Introduction

HNSCC represent a classification of malignant neoplasms that predominantly arise in the mucosal epithelium of anatomical sites including the pharynx, oral cavity, nasal cavity, and paranasal sinuses. The etiology of HNSCC is multifactorial, with significant associations to behaviors such as tobacco smoking and excessive alcohol consumption, as well as infection by the human papillomavirus (HPV). Globally, HNSCC is the sixth most diagnosed cancer, with an estimated 890,000 new cases and 450,000 mortalities occurring each year1,2. Despite notable advancements in treatment and diagnostic approaches, the overall effectiveness of managing HNSCC remains unsatisfactory, particularly in cases where the tumor has invaded and metastasized. Therefore, it is imperative to further investigate the factors that influence treatment response and prognosis in patients with advanced HNSCC.

In recent years, cancer-associated fibroblasts (CAFs) have gained significant attention as one of the most prominent cell populations in the tumor microenvironment (TME). These cells play a crucial role in interacting with diverse nonmalignant cells, exerting a profound impact on various aspects of cancer biology, including tumor progression, metastasis, and response to treatment3,4. In the HNSCC, CAFs are actively involved in remodeling the tumor stroma. Their actions contribute to creating an environment that enhances cancer invasion by modulating the composition of the extracellular matrix (ECM), facilitating angiogenesis, and inducing immunosuppression5. Furthermore, CAFs can enhance the malignant characteristics of HNSCC through the transfer of miRNAs via extracellular vesicles6. Although present methods have enabled the identification of different CAFs subtypes, targeting CAFs with single-molecule or single-category inhibitors alone still falls short of delivering satisfactory therapeutic outcomes, especially in advanced HNSCC7. To address this obstacle, it is imperative to develop a profound comprehension of the molecular and functional heterogeneity exhibited by CAFs subtypes. By employing a range of multi-omics methodologies, we can delve into the unique attributes and roles played by different CAFs subtypes. Moreover, exploring interventions aimed at disrupting the intercellular communication and crosstalk between CAFs and other constituents of TME can effectively impede their pro-tumorigenic effects8.

In this study, we have made a groundbreaking discovery of a novel subtype of CAFs named SFRP2_CAFs through an extensive analysis of multicenter cohorts and a large volume of single-cell specimens obtained from different researches. Our findings demonstrate that this subtype is associated with poor overall survival and exhibits characteristics of both myofibroblastic CAFs and inflammatory CAFs (myCAFs and iCAFs) with these features being particularly pronounced in individuals at metastatic site of the HNSCC. Furthermore, we have successfully identified the origin of this subtype and unraveled the underlying mechanisms governing tumor invasion and metastasis. By constructing a novel gene regulatory network, we have also revealed key transcription factors involved in the regulation of this cellular subtype. Targeting SFRP2_CAFs will presents a promising therapeutic approach for effectively treating patients afflicted with advanced HNSCC.

Methods

Cell culture

THP-1 (human acute monocytic leukemia cell line), SCC25 (human head and neck squamous cell carcinoma cell line) and HFL-1 (human lung fibroblasts cell line) were obtained from National Collection of Authenticated Cell Cultures. THP-1 was incubated in the 1640 medium containing 0% FBS, 1% antibiotic-antimycotic solution, and 0.05 mM β- mercaptoethanol during the experiment. SCC25 was incubated in the DMEM/F12 medium containing 10% FBS and 1% antibiotic-antimycotic solution during the experiment. HFL-1 was incubated in the F12K medium containing 10% FBS and 1% antibiotic-antimycotic solution during the experiment.

Cell induction

We utilize THP-1 cells to induce TAMs and HFL-1 cells to induce CAFs. After removing the floating cells from the culture medium, we add the conditioned medium from treated SCC25 cells separately to THP-1 and HFL-1 cells. Following a 72 h incubation period, THP-1 cells will be induced into TAMs, while HFL-1 cells will be induced into CAFs.

In vitro cell metastasis, proliferation, and wound healing assays

For Transwell assay, cells were seeded in transwell inserts at a density of 10,000 cells per insert (three replicates per group) and incubated for 96 h. Crystal violet staining was employed to label the cells, under an inverted microscope, observe the cells on the upper chamber surface, capturing three fields of view from each sample. For CCK8 assay, the transfected cells were plated in 96-well plates at a density of 5000 cells per well (three replicates per group). The optical density (OD) value at 450 nm was measured using a CCK8 assay kit after 48 h, 72 h, and 96 h of incubation. For wound healing assay, 5× 105 cells/well (three replicates per group) were plated into a 6-well plate and incubated to reach confluence. The monolayer was scratched using a tip and washed with serum-free medium to remove detached cells. cells were photographed at 0 h, 24 h and 48 h later, capturing three fields of view from each sample.

Quantitative real‑time PCR assay

To perform reverse transcription, 2 μg of total RNA was utilized to synthesize cDNA using a cDNA Synthesis Kit. β-actin served as the internal control. The CCL2 forward sequence of primer was 5’-CAGCCAGATGCAATCAATGCC-3’, and reverse sequence of primer was 5’-TGGAATCCTGAACCCACTTCT-3’. The CD163 forward sequence of primer was 5’-GACGCATTTGGATGGATCATGT-3’, and reverse sequence of primer was 5’-CCCACCGTCCTTGGAATTTGA-3’. The NFIX forward sequence of primer was 5’-ATGCGGACATCAAACCACT-3’, and reverse sequence of primer was 5’-ATACTCTCACCAGCTCCGTCA-3’. The ACTA2 forward sequence of primer was 5’-AAAAGACAGCTACGTGGGTGA-3’, and reverse sequence of primer was 5’-GCCATGTTCTATCGGGTACTTC-3’.

Western blotting

After cell protein extraction, the samples were separated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto a polyvinylidene difluoride (PVDF) membrane. The PVDF membrane was blocked with 5% milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) at room temperature for 1 h. Subsequently, the membrane was incubated overnight at 4 °C with the primary antibody. The antibodies including MIF (1:2000, Cell Signaling Technology, USA), β-actin (1:1000, Cell Signaling Technology, USA), NF-kB (1:2000, Cell Signaling Technology, USA), BCL2 (1:2000, Cell Signaling Technology, USA) and CCL2 (1:2000, Cell Signaling Technology, USA). The bands signal was visualized by luminescent image analyzer (Bio-Rad, USA).

