Tumor-Initiating Stem Cells Fine-tune the Plasticity of Neutrophils to Sculpt a Protective Niche

Summary

The heterogeneous nature of tumor-associated neutrophils (TANs) has been recognized, but how different cell states of TANs emerge, evolve, distribute, and impact cancer immunotherapy efficacy remain elusive. Using single-cell RNA sequencing, spatial transcriptomics, and genetic manipulations, we show that anti-PDL1 + CD40 agonist immunotherapy can induce interferon responses in TANs, allowing them to regain anti-tumor activities in squamous cell carcinomas (SCC). In contrast, TANs residing at the tumor-stroma interface can preserve their immune suppressive state. Importantly, we identify a group of SOX2^High tumor-initiating stem cells (tSCs) at the tumor-stroma interface that upregulate fatty acid desaturase 1 (Fads1) to produce arachidonic acid (AA). This tSC-specific pathway enhances the prostaglandin E₂ (PGE₂) signaling in TANs, which can disrupt the interferon response and prevent the interferon-induced anti-tumor functions in TANs. By fine-tuning the plasticity of neutrophils, tSCs shape neutrophil heterogeneity and sculpt a protective micro-niche to survive from immunotherapy and drive cancer relapse.

eTOC blurb

Guo et al. show that anti-PD-L1 + CD40 agonist therapy activates anti-tumor functions in neutrophils within the tumor stroma. Conversely, SOX2^High tumor-initiating stem cells at the tumor–stroma interface suppress interferon responses and maintain immunosuppressive functions of neutrophils via the arachidonic acid–prostaglandin E2 axis, forming a protective niche.

Graphical Abstract

Introduction

Most solid tumors are complex ecosystems in which cancer cells interact with various cell types within the tumor microenvironment (TME) to drive tumorigenesis, metastasis, and immune evasion. Given the important roles of these communications, targeting these dialogues represents a promising strategy to control malignancy. In particular, disrupting immune suppressive interactions through various immunotherapies holds great potential, as evidenced by the initial success of immune checkpoint blockade (ICB) therapy in many cancers ^1,2. However, due to a lack of comprehensive understanding of these interactions between cancer cells and the TME, the effects of most immunotherapy treatment are not sustained, and many patients experience cancer relapse despite their initial responses ^3,4. A puzzling question is how the dynamic spatial-temporal rearrangement of the TME induced by immunotherapy treatment is overcome by cancer cells to give rise to relapsed tumors.

The heterogeneous composition and hierarchical organization of tumor populations add additional layers of complexity to the intertwined relationship between cancer cells and the TME ⁵. It is increasingly clear that many solid tumors, such as squamous cell carcinomas (SCCs) originating in various tissues, are initiated and maintained by a group of tumor-initiating stem cells (tSCs) with strong stemness signatures ⁶. These tSCs are known to activate special molecular programs to drive tumorigenesis and resistance to various therapies ^7,8. Importantly, recent studies have demonstrated that tSCs can also survive robust anti-tumor immune responses evoked during immunotherapy which can otherwise clear differentiated cancer cells ^9–11. To achieve this, immune modulatory molecules such as CD80 ¹⁰ and CD276 ¹² are specifically enriched in tSCs. These surface proteins provide tSCs with additional immune suppressive ligands to directly dampen the anti-tumor activity of cytotoxic T cells. Despite these advances, the field has only begun to uncover the unique immune resistance mechanisms that protect tSCs.

Accumulating research investigating how normal tissue stem cells adapt to inflammation sheds key insights into understanding tSC-specific immune resistance. As long-lived cells, frequent exposure to inflammation could impact the fitness of tissue stem cells ^13,14, debilitating their regenerative potential. Thus, stem cells residing in various barrier tissues are believed to be protected by an “immune privileged” niche, which is composed of a myriad of immune suppressive cells ^15,16. It is possible that tSCs are shielded within a similar niche in the TME ¹⁷. However, our understanding of the composition, organization, and function of the special niche protecting tSCs is still unfolding.

Among various cell types in the TME that can interact with tSCs, neutrophils are the less understood components ^18,19. Although the abundant accumulation of neutrophils in the TME has been well recognized, the functions of tumor-associated neutrophils (TANs) remain elusive ^20,21. TANs have long been believed to be immune suppressive and have thus been referred to as “polymorphonuclear myeloid-derived suppressor cells” (PMN-MDSCs) ^22,23. Single-cell analysis has uncovered multiple cell states of TANs, especially a subset of dcTRAIL-R1⁺CD14^high neutrophils, which undergo deterministic programming to acquire pro-tumor functions ²⁴. However, recent studies also identify various anti-tumor functions of TANs ^25,26,27, and robust antigen presentation potential can be activated in a subset of TANs ^28,29. Importantly, effective tumor control induced by immunotherapy is associated with strong neutrophil signatures^30,31. These seemingly contradictory findings highlight the unexpected degree of plasticity and heterogeneity of TANs²⁰. Here, we investigate how effective immunotherapy influences TAN plasticity and how tSCs evade TAN-mediated anti-tumor immunity, ultimately contributing to tumor relapse following immunotherapy.

Results

Immunotherapies remodel the neutrophil landscape in SCCs.

In this study, we aim to define how immunotherapy impact TANs in SCCs, which harbor high mutational burdens and initially respond to immune checkpoint blockade ^32–34. Specifically, we induced skin SCCs by exposing the mouse skin epithelium to chemical mutagen 7,12-dimethylbenz[a]anthracene (DMBA) followed by repeated applications of phorbol ester 12-O-tetradecanoylphorbol 13-acetate (TPA)³⁵ (Figure 1A). Compared to previous studies which mostly profiled neutrophils infiltrating transplanted mouse tumors, the DMBA + TPA treatment induce autochthonous tumors whose developmental trajectory, heterogeneous composition, and complex ecosystem can better mimic human epithelial cancers. With this model, we next treated the tumor-bearing mice with PDL1 blocking and CD40 agonist antibodies, which were recently demonstrated to induce anti-tumor activities in neutrophils ^31,36 (Figure 1A). We first confirmed that the combination of anti-PDL1 + CD40 agonist can induce effective tumor clearance (Figure 1A). Next, immediately after the treatment, we isolated total CD45⁺ immune cells from treated and untreated tumors, then subjected these cells to the split-pool combinatorial barcoding-based single cell RNA-sequencing (scRNA-seq) ³⁷ (Figure 1B).

Figure 1. Immunotherapy induces distinct responses in different neutrophil subpopulations.

A. Schematic of experimental procedures and growth curve of spontaneous skin SCCs after anti-PDL1 + CD40 agonist immunotherapy treatment. n = 5 in each group.

B. UMAP showing the immune cell types identified by scRNA-seq in skin SCCs.

C. UMAP showing the TAN subpopulation clustering in skin SCCs.

D. UMAP showing the signature and immune suppressive genes expressed in various TAN clusters.

E. UMAP and stacked bar chart showing the changes in the composition of TAN subpopulations induced by the anti-PDL1 + CD40 agonist treatment.

F. Chord diagram showing the upregulated pathways induced in each cluster of TANs after the anti-PDL1 + CD40 agonist treatment.

G. Bubble heatmap showing transcripts of various immune stimulatory genes induced in cluster 1 – 3 (T1 to T3), and immune suppressive genes induced in cluster 4 (T4) after anti-PDL1 + CD40 agonist treatment.

Two-way ANOVA and Sidak’s multiple comparisons tests were used in (A), and results in (A) are presented as mean ± SEM. **p < 0.01.

See also Figure S1 and Table S1

Among these 40,500 immune cells sequenced, using markers, such as Mmp9, S100a8 or Csf3r, a total of 14,006 cells were identified as neutrophils (Figure S1A). Consistent with previously published scRNA-seq results ^24,38, we have identified similar subsets of TANs in spontaneous SCCs (Figure 1C). For example, many cells in cluster 1 (T1) expressed Tnfrsf23 (gene encoding dcTRAIL-R1) ²⁴, Siglecf ³¹, as well as high level of Cd14 ³⁸ and Cd101 ²⁴ (Figure 1D), suggesting that this subset mainly contained mature neutrophils that were thought to undergo deterministic programming to acquire pro-tumor and immune suppressive functions ^24,38. In contrast, Tnfrsf23 and Cd101 could only be detected in a few cells in cluster 4 (T4) (Figure 1D), suggesting cells in this cluster were mostly immature neutrophils. We also identified additional transition states (T2 and T3) which still expressed markers that could identify TANs with pro-tumor functions, but fewer cells did so when compared to T1 cluster (Figure 1D). Each subset of TANs expressed specific signature genes (Figure S1B; Table S1), and the cells in the transitional state shared certain markers with both T1 and T4, such as Csf1 and Lpl (Figure S1B; Table S1). Further supporting the developmental trajectory of these clusters, RNA velocity analysis confirmed that T1 cluster was the major mature subset, whereas the other three clusters developed towards T1 (Figure S1C). We then examined additional functional features of each TAN cluster. Interestingly, in addition to sharing common immune suppressive mechanisms, such as expression of Ptgs2 ³⁹ (Figure 1D), the neutrophils in each cluster activate distinct strategies to further dampen anti-tumor immunity. For example, neutrophils in the T1 and T2 cluster express Cd274 and Cd300ld ⁴⁰ (Figure 1D), while neutrophils in the T4 cluster expressed high levels of Arg1, which could deplete arginine in the TME to block T cell functions ⁴¹ and Mmp12, which could inactivate numerous chemokines involved in leukocyte migration (Figure 1D).

Next, we sought to investigate how immunotherapy affects each TAN subpopulation. The expansion of IFNγ and GZMB-producing T cells suggested that anti-PDL1 + CD40 agonist treatment could effectively alter the immune landscape of SCCs (Figure S1D). Interestingly, when we compared the representation of each TAN subpopulation before and after the treatment, we found that T1 and T2 populations expanded proportionally in treated tumors, whereas the T4 cluster maintained its representation after the treatment (Figure 1E). We then performed pseudo-bulk analysis and profiled the gene expression changes within these persisting populations before and after immunotherapy. Surprisingly, after the immunotherapy, the Tnfrsf23⁺ and Cd14⁺ pro-tumor neutrophils (in both T1 and T2) ^24,38 began to upregulate pathways that can boost anti-tumor immunity, such as genes involved in interferon responses and T cell regulation (Figure 1F). However, these changes were not observed in the T4 cluster (Figure 1F). Instead, the TANs in the T4 cluster upregulated pathways involved in responding to reactive oxygen stresses and regulating tissue regeneration (Figure 1F). Looking into specific genes, it was reported that anti-tumor neutrophils express high levels of Ly6e ³⁰ and Sell ³¹. Consistent with these findings, the neutrophils in T1 and T2 down-regulated Tnfrsf23 and immune suppressive ligands (e.g. Cd300ld) (Figure S1E), but upregulated Ly6e and Sell, and several other genes involved in anti-tumor immunity, such as Il12a, Tnf, and Gbp3 (Figure 1G). In contrast, the neutrophils in T4 seemed to be resistant to these reprogramming effects from immunotherapy (Figure 1G). Instead, T4 TANs upregulated metabolic immune checkpoint molecules (e.g. Arg1 ⁴¹ and Il4i1 ⁴²), Treg attracting chemokines (e.g. Ccl22 ⁴³), and growth factors (e.g. Fgf14⁴⁴) (Figure 1G).

Immunotherapy-stimulated interferons induce a subset of TANs to gain anti-tumor functions.

Given that the anti-PDL1 + CD40 agonist can elicit distinct responses in different subsets of TANs, we sought to understand how different TAN subpopulations impacted the outcomes of immunotherapy. Interestingly, when we re-analyzed the published scRNA-seq data profiling TANs in grafted tumors and then projected their signatures on our dataset, we found that our spontaneous SCC tumors harbor more heterogeneous neutrophil cell states (Figures S2A-S2C). In particular, the grafted tumors, despite the distinct tissue of origin, appear to lack the Arg1^high T4 TANs (Figures S2A, S2D and S2E). The grafted tumors also induce more uniform and synchronized interferon-responsive or anti-tumor phenotypes in neutrophils (Figures S2B and S2C) ³¹. Thus, we decided to switch the model from spontaneous SCCs to grafted SCC models since it allows us to rigorously investigate the functional significance of different neutrophil cell states during immunotherapy treatment.

