.01 (Student's t-test). Figure 6. Ferroptosis induced active-immune TME and sensitizes the tumors to PD-L1 treatment. A) A schema to illustrate the experiment. 1 × 10 6 of MTCQ1-2 cells were inoculated into the subcutaneous region of the B6 mice. Intratumoral injection of FIN56 (100 mg kg −1 ) was administered on the 15th day and 17th day. Mice were sacrificed on the 20th day and tumors were harvested. B) Ferroptotic inducer treated HNSCC tumor tissues were collected for IHC staining. Left, the average infiltrated cells numbers were counted (n = 6). The data is presented in mean ± S.D; *p < 0.05, **p < 0.01 (Student's t-test). Right, representative images of the IHC were shown. Scale bar, 100 μm. C) Tumor-infiltrated CD45 + immune cells were assessed by flow cytometry. Population of CD4 + T cells, CD8+ T cell, macrophage, DC, and B cell in tumors from the indicated groups. DMSO (n = 5), FIN56 (n = 4). The data is presented in mean ± S.D: *p < 0.05, **p < 0.01 (Student's t-test). D) Pd-l1 expression and percentage of pd-l1 + cells of CD4 + T cells, CD8 + T cell, macrophage, DC, and B cell in tumors from the indicated groups. DMSO (n = 5), FIN56 (n = 4). The data is presented in mean ± S.D (Student's t-test). E) A schema of syngeneic tumors 1 × 10 7 MOCL2-1 cells were subcutaneously injected into the flank of C57BL/6 mice. Tumor-bearing mice received two intra-tumor injection of 100 mg kg −1 FIN56 on the 12th and 14th day after tumor cell injection. On the 14th day 100 μg anti-PD-L1 antibody (Bio X Cell) was administered intraperitoneally to each mouse. Antibodies were administered every 3 days until the 32nd day. The mice were sacrificed on the 38th day post-tumor cell injection. F) Left, photo of the harvested tumors. Middle, quantification of tumor weight. DMSO with isotype antibody control (n = 4), DMSO with anti-PD-L1 antibody (n = 4), FIN56 with isotype antibody control (n = 3), FIN56 with anti-PD-L1 antibody (n = 4). The data is presented in mean ± S.D; *p < 0.05 (Student's t-test). Right, a waterfall plot to show the tumor volume change by quantifying the ratio of tumor volume at the 38th day to the 12th day. **p < 0.01 (Student's t-test). www.advancedsciencenews.com www.advancedscience.com ( Figure 6C). For ferroptotic stress upregulates PD-L1 in HNSCC cells (Figures 3,4), we examined the influence ferroptotic stress on Pd-l1 expression of mouse immune cells. The mean fluorescent intensity and the Pd-l1-positive subpopulations were not different among the FIN56-treated and control groups of the syngeneic murine HNSCC tumors ( Figure 6D).
Next, we examine the combination of the ferroptosis inducer with an anti-PD-L1 antibody ( PD-L1) in a syngeneic murine HNSCC model MOCL2 cell line/C57BL6/J mice. [33] Activation of NF-B and induction of PD-L1 expression by sublethal FIN56 was validated in MOCL-2 cells in vitro ( Figure S9A-C, Supporting Information). Upregulated PD-L1 expression in MOCL2-formed tumors under FIN56 treatment was observed ( Figure S9D, Supporting Information). We further treated tumor-bearing mice with PD-L1, FIN56, a combination of PD-L1 and FIN56, or an IgG isotype control. The PD-L1 treatment did not show a significant suppressive effect on syngeneic MOCL2 tumors, and modest tumor suppression was observed in the FIN56 group. The combination of PD-L1 and FIN56 revealed the highest tumor suppression among the four groups ( Figure 6E,F). In summary, the results indicate that ferroptosis inducers inhibit HN-SCC growth, influence the infiltrated immune cells and demonstrate a combinatory effect with the PD-L1 treatment.