Bulk sequence and single-cell sequence databases pre-processing

STAR9 and CellRanger (version 7.1.0, https://www.10xgenomics.com/) methods were used to process fastq.gz format from bulk RNA and single cell RNA sequence datasets, respectively. The R package “combat-seq”10 was utilized to remove batch effects during the integration of meta cohorts (including TCGA-HNSC, GSE41613, and GSE65858). For single-cell analysis, we employed the R package “Seurat”11 and Python package “scanpy”12. The following criteria were applied to filter out unwanted cells during the analysis: (1), Cells < 500 were considered low quality and removed. (2), Cells detected genes higher than 4000 were considered potential doublets. (3), Cells with more than 10% mitochondrial genes were removed as they indicate poor cellular health. (4), Cells expressing both tumor and immune markers were identified as doublets and excluded from the analysis. The parameter of Find Variable Features in Seurat pipeline was set to 2000. Cell cluster visualization was used in Uniform Manifold Approximation and Projection (UMAP) coordinate. Harmony method13 was performed to remove single cell batch effects. The resolution for Find Clusters in Seurat was depended on R package clustree14 analysis results. FindAllMarkers function from the Seurat pipline to identify highly expressed genes among various cell clusters. A total of 702446 cells were included in this study.

Survival analysis

To evaluate the difference in overall survival (OS) between high-expression and low-expression subgroups, Kaplan–Meier survival curves and log-rank tests were employed. A multivariate Cox regression model was utilized to determine the prognostic fibroblast subtypes in HNSCC patients.

Analyses for the genomic data

The Wilcoxon rank-sum test was used to identify differentially expressed genes between the relative groups. Following the identification of differentially expressed genes, Gene Set Enrichment Analysis (GSEA) was conducted using the clusterProfiler15 R package.

Statistical analysis

The Operating system was Linux. All computational and statistical analyses were conducted using Python-3.8.17 (https://www.python.org/), R-4.2.2 (https://www.r-project.org/), and GraphPad Prism 9.4 software. P-value less than 0.05 was considered statistically significant, indicating a significant difference or association in the analyzed data.

Result

SFRP2_CAFs are commonly found in HNSCC patients and contribute to the poor clinical outcome

We collected pairs of HNSCC tissues from the Gene Expression Omnibus (GEO) single cell database GSE181919. These paired samples consist of primary tumors and corresponding metastatic lesions. To partition suitable cell clusters, we applied the R package clustree14. The results revealed the presence of 21 stable cell clusters within this database. By identifying highly expressed genes and utilizing the CellMarker 2.0 and scTYPE database16,17, we have successfully annotated the 21 cell clusters into 20 distinct cell types, they are T exhausted cells, CD8 + NK T-like cells, Plasma Dendritic cells, T regulator cells, Basal cells, T prolife cells, SPP1_Tumor associated macrophages (TAMs) and SFRP2_CAFs et al. (Fig. 1A). In another single-cell dataset GSE188737 involving corresponding paired samples of primary tumors and metastatic lesions, we identified 17 cell types using the same approach. In contrast to the cell types annotated in GSE181919, the analysis of GSE188737 has yielded the additional annotation of neutrophils, smooth muscle cells, and fibroblasts (Fig. 1B). During comprehensive analysis of various HNSCC single-cell datasets, we successfully annotated the presence of the cell type “SFRP2_CAFs” across all identified cell clusters (Supplementary Fig. 1A–D). Notably, our investigation uncovered an intriguing phenomenon where the marker ACTA2 for myCAFs and the marker IL-11 for iCAFs18 coexisted within the SFRP2_CAFs population (Supplementary Fig. 1E–H) in metastatic lesions. However, this coexistence was weak in the non-metastatic single-cell dataset GSE215403 and GSE164690 (Supplementary Fig. 1I–J). This observation indicates that SFRP2_CAFs are consistently present in both primary and metastatic lesions of HNSCC, emphasizing their uniform expression pattern during the progression of HNSCC and the characteristics of metastatic lesions differ from those of primary lesions, which may lead to potential functional differences. Next, we aim to explore the cellular origin of SFRP2_CAFs. Given the accessibility of raw data and UMAP plots of cell types, we have selected GSE188737 to investigate the origin of SFRP2_CAFs. First, we used the Palantir method19 to identify a specific starting cell based on the lowest expression level of SFRP2 (blue). This allowed us to unravel the transcriptional differentiation of SFRP2_CAFs. Through this approach, we identified the terminal cells (highlighted in red) predicted to represent the stage of the transition from smooth muscle cells to fibroblasts, progressing towards SFRP2_CAFs (Fig. 1C). Then, we utilized RNA velocyto and scvelo analysis20,21 to compute RNA velocity. The resulting vector field and extrapolated pseudotime confirmed a trajectory starting within the smooth muscle and terminating in the SFRP2_CAFs (Fig. 1D). Employing the same methodology as described above, we similarly concluded, in GSE215403, that SFRP2_CAFs originate from smooth muscle cells (Supplementary Fig. 1K, L).