To first confirm the functional conversion can be induced in TANs by immunotherapy, we generated S100A8-Cre; R26-LSL-DTR (Neu^DTR) mice and grafted the immunogenic SCC cells (PDVC57 line) on either Cre⁺DTR⁺ mice or Cre⁻ littermate controls. With these mice, we can deplete neutrophils in tumor-bearing mice at different time points when injecting diphtheria toxin (DT). Consistent with the known roles of TANs in suppressing anti-tumor immunity during tumorigenesis ^23,39, when we depleted neutrophils before immunotherapy (Figure S3A), the activities of tumor infiltrating T cells were enhanced (Figure 2A), without affecting the frequency of other myeloid cell compartments (Figure S3B). However, when DT was injected during anti-PDL1 + CD40 agonist treatment, neutrophil depletion significantly blunted anti-tumor activities in T cells (Figure 2B). As a result, the tumor control induced by immunotherapy was significantly blunted (Figure 2C). To further confirm that the reduction in immunotherapy efficacy is specifically due to neutrophil depletion, we also employed LY6G antibody to deplete the neutrophils and observed similar impacts on the efficacy of anti-PDL1 + CD40 agonist (Figure S3C). These results suggested that, while most TANs were PMN-MDSC-like and were immune suppressive during tumorigenesis, many TANs became immune-stimulatory during immunotherapy treatment. Further bolstering this conclusion, we found that, after immunotherapy treatment, many TANs upregulated LY6E and activated class II MHC, but downregulated Siglec-F (Figure 2D). All these molecular changes have been shown to closely associate with anti-tumor functions of TANs ^28,30,31.

Figure 2. The interferon responses triggered by immunotherapy can restore the anti-tumor functions in TANs.

A. Experimental scheme and flow cytometry quantification of T cell responses when TANs were depleted before immunotherapy treatment. n = 5 for Cre⁻ and n = 6 for Cre⁺ group.

B. Experimental scheme and flow cytometry quantification of T cell responses when TANs were depleted during the anti-PDL1 + CD40 agonist treatment. n = 15 for Cre⁻ and n = 12 for Cre⁺ group

C. Growth curve of SCC tumors with or without TAN depletion during immunotherapy treatment. n = 18 for Cre⁺ group and n = 10 for Cre⁻ group.

D. Representative flow cytometry histogram showing the upregulation of markers associated with anti-tumor activity (e.g. LY6E or MHCII), but downregulation of immune-suppressive markers (e.g. Siglec-F) in TANs after anti-PDL1 + CD40 agonist treatment. Note that these changes cannot be detected in neutrophils in other tissues (bone marrow or spleen) from the same tumor-bearing mice.

E. Experimental scheme, flow cytometry quantification of LY6E and MHCII expression (left) and qPCR (right) quantification of interferon stimulated gene (Gbp2 and Ifit3) expression in lineage traced pre-existing TANs (Tomato⁺) before or after the immunotherapy treatment. Gbp2 and Ifit3 expression levels were first normalized to 18S rRNA and then further normalized to the average value of the naive group. n = 4 for naive and n = 6 for treated group in flow cytometry, n = 3 for each group in qPCR.

F to H. Flow cytometry quantification of (F) MHCII expression in neutrophils or (G) IFNγ production in CD4⁺ T cells, and (H) the growth of SCC tumors before and after immunotherapy treatment when both IFNα and IFNγ receptors are deleted specifically in neutrophils (Ifnar1/Ifngr1 dKO). n = 20 control group, n = 10 for the dKO group.

Graphs from A to H show representative results from one of the three repeats for each experiment and was presented as mean ± SEM. Student’s t tests (A, B and E), Mann-Whitney U Tests (D), and two-way ANOVA and Sidak’s multiple comparisons tests (C, F, G and H) were used for statistical comparison. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, non-significant.

See also Figures S2 and S3

The shift in neutrophil functions during immunotherapy could be due to biogenesis of new anti-tumor neutrophils which are recruited to replace the original suppressive TANs in the TME. However, analyzing the phenotypes of neutrophils in the bone marrow or spleen from the same tumor-bearing mice showed that the anti-tumor phenotype of TANs was not observed outside the TME (Figure 2D). To further test whether immunotherapy can convert the functional states of TANs, we designed a lineage tracing assay to distinguish the original TANs that were present in the TME before treatment from any newly generated neutrophils after immunotherapy (Figure 2E). Driven by this goal, we grafted PDVC57 cells on Ly6G-CreER; R26-LSL-tdTomato mice which could genetically label the neutrophils with Tomato upon receiving tamoxifen ⁴⁵ (Figures 2E and S3D). Following genetic labeling of pre-existing TANs, we treated these mice with anti-PDL1 + CD40 agonist and then analyzed Tomato⁺ neutrophils. Both flow cytometry and qPCR analysis of sorted Tomato⁺ TANs confirmed that these Tomato⁺ neutrophils effectively acquired markers associated with anti-tumor functions (e.g. LY6E or MHCII) and activated interferon-stimulated genes (e.g. Gbp2 or Ifit3) (Figure 2E). This finding confirmed our speculation that the TANs that were already present within the TME could directly regain anti-tumor functions. To further test this hypothesis, we isolated the CD14^High, CD14^Mid, and CD14^Low TANs from naive tumors (Figure S3E). We then treated these different subsets of TANs with combined IFNα and IFNγ in culture. Interestingly, we found that the transcripts of MHCII gene (e.g, H2-Aa) and Cd74, or LY6E proteins could be most efficiently induced in the CD14^High cells (Figures S3E and S3F). In addition, we generated S100A8-Cre; Ifnar1^flox/flox; Ifngr1^flox/flox double knockout (Ifnar1/Ifngr1 dKO) mice to specifically block interferon responses in neutrophils. We found that ablation of interferon signaling specifically in neutrophils blocked the phenotype switch of neutrophils (Figures 2F and S3G) and reduced the anti-tumor activities in T cells (Figures 2G and S3H). As a result, the immunotherapy-induced tumor clearance was abolished (Figure 2H).

Distinct spatial distributions of neutrophil responses during immunotherapy treatment

The distinct functional states of different TAN clusters prompted us to hypothesize that these neutrophil subpopulations might be positioned in separate areas of the TME, and their location could dictate how they respond to immunotherapy. To test this idea, we employed GeoMx spatial transcriptomics analysis to determine the distribution patterns of neutrophil responses in the spontaneous mouse SCCs. Briefly, we induced autochthonous skin SCCs on the neutrophil reporter ‘Catchup^IVM-red’ mice⁴⁶. Next, we subjected naive or treated tumors to the GeoMx spatial transcriptome profiling (Figure 3A). When using the relative location of cancer cells (K14⁺), blood vessels (CD31) or the cancer associated fibroblasts (CD140a) to distinguish different tumor regions ^47,48 (Figures S4A and S4B), we identified that most Tomato⁺ TANs were located either within the stroma (stromal TANs) or at the tumor-stroma interface (interface TANs) (Figure 3A). Thus, we selected Regions of Interest (ROI) within these two areas, sequenced the Tomato⁺ neutrophils from selected ROIs, and mapped their gene expression pattern to their locations. Interestingly, we found that, similar to the T1 and T2 clusters, the neutrophils in the stroma exhibited global upregulation of interferon responses and other anti-tumor immunity-related pathways (Figures 3B and 3C). In contrast, neutrophils located at the tumor-stroma interface showed downregulated interferon responses compared to those found in the stroma (Figures 3B and 3C).

Figure 3. Spatial distribution of distinct neutrophil responses during immunotherapy treatment.

A. Experimental scheme and representative IF images of skin SCCs collected for selecting different ROIs either within the tumor (Tu) or in the stroma (St) for spatial transcriptomic analysis.

B. Heatmap showing scaled expression of anti-tumor immunity-related genes in neutrophils located at the tumor-stroma interface or in stroma before and after anti-PDL1 + CD40 agonist treatment.

C. Chord diagram showing the down-regulated pathways in neutrophils located at the tumor-stroma interface compared to the neutrophils in stroma during immunotherapy treatment.

D and E. Representative IF images and quantifications of (D) Gbp2⁺ or (E) MHCII⁺ neutrophils among all the MPO⁺ cells located at the tumor-stroma interface (arrow) or in stroma (arrowhead) after immunotherapy treatment. The tumor-stroma interface was labelled with white dotted line. n = 30 in D, n = 36 in E.

Scale bars: 50 μm. Mann-Whitney U Tests were used for statistical comparison in D and E, and data are presented as mean ± SEM. *p < 0.05, **p < 0.01.

See also Figure S4

Since the data quality from GeoMx was suboptimal, we next validated the pattern of interferon responses in TANs revealed by spatial transcriptome profiling. First, we used RNAscope to image the interferon response in TANs from different regions. This confirmed that interferon-stimulated genes (e.g. Gbp2) were strongly induced in TANs expanding in the stroma after immunotherapy, whereas most neutrophils residing at the tumor-stroma interface were less responsive (Figure 3D). We further confirmed this spatial pattern of anti-tumor responses in TANs through immunofluorescent staining for MHCII (Figure 3E). Collectively, these results suggest that the tumor-stroma interface represents a unique micro-niche that can actively block a subset of neutrophils from responding to immunotherapy-induced reprogramming, so these neutrophils may retain their immune suppressive activities.

Sox2 amplification endows cancer cells with the capacity to block TANs from responding to interferons

Next, we sought to uncover the mechanisms specifically activated at the tumor-stroma interface that were responsible for blocking the TANs from responding to immunotherapies. The tumor-stroma interface in SCCs is an interesting location, as it is where a group of TGFβ-responsive tSCs accumulate ^47,48. These tSCs can be specifically recognized and distinguished from differentiated SCC cells by their expression of ITGA6 ^47–49 and immune modulatory molecule CD80 ¹⁰. Importantly, these cells were shown to be able to survive immunotherapy and give rise to relapsed tumors¹⁰. Thus, we speculate that tSCs might be equipped with molecular programs to modulate neutrophils. To identify the tSC-specific pathways, we isolated ITGA6⁺ CD80⁺ tSC (Figure S4C) and ITGA6⁻ non-tSCs from DMBA + TPA induced skin SCCs for bulk RNA-seq. We then compared the transcriptome of different tumor populations. Through this analysis, we identified 1,170 genes that were up-regulated by at least two-fold in tSCs compared to non-tSCs (Table S2). These genes include Hmga2, Wnt7b, and Trp73, which were previously shown to mediate stem cell functions. Gene Set Enrichment Analysis (GSEA) confirmed pathways, such as stemness, microenvironment regulation, and quiescence were specifically enriched, whereas many immune stimulatory pathways were downregulated in tSCs (Figure S4D; Table S3). Given that most stem cell-specific functions are activated by transcription factors (TFs), we specifically focused on the TFs that were mostly enriched in tSCs (Figure 4A; Table S4). To this end, we selected top TFs that are mostly enriched in tSCs and individually amplified these TFs in the PDVC57 cells (Figure 4B). Following grafting and immunotherapy, we analyzed the phenotypes of TANs. This approach showed that, only in tumors with high level of SOX2 (SOX2^High), TANs maintained the CD14^Low LY6E^Low phenotypes (Figure 4B) and were unable to activate MHCII antigen presentation (Figure 4C).

Figure 4. Sox2 amplification allows SCC cells to dampen interferon responses in TANs.

A. MA plot showing the up-regulated TFs in sorted ITGA6⁺ CD80⁺ tSCs from skin SCCs compared to differentiated cancer cells. Genes with significant differential expression (adjusted p-value < 0.05) are shown as pink dots.

B. Experimental scheme and representative flow cytometry plots quantifying upregulation of CD14 and LY6E in TANs isolated from tumors formed by SCC cells individually amplifying various tSC-specific TFs.

C. Flow cytometry quantification of the MHCII expression in TANs from SCC tumor cells with or without amplifying Sox2, before and after anti-PDL1 + CD40 agonist treatment. n = 8 in SOX2^Low group and n = 7 in SOX2^High group.

D. Heatmap showing the differentially expressed genes in neutrophils isolated from SCC tumors with (SOX2^High) or without (SOX2^Low) amplifying Sox2, before (Naive) and after (Treated) anti-PDL1 + CD40 agonist treatment.

E. Leading edge plots showing the downregulated pathways in neutrophils isolated from SOX2^High SCC tumors compared to the neutrophils isolated from the SOX2^Low tumors after the anti-PDL1 + CD40 agonist treatment.