The Ferroptosis Signature Correlates with an Immune-Active State of HNSCC
We investigated the relevance of the ferroptosis signature to the immune environment of HNSCC. We first applied the TCGA-HNSC data set to investigate the correlation between ferroptosis signature/GPX4 and T cell infiltration/activation. A higher expression of CD4, CD8A, CD3D, CD28, GZMB, IFNG, PRF1, and IL2 was observed in the ferroptosis_high patient group (Figure S10A, Supporting Information). A positive correlation was shown between the ferroptosis signature and the CD8 + cytotoxic T cell signature ( Figure S10B, Supporting Information). Further analysis by TIMER 2.0 demonstrated that GPX4 expression was negatively correlated with the level of CD8 T cell infiltration in most analyzes for HPV-negative patients, while a positive association between GPX4 and T cell infiltration was noted in most analyses for HPV-positive patients ( Figure S10C, Supporting Information). The correlation between individual genes showed a distinct correlation pattern of correlation in the HPV negative and positive subgroups. The T-cell infiltration score was significantly higher in the HPV positive group compared to the HPV-negative patients ( Figure S10D, Supporting Information). In HPV-negative HNSCC, there was a trend that GPX4 negatively correlated with CD8A expression and significantly correlated with the T cell activation marker IFNG or GZMB negatively. On the contrary, the HPV positive patient group exhibited a trend of positive correlation between GPX and T cell activation markers, and GZMB reached statistical significance ( Figure  S10E, Supporting Information).
Next, we applied multispectral immunofluorescent imaging to investigate the correlation between ferroptotic stress and immune microenvironment in the HNSCC sample. The lipid peroxidation marker 4-hydroxynonenal (4-HNE) was applied to indicate the ferroptotic stress in HNSCC samples. We conducted two sets of samples for investigation of T cells (set#1) (Figure 7A-C) and myeloid linage cells (set#2) (Figure 7D-F). The workflow is illustrated in Figure 7A,D, and the representative multispectral immunofluorescent images are shown in Figure 7B,E. A significant higher density of CD8 + cell and CD8 + PD1 + cells was demonstrated in ROIs with higher 4-HNE in tumor cells (Figure 7C). A lower density of DC and macrophages was shown in ROIs with higher 4-HNE in tumor cells ( Figure 7F). Among the myeloid immune cell types we screened, macrophages were the dominant PD-L1 expressing cell ( Figure S11, Supporting Information). We also examined the PD-L1 expression in tumor and immune cells of the ROIs with high versus low 4-HNE signals. An increased PD-L1 expression was noted in tumor cells of ROIs with high 4-HNE ( Figure 7G). In contrast, a reduced percentage of PD-L1 + DC and macrophages was shown in ROIs with high 4-HNE ( Figure 7H).
We summarize the main findings of this study in Figure 8. In HNSCC, ferroptotic stress generates lipid ROS to activates NF-B, which subsequently activates CD274 transcription. Furthermore, ferroptosis-induced calcium signaling also contributes to the up-regulation of PD-L1. HNSCC cells with ferroptosis signature reveal an inflamed phenotype together with PD-L1 expression, indicating the potential of priming HNSCC with ferroptosis inducers to improve the efficacy of PD-L1.

Discussion
The antitumor effect of ferroptosis has been extensively investigated, especially in drug-resistant mesenchymal-like cancer stem cells. [4][5][6][7] Here, we validated the antitumor activities of ferroptosis inducers in vivo. In addition to the cytotoxic activities of ferroptosis inducers, ferroptosis has been found to induce immunogenic cell death with the release of immunoregulating molecules such as HMGB1 [34] and ATP. [35] Ferroptotic cancer cells modulate tumor microenvironments to an immunoactive state. [36] In this study, we observed that ferroptotic stress modulates tumorinfiltrated immunocytes in murine model and HNSCC samples. On the basis of the evidence presented above, we suggest that ferroptosis inducers suppress tumor growth through a direct tumor suppressing effect and microenvironmental modulation.
Several strategies have been reported to prime the tumor to facilitate the effectiveness of ICIs, including chemotherapy, [37,38] toll-like receptor agonist, [39] STING1 agonist, [40] and radiotherapy. [41] The main finding of our study is that ferroptotic stress in HNSCC cells upregulates PD-L1, which provides the potential to prime HNSCC to potentiate the efficacy of ICIs. The combinatory effect of ferroptosis inducers and anti-PD-L1 antibody was validated in the syngeneic murine model of this study. Intriguingly, a recent study demonstrated that immunotherapy-activated CD8 + T cells enhance ferroptosis in tumor cells, which contributes to the anti-tumor efficacy of immunotherapy. [42] This finding together with our result strengthen the potential synergy of ferroptosis inducer and immunotherapy. Therefore, the development of optimal-dosed ferroptosis inducer as a priming strategy for augmenting antitumor immune response is a promising strategy for cancer immunotherapy. An interesting question is whether ferroptotic stress in tumor cells affects PD-L1 expression in immunocytes.