Fig. 1: SFRP2_CAFs contribute to the poor clinical outcome of HNSCC.
figure 1

A In the study of HNSCC single-cell dataset GSE181919, 20 distinct cell types were identified. B In another HNSCC single-cell dataset GSE188737, 17 distinct cell types were identified. C UMAP representation highlighting the starting cell (blue) with the lowest SFRP2 expression (left), the terminal cells (red) as predicted by Palantir in GSE188737 (median) and cell type annotation (right). D Velocities derived from the dynamical model for GSE188737 HNSCC single cell dataset are visualized as streamlines in a UMAP-based embedding. E Seven Ecotyper-Fibroblast subtypes were identified, with three subtypes significantly associated with multivariate cox analysis (Fibroblast_S03, p = 0.02124; Fibroblast_S01, p = 0.00455; and Fibroblast_S08, p = 0.00333). F–H Correlation analysis revealed a significant association between SFRP2, SFRP2_CAFs, and Ecotyper-Fibroblasts_S03 (R = 0.75, p < 2.2e−16, R = 0.823, p < 2.2e−16, R = 0.809, p < 2.2e−16). E Three subtypes significantly associated with multivariate cox analysis (Fibroblast_S03, p = 0.02124; Fibroblast_S01, p = 0.00455; and Fibroblast_S08, p = 0.00333). I–L Elevated expression of SFRP2 was consistently associated with poor overall survival across different datasets (meta; p = 0.019, HR = 0.75, 95% CI = 0.6–0.96. TCGA-HNSC; p = 0.002, HR = 0.56, 95% CI = 0.38–0.81. GSE65858; p = 0.035, HR = 0.72, 95% CI = 0.53–0.98. GSE41613; p = 0.022, HR = 0.28, 95% CI = 0.09–0.9). M Gene Set Enrichment Analysis (GSEA) analysis revealed high expression of SFRP2 associated with myogenesis, angiogenesis, and epithelial-mesenchymal transition pathways activated.

Full size image

Subsequently, we employed SFRP2_CAFs genes list to analyze a meta of bulk mRNA-sequencing data derived from 886 patients diagnosed with HNSCC, which was acquired by aggregating samples from three distinct cohorts, namely TCGA-HNSC, GSE65858 and GSE41613. Ecotyper demonstrates a remarkable ability to accurately predict the cellular composition of tumor microenvironmental cells within bulk tissues22. Leveraging this deconvolution approach, we revealed the presence of seven distinct fibroblast subtypes (Fig. S1M). Subsequently, a comprehensive multivariate Cox analysis was conducted, incorporating all seven fibroblast cellular compartments to assess their impact on survival. We found only the signatures of Fibroblast_S03 (p = 0.02124), Fibroblast_S01 (p = 0.00455), and Fibroblast_S08 (p = 0.00333) demonstrated prognostic significance (Fig. 1E). which suggest a candidate involvement of these fibroblast types in disease progression. Remarkably, a substantially significant correlation was observed among the expression levels of SFRP2, SFRP2_CAFs, and Fibroblast_S03 (Fig. 1F–H, R = 0.75, p < 2.2e−16; R = 0.823, p < 2.2e−16; R = 0.809, p < 2.2e−16). This finding highlights the ability of SFRP2 to quantify CAFs cells within the microenvironment of HNSCC. Because SFRP2 serves as a reliable marker for representing the expression of CAFs, we then investigated the clinical significance of SFRP2 across multiple cohorts, including a meta of bulk mRNA-sequencing data, TCGA-HNSC, GSE65858 and GSE41613. Kaplan-Meier curve revealed that elevated expression of SFRP2 is indicative of poor overall survival (Fig. 1I–L, meta; p = 0.019, HR = 0.75, 95% CI = 0.6–0.96. TCGA-HNSC; p = 0.002, HR = 0.56, 95% CI = 0.38–0.81. GSE65858; p = 0.035, HR = 0.72, 95% CI = 0.53–0.98. GSE41613; p = 0.022, HR = 0.28, 95% CI = 0.09–0.9). These results collectively indicate that SFRP2_CAFs is associated with unfavorable prognosis across multiple distinct HNSCC cohorts. Additionally, we partitioned the meta cohort into two groups, SFRP2_High and SFRP2_Low based on the median expression level of SFRP2. The GSEA results showed that, when compared to the SFRP2_Low group, the SFRP2_High group demonstrated significant activation in pathways associated with myogenesis, angiogenesis, and epithelial-mesenchymal transition (EMT) (Fig. 1M).

Cross-talk between SFRP2_CAFs and SPP1_TAMs are major contributors to the tumor microenvironment of HNSCC metastatic lesions

Next, we aim to unravel the intricate complexity of SFRP2_CAFs in the tumor microenvironment of HNSCC metastatic lesions. In GSE181919, we assessed the interactions between cell types using CellChat23, a tool for estimating ligand-receptor interaction strength. We identified all the pairs of interacting cell types and quantified the strength of each interaction (Fig. 2A). It is evident that the interaction between SFRP2_CAFs and SPP1_TAMs is the strongest, followed by the interaction between SFRP2_CAFs and endothelial cells. To gain a deeper understanding of the cell type interactions, we constructed a directed cell-network based on the interaction matrix. The analysis unveiled a hierarchical interaction network, where SFRP2_CAFs occupy the max counts position for transmitting signals to other cell types while SPP1_TAMs are situated at the most counts position for receiving signals from both CAFs and other cell types (Fig. 2B, C). In the GSE188737 scRNA-seq dataset, we also observed that SFRP2_CAFs exhibit the highest degree of cellular interactions, including the interactions with SPP1_TAMs (Fig. 2D). As we anticipated, the same results of cell-network interaction demonstrate that SFRP2_CAFs occupy the predominant position in transmitting signals to other cell types and SPP1_TAMs are located at the highest count position for receiving signals from both CAFs and other cell types (Fig. 2E, F). Subsequently, we were interested in investigating the potential co-localization relationship between SFRP2_CAFs and SPP1_TAMs in the spatial dimension. To explore this, we conducted an analysis of the GSE181300 spatial transcriptome dataset. Leveraging the genes sets defined by SFRP2_CAFs and SPP1_TAMs, the RCTD deconvolution24 results indicate a noticeable adjacency or co-localization niches between SFRP2_CAFs and SPP1_TAMs within the invasive situ of the tumor. Moreover, these cell types were observed surrounding the tumor cells (Fig. 2G–I). However, such interactions between SFRP2_CAFs and SPP1_TAMs niches were less observed within the inner regions of the tumor (Fig. 2J–L). Based on the ligand-receptor analysis results, SFRP2_CAFs and SPP1_TAMs primarily establish communication through the MIF-CD74 pathway (Fig. 2M, N).

Fig. 2: Cross-talk between SFRP2_CAFs and SPP1_TAMs are major contributors to the tumor microenvironment of HNSCC metastatic lesions.
figure 2

A–C In GSE181919, cell communication revealed that SFRP2_CAFs and SPP1_TAMs play a dominant role in the tumor microenvironment of HNSCC metastases (the “communication score” between SFRP2_CAFs and SPP1_TAMs = 42, much higher than the “communication score” between any other pair of cells). Moreover, SFRP2_CAFs are the most active cell type in signaling within the entire microenvironment, while SPP1_TAMs are the most responsive cell type. D–F The same phenomenon was observed in single-cell data from GSE188737. G–I Spatial plots showed cell abundance and location of selected cell types estimated by RCTD deconvolution in invasive regions of the tumor. They are SFRP2_CAFs, SPP1_TAMs and Cancer cells. J–L Spatial plots showed cell abundance and location of selected cell types estimated by RCTD deconvolution in the inner regions of the tumor. M, N The communication between SFRP2_CAFs and SPP1_TAMs are primarily mediated through the MIF-CD74 ligand-receptor interaction in both GSE181919 and GSE188737.

Full size image

We further examined whether consistent outcomes were observed across two additional single-cell datasets (GSE173468 and GSE234933) pertaining to metastatic lesions of HNSCC. As expected, both single-cell datasets revealed that the communication between SFRP2_CAFs and SPP1_TAMs predominantly shape the tumor microenvironment, and we identified MIF-CD74 interaction as a key nexus between these two cell types (Supplementary Fig. 2A–D). It is noteworthy that in non-metastatic single cell dataset of HNSCC (GSE164690 and GSE215403), the communication between SFRP2_CAFs and Endothelial cells constitutes the main cellular crosstalk within the tumor microenvironment (Supplementary Fig. 2E, F). Despite the interaction between SFRP2_CAFs and SPP1_TAMs do not represent the strongest cell-cell crosstalk in non-metastatic lesions, it is significant to note that these cell types primarily interact through the MIF-CD74 ligand-receptor axis (Supplementary Fig. 2G, H). The above-mentioned results suggest that the augmented communication between MIF-CD74-mediated SFRP2_CAFs and SPP1_TAMs in metastatic lesions of HNSCC might be a pivotal factor in tumor invasion and metastasis. To eliminate the potential confounding effects of HPV status on the interaction between SFRP2_CAFs and SPP1_TAMs, we re-analyzed the dataset GSE234933, which includes patients with both HPV-negative and HPV-positive status. Utilizing the Milo method25, our findings revealed no significant differences in the cellular abundance of SFRP2_CAFs and SPP1_TAMs between HPV negative and HPV positive samples (Supplementary Fig. 2I). Correspondingly, B-cell abundance was significantly elevated in the HPV positive group, aligning with prior knowledge26,27. Furthermore, we continued to explore the expression programs of SFRP2_CAFs and SPP1_TAMs in HPV negative and HPV positive populations, followed by ssGSEA28 and correlation analysis. The results indicated a uniformity in the expression programs of these cell types across HPV negative and HPV positive HNSCC (Supplementary Fig. 2J, K).

SFRP2_CAFs secrete chemokine CCL2 to recruit SPP1_TAMs

Given the strong degree of communication between SFRP2_CAFs and SPP1_TAMs, it is crucial to delve into the mechanisms underlying their interaction. We aim to explore whether there exists a signaling factor or molecule transmitted or secreted by SFRP2_CAFs that not only recruits SPP1_TAMs but also enhances metastasis or invasion in HNSCC. Geneformer is a context-aware, attention-based deep learning model that comprises a large-scale pretraining library of 30 million human single-cell transcriptomes from various tissues29. Leveraging this extensive human single-cell deep learning pretraining library enables us to simulate the effects of gene perturbations in SFRP2_CAFs within the HNSCC single-cell dataset to drive cells from a metastatic state to a primary state. Initially, we differentiate between metastatic and primary cells in HNSCC using a fine-tuned model (Supplementary Fig. 3A–F). Due to the continued challenge of distinguishing SFRP2_CAFs between primary and metastatic sites even after fine-tuned in GSE181919 (Supplementary Fig. 3D), we proceeded with silico gene knockout on GSE188737 and GSE234933. Our hypothesis was that genes capable of triggering a shift from a metastatic to a primary state upon knockout could serve as potential targets driving invasion or metastasis in HNSCC (Fig. 3A). In GSE188737, we identified 402 potential targets that could facilitate the transition of HNSCC from a metastatic to a primary state through silico knockout. Similarly, in GSE234933, we discovered 430 potential targets. Taking the intersection of these sets, we obtained a total of 80 commonly expressed loci. Further refinement involved intersecting these 80 shared loci with genes associated with SFRP2_CAFs, leading to the identification of 9 genes: RARRES2, MT1E, AQP1, TGFBR2, CTNNB1, C1S, LAMP1, A2M, and CCL2 (Fig. 3B, C). Among these factors, we are particularly interested in CCL2, also known as monocyte chemoattractant protein-1 (MCP-1), which is a well-established chemokine involved in the recruitment of macrophages and other mononuclear cells to sites of inflammation or tumorigenesis30,31. We ask whether SFRP2_CAFs, as the primary cell type in transmitting signals to others, can recruit SPP1_TAMs by secreting the chemokine CCL2. We performed GSEA on the datasets GSE181919 and GSE188737 to gain insights into SFRP2_CAFs biological function and observed a significant activation of cytokine-cytokine-receptor signaling pathways in both datasets, among them, CCL2 exhibited a significant positive regulatory effect (Fig. 3D, GSE181919, p = 0.0009481; Fig. 3F, GSE188737, p = 8.994e−05). In GSE234933, we also found that CCL2 exhibited the same regulatory effect in SFRP2_CAFs (Supplementary Fig. 3G). Furthermore, in our analysis of GSE181919, we found that CCL2 exhibited the highest expression in SFRP2_CAFs (Fig. 3E). In GSE188737, SFRP2_CAFs were ranked as the second highest in CCL2 expression across all cell types (Fig. 3G). Although smooth muscle cells showed the highest expression of CCL2 in GSE188737, our previous analysis revealed that as the tumor progresses, smooth muscle cells undergo a transformation into SFRP2_CAFs (Fig. 1C, D and Supplementary Fig. 1K, L).

Fig. 3: SFRP2_CAFs secrete chemokine CCL2 to recruit SPP1_TAMs.
figure 3

A Using the Geneformer fine-tuned module to distinguish between metastatic and primary cells in HNSCC. The embedding positions of cellular states are defined. Subsequently, silico deletion or activation of random genes is carried out to determine which gene alterations would cause cellular states from metastatic site (gray cloud) to lean towards the primary site (green cloud). B After silico knockout, cellular embedding shows a negative value indicating movement away from the metastatic foci state and a positive value indicates movement towards the primary foci state. The positions of RARRES2, MT1E, AQP1, TGFBR2, CTNNB1, C1S, LAMP1, A2M, and CCL2 among the 402 silico targets in GSE188737. C The positions of RARRES2, MT1E, AQP1, TGFBR2, CTNNB1, C1S, LAMP1, A2M, and CCL2 among the 430 silico targets in GSE234933. D CCL2 exhibited a significant positive regulatory effect of cytokine-cytokine-receptor signaling pathways in SFRP2_CAFs, GSE181919, p = 0.00009481. E CCL2 expression in all cell types of GSE181919. F CCL2 exhibited a significant positive regulatory effect of cytokine-cytokine-receptor signaling pathways in SFRP2_CAFs, GSE188737, p = 8.944e−05. G CCL2 expression in all cell types of GSE188737. H, I CCL2 expression in SFRP2_CAFs and SPP1_TAMs within GSE181919 and GSE188737. J CCL2 expression in CA (the primary tumor site) and LN (the metastatic tumor site) within GSE181919. K CCL2 expression in M (the metastatic tumor site) and P (the primary tumor site) within GSE188737. L, M Bar graph displaying the expression of ACTA2 and CCL2 in HFL-1 and CAFs. N Bar graph displaying the expression of CD163 in THP-1 and TAMs. O (Left) The transwell assay results of the TAM recruitment experiment showed CAFs could secrete CCL2 to enhance the recruitment of TAMs. (Right) The cell counting for transwell assay. P value = (* <0.05; ** <0.01; *** <0.001, **** <0.0001).

Full size image

We propose that CCL2 serves a biomarker for SFRP2_CAFs, challenging the previous notion that it is solely a specific marker for tumor-associated macrophages32,33,34. To validate this hypothesis, we performed t-tests analysis. The results revealed that SPP1_TAMs exhibited minimal expression of CCL2 comparison to SFRP2_CAFs (Fig. 3H, GSE181919, p <0.001; Fig. 3I, GSE188737, p <0.001; Supplementary Fig. 3H, GSE234933, p <0.001). We then compared the expression of CCL2 within SFRP2_CAFs in metastatic lesions and their corresponding primary lesions. Our findings revealed significantly higher CCL2 expression in the metastatic lesions compared to the primary lesions (Fig. 3J, GSE181919, p <0.01; Fig. 3K, GSE188737, p <0.001; Supplementary Fig. 3I, GSE234933, p <0.001). These findings indicate that the secretion of CCL2 by SFRP2_CAFs plays a crucial role in recruiting SPP1_TAMs and may contribute to the invasion and metastasis of HNSCC. Transwell experiments were conducted to validate our computational analysis. We replicated the biological interaction of SPP1_TAMs and SFRP2_CAFs using THP-1 cell-induced TAMs and HFL-1 cell-induced CAFs (Fig. 3L, N). We observed that when TAMs were exposed to the conditioned medium of CAFs cell culture, there was an enhancement in TAM recruitment. Recombinant CCL2 protein further intensified this recruitment effect (Fig. 3O, p <0.05, p <0.01).

CCL2 enhances the interaction between MIF-CD74 ligands-receptors present on CAFs and TAMs, result in increasing the invasive and metastatic abilities of HNSCC

We then sought to ask whether the secretion of CCL2 by CAFs would have an impact on the ligand-receptor interaction of MIF-CD74. The Western blotting results revealed that the addition of CAFs conditioned medium to TAMs led to increased expression of MIF and CCL2 was found to enhance this signature to a greater extent (Fig. 4A). In our next investigation, we explored the relationship between CD74, SFRP2_CAFs, and SPP1_TAMs. Based on SPP1_TAMs median value acquired from Gene Set Variation Analysis (GSVA) method and SFRP2 median expression, we divided the meta cohorts into two groups: high and low expression. We observed significant differences in the abundance of CD74 between the high and low expression groups of SFRP2 as well as the high and low expression groups of SPP1_TAMs (Fig. 4D, p = 1.25e−03; Fig. 4E, p = 1.32e−83). Moreover, correlation analysis revealed a strong association between CD74 and both SFRP2_CAFs and SPP1_TAMs (Fig. 4B, R = 0.147, p = 1.3e−05; Fig. 4C, R = 0.721, p < 2.2e−16). These findings suggest that CD74 may exhibit shared characteristics with cellular states of SFRP2_CAFs and SPP1_TAMs. Scvelo can rank genes based on degree of fitting to determine key genes between sequential cell populations, hence we utilized RNA velocyto and scvelo20,21 to compute RNA velocity. We place particular emphasis on exploring the temporal dynamics of CD74 within SFRP2_CAFs and SPP1_TAMs in the tumor microenvironment. The obtained results clearly demonstrate the temporal dynamics of the marker CCL2 associated with CD74 (Fig. 4F), besides, a noticeable resemblance in the velocity kinetics between CD74 and the specific markers of SPP1_TAMs are observed (Fig. 4G). Subsequently, we aimed to ask whether the temporal dynamics rate of change in CD74 were consistent across all cell types within the HNSCC tumor microenvironment. To investigate this, we conducted a velocity relevance analysis of CD74. The findings highlight a distinctive association between the kinetics of CD74 specifically in only SFRP2_CAFs and SPP1_TAMs, compared to all other cell types (Fig. 4H).

Fig. 4: CCL2 enhances the interaction between MIF-CD74 receptor ligands present on CAFs and TAMs, thereby increasing tumor-promoting abilities of HNSCC.
figure 4

A (left) The western blotting results of MIF across three groups. (right) The graphical representation of the western blotting results. B The correlation (R = 0.147, p = 1.3e−05) between CD74 and SFRP2 in meta cohort. C The correlation (R = 0.721, p < 2.2e−16) between CD74 and SPP1_TAMs in meta cohort. D CD74 expression in the meta cohort’s high and low SFRP2 expression groups (p = 1.25e−03). E CD74 expression in the meta cohort’s high and low SPP1_TAMs expression groups (p = 1.32e−83). F CCL2 and CD74 as SFRP2_CAFs putative driver genes (left) are identified by high likelihoods. Phase portraits of temporal dynamics along latent time (median) and expression (right) for these driver genes characterize their activity. G SPP1 and CD74 as SPP1_TAMs putative driver genes (left) are identified by high likelihoods. Phase portraits of temporal dynamics along latent time (median) and expression (right) for these driver genes characterize their activity. H CD74 kinetics in all cell types showed only a unique and exclusive similarity between SFRP2_CAFs and SPP1_TAMs. I (left) The transwell assay results showed remarkable enhancement in the invasive capacity of HNSCC cells when co-cultured with CAFs and TAMs. J The cell counting for transwell assay. K The wound healing assay results showed remarkable variations in the migration capacity of HNSCC cells when co-cultured with CAFs and TAMs. L The wounding area for wound healing assay. M The CCK8 assay results showed remarkable enhancement in the cytotoxicity capacity of HNSCC cells when co-cultured with CAFs and TAMs. N NF-κB exhibited a significant positive regulatory effect of immune response signaling pathways in TAMs, p = 3.959e−06. O The western blotting results of NF-κB and BCL2 across three groups. P The Elisa results of CXCL8 across three groups. P value = (* <0.05; ** <0.01; *** <0.001, **** <0.0001).

Full size image

Building upon the fact that CAFs are capable of secreting CCL2 to recruit TAMs, we explored the phenotypic changes of tumor cells when co-cultured with both CAFs and TAMs. The transwell experiment results demonstrated a remarkable enhancement in the invasive capacity of HNSCC cells when co-cultured with CAFs and TAMs ((Fig. 4I, J), p <0.01, p <0.001) and the cells’ cytotoxicity enhancement was further observed in CCK8 assay when co-cultured with CAFs and TAMs (Fig. 4M). In the wound healing assay conducted at three-time intervals (0 h, 24 h, and 48 h), our findings demonstrated that co-culturing with CAFs and TAMs resulted in the most substantial closure of the wound, indicating an enhancement of tumor cell migration ability in the presence of both CAFs and TAMs (Fig. 4K, L). In HNSCC, TAMs frequently engage in an immune cascade reaction. The activation of this signaling pathway is positively modulated by the nuclear transcription factor NF-kB, which also serves as an upstream regulator of the classical anti-apoptotic factor BCL235,36. In this study, a similar immune response phenomenon was observed in TAMs (Fig. 4N, p = 3.959e−06). Consequently, we employed western blot analysis to quantify the expression levels of NF-κB and BCL2 in TAMs. The results revealed a significant increase in the expression levels of NF-κB and BCL2 in TAMs under the influence of CCL2 (Fig. 4O). Moreover, we seek to explore whether the inflammatory factors controlled by NF-κB in TAMs can impact the extracellular environment. Previous studies have indicated that the NF-κB signaling pathway in HNSCC can stimulate the transcriptional activation of CXCL8 hence induce immune suppression37,38. Our Elisa results confirm a substantial elevation in the levels of CXCL8 in the extracellular milieu due to the interaction between CAFs and TAMs (Fig. 4P). These collective outcomes highlight that CAFs secrete CCL2, intensifying communication between CAFs and TAMs, thereby enhancing tumor cell invasion and migration capacities, cytotoxic effects, inflammatory responses, along with anti-apoptotic benefits.

Inhibiting CCL2 leads to a decrease in MIF expression and attenuates the invasive and metastatic abilities of HNSCC

It has been established that CAFs secrete CCL2 to recruit TAMs and augment the cellular communication facilitated by MIF-CD74 between CAFs and TAMs. Given these compelling observations, our next objective is to study whether inhibiting CCL2 can reverse the observed phenotypic effects. We employed the CCL2 inhibitor MCE, the principal component of which is Emapticap pegol (NOX-E36)39, to investigate the experimental outcomes. In the transwell assay, our findings demonstrated that MCE effectively mitigates the recruitment of TAMs induced by the secretion of CCL2 from CAFs (Fig. 5A). Similarly, in the group treated with MCE, a significant decrease was observed in the expression level of MIF (Fig. 5B). In the transwell assay conducted on tumor cells, treatment with the CCL2 inhibitor MCE successfully reverses the enhanced invasive effect of tumor cells following co-culture with CAFs and TAMs (Fig. 5C). We also observed that MCE reduces the heightened cytotoxic effects of CAFs and TAMs co-culture on HNSCC cells in the CCK8 assay (Fig. 5D). Furthermore, in the wound healing assay conducted at three-time intervals (0, 24, and 48 h), MCE was found to mitigate the increased migratory capacity of HNSCC cells induced by the co-culture of CAFs and TAMs (Fig. 5E, F).

Fig. 5: Inhibiting CCL2 leads to a decrease in MIF expression and attenuates the invasive and metastatic abilities of HNSCC.
figure 5

A (Left) The transwell assay results of the TAM recruitment experiment showed CCL2-CCR2 inhibitor MCE could mitigates the recruitment of TAMs induced by the secretion of CCL2 from CAFs. (Right) The cell counting for transwell assay. B The western blotting results of MIF across three groups. C (Left) The transwell assay results demonstrated that MCE effectively reverses the invasive capacity of HNSCC cells when co-cultured with CAFs and TAMs. (Right) The cell counting analysis further supported these findings. D The CCK8 assay results showed MCE reduces the heightened viability of HNSCC cells when co-cultured with CAFs and TAMs. E (Left) The wound healing assay results showed MCE could mitigate the increased migratory capacity of HNSCC cells induced by the co-culture of CAFs and TAMs. F The wounding area for wound healing assay. P value = (* <0.05; ** <0.01; *** <0.001, **** <0.0001).

Full size image

The CCL2 expression in SFRP2_CAFs may be regulated by NFIX

For subsequent analyses, we performed SCENIC analysis40 on the GSE181919 and GSE188737 single-cell datasets to gain a comprehensive understanding of the transcription factors regulating SFRP2_CAFs. We extracted a total of 379 and 377 transcription factors from the respective repositories. By applying correlation and clustering analysis to these transcription factors, the GSE181919 dataset was divided into 12 distinct modules labeled as M1-12 (Fig. 6A). Similarly, the GSE188737 dataset yielded 14 modules labeled as M1-14 (Fig. 6E). Interestingly, when examining the cell annotation information, we observed a significant enrichment of SFRP2_CAFs within the M1 module in both datasets (Fig. 6B, F). This observation suggests that the transcription factors represented by the M1 cluster may have specificity towards SFRP2_CAFs.

Fig. 6: The CCL2 expression in SFRP2_CAFs may be regulated by NFIX.
figure 6

A SCENIC transcription factors correlation and clustering results showed 12 distinct modules labeled as M1-12 in GSE181919. B The UMAP embedding plot for M1 module in GSE181919. C Integrating the JASPAR database with the SCENIC results revealed 36 common transcription factors in GSE181919. D The UMAP embedding plot for SFRP2_CAFs annotation and NFIX in GSE181919. E SCENIC transcription factors correlation and clustering results showed 14 distinct modules labeled as M1-14 in GSE188737. F The UMAP embedding plot for M1 module in GSE188737. G Integrating the JASPAR database with the SCENIC results revealed 37 common transcription factors in GSE188737. H The UMAP embedding plot for SFRP2_CAFs annotation and NFIX in GSE188737. I Dot plot of CCL2 features according to the GSE181919 cell types, the scaled mean expression value within each cell type, visualized by dot color; the dot size represents the fraction of cells in the cluster with corresponding values. J Dot plot of CCL2 features according to the GSE188737 cell types, the scaled mean expression value within each cell type, visualized by dot color; the dot size represents the fraction of cells in the cluster with corresponding values. K The correlation (R = 0.304, p = 3.38e−12) between CCL2 and NFIX in TCGA-HNSC cohort. L The bar graph demonstrates that the transfection efficiency of siNFIX is excellent in CAFs. M, N The qPCR and western blot results indicate alterations in the expression levels of CCL2 following the siNFIX perturbation. P value = (* <0.05; ** <0.01; *** <0.001, **** <0.0001).

Full size image

To identify potential transcription factor binding to CCL2 in SFRP2_CAFs, we obtained the JASPAR database, which encompassing transcription factor binding profiles for all human genes41. We define the region located 500 bp upstream and 200 bp downstream of the CCL2 gene as potential transcription factor binding sites (TFBS) involved in regulating CCL2 expression. By integrating the JASPAR database with the identified TFBS, we uncovered 106 potential transcription factors that have the potential to bind to CCL2. The intersection analysis revealed 36 common transcription factors in the GSE181919 dataset and 37 common transcription factors in the GSE188737 dataset between the SCENIC and JASPAR analyses, respectively (Fig. 6C, G). Among these transcription factors, only NFIX, HOXA1, and HOXA7 were found exclusively in both M1 module. Subsequent analysis using UMAP embedding demonstrated that NFIX displayed highest expression in SFRP2_CAFs (Fig. 6D, H). These findings strongly suggest a potential regulatory role of NFIX in SFRP2_CAFs. Next, we examined the expression patterns of NFIX and CCL2 across different cell types. In the GSE181919 dataset, we observed that NFIX and CCL2 were specifically associated in SFRP2_CAFs (Fig. 6I). In the GSE188737 dataset, besides their association in SFRP2_CAFs, they also showed associations in Fibroblasts, Endothelial cells, and Smooth muscle cells (Fig. 6J). In addition, in analysis of the TCGA-HNSC bulk data, NFIX and CCL2 displayed a significant correlation with each other (Fig. 6K, R = 0.304, p = 3.38e−12). Subsequently, siNFIX transfection into CAFs was performed, unveiling a high transfection efficiency (Fig. 6L). We then explored the expression of CCL2 post siNFIX transfection in CAFs. Both RT-qPCR and Western blot analyses illustrated that disrupting NFIX expression led to a decrease in CCL2 expression (Fig. 6M, N). These results provide further evidence supporting the NFIX as a potential transcription factor for CCL2.

Transcription factor NFIX perturbation regulate the identity of SFRP2_CAFs

In our study, CCL2 has been identified as a specific biomarker for SFRP2_CAFs, while NFIX is potentially a specific transcription factor for the CCL2 and SFRP2_CAFs. To validate these findings, we further conducted silico perturbation analyses using CellOracle42 to construct a gene regulatory and network in HNSCC. Considering the network structure is constructed using a list of directed edges connecting transcription factors to their target genes, we can assign numerical values or ranks to the nodes in the graph based on centrality metrics to determine the ranking of target genes within specific cell types. Consistent with our hypothesis, CCL2 demonstrates the highest weight among SFRP2_CAFs in single-cell datasets GSE181919 (Fig. 7A), GSE188737 (Fig. 7D). and external validation dataset GSE234933 (Fig. 7G). Similarly, NFIX attains the top weight among SFRP2_CAFs in GSE181919 (Fig. 7B) and the external validation dataset GSE234933 (Fig. 7H), while securing the fifth position in GSE188737 (Fig. E). These results indicate that the activation of NFIX-CCL2 regulatory axis holds a dominant position in SFRP2_CAFs, compared to other cell types in TME. To confirm the specificity of NFIX as a transcription factor for SFRP2_CAFs, we performed transcription factors silico perturbation using the same methodology. By downregulating the average expression levels of NFIX to 0, our results reveal that under this perturbation condition, The population of SFRP2_CAFs undergoes identity differentiation changes in all tested datasets, including GSE181919 (Fig. 7C), GSE188737 (Fig. 7F), and the external validation dataset GSE234933 (Fig. 7I). Similarly, when upregulating the average expression level of NFIX to twice its original level, an opposite change was observed in all datasets (Supplementary Fig. 4A–C). Such changes were not observed in other cell types, highlighting the crucial regulatory role of the transcription factor NFIX in the developmental process of SFRP2_CAFs.

Fig. 7: Transcription factor NFIX perturbation regulate the identity of SFRP2_CAFs.
figure 7

A CCL2 gene’s network score in GSE181919. Note: degree centrality is defined as the number of links incident upon gene nodes; betweenness centrality quantifies a gene node acts as a bridge along the shortest path between two other gene nodes, eigenvector centrality is defined as relative scores to all nodes in the network based on the concept that connections to high-scoring gene nodes contribute more to the score of the gene node in question than equal connections to low-scoring gene nodes. Weight is calculated by the R package Igraph. B NFIX gene’s network score in GSE181919. C CellOracle simulation cell-state transitions in the NFIX knockout (KO) simulation; The resulting cell-state transition vectors were summarized and projected onto a force-directed graph; red box; GSE181919. D CCL2 gene’s network score in GSE188737. E NFIX gene’s network score in GSE188737. F CellOracle simulation cell-state transitions in the NFIX knockout (KO) simulation; The resulting cell-state transition vectors were summarized and projected onto a force-directed graph; red box; GSE188737. G CCL2 gene’s network score in GSE234933. H NFIX gene’s network score in GSE234933. I CellOracle simulation cell-state transitions in the NFIX knockout (KO) simulation; The resulting cell-state transition vectors were summarized and projected onto a force-directed graph; red box; GSE234933.

Full size image

Discussion

Research findings have demonstrated that CAFs play a crucial role in governing the HNSCC tumor microenvironment43,44. For instance, high expression of OXTR in CAFs contributes to local-regional metastasis45. However, the investigation into the characteristics of CAFs has been hampered by limitations such as small sample sizes and insufficient multidimensional data. In our study, we aimed to overcome these constraints by conducting a targeted clinical-driven multi-center sample and multi-omics analysis on HNSCC. This comprehensive approach led us to uncover a crucial phenomenon in tumor invasion and metastasis—an exceptional biomarker that serves as an indicator of CAFs abundance and is closely associated with prognosis. Specifically, we made a significant observation regarding the expression profiles of SFRP2, which corresponds to CAFs identified in the HNSCC single-cell dataset, and Fibroblast_S03 in the bulk ecosystem. The striking resemblance between these two cell types allowed us to quantify the expression patterns by utilizing SFRP2 as a biomarker. Our findings demonstrate that SFRP2_CAFs cells originate from smooth muscle cells, which aligns with previous research on CAFs in different cancer types46,47. Moreover, we have observed that this specific cell type exhibits characteristics of both myCAFs and iCAFs, with more prominent features observed in metastatic HNSCC. These findings offer valuable insights into the pro-tumorigenic role underlying the developmental aspects of SFRP2_CAFs. Prior scholarly investigations have demonstrated that SFRP2 can enhance the invasive and metastatic capabilities of melanoma and pancreatic cancer48,49. Furthermore, our findings suggest that SFRP2 functions not solely as a robust prognostic indicator but also exhibits significant associations with myogenesis, angiogenesis, and the EMT. These three pathways are recognized as pivotal signaling cascades intricately linked to the processes of tumor invasion and metastasis50,51,52.

Cell communication analysis facilitates a more comprehensive characterization of cell populations exhibiting elevated expression levels of SFRP2 in the tumor microenvironment. In our investigation, we noted that the interaction between SFRP2_CAFs and SPP1_TAMs exhibited the highest intensity. We have, for the first time, brought to light the presence of invasive niches constituted by SFRP2_CAFs and SPP1_TAMs in HNSCC. Furthermore, our observations revealed that SPP1_TAMs are the predominant immune cells in proximity to SFRP2_CAFs, which is consistent with previous research conducted in colorectal cancer53. In our study, we have discovered that SFRP2_CAFs represent the most potent signaling population within the tumor microenvironment (TME). These cells may regulate the function and behavior of neighboring cells by releasing specific signaling molecules. On the other hand, SPP1_TAMs serve as the most responsive population within the TME, efficiently perceiving and responding to the signals released by SFRP2_CAFs. Those findings suggest that the close interaction between SFRP2_CAFs and SPP1_TAMs in the metastatic lesion contributes to creating a favorable external environment for tumor growth, invasion, and dissemination. In the past, researchers have identified two distinct types of HNSCC, namely HPV negative and HPV positive HNSCC54. Our study indicates that the abundance and gene expression profiles of SFRP2_CAFs and SPP1_TAMs are highly consistent between HPV negative and HPV positive HNSCC. This suggests that the interaction between SFRP2_CAFs and SPP1_TAMs is conserved in HNSCC. Notably, the MIF-CD74 ligand-receptor interaction serves as the main communication mediator between SFRP2_CAFs and SPP1_TAMs in HNSCC. Kang et al.55 reported that MIF could stimulate the proliferation in oral squamous cell carcinoma cells. In our study, we unveiled that SFRP2_CAFs exhibit significantly higher expression levels of CCL2 compared to SPP1_TAMs, a well-known chemokine involved in TAMs recruitment56. Considering these findings, we postulate that MIF-CD74 ligand-receptor interaction plays a critical role in promoting invasion, and its regulation is influenced by CCL2.

To substantiate this hypothesis, we conducted co-culture experiments. In vitro findings clearly revealed that CAFs actively secrete CCL2, which subsequently recruits TAMs and upregulates the expression of MIF, thereby facilitating tumor invasion, metastasis, cytotoxic effects, and anti-apoptotic benefits. Notably, when a CCL2 inhibitor was administered, it successfully attenuated tumor-promoting effects. These results strongly support the regulatory influence of CAF-secreted CCL2 on MIF. Currently, the functional role of CD74 in HNSCC has not been extensively explored. Limited research suggests that upregulation of CD74 promotes drug resistance and inhibits apoptosis in lung cancer cells57. In our research, we made the significant finding that CD74 serves as a shared biological marker for both CAFs and TAMs. Moreover, an intriguing similarity was observed in the velocity profiles of CD74 between SFRP2_CAFs and SPP1_TAMs. Subsequent investigation further revealed that the dynamic kinetics of CD74 exhibited a notable correlation exclusively within the cell types of SFRP2_CAFs and SPP1_TAMs. These intriguing findings subtly suggest that the interaction between MIF and CD74 might be established during the developmental transition of CAFs and TAMs.

Another conclusion of this study is that NFIX may play a role in the regulation of CCL2. While previous studies have highlighted the significant involvement of