F. Quantitative PCR measuring the expression of interferon stimulated genes (Gbp2 and Ifit3) in neutrophils isolated from naive SOX2^Low or SOX2^High tumors following in vitro IFNα + IFNγ treatment for 4 hr. Expression levels of Gbp2 and Ifit3 were first normalized to 18S rRNA and then further normalized to the average value of the SOX2^High group. n = 8 in each group.

G and H. Experimental scheme, representative IF images and quantification of (G) the percentage of MHCII⁺ TANs among all the MPO⁺ cells or (H) the number of CD4⁺ T cell-neutrophil interactions per imaged area in SOX2^High (arrow) or SOX2^Low regions (arrowhead) when SOX2^Low and SOX2^High cells were mixed at 7:3 ratio for grafting. Scale bars: 50 μm. n = 40 in G, n = 30 in H.

I. Experimental scheme and flow cytometry quantification of IL2 production in OTII CD4⁺ T cells after co-culture with OVA-loaded neutrophils isolated from SCC tumors formed by SOX2^Low or SOX2^High cells. n = 6 in each group.

J and K. Representative IF images and quantification of (J) the HLA-DR⁺ TANs among all the MPO⁺ cells or (K) the number of CD4⁺ T cells per imaged area in SOX2^High (arrow) or SOX2^Low regions (arrowhead) in HNSCCs samples collected from immunotherapy-treated patients. The tumor-stroma interface was labelled with white dotted line. Scale bars: 100 μm. n = 30 in J, n = 50 in K.

Graphs in C, F, G, H, I, J and K show representative results from one of the three repeats for each experiment and results are presented as mean ± SEM. Two-way ANOVA and Sidak’s multiple comparisons tests (C), Student’s t tests (F, G, H, I, J and K) were used for statistical comparison. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns, non-significant.

See also Figures S4-S6 and Tables S2-S4.

SOX2 is a TF that plays a key role in embryonic stem cells and becomes reactivated in many epithelial cancers ^50–52. SOX2 is specifically enriched in tSCs and is responsible for initiating and maintaining SCCs ⁵³. Although many tumor promoting functions of SOX2 have been identified, whether SOX2 can activate immune modulatory capacities in tSCs is unclear. To address this question, we grafted either SOX2^Low or SOX2^High SCC cells on the ‘Catchup^IVM-red’ mice, and then isolated Tomato⁺ neutrophils from these tumors for transcriptome analysis. We found that interferon responses were robustly induced in neutrophils in SOX2^Low SCC tumors after immunotherapy, but these pathways were significantly dampened in neutrophils infiltrating SOX2^High tumors (Figures 4D and 4E). It was still possible that these blunted interferon responses resulted from lacking sufficient interferons production in the TME of the SOX2^High tumors. To test this, we isolated TANs from naive SOX2^Low or SOX2^High tumors and then treated them with interferons (IFNα + IFNγ) in vitro. Whereas the neutrophils from SOX2^Low tumors effectively activated interferon stimulated genes, the interferon responsive potential of TANs isolated from SOX2^High SCCs was significantly blunted (Figure 4F). Importantly, when we isolated neutrophils from bone marrow or spleen from the same mice bearing the SOX2^High SCCs, these cells still respond (Figures S5A and S5B). We next mixed SOX2^High with SOX2^Low cells and then grafted the mixed SCC cells. We found that SOX2^High cells could effectively block their surrounding TANs from becoming immune stimulatory (Figure 4G). Correlating to this pattern, significantly fewer interactions between CD4⁺ T cells and neutrophils were observed in the tumor regions formed by SOX2^High cells (Figure 4H). We then designed a co-culture assay to functionally determine the impact of SOX2^High cancer cells on neutrophils. Briefly, we isolated neutrophils from the immunotherapy treated SOX2^Low or SOX2^High tumors. We then loaded these neutrophils with the OVA peptide (323–339) and co-cultured them with naive CD4⁺ T cells isolated from OTII mice (Figure 4I). By measuring the T cell activities, we found that the immunotherapy-induced T cell stimulatory activity of neutrophils is blunted by SOX2^High SCC cells (Figure 4I). Importantly, these effects of SOX2 are conserved across different epithelial cancers. When we amplified Sox2 in head and neck SCC (HNSCC) cells and treated these tumors with the anti-PDL1 + CD40 agonist, the SOX2^High HNSCC cells could also efficiently block neutrophil responses, leading to dampened T cell immunity (Figures S5C-S5H).

We next sought to confirm the human relevance of our findings. To this end, we first analyzed the published scRNA-seq data profiling human tumors collected from immunotherapy-treated patients ⁵⁴. We focused on the cell states of TANs and confirmed the presence of a subset of TANs (Cluster 0) characterized by high expression of ARG1 and low expression of interferon stimulated genes or genes for antigen presentation (Figure S6A). We further analyzed the published spatial transcriptomics data profiling both human HNSCCs⁵⁵ and CSCCs⁵⁶. We found that neutrophils proximal to SOX2^High cells exhibit significantly lower expression of MHCII antigen presentation genes compared to neutrophils located farther away (Figures S6B and S6C). Finally, we obtained both human HNSCC (Figures 4J and 4K) and CSCC (Figure S6D) tissue samples, including HNSCCs from patients who had received and responded to immune checkpoint blockade therapy. Imaging analysis revealed that a smaller number of TANs near SOX2^High cancer cells expressed HLA-DR compared to TANs in the stroma (Figures 4J and S5D). Correspondingly, these SOX2^High tumor regions also showed reduced infiltration of CD4⁺ T cells (Figure 4K).

Collectively, these key findings, from both mouse and human SCCs, suggested that Sox2 amplification in SCC cells can effectively imprint neutrophils within their surroundings, blunting the neutrophils’ capacity to respond to interferons and preventing them from being reprogrammed by immunotherapy treatment.

SOX2^High tSCs shape the cell states of TANs which provide reciprocal protections to tSCs to drive cancer relapse following immunotherapy

Our functional assays using grafted homogeneous SCC cells provided evidence supporting the critical roles of SOX2 in facilitating cancer cells to modulate TANs. Given that, in autochthonous SCCs, Sox2 is mostly amplified in tSCs ^53,57, we next explored how SOX2^High tSCs regulate neutrophils in spontaneous cancer models. To this end, we generated Sox2 cKO mice (K14CreER; Sox2^flox/flox) and used DMBA + TPA to induce autochthonous SCCs (Figure 5A). We first verified that ablating Sox2 in basal epithelium can abolish the cancer relapse after the immunotherapy (Figure 5A). To investigate how the loss of SOX2 in tSCs impact the cell states of TANs, we subjected total CD45⁺ cells isolated from both naive and treated Sox2 cKO tumors to scRNA-seq. Compared to the WT tumors in which a subset of interferon non-responsive TANs (T4 cluster) was preserved after immunotherapy (Figures 1E–1G), the neutrophils infiltrating Sox2 cKO tumors completely lost their heterogeneity (Figure 5B). Importantly, almost all the neutrophils in the Sox2 cKO tumors became interferon responsive with similar signatures of the T1 cluster in treated WT tumors (Figure 5B). Furthermore, compared to the T1 cluster in WT tumors, the interferon responses and immune stimulation signatures were enhanced in TANs when Sox2 was deficient in tSCs (Figure 5C).

Figure 5. SOX2 is critical for tSCs to shape neutrophil cell states, block neutrophil-T cell interactions, exclude T cells and drive cancer relapse.

A. Model schematics and growth curve of spontaneous skin SCCs with or without ablating Sox2 in basal epithelium before the anti-PDL1 + CD40 agonist treatment. n = 12 in control, n = 14 in Sox2 cKO.

B. UMAP and stacked bar chart showing the changes in the composition of neutrophil subpopulations induced by the anti-PDL1 + CD40 agonist treatment when Sox2 is ablated in tSCs.

C. UMAP showing expression level of various immune stimulatory genes expressed in different TAN clusters in WT or Sox2 cKO SCCs with or without anti-PDL1 + CD40 agonist treatment.

D to F. Representative IF images and quantification of (D) the Gbp2⁺ or (E) the MHCII⁺ neutrophils among all the MPO⁺ cells, and (F) the number of interactions between neutrophils and CD4⁺ T cells in the imaged area in WT (arrow) or Sox2 cKO (arrowhead) SCC tumors after anti-PDL1 + CD40 agonist treatment. The tumor boarder was labelled with white dotted line. Scale bars: 50 μm. n = 60 in D, n = 100 in E, n = 90 in F.

G and H. Growth of SCC tumors formed by (G) SOX2^High or SOX2^Low SCC cells with or without anti-PDL1 + CD40 agonist treatment, or by (H) SOX2^High cells with or without neutrophil depletion after anti-PDL1 + CD40 agonist treatment. n = 8 in each group in G and n = 10 in each group in H.

I and J. Representative IF images and quantification of (I) the number of CD4⁺ T cells in the imaged tSC-enriched regions or (J) the percentage of dead (active Caspase3⁺) SOX2⁺ tSCs in spontaneous SCCs with (arrow) or without (arrowhead) neutrophil depletion after anti-PDL1 + CD40 agonist treatment. The tumor boarder was labelled with white dotted line. Scale bars: 100 μm. n = 60 in I, n = 100 in J.

Representative results in A, D to F, I and J are from one of the three repeats, and presented as mean ± SEM. Mann-Whitney U Tests (D to F, I and J) and Two-way ANOVA and Sidak’s multiple comparisons tests (A, G and H) were used for statistical comparison. *p < 0.05; **p < 0.01; ****p < 0.0001; ns, non-significant.

See also Figure S7.

We next investigated the functional consequence of this stem cell-neutrophil crosstalk during immunotherapy treatment. We found that many Gbp2⁺ neutrophils infiltrated the Sox2-deficient tumors after immunotherapy (Figure 5D). These neutrophils also activated MHCII, indicating their potential to prime CD4⁺ T cells (Figure 5E). As a result, we detected dynamic interactions between neutrophils and T cells in both the stroma, and within the tumor mass (Figure 5F). Importantly, whereas most CD4⁺ T cells were excluded from infiltrating the tSC-enriched regions in WT tumors, we can now detect CD4⁺ T cell infiltration into the Sox2 cKO tumors (Figure 5F). Single-cell analysis also revealed no significant increases in the proportions of CD4⁺ or CD8⁺ T cells recruited into the treated Sox2 cKO tumors (Figures S7A and S7B). These findings suggest that the observed increase in neutrophil-CD4⁺ T cell interactions within the tumor mass is not due to greater T cell recruitment. Instead, it likely results from enhanced engagement between T cells and neutrophils.

Based on these results, we then examined the growth of SOX2^Low or SOX2^High tumors after immunotherapy treatment. As we anticipated, the SOX2^Low and SOX2^High SCC tumors grew similarly in mice treated with control antibodies (Figure 5G), as most TANs in untreated tumors are immunosuppressive. However, while the growth of both groups of tumor could be initially reduced by the anti-PDL1 + CD40 agonist treatment, only SOX2^High tumors quickly relapsed (Figure 5G). Next, we explored whether the enhanced ability of SOX2^High SCC cells to survive and relapse following immunotherapy depends on protection provided by TANs. Depletion of neutrophils markedly reduced the tumor relapse capacity of SOX2^High SCC cells (Figure 5H). To further confirm that TANs specifically protect SOX2^High SCC cells, we performed neutrophil depletion in SCC tumors formed by mixing SOX2^Low and SOX2^High SCC cells. This resulted in increased migration of CD4⁺ T cells into SOX2^High tumor regions (Figure S7C) and elevated death of SOX2^High cells (Figure S7D). To further substantiate the role of this crosstalk in a more relevant model, we depleted neutrophils in DMBA + TPA-induced spontaneous SCCs. Consistent with what we have found so far, neutrophil depletion enhanced CD4⁺ T cell infiltration into tumor regions enriched for SOX2^High tSCs (Figure 5I) and increased death of SOX2^High tSCs (Figure 5J).

SOX2 activates FADS1 in tSCs to block the interferon responses in TANs.

Built on these findings, we speculated that Sox2 amplification must induce the secretion of certain factors from tSCs that can disrupt the interferon responsiveness in neutrophils. To identify such factors, we isolated Sox2^−/− and Sox2 OE SCC cells, and subjected these cells to bulk RNA-seq. We were particularly intrigued by the expression of Fads1 in SOX2^High cells (Figure 6A). FADS1 is a Δ5 desaturase that can catalyze linoleic and linolenic acids metabolism to produce polyunsaturated fatty acids (PUFA), such as arachidonic acid (AA) ⁵⁸, which is a highly bioactive molecule that has profound impacts on neutrophils ^59,60. We confirmed the direct regulation of Fads1 gene expression by SOX2 using CUT & RUN-sequencing which identified the direct binding of SOX2 at the promoter region of Fads1 gene locus (Figure 6B).

Figure 6. SOX2 activates FADS1 to block the interferon-induced anti-tumor functions in TANs.

A. Heatmap showing the differentially expressed genes in Sox2 OE SCC cells compared to Sox2 KO cells.

B. IGV image showing the SOX2 or control antibody CUT & RUN.

C to E. Flow cytometry quantification of the (C) LY6E, (D) MHCII and (E) Siglec-F expression in TANs in immunotherapy-treated SCC tumors formed by SOX2^High cells with or without silencing Fads1. n = 5 in each group.

F. Quantitative PCR measuring Ifit3 expression in TANs isolated from naive SCC tumors formed by SOX2^High cells with or without silencing Fads1 following treatment with IFNα + IFNγ in vitro for 4 hr. Expression levels of Ifit3 were first normalized to 18S rRNA and then further normalized to the average value of the group without Fads1 silencing. n = 9 in each group

G to I. Flow cytometry quantification of the (G) LY6E, (H) Siglec-F, and (I) MHCII expression in TANs in immunotherapy-treated SCC tumors formed by SCC cells with or without amplifying Fads1. n = 6 in each group.

J. Quantitative PCR measuring Gbp2 expression in TANs isolated from naive SCC tumors with or without amplifying Fads1 following treatment with IFNα + IFNγ in vitro for 4 hr. Expression levels of Gbp2 were first normalized to 18S rRNA and then further normalized to the average value of the vector control group. n = 12 in each group.

K to N. Flow cytometry quantification of CD4⁺ T cell infiltration (K) and IFNγ production (L), as well as CD8⁺ T cell infiltration (M) and IFNγ production (N) in immunotherapy-treated SCC tumor cells formed by SCC cells with or without amplifying Fads1. n = 6 in each group.

Bar graphs from C to N show representative results from one of the three repeats for each experiment and are presented as mean ± SEM. Student’s t tests (C to J) and one-way ANOVA followed by Tukey’s multiple-comparison tests (K to N) were used for statistical comparison. *p < 0.05; **p < 0.01.

To explore the functional significance of FADS1 in modulating TANs, we first performed CRISPR gene editing and generated Fads1^−/− PDVC57 cells in which Sox2 was also amplified. As we expected, silencing Fads1 in SOX2^High SCC tumors reactivated the immunotherapy-induced anti-tumor phenotypes in TANs. Even though SOX2 was still overexpressed in the same cancer cells, the TANs infiltrating these tumors could now respond to immunotherapy, upregulate LY6E or class II MHC, but downregulate Siglec-F (Figures 6C-6E). Importantly, when Fads1 was silenced, the neutrophils from these SOX2^High tumors could now respond to interferons when isolated and treated with interferons in vitro (Figure 6F).

In parallel, we amplified Fads1 in SOX2^Low PDVC57 cells. Fads1 amplification in SCC cells alone was sufficient to enable cancer cells to manipulate TANs and block them from responding to interferons. For example, the TANs in the FADS1^High tumors maintained LY6E^Low and Siglec-F^High immune suppressive phenotypes (Figures 6G and 6H), and these TANs could not activate class II antigen presentation (Figure 6I). In addition, we isolated neutrophils from naive FADS1^High tumors and treated them with interferons in vitro. Compared to the neutrophils from control tumors, the neutrophils isolated from the FADS1^High tumors could not efficiently induce interferon-stimulated genes (Figure 6J). As a result, similar to the T cells infiltrating the SOX2^High tumors, both T cell numbers and their anti-tumor activities were significantly blunted in the FADS1^High tumors (Figures 6K-6N). Therefore, we conclude that FADS1 is the key downstream effector activated by SOX2 in tSCs to modulate the interferon responsive potentials of TANs.

SOX2^High tSCs produce AA to activate PGE₂ signaling and block the interferon responsiveness in TANs

The significance of the SOX2-FADS1 axis for modulating TANs prompted us to further investigate the role of AA. We first investigated whether Sox2 amplification induced SCC cells to produce more AA. For this purpose, we focused on the tumor interstitial fluid (TIF), since the composition of metabolites and signaling factors in TIF faithfully reflect the molecular components present in the TME ⁶¹. As we expected, quantitative mass spectrometry analysis showed that AA was significantly enriched in the TIF from SOX2^High tumors (Figure 7A). We next intratumorally injected AA into the SCCs formed by SOX2^Low PDVC57 cells, isolated TANs from treated tumors, and stimulated these neutrophils in vitro with interferons. This key experiment confirmed that the presence of a high amount of AA in the TME can effectively blunt the interferon responsive capacity of TANs (Figure 7B).

Figure 7. AA induces PGE₂ signaling to disrupt the interferon response in TANs.

A. Quantitative mass spectrometry measurement of AA in TIF extracted from naive SCC tumors formed by SOX2^Low or SOX2^High tumors. Relative AA levels were normalized to the average amount in SOX2^Low tumors. n = 6 in each group.

B. Experimental scheme and quantitative PCR measuring the Gbp2 and Ifit3 expression in TANs isolated from SCC tumors with or without intratumoral injection of AA for 5 days. Isolated neutrophils were treated with IFNα + IFNγ in vitro for 4 hrs to measure their interferon responsive capacity. Expression levels of Gbp2 and Ifit3 were first normalized to 18S rRNA and then further normalized to the average value of vehicle control. n = 5 in each group.

C and D. Quantitative PCR measuring the (C) Gbp2 and (D) Ifit3 expression in cultured primary neutrophils treated with vehicle or different AA metabolites for 48 hrs followed by IFNα + IFNγ treatment for 4 hr to measure their interferon responsive capacity. Expression levels of Gbp2 and Ifit3 were first normalized to 18S rRNA and then normalized to the average value of the vehicle control group. n = 4 in each group.

E. Quantitative PCR measuring the Ptgs2 expression in TANs isolated from naive SCC tumors formed by SOX2^Low or SOX2^High SCC cells. Expression levels of Ptgs2 were first normalized to 18S rRNA and then normalized to the average value of the SOX2^Low group. n = 8 in each group.

F. Western blots probing the phosphorylation of JAK1, JAK2 and STAT1 in cultured primary neutrophils after treatment with vehicle or different AA metabolites for 48 hrs followed by IFNα + IFNγ in vitro treatment for 30 mins to quantify the impacts of PGE₂ on the interferon signaling transduction in neutrophils. Actin was used as a loading control.

G to I. Flow cytometry quantification of the (G) MHCII and (H) LY6E on LY6G⁺ TANs or (I) the LY6G- myeloid cells in SCC tumors formed by SOX2^High SCC cells on control or the Ptger2/Ptger4 myeloid cell-specific dKO mice. n = 8 in control and n = 6 in dKO group.

J. Growth of SOX2^High or SOX2^Low SCC tumors following combinational treatment of Cox2 inhibitors (Celecoxib or Aspirin) and immunotherapy. n = 10 in each group.

Graphs show pooled results (A) or representative results (B to J) from one of the three repeats for each experiment and are presented as mean ± SEM. Student’s t tests (A, B, E, G, H, I), one-way ANOVA followed by Tukey’s multiple-comparison tests (D and E), and Two-way ANOVA and Sidak’s multiple comparisons tests (J) were used for statistical comparison. *p < 0.05; **p < 0.01; ***p < 0.001.

Based on these results, we explored the underlying mechanisms of how AA impacted the interferon responsive potentials of neutrophils. As an important unsaturated fatty acid, AA is metabolized into either prostaglandins by cyclooxygenases (COXs) or leukotrienes by lipoxygenases (LOXs) which can further process leukotrienes to generate lipoxin ⁵⁹. To test the impact of these AA metabolites, we treated cultured primary neutrophils with these metabolites. This assay identified prostaglandin E₂ (PGE₂) as the critical downstream product of AA that can significantly dampen the interferon responses in neutrophils (Figures 7C and 7D). Further supporting this function, we found that the TANs from SOX2^High SCCs express higher level of Ptgs2, the gene encoding for the rate-limiting enzyme for producing PGE₂ (Figure 7E). Interestingly, exposing the bone marrow derived neutrophils to PGE₂ disrupted the signaling transduction of the interferon pathway, as demonstrated by the reduced phosphorylation of major signaling molecules, especially STAT1 (Figure 7F).

Next, we explored the functional significance of PGE₂ signaling in suppressing the immunotherapy-induced neutrophil reprogramming. To this end, we have first deleted PGE₂ receptors, both EP2 and EP4, specifically in myeloid cells, using the Ptger2/Ptger4 conditional double knockout (dKO: Lyz2Cre; Ptger2^flox/flox; Ptger4^flox/flox) mice ⁶². Next, by grafting the SOX2^High SCC cells on these mice, we demonstrated that immunotherapy-induced reprogramming, such as elevated LY6E and MHCII expression, can be efficiently restored specifically in TANs (Figures 7G and 7H), but not in other LY6G negative myeloid cells (Figure 7I). To further support the role of SOX2-induced PGE₂ signaling in promoting cancer relapse after immunotherapy, we treated mice bearing SOX2^High SCC tumors with COX-2 inhibitors (e.g. aspirin or celecoxib) to block PGE₂ production. Combination treatment with immunotherapy and COX-2 inhibition led to significantly reduced tumor growth and blunted relapse of the SOX2^High tumors (Figure 7J), underscoring the importance of PGE₂ signaling in blocking the immunotherapy-induced interferon response in neutrophils.

Discussion

tSCs comprise a vital cellular reservoir for repopulating the tumors when effective therapies remove the bulk of the tumor mass ^8,63. Thus, tSCs must be equipped with special molecular programs to orchestrate therapy resistance and cancer relapse. Unlike other conventional therapies, effective immunotherapies are expected to elicit long-lasting changes in the TME ⁶⁴, such as tissue residency programs ⁶⁵ and long-term memory in the immune cells ⁶⁶. Thus, it remains unclear how tSCs can overcome these enduring effects from immunotherapy and give rise to the relapsed tumors. In this study, we show that tSCs can block neutrophils at the tumor-stroma interface from responding to interferons and prevent these neutrophils from being reprogrammed by immunotherapy. This important modulation maintains the neutrophils in a suppressive cell state and allows them to form a micro-niche that protects tSC from being attacked by T cells during immunotherapy treatment.

Investigating the mechanisms, we found that SOX2 can activate FADS1 to produce arachidonic acid, which blunts the interferon-responsive capacity in the nearby neutrophils. We further identified PGE₂ as the critical downstream metabolites of AA that mediated the inhibitory effects on interferon response. Recent studies investigating the impacts of PGE₂ on tumor infiltrating T cells have provided clues of how PGE₂ might interfere with interferon responses. It has been shown that PGE₂ can disrupt IL2 signaling in T cells by downregulating the IL-2Rγc chain ⁶⁷. Based on this finding, future studies are needed to investigate whether PGE₂ can also disrupt the interferon receptor complex formation. Importantly, this study also underscored the promising treatment of combining Cox-2 inhibitors with various immunotherapies to enhance treatment efficacy and prevent cancer relapses.

Furthermore, our study has important implications for understanding the biology of TANs. Only recently have we begun to understand many roles of TANs during immunotherapy. How TANs switch phenotypes from being predominantly immune suppressive to supporting anti-tumor immunity during the immunotherapy treatment is still unknown. Interestingly, it appears that the synergistic effect of both type I and II interferons is required to reprogram TANs ³⁰. Future research is required to untangle the distinct and combined effects of IFNα and IFNγ on TANs. Comprehensive epigenetic profiling is also required to understand the impacts of interferons on the plasticity and cell states of TANs.

In closing, we have identified an intimate crosstalk between a subset of TANs and tSCs at the tumor-stroma interface. Thus, this study provides insights for designing strategies that can disrupt the dialogue between neutrophils and tSCs, exposing tSCs to enhanced immunity, hence preventing cancer relapse.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for reagents and resources should be directed to Dr. Yuxuan Miao (miaoy@uchicago.edu)

Materials availability

This study did not generate unique new reagents.

Data and code availability

Single cell RNA sequencing (GSE278435), bulk RNA sequencing (GSE278429, GSE278430, GSE303056), spatial transcriptomics profiling (GSE278436), and CUT & RUN-seq (GSE278426) data reported in this paper have been deposited at Gene Expression Omnibus (GEO) and are publicly available. Previously published datasets analyzed in this study were obtained from GEO under accession numbers GSE208253 ⁵⁵, GSE144239 ⁵⁶, GSE243466 ²⁴, GSE224399 ³¹, GSE224400 ³¹, and GSE243013 ⁵⁴. Tumor growth, microscopy data and flow cytometry data reported in this paper will be shared upon request. This paper does not report original code and any additional information or data in this paper will be shared upon request.

STAR★Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Animals

K14CreER mice were generated by Dr. Elaine Fuchs lab and backcrossed to C57/BL6J background for ten generations ⁶⁸. LyzMCre; Ptger2^flox/flox; Ptger4^flox/flox (Ptger2/Ptger4 dKO) mice were generated by Dr. Narumiya’s lab ⁶². C57BL/6-Ly6g(tm2621(CretdTomato)Arte); R26-LSL-tdTomato (Catchup^IVM-red) mice were generated by Dr. Matthias Gunzer’s lab. Ly6G-CreER; R26-LSL-tdTomato (C57BL/6-iLy6GtdTom) mice were generated by Dr. Andrés Hidalgo’s lab. These mice have been described before ⁴⁵. Wild-type C57BL/6J, B6(Cg)-Ifnar1^tm1.1Ees/J(Ifnar1^flox/flox), C57BL/6N-Ifngr1^tm1.1Rds/J (Ifngr1^flox/flox), B6.Sox2^tm1.1Lan/J (Sox2^flox/flox), B6.Cg-Tg(S100A8-cre,-EGFP)1Ilw/J (S100A8-Cre), C57BL/6-Gt(ROSA)^{26Sortm1(HBEGF)Awai}/J (R26-LSL-DTR) mice, were obtained from The Jackson Laboratory. To induce Sox2 conditional knockout, K14-CreER; Sox2^flox/flox mice were treated with daily intraperitoneal (i.p.) injection of 1 mg tamoxifen for three consecutive days. To lineage trace neutrophils, Ly6G-CreER; R26-LSL-tdTomato mice were treated with one dose of 4 mg tamoxifen by i.p. injection.

All mice were maintained in an Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC)-an accredited animal facility. Procedures were performed using IACUC-approved protocols with protocol number 72637. All mice were maintained and bred under specific-pathogen free conditions at The University of Chicago. All the procedures are in accordance with the Guide of the Care and Use of Laboratory Animals.

Cell lines

Mouse skin SCC line PDVC57 for tumor grafting and HEK293T cells for packaging lentivirus were cultured in Dulbecco’s modified eagle medium (DMEM) in 10% FBS, 100 units/mL streptomycin and 100 μg/mL penicillin with 2mM glutamine. Mouse oral SCC line MOC1 was cultured in Iscove’s Modified Dulbecco’s Medium (IMDM)/F12 (2:1) media with 5% FBS, 100 units/mL streptomycin, 100 μg/mL penicillin, 5 μg/mL insulin, 40 ng/mL hydrocortisone, and 5 ng/mL EGF.

Human samples

Patient specimens analyzed in this study were deidentified and obtained under the IRB protocol number 22–1951 (CSCC) and IRB protocol number 8980 (HNSCC) approved by the University of Chicago Medicine, in compliance with federal and state regulations and NIH guidelines. CSCC tumors were fixed, embedded, and sectioned for immunofluorescence (IF) imaging. HNSCC samples used in this study were formalin fixed, and paraffin embedded (FFPE).

METHOD DETAILS

Tumor formation and treatment

A two-stage chemical carcinogenesis protocol was used to induce skin SCCs in mice. The dorsal skin of various mouse strains was shaved and topically treated with 200 nM 7,12-dimethylbenz[a]anthracene (DMBA) once a week for 4 weeks to initiate tumorigenesis. Four weeks later, tumor promotion was induced by weekly (twice a week) applications of 200 μL of 35 μM 12-O-tetradecanoylphorbol-13-acetate (TPA) for 20 weeks. For tumor transplantation, 5×10⁵ PDVC57 mouse skin SCC cells or 2×10⁶ MOC1 mouse oral SCC cells were mixed with Cultrex Basement Membrane Extract, PathClear, Type 3 Matrigel (Bio-Techne) and injected subcutaneously. Tumors were allowed to grow for two weeks prior to immunotherapy. The tumor growth was measured twice a week, and tumor volumes were calculated using the formula: π/2 × length × width × height.

For immunotherapy treatment on both DMBA + TPA-induced spontaneous tumors or grafted SCCs, tumor bearing mice were treated intraperitoneally (i.p.) with a combination of 100 μg PD-L1 blocking antibody (Clone 10F.9G2, BioXCell) and 100 μg CD40 agonist antibody (Clone FGK4.5, BioXCell) for a total of 2 doses every other day. Tumors were collected 2 days after the last treatment for downstream experiments. For intratumoral treatment with AA, tumors were allowed to grow for three weeks prior to treatment. AA was mixed in 5% DMSO/PBS solution and 1.5 mg AA was injected into each tumor once a day for 5 consecutive days⁶⁹.

Cell sorting and flow cytometry

To sort immune cells from DMBA + TPA induced tumors, tumors were collected at designated time points, minced and digested with 2 mg/mL type 4 collagenase (GIBCO) and 20 U/mL DNase I (Roche) in RPMI-1640 (GIBCO) for 60 minutes at 37 °C. Digested tissues were then passed through a 70 μm cell strainer, subjected to ACK lysis for 1 minute to lyse the red blood cells and resuspended in PBS. Single cells suspension was incubated with Fc Block TruStain FcX (Clone 93, Biolegend) in PBS with 5% normal rat serum and 5% normal mouse serum, and then stained with a cocktail Ab at predetermined concentration in FACS buffer (PBS with 5% FBS, 5 mM EDTA and 1% HEPES). DAPI was used to exclude dead cells. The sorting was performed on BD Symphony S6 Cell Sorter.

To isolate ITGA^highCD80⁺ tSCs from DMBA + TPA induced spontaneous SCCs at designated time points, tumors were collected, minced and digested with 2 mg/mL type 4 collagenase (GIBCO) and 20 U/mL DNase I (Roche) in RPMI-1640 (GIBCO) for 30 minutes followed by 0.25% Trypsin treatment for 10 minutes at 37 °C. Digested tissues were then passed through a 70 μm cell strainer, subjected to ACK lysis for 1 minute to lyse the red blood cells and resuspended in PBS. Cells were first incubated with biotin-conjugated antibodies against CD140a, CD45, CD11b, CD31, CD117, and CD64 for 15 minutes, followed by incubation with 20 μL of magnetic streptavidin beads for an additional 15 minutes. Cells bound to the beads were removed by magnetic separation to facilitate clean cell sorting. The remaining cells were then stained with antibodies against CD49f (ITGA), CD80, and along with PE-streptavidin, in FACS buffer. DAPI was added to exclude dead cells. Live, integrin ITGA^highCD80⁺ tSCs were sorted using a BD Symphony S6 cell sorter. Sorted cells were immediately processed for RNA extraction and downstream RNA sequencing.

To profile immune population and measure tumor-infiltrating T cell activity in grafted tumors, the tumors were minced and digested in 2 mg/mL type 4 collagenase (GIBCO) and 20 U/mL DNase I (Roche) in RPMI-1640 for 60 minutes at 37 °C with shaking. Digested tissues were then passed through a 70 μm cell strainer and red blood cells were lysed using ACK lysis buffer for 1 min. After preparation of single cell suspension, cells were then directly stained with Zombie Aqua (Biolegend) to exclude dead cells, blocked with TruStain FcX, and then stained with a cocktail of Abs for surface antigens at pre-determined concentrations. For intracellular cytokine staining, the single cell suspension of various tumors was stimulated with Cell Stimulation Cocktail (plus protein transport inhibitors) (Thermo Fisher Scientific) at 37 °C for 4 hours, then followed by fixable cell death staining, Fc block, cell surface marker staining, fixation and permeabilization with Cytofix/Cytoperm (BD Bioscience) and stained with Abs recognizing intracellular antigens in Perm/Wash buffer overnight at 4 °C. The stained cells were profiled on LSR-Fortessa analyzer (BD Biosciences) and analyzed with FlowJo software.

Neutrophil isolation

The neutrophil isolation methods were adopted from previously published protocols⁷⁰. Briefly, to isolate neutrophils from grafted tumors, the tumors were minced and digested in 2 mg/mL type 4 collagenase (GIBCO) and 20 U/mL DNase I (Roche) in RPMI-1640 for 60 minutes at 37 °C with shaking. Digested tissues were then passed through a 70 μm cell strainer and red blood cells were lysed using ACK lysis buffer for 1 min. After preparation of single cell suspension, cells were resuspended in 8 ml of 40% Percoll and overlayed onto 3 ml of 70% Percoll in a 15 ml conical centrifuge tube. After centrifugation for 30 minutes at 900 × g without brake at room temperature, cells from the interface were harvested, washed twice with FACS buffer and resuspended in FACS buffer. Cells were blocked with TruStain FcX for 10 min, then incubated with 20 μL biotin-anti-LY6G antibody for 15 min followed by incubation with 20 μL magnetic strep-beads for 15 min. Samples were washed 4 times with FACS buffer and resuspended in RPMI-1640 for in vitro interferon treatment.

To isolate neutrophils from mouse bone marrow, femurs were harvested, and both ends were cut using sterile scissors. The bone marrow was flushed from the femurs using a syringe filled with RPMI-1640 medium and passed through a 40 μm cell strainer into sterile tubes. To isolate neutrophils from mouse spleen, the tissue was harvested, mechanically dissociated, passed through a 40 μm cell strainer, and collected into sterile tubes. The collected cells were treated with ACK lysis buffer for 1 minute to lyse red blood cells. Following lysis, cells were blocked and subjected to the same isolation procedure described for tumor samples.

Neutrophil culture and treatment

Freshly isolated neutrophils were cultured in RPMI-1640 with 10% FBS, 100 units/mL streptomycin and 100 μg/mL penicillin, 55 μM beta-mercaptoethanol (GIBCO), 50 U/mL GM-CSF (Biolegend). For in vitro IFN treatment, 5 ng/mL of IFN-α and 5 ng/mL of IFNγ were added to the culture media and incubated for 4 hours prior to harvesting cells for further analysis. For the neutrophil-OT-II CD4⁺ T cell co-culture, neutrophils were isolated from different tumors, then pulsed with 15 μg/mL OVA₃₂₃-₃₃₉ peptide (Invivogen) for 4 hours, washed, and seeded into round-bottom, 96-well, non-tissue culture-treated plates. OT-II CD4⁺ T cells were isolated by negative selection from the spleen of the OT-II transgenic mice using the MojoSort Mouse CD4 Naive T Cell Isolation Kit (BioLegend). Neutrophils and T cells were co-cultured at 1:1 ratio, 50,000 cells per well in the presence of 10 ng/mL IL2 (R&D Systems). After 5 days of culture, T cell activation and cytokine production were stained and analyzed by flow cytometry, as described above. For in vitro metabolite treatment, neutrophils were cultured with 10 μM PGE₂, 1 μM LTB₄, or 1 μM LXA₄ for 48 hrs. Treated cells were then washed and subjected to recombinant interferon treatment as described above.

Western blot

Neutrophils were cultured with 10 μM PGE₂ for 48 hrs and then stimulated with 5 ng/mL IFNα and 5 ng/mL IFNγ for 30 minutes. Cells were collected and lysed directly in 1× Laemmli buffer supplemented with 100 mM DTT and 1x Halt^™ Protease and Phosphatase Inhibitor Cocktail (Thermo Fisher Scientifc), followed by boiling at 100 °C for 10 minutes. Lysates were centrifuged at 13,000 × g for 5 minutes at room temperature, and the supernatants were loaded onto 4–12% NuPAGE^™ Bis-Tris Mini Protein Gels (Invitrogen). Electrophoresis was performed at 200 V for 30 minutes, followed by protein transfer to 0.45 μm nitrocellulose membranes (Cytiva). Membranes were blocked with 5% BSA (Fisher Scientific) in TBST (0.1% Tween 20 in Tris-buffered saline, Thermo Scientific) for 1 hour at room temperature, then incubated overnight at 4 °C with primary antibodies. After washing, membranes were incubated with appropriate HRP-conjugated secondary antibodies for 1 hour at room temperature. Signal was developed using SuperSignal^™ West Pico PLUS Chemiluminescent Substrate (Thermo Scientific) and imaged using the Tanon Chemi Dog 5200T automatic luminescence imaging system (Tanon).

Neutrophil depletion

The neutrophils in S100A8-Cre; R26-LSL-DTR mice were depleted by i.p. injection of 0.25 μg diphtheria toxin (DT, Sigma-Aldrich) every 2 days. To quantify immune cell changes after neutrophil depletion, three doses of DT were injected. To measure tumor growth after neutrophil depletion, seven doses of DT were injected. For neutrophil depletion in wildtype C57BL/6 mice, mice were injected i.p. with 500 μg anti-LY6G antibody (BioXCell) for the first two doses, followed by 200 μg antibodies for the next eight doses. For depleting neutrophils in SOX2^Low or SOX2^High tumors and measure tumor growth, LY6G antibodies were administered every other day, starting three days before immunotherapy treatment. For depleting neutrophils in grafted tumors mixing SOX2^High and SOX2^Low cells or in the spontaneous SCCs, antibodies were administered on day two and day four after the last dose of immunotherapy. Tumors were then collected for imaging analysis

In vivo COX2 inhibition

Aspirin (Cayman Chemical) was administered via drinking water at a concentration of 600 μg/mL.^71,72. Celecoxib (Selleckchem) was administered via daily i.p. injections of 200 μL at 500 μg/mL in 12.5% DMSO/PBS, as previously described⁷¹. All drinking water contained sweetener was replaced every other day. Treatment began one day prior to tumor cell injection. Immunotherapy was administered on day 14 and 16 post tumor grafting.

TIF collection and AA quantification

To collect TIF from grafted tumors, tumors were dissected into small pieces (3–5 mm³) and placed on 40 μm cell strainers fitted into 50 mL conical centrifuge tubes. Samples were centrifuged at 300 × g for 10 minutes at 4 °C to extract interstitial fluid. The resulting TIF was collected and AA in TIF was quantified using liquid chromatography–mass spectrometry (LC/MS).

Cloning and transducing cells

Mouse Sox2 and Fads1 full open reading frames (ORFs) were amplified from PDVC57 cDNA through PCR, then cloned into lentiviral vectors using NEBuilder^® HiFi DNA Assembly kit (New England Biolabs) following the manufacturer’s instructions.

Lentivirus was produced by co-transfecting the main plasmids, psPAX2, and pMD2.G using Lipofectamine 3000 (Thermo Fisher), according to the manufacturer’s instructions. The medium was replaced 4 hours post-transfection. Viral supernatants were collected 48 hours post-transfection, filtered through a 0.45 μm filter.

For transduction, target cells were seeded in 6-well plates. Viral supernatant was added with 10 μg/mL polybrene. Plates were centrifuged at 1,100 × g for 30 minutes at 37 °C. After centrifugation, the viral medium was replaced with fresh medium, and transduced cells were cultured for 48 hours before further analysis or selection.

Bulk RNA-Seq

Total RNAs from FACS-sorted cells were extracted using Quick-RNA Microprep Kit (Zymo Research) following the manufacturer’s protocol. RNA-seq libraries were prepared with the NEBNext Single Cell/Low input RNA library prep kit for Illumina (NEB) following the manufacturer’s protocol. The libraries were sequenced on the Illumina Novaseq platform.

Raw FASTQ files were first trimmed and filtered by cutadapt (v3.2)⁷³. Estimated transcript counts for the mouse mm10 genome assembly (GENCODE vM24) were obtained using the pseudo-aligner Kallisto (v0.44.0)⁷⁴. Gene level abundance was summarized and differential gene expression was performed using the DESeq2 R package (v1.30.0) in R (v4.1.1). Genes with Benjamini-Hochberg method adjusted P values < 0.05 were regarded as significantly differentially expressed.

Gene Set Enrichment Analysis (GSEA)

GSEA was performed using clusterProfiler R package (v4.6.0, https://guangchuangyu.github.io/software/clusterProfiler)⁷⁵. MSigDB (Molecular signature database, v7.4) gene set collections (http://www.gsea-msigdb.org/gsea/msigdb/index.jsp)⁷⁶ was used in the analysis. Genes were ranked by the fold change value obtained from DESeq2. Enriched pathways with Benjamini-Hochberg method adjusted P values < 0.25 were considered to be significant.

Single-cell RNA-seq

Tumors were collected from naive mice or on day one after the last dose of immunotherapy and total CD45 positive immune cells were sorted. Immediately after sorting, the sorted immune cells were fixed with Evercode Cell Fixation kit (Parse Biosciences). The fixed cells were prepared using Evercode WT v2 kit (Parse Biosciences) according to manufacturer’s instructions. Input cell numbers and volumes were calculated using WT_100K_Sample_Loading_Table_V1.3.0 (Parse Biosciences). All the samples were processed together with Evercode WT v2 kit (Parse Biosciences) and 8 sub-libraries were generated. The quality of sub-libraries was checked by the Agilent Tape Station system (Agilent) and sequenced by Illumina NovaSeqX.

The raw FASTQ files were processed by spipe.v1.1.1(Parse Biosciences) using mm10 as a reference. Count matrices were imported to Seurat (v.4.3.0).⁷⁷ Cells with <50 detected genes or >10000 detected genes or > 7.5% mitochondrial genes were filtered-out from the dataset. We then used the standard Seurat pipeline by running NormalizeData(), FindVariableFeatures(selection.method =”vst”, nfeatures =4000), and ScaleData(). For the first round of clustering, Principal component analysis (PCA) was performed and the top 50 PCs with a resolution = 0.6 were applied. RunUMAP(dim=1:40, n.neighbours =30) was used to visualize the data. Cell types were annotated using SingleR (V1.7.1)⁷⁸ package with ImmGen data as reference. Neutrophils were subset, and standard pipeline was applied to neutrophil population by running FindVariableFeatures(selection.method =”vst”, nfeatures =4000), and ScaleData(). For clustering the neutrophils, PCA was re-performed and the top 40 PCs with a resolution = 0.2 were applied. For visualization, RunUMAP(dim=1:50, n.neighbours =30) was used.

For RNA velocity analysis, spliced and unspliced count matrices were first obtained from spipe output and exported as h5ad files for subsequent input of scVelo. Velocity analysis was then performed with scVelo (0.2.5)⁷⁹ using the default stochastic model and velocity vectors were projected into the UMAP generated from previous Seurat pipeline.

For pseudo-bulk analysis, cells were randomly grouped into 3 pseudo replicates. Gene counts were summarized and DESeq2 R package (v1.30.0)⁸⁰ in R (v4.1.1) was used for differential gene analysis.

Public single-cell RNA-seq data analysis

Single-cell RNA sequencing (scRNA-seq) data were downloaded from the GEO and processed into Seurat objects. Tumor-associated neutrophils were subset from the dataset using author-provided metadata. Batch effects arising from different sequencing runs were corrected using the Harmony algorithm. Subsequently, standard Seurat analysis workflows were performed as described above, and gene expression patterns were visualized accordingly.

UMAP projection and cross-dataset label transfer

Single-cell RNA sequencing data of neutrophils from grafted lung (KP19), colon (MC38), and pancreatic (KPC) tumors were obtained from the GEO and processed into Seurat objects. To compare neutrophil states across datasets, cross-dataset projection was performed using the Seurat package. A reference UMAP embedding was generated from neutrophils in our DMBA + TPA-induced SCC dataset using principal component analysis. Neutrophils from the grafted tumor models were then projected into this reference UMAP space using the FindTransferAnchors and MapQuery functions, which identify shared features across datasets and transfer cell type annotations. Projected cells were visualized using the ref.umap coordinates, and the transferred labels were used for comparative analysis. Cluster-specific color schemes and harmonized axis ranges were applied to facilitate direct visual comparison across datasets.

Spatial transcriptomic data analysis

Processed spatial transcriptomic data were obtained from the GEO and imported into Seurat to generate spatial Seurat objects. Data normalization was performed using Seurat’s SCTransform workflow. To identify regions of high SOX2 expression, tissue spots were ranked based on SOX2 transcript levels, and the top 10% were designated as “SOX2-high”. Euclidean distances from each spot to the nearest SOX2-high spot were calculated using a k-nearest neighbor model. Based on these distances, spots were categorized into “near” (lowest 20%), “mid” (middle 20–60%), and “far” (highest 40%) proximity groups. To explore the spatial association between SOX2-high regions and neutrophil-related gene expression, a neutrophil-enriched subset was defined by selecting spots with high S100A8 expression (top 20%). Differential gene expression analysis between SOX2-near and SOX2-far neutrophil subsets was performed using Seurat’s FindMarkers function. Expression patterns of selected neutrophil marker genes were visualized using violin plots.

CUT&RUN

CUT&RUN was performed using CUTANA^™ ChIC/CUT&RUN Kit version 3 (EpiCypher) following the manufacture’s protocol. For each CUT&RUN reaction, 500,000 PDV-Sox2 cells were first cross-linked using 1% formaldehyde for 1 min at room temperature and then quenched by adding 2.5 M Glycine to a final concentration of 125 mM. Crosslinked cells were permeabilized, immobilized onto Concanavalin-A beads and incubated overnight (4 °C with gentle rocking) with either rabbit IgG or SOX2 antibody. DNA was extracted after chromatin digestion and release. Sequencing libraries were prepared with the NEBNext^® Ultra^™ II DNA Library Prep Kit for Illumina (NEB) according to the manufacture’s protocol and sequenced by Illumina NovaSeqX.

Raw FASTQ files were trimmed and filtered using Cutadapt (v3.2), followed by alignment to the mm10 mouse reference genome with Bowtie2 (v2.3.4.3). Duplicate reads were removed using Picard (v2.18.7), and peaks were identified with MACS2 (v2.2.7.1). BigWig track files were generated using bamCoverage from deepTools (v3.5.6), normalized to reads per kilobase per million mapped reads (RPKM), and visualized in IGV.

NanoString GeoMx Mouse Whole Genome Transcriptome Profiling

NanoString’s GeoMx Digital Spatial Profiler (DSP) platform was used to perform whole-genome transcriptome profiling on mouse tissue samples. Fresh frozen tumor sections were mounted onto slides, followed by immunofluorescent staining with specific antibodies targeting protein markers of interest to delineate distinct tissue regions. After staining, high-resolution digital images were acquired to visualize and identify regions of interest (ROIs).

ROIs were selected based on morphological and biological relevance. These regions were then subjected to UV light-based spatially resolved photocleavage to release oligonucleotide tags bound to the target mRNA transcripts. The released tags were collected into 96-well plates.

Library preparation was performed according to the NanoString GeoMx-NGS Readout Library Prep manual. Briefly, the released tags were dried and reconstituted in 10 μL nuclease free water, and 4 μL was used in a PCR reaction. NanoString barcoded primers were used to amplify the tags and add Illumina adaptor sequences and sample demultiplexing barcodes. PCR products were pooled in equal volumes and purified with two rounds of AMPure XP beads (Beckman Coulter). Libraries were sequenced on an Illumina NextSeq 550 platform.

FASTQ files were first processed by the NanoString GeoMx NGS Pipeline v2.0. Briefly, raw reads were pre-processed by trimming low-quality bases and adapter sequences. Filtered reads were then stitched and aligned, followed by extraction of barcode and UMI sequences. Barcodes were matched to a reference set of known probe barcodes, allowing for a maximum of one mismatch. To remove duplicates, reads sharing the same barcode were collapsed based on unique molecular identifiers (UMIs).

Raw data from NanoString GeoMx NGS Pipeline were then imported to R using GeomxTools package to generate GeoMxSet object. The raw count data was first normalized with “q_norm” method. To facilitate data processing, GeoMxSet object was further coerced into Seurat object and standard Seurat pipelines were applied for analysis.

Immunofluorescence staining for frozen tumor samples

Tumors freshly collected from mice or from human patients were fixed in 1% paraformaldehyde for 1 hour at 4 °C and washed three times with cold PBS. After incubation in 30% sucrose at 4 °C overnight, tumor tissues were embedded in OCT (Tissue Tek), frozen, and sectioned (10–15 μm). Cryosections were permeabilized, blocked, and stained with the following primary antibodies: CD4 (rat, 1:100, BioLegend), K14 (chicken, 1:1000, BioLegend), SOX2 (rabbit, 1:400, Cell Signaling Technology), SOX2 (rat, 1:100, Thermo Fisher Scientific), RFP (rat, 1:1000, Proteintech), MPO (goat, 1:40, R&D), CD74 (sheep, 1:20, R&D), I-A/I-E (rat, 1:200, BioLegend), Cleaved caspase3 (rabbit, 1:500, Cell Signaling Technology), CD140a (rat, 1:100, BioLegend), CD31 (rat, 1:100, BioLegend) or CD45 (rabbit, 1:200, Cell Signaling Technology). The samples were then stained with corresponding secondary antibodies conjugated with Alexa Fluor 488, Rhodamine Red, or Alexa Fluor 647 (Jackson ImmunoResearch Laboratories) and imaged on Leica Stellaris 8 Laser Scanning Confocal microscope.

RNA scope staining combined with Immunofluorescence

For the RNA scope staining, cryosections were dehydrated, treated with hydrogen peroxide and Protease III, hybridized with AMPs and developed the HRP-C1 channel with the Gbp2 probe according to the RNAscope Multiplex Fluorescent Reagent Kit Assay (ACDbio). Immediately after RNA labeling, the cryosections were blocked and stained with the primary antibodies against K14 and MPO at 4 °C overnight and then incubated with the related secondary antibodies. The images were collected by Leica Stellaris 8 Confocal microscope and analyzed using Fiji/ImageJ software.

Immunofluorescence staining of FFPE slides of human HNSCC samples.

FFPE slides were deparaffinized by incubation at 60 °C for 1 hour. Rehydration was performed through sequential washes: three times in xylene, followed by 100% ethanol, 95% ethanol, 75% ethanol, and distilled water (ddH2O), each for 10 minutes. Antigen retrieval was performed using a low-pH antigen retrieval buffer (Thermo Fisher Scientific). Slides were then blocked at room temperature for 1 hour and incubated overnight at 4 °C in a humidified chamber with the following antibodies: pan cytokeratin (rabbit, 1:100, Abcam), pan cytokeratin (mouse, 1:200, Novus Biologicals), SOX2 (rat, 1:100, Thermo Fisher Scientific), MPO (goat, 1:40, R&D Systems), CD4 (rabbit, 1:500, Abcam), or HLA-DR (mouse, 1:500, Thermo Fisher Scientific). After primary antibody incubation, slides were washed three times with PBS, each for 5 minutes, and then incubated with secondary antibodies conjugated with Alexa Fluor 488, Rhodamine Red, or Alexa Fluor 647 (Jackson ImmunoResearch Laboratories). Slides were mounted using Prolong Gold Antifade Mountant (Thermo Fisher Scientific), carefully avoiding air bubbles during coverslip placement. Fluorescent images were acquired using a Leica Stellaris 8 Laser Scanning Confocal microscope.

Immunofluorescence quantification

The immunofluorescence data was collected from at least four tumor samples each group, with each sample sectioned and imaged separately. Multiple images were taken from different regions within each individual section, ensuring that the quantification is based on a diverse set of data points from each sample. The quantification of CD4-neutrophil interactions was performed by measuring the number of CD4⁺ cells overlapping with MPO⁺ cells and normalizing to each field of view. The quantification of CD4⁺ cells was performed by measuring the number of CD4⁺ cells and normalizing to each field of view. The proportions of Gbp2 mRNA, MHCII⁺, or HLA-DR⁺ neutrophils were calculated relative to the total number of MPO⁺ cells in each field of view. The proportion of SOX2⁺Caspase3⁺ cells was calculated relative to the total number of SOX2⁺ cells in each field of view.

RNA purification and qRT-PCR

The total RNA was purified using Quick-RNA Microprep Kit (Zymo Research) in accordance with manufacturer’s instructions. For real-time qRT-PCR analysis of any target genes, equal amount of RNAs were reverse transcribed using Maxima First Strand cDNA Synthesis Kit (Thermo Fisher). cDNAs were then mixed with specific gene primers and PowerTrack^™ SYBR Green Master Mix (Thermo Fisher) and the qRT-PCR was performed on the QuantStudio^™ 3 Real-Time PCR System (Thermo Fisher). Relative gene expression levels were calculated using the ΔΔCt method, with the control group serving as the baseline and set to 1.

CRISPR mediated gene knockout

To generate gene knockouts, the lentiCRISPRv2 system was employed. Specific single guide RNAs (sgRNAs) targeting the gene of interest were designed using the CRISPick (https://portals.broadinstitute.org/gppx/crispick/public) and cloned into the lentiCRISPRv2 vector (Addgene #52961)⁸¹.

HEK293T cells were transfected with the lentiCRISPRv2 construct, along with packaging plasmids (psPAX2 and pMD2.G) to produce lentiviral particles. After 48 hours, viral supernatants were collected, filtered, and used to transduce target cells in the presence of 10 μg/mL polybrene to enhance infection efficiency.

Following transduction, cells were selected with puromycin (2 μg/mL) for 3days to enrich successfully transduced cells. Single clones were generated by limiting dilution.

QUANTIFICATION AND STATISTICAL ANALYSIS

Data are presented as mean ± SEM or mean ± 95% confidence interval. P values were determined by using two-tailed Student’s t test, Mann-Whitney U Tests or ANOVA test as indicated in corresponding figure legends. Statistical analyses were performed using Prism 9 (GraphPad). Experiments were performed in an open-label manner. Significant difference between two groups were noted by asterisks (* p < 0.05; ** p < 0.01: *** p < 0.001).

Supplementary Material

2

Table S1. Signature genes of TAN clusters, related to Figure 1.

3

Table S2. Differential expression analysis of tSCs compared to non-tSCs, related to Figure 4.

4

Table S3. GSEA results of tSCs compared to non-tSCs, related to Figure 4.

5

Table S4. TFs enriched in tSCs, related to Figure 4.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Biotin conjugated anti-CD45, rat monoclonal (Clone 30-F11)	Biolegend	Cat#103104; RRID: AB_312969
Biotin conjugated anti-CD31, rat monoclonal (Clone MEC13.3)	Biolegend	Cat#102504; RRID: AB_312911
Biotin conjugated anti-CD117, rat monoclonal (Clone 2B8)	Biolegend	Cat#105804; RRID: AB_313213
Biotin conjugated anti-CD140a, rat monoclonal (Clone APA5)	Biolegend	Cat#135910; RRID: AB_2043974
Alexa Fluor 700 anti-mouse CD45 rat monoclonal (Clone 30-F11)	Biolegend	Cat# 103128; RRID: AB_493715
PE/Cy7 anti-mouse CD4 rat monoclonal (Clone GK1.5)	Biolegend	Cat# 100421; RRID: AB_312706
APC/Cy7 anti-mouse CD8b rat monoclonal (Clone YTS156.7)	Biolegend	Cat# 126619; RRID: AB_2563950
PE anti-Granzyme B rat monoclonal (Clone NGZB)	eBioscience	Cat# 12-8898-80; RRID: AB_10853811
Anti-Cleaved Caspase-3(Asp175) Rabbit monoclonal (Clone 5A1E)	Cell Signaling Technology	Cat# 9664; RRID: AB_2070042
Anti-mouse PD-L1 (B7-H1) (Clone 10F.9G2)	bioxcell	Cat# BE0101; RRID: AB_10949073
Anti-mouse CD40 (Clone FGK4.5)	bioxcell	Cat# #BE0016-2; RRID: AB_1107601
PerCP/Cyanine5.5 anti-mouse/human CD11b (Clone M1/70)	Biolegend	Cat# 101228; RRID: AB_893232
PE/Cyanine7 anti-mouse Ly-6G (Clone 1A8)	Biolegend	Cat# 127618; RRID: AB_1877261
APC anti-mouse CD101 (clone Moushi101)	eBioscience	Cat# 17-1011-82; RRID: AB_2573378
BV711 anti-mouse CD62L Antibody (MEL-14)	Biolegend	Cat# 104445; RRID: AB_2564215
PE anti-mouse Siglec-F (clone E50-2440)	BD Biosciences	Cat# 552126; RRID: AB_394341
APC/Cyanine7 anti-mouse CD14 Antibody (clone Sa14-2)	Biolegend	Cat# 123318; RRID: AB_10897102
Brilliant Violet 421™ anti-mouse I-A/I-E (M5/114.15.2)	Biolegend	Cat# 107632; RRID: AB_2650896
Brilliant Violet 605™ anti-mouse LY6A/E (Sca-1) Antibody (Clone D7)	Biolegend	Cat# 108134; RRID: AB_2650926
Alexa Fluor® 488 anti-mouse CD74 (CLIP) Antibody (Clone ln1/CD74)	Biolegend	Cat# 151005; RRID: AB_2750325
Mouse DcTRAILR1/TNFRSF23 Biotinylated Antibody	R&D Systems	Cat# BAF2378; RRID: AB_416745
APC/Cy7 anti-mouse CD8b, rat monoclonal (clone YTS156.7.7)	Biolegend	Cat# 126620; RRID: AB_2563951
PE/Cy7 anti-mouse CD4, rat monoclonal (clone GK1.5)	Biolegend	Cat# 100422; RRID: AB_312707
BV711 anti-mouse TCR β Chain, armenian hamster monoclonal (clone H57-597)	Biolegend	Cat# 109243; RRID: AB_2629564
FITC anti-mouse IFN-γ, rat monoclonal (clone XMG1.2)	Biolegend	Cat# 505806; RRID: AB_315400
PerCP/Cy5.5 anti-mouse Ki67, rat monoclonal (clone 16A8)	Biolegend	Cat# 652424; RRID: AB_2629531
Granzyme B Monoclonal Antibody (GB11), PE (clone GB11)	Thermo Fisher	Cat# GRB04; RRID: AB_2536538
APC anti-mouse TNF-α (clone MP6-XT22)	Biolegend	Cat# 506308; RRID: AB_315429
SOX2 (D9B8N) Rabbit mAb	Cell Signaling Technology	Cat# #23064; RRID: AB_2714146
Anti-GFP, rabbit polyclonal	Abcam	Cat# ab290; RRID: AB_2313768
Purified anti-Keratin 14, chicken polyclonal (clone Poly9060)	Biolegend	Cat# 906004; RRID: AB_2616962
AF488 anti-rabbit IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 711-545-152; RRID: AB_2313584
AF488 anti-chicken IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 703-545-155; RRID: AB_2340375
AF488 anti-goat IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 705-545-147; RRID: AB_2336933
AF647 anti-rabbit IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 711-605-152; RRID: AB_2492288
AF647 anti-chicken IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 703-605-155; RRID: AB_2340379
RRX anti-rat IgG, donkey polyclonal	Jackson ImmunoResearch Laboratories	Cat# 712-295-150; RRID: AB_2340675
Anti-GFP, rabbit polyclonal	Abcam	Cat# ab290; RRID: AB_2313768
Human/Mouse Myeloperoxidase/MPO Antibody	R&D Systems	Cat# AF3667; RRID: AB_2250866
TruStain FcX™ (anti-mouse CD16/32) Antibody (clone 93)	Biolegend	Cat# 101320; RRID: AB_1574975
SOX2 Monoclonal Antibody (clone Btjce)	Thermo Fisher	Cat# 14-9811-82; RRID: AB_11219471
Mouse CD74 Antibody	R&D Systems	Cat# AF7478
anti-mouse I-A/I-E (M5/114.15.2)	Biolegend	Cat# 107602; RRID: AB_313317
Phospho-Stat1 (Ser727) Antibody	Cell Signaling Technology	Cat# #9177; RRID: AB_2197983
Phospho-Stat1 (Tyr701) (58D6) Rabbit mAb	Cell Signaling Technology	Cat# 9167; RRID: AB_561284
Stat1 Antibody	Cell Signaling Technology	Cat# 9172; RRID: AB_2198300
Phospho-Jak2 (Tyr1008) (D4A8) Rabbit mAb	Cell Signaling Technology	Cat# 8082; RRID: AB_10949104
Jak2 (D2E12) XP® Rabbit mAb	Cell Signaling Technology	Cat# 3230; RRID: AB_2128522
β-Actin (13E5) Rabbit mAb	Cell Signaling Technology	Cat# 4970; RRID: AB_2223172
Phospho-Jak1(Tyr1034/1035) (D7N4Z) Rabbit mAb	Cell Signaling Technology	Cat# 74129; RRID: AB_2799851
Jak1 Antibody	Cell Signaling Technology	Cat# 3332; RRID: AB_2128499
anti-mouse LY6G	bioxcell	Cat# BE0075-1; RRID: AB_1107721
rat IgG2a isotype control	bioxcell	Cat# BE0089; RRID: AB_1107769
Anti-pan Cytokeratin Antibody (clone KRT/1877R)	Abcam	Cat# ab234297; RRID: AB_2895302
Cytokeratin, pan Antibody (clone AE-1/AE-3)	Novus Biologicals	Cat# NBP2-29429; RRID: AB_3068002
HLA-DR Monoclonal Antibody (clone LN3)	Thermo Fisher Scientific	Cat# 14-9956-82; RRID: AB_468639
Anti-CD4 antibody (clone EPR6855)	Abcam	Cat# ab133616; RRID: AB_2750883
Purified anti-mouse CD140a Antibody (clone APA5)	BioLegend	Cat# 135901; RRID: AB_1953327
Purified anti-mouse CD31 (clone MEC13.3)	BioLegend	Cat# 102502; RRID: AB_312909
CD45 Monoclonal Antibody (clone D3F8Q)	Cell Signaling Technology	Cat# 70257; RRID: AB_2799780
Biotin anti-mouse CD64 (FcγRI) Antibody	BioLegend	Cat# 139318; RRID: AB_2566557
PE anti-mouse IL-2 Antibody (clone JES6-5H4)	BioLegend	Cat# 503807; RRID: AB_315301
Biotin anti-mouse/human CD11b Antibody	Biolegend	Cat# 101204; RRID: AB_312787
Bacterial and virus strains
Biological samples
Chemicals, peptides, and recombinant proteins
Tamoxifen	Sigma	Cat# T5648
Diptheria toxin	Sigma	Cat# D0564
Cultrex BME, Type 3 gels	Bio-Techne	Cat# 3632-005-02
arachidonic acid	MP BIOMEDICALS	Cat# 1713889
beta-mercaptoethanol	GIBCO	Cat# 21985023
insulin	Sigma	Cat# I6634
Hydrocortisone	Sigma	Cat# H0135
Epidermal Growth Factor (EGF), human recombinant	Sigma	Cat# 01-107
Recombinant Mouse GM-CSF	Biolegend	Cat# 576306
Recombinant Mouse IFN-α	Biolegend	Cat# 752802
Recombinant Mouse IFN-γ	Biolegend	Cat# 575302
RNAscope Probe-Mm-Gbp2-C2	Advanced Cell Diagnostics	Cat# 572491-C2
Recombinant Mouse IL-2 Protein	R&D Systems	Cat# 402-ML-100/CF
OVA 323-339	Invivogen	Cat# vac-isq
Aspirin	Cayman Chemical	Cat# 70260
Celecoxib	Selleckchem	Cat# S1261
DMBA	Sigma	Cat# D3254
TPA	Sigma	Cat# P8139-5MG
IHC Antigen Retrieval Solution - Low pH	Thermo Fisher Scientific	Cat# 00-4955-58
Prolong Gold Antifade Mountant	Thermo Fisher Scientific	Cat# P36930
Collagenase, Type IV	Gibco	Cat# 17104019
DNase I	Roche	Cat# 4536282001
Critical commercial assays
GeoMx WTA Starter Mm	Nanostring	Cat# 999064
Evercode™ WT v2	ParseBiosciences	Cat# ECW02030
Evercode™ Cell Fixation v2	ParseBiosciences	Cat# ECF2001
UDI Plate - WT	ParseBiosciences	Cat# UDI1001
NEBNext Single Cell/Low input RNA library prep kit for Illumina	New England Biolabs	Cat# E6420S
MojoSort Mouse Ly-6G Selection Kit	Biolegend	Cat# 480124
MojoSort Mouse CD4 NaiveT Cell Isolation Kit	Biolegend	Cat# 480040
RNAscope Multiplex Fluorescent Reagent Kit v2	Advanced Cell Diagnostics	Cat# 323100
BD Cytofix/Cytoperm™ Fixation/Permeabilization Kit	BD	Cat# 554714
Cell Stimulation Cocktail (plus protein transport inhibitors)	Thermo Fisher Scientific Inc.	Cat# 00-4975-03
NEBuilder® HiFi DNA Assembly Master Mix	New England Biolabs	Cat# E2621S
Lipofectamine™ 3000 Transfection Reagent	Thermo Fisher Scientific Inc.	Cat# L3000008
CUTANA™ ChIC/CUT&RUN Kit version 3	EpiCypher	Cat# 14-1048
NEBNext® Ultra™ II DNA Library Prep Kit for Illumina	New England Biolabs	Cat# E7645S
Deposited data
scRNAseq of CD45+ cells in skin SCC	This paper	GEO: GSE278435
NanoString GeoMx of skin SCC	This paper	GEO: GSE278436
Bulk RNAseq data of neutrophils in skin SCC	This paper	GEO: GSE278430
Bulk RNAseq data or SOX2Low and SOX2High SCC cells	This paper	GEO: GSE278429
SOX2 Cut and Run sequencing	This paper	GEO: GSE278426
Bulk RNAseq data or ITGA6HiCD80+ and ITGA6LowCD80− SCC cells	This paper	GEO: GSE303056
Spatial transcriptome of human HNSCC	Arora et al. 55	GEO: GSE208253
Spatial transcriptome of human CSCC	Ji et al.56	GEO: GSE144239
scRNAseq of neutrophils in grafted pancreatic cancer	Ng et al.24	GEO: GSE243466
scRNAseq of CD11b+ cells in grafted lung cancer	Gungabeesoon et al.31	GEO: GSE224399
scRNAseq of CD45+ cells in grafted colon cancer	Gungabeesoon et al.31	GEO: GSE224400
scRNAseq of immune cells from lung cancer patients treated with immune checkpoint blockade	Liu et al. 54	GEO: GSE243013
Experimental models: Cell lines
Mouse skin SCC PDVC57	Bremner et al.82	N/A
Mouse oral SCC MOC1	Kerafast	EWL001-FP
Mouse skin SCC PDV	Bremner et al. 82	N/A
Experimental models: Organisms/strains
Mouse: C57BL/6J	The Jackson Laboratory	000664
Mouse: C57BL/6N-Ifngr1tm1.1Rds/J (Ifngr1flox/flox)	The Jackson Laboratory	025394
Mouse: C57BL/6-Ly6g(tm2621(CretdTomato)Arte); R26-LSL-tdTomato (CatchupIVM-red)	Hasenberg et al.46	N/A
Mouse: C57BL/6-Ly6G-CreER; R26-LSL-tdTomato	Ballesteros et al.45	N/A
Mouse: B6.Cg-Tg(S100A8-cre,-EGFP)1Ilw/J	The Jackson Laboratory	021614
Mouse: C57BL/6-Gt(ROSA)26Sortm1(HBEGF)Awai/J (R26-LSL-DTR)	The Jackson Laboratory	007900
Mouse: B6.Tg(KRT14-cre/ERT)20Efu/J (K14CreER)	The Jackson Laboratory	005107
Mouse: B6.Sox2tm1.1Lan/J (Sox2flox/flox)	The Jackson Laboratory	013093
Mouse: B6(Cg)-Ifnar1tm1.1Ees/J (Ifnar1flox/flox)	The Jackson Laboratory	028256
Mouse: Lyz2Cre; Ptger2flox/flox; Ptger4flox/flox (Ptger2/Ptger4 dKO)	Thumkeo et al. 62	N/A
Oligonucleotides
sgSox2_F: CACCGATAAGTACACGCTTCCCGG	This Study	N/A
sgSox2_R: AAACCCGGGAAGCGTGTACTTATC	This Study	N/A
sgFadsI_F: CACCGAGGCCCATTCGCTCTACTG	This Study	N/A
sgFads1_R: AAACCAGTAGAGCGAATGGGCCTC	This Study	N/A
H2-Aa qPCR forward: GGAGGTGAAGACGACATTGAGG	This Study	N/A
H2-Aa qPCR reverse: CTCAGGAAGCATCCAGACAGTC	This Study	N/A
Gbp2 qPCR forward: CTGCAC TATGTGACGGAGCTA	This Study	N/A
Gbp2 qPCR reverse: GAGTCCACACAAAGGTTGGAAA	This Study	N/A
18S rRNA qPCR forward: CTTAGAGGGACAAGTGGCG	This Study	N/A
18S rRNA qPCR reverse: ACGCTGAGCCAGTCAGTGTA	This Study	N/A
Ifit3 qPCR forward: GTGGACTGAGATTTCTGAACTGC	This Study	N/A
Ifit3 qPCR reverse: GCTTCCAGAGATTCCCGGTT	This Study	N/A
Cd74 qPCR forward: CTCCATGGATGGCGTGAACT	This Study	N/A
Cd74 qPCR reverse: GTGGCTCTTTAGGTGGAGCC	This Study	N/A
Ptgs2 qPCR forward: GCGACATACTCAAGCAGGAGCA	This Study	N/A
Ptgs2 qPCR reverse: AGTGGTAACCGCTCAGGTGTTG	This Study	N/A
Recombinant DNA
lentiCRISPRv2	Sanjana et al.81	Addgene #52961
psPAX2	Trono Lab	Addgene #12260
pMD2.G	Trono Lab	Addgene #12259
pLKO-PGK-Sox2-IRES-GFP	This Study	N/A
pCDH-Fads1	This Study	N/A
pCDH-Sox9	This Study	N/A
pCDH-Klf5	This Study	N/A
pCDH-Gata3	This Study	N/A
pCDH-Grhl3	This Study	N/A
Software and algorithms
R_4.1.1	R Core	https://www.r-project.org/
Seurat_4.3.0	Stuart et al.77	https://github.com/satijalab/seurat/releases/tag/v4.3.0
scVelo_0.2.5	Bergen et al.79	https://scvelo.readthedocs.io/en/stable/
FlowJo	https://www.flowjo.com	N/A
Graphpad Prism v.9	GraphPad Prism	RRID: SCR_002798
Fuji (Image J)	https://fiji.sc/	N/A
spipe.v1.1.1	ParseBiosciences	https://www.parsebiosciences.com
SingleR V1.7.1	Aran et al. 78	https://bioconductor.org/packages/SingleR
ComplexHeatmap V2.15.1	Gu et al.83	https://bioconductor.org/packages/release/bioc/html/ComplexHeatmap.html
Cutadapt (v3.2)	Martin et al.73	https://cutadapt.readthedocs.io/en/v3.4/
Pseudo-aligner Kallisto (v0.44.0)	Bray et al.74	https://github.com/pachterlab/kallisto
DESeq2 R package (v1.30.0)	Love et al.80	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
clusterProfiler package (v4.6.0)	Yu et al.75	https://guangchuangyu.github.io/software/clusterProfiler/
MSigDB (Molecular signature database, v7.4)	Castanza et al.76	https://www.gsea-msigdb.org/gsea/msigdb
NanoStringNCTools (v1.12.0.)	Bioconductor	https://www.bioconductor.org/packages/release/bioc/html/NanoStringNCTools.html
GeomxTools (v3.8.0.)	Bioconductor	https://www.bioconductor.org/packages/release/bioc/html/GeomxTools.html
GeoMxWorkflows (v1.10.0)	Bioconductor	https://www.bioconductor.org/packages/release/workflows/html/GeoMxWorkflows.html
NanoString GeoMx NGS Pipeline v2.0	NanoString	N/A
Bowtie2 (v2.3.4.3)	Langmead and Salzberg	https://bowtie-bio.sourceforge.net/owtie2/index.shtml
Picard (v2.18.7)	Broad Institute	http://github.com/broadinstitute/picard/releases/tag/2.7.1
MACS2 (v2.2.7.1)	Zhang et al.84	https://pypi.org/project/MACS2/
deepTools (v3.5.6)	Ramírez et al.85	https://github.com/deeptools/deepTools
IGV	Robinson et al.86	https://igv.org
Other
BD LSR-Fortessa analyzer	BD Biosciences	N/A
BD Symphony S6 Cell Sorter	BD Biosciences	N/A

Highlights.

• Immunotherapy elicits distinct responses in different TAN subpopulations.

• IFNs reprogram most TANs in the stroma to restore their anti-tumor activities.

• SOX2^High tSCs impairs IFN responses in TANs at the tumor-stroma interface.

• tSCs produce AA to activate PGE₂ signaling, suppressing IFN responses in TANs.