www.advancedsciencenews.com www.advancedscience.com
Recent studies have shown that ferroptosis affects macrophage infiltration and polarization in the murine liver cancer model. [43] However, we did not observe that ferroptotic stress in HNSCC cells influences PD-L1 expression in immunocytes in our study. The possible explanation is that our study observed ferroptotic stress in tumor cells rather than directly in immune cells, and the distinct signaling pathways between tumor cells and immune cells should also be considered. Further investigation of ferroptotic stress in infiltrated immune cells of different tumors is mandatory.
In this study, we found that both ROS-driven NF-B activation and calcium influx contribute to ferroptosis-induced PD-L1 expression. ROS has been well known to activate NF-B signaling, [44] and PD-L1 is one of the major immune-related molecules upregulated by NF-B pathway. [45] Besides, the increased intracellular calcium signal is a common response to stress. [46] Here we showed that prolonged ferroptotic stress induced calcium influx after hours, which is consistent with the previous finding that ferroptosis caused a sustained increase in cytosolic calcium that is related to plasma membrane dam-age rather than ion channel dependent. [30] There are other sources of calcium ion storage in cells such as the ER [47] and mitochondria [48] which may contribute to the increase in cytosolic calcium. Our result indicates that extracellular calcium is the main source of ferroptosis-induced calcium influx, and the role of intracellular calcium storage organelles in the ferroptosisinduced calcium signal appears to be minor and remains not yet fully determined.
Another interesting finding of the study is that the ferroptosisinduced inflammation signature or PD-L1 expression is more prominent in the HPV-negative HNSCC group than in the HPVpositive group. HPV-positive HNSCC is generally considered an inflamed tumor with an immune-activated microenvironment compared to HPV-negative patients. [49] Our study implies that in patients with HPV negative HNSCC, analysis of the ferroptosis signature may be helpful to categorize patients as immune active and immune inactive, which will be informative in selecting the appropriate cases for ICI therapy. It may also indicate the potential of the application of ferroptosis-inducing agents in HPV-negative HNSCC to create an immune-active environment.  www.advancedsciencenews.com www.advancedscience.com Animal Experiments: The animal experiment was approved by the Institutional Animal Care and Utilization Committee of Taipei Veterans General Hospital (IACUC certificate No. 2019-040). To evaluate the antitumor effect of the ferroptosis inducer, 1 × 10 6 HSC3 cells or primary HNSCC cells were subcutaneously injected into the flanks of nude mice. Tumorbearing mice received two intratumor injections of 100 mg kg −1 RSL3 on the 5th and 7th days after tumor cells. The mice were sacrificed on the 28th day post-tumor-cell injection. For patient-derived xenografts (PDX), the HNSCC specimens were first rinsed twice and immersed in Matrigel (Becton-Dickinson) at 37°C. The tumors were cut into 1 mm 3 pieces and subcutaneously implanted in 4 weeks old female nude mice to establish PDX. For investigation of the ferroptotic drug-induced immune cell infiltration in the syngeneic HNSCC mouse model, 1 × 10 6 MTCQ1-2 cells were subcutaneously injected into the flanks of C57BL/6 mice. The tumorbearing mice received two intratumor injections of 100 mg kg −1 FIN56 on the 15th and 17th days after inoculation of tumor cells. The mice were sacrificed on the 20th day post tumor cell injection and tumor sample were collected for IHC and immunophenotyping of infiltrated immune cells. For investigation of the anti-tumor effect of the ferroptotic inducer combined with an anti-PD-L1 antibody in the syngeneic HNSCC mouse model, 1 × 10 7 MOCL2-1 cells were subcutaneously injected into the flanks of C57BL/6 mice. The tumor-bearing mice received two intratumor injections of 100 mg kg −1 FIN56 on the 12th and 14th days after tumor cell injection. On the 14th day, 100 μg anti-PD-L1 antibody (Bio X Cell) was administered intraperitoneally to each mouse. Antibodies were administered every 3 days until the 32nd day. The mice were sacrificed on the 38th day post-tumor-cell injection. The volume of the tumors was measured regularly and the weight of the tumors was measured after they were harvested.
Statistical Analysis: Statistical analyzes were performed using Graph-Pad Prism 8 (GraphPad Software). The two-sided independent Student's ttest was used to compare continuous variables between two groups. The Pearson correlation test was used to analyze the correlation between two continuous factors. All statistical data were derived from at least three independent biological replicates, and each experiment contained at least two technical replicates. p ≤ 0.05 was considered statistically significant (*: ≤0.05, **: ≤0.01, ***: ≤0.001).

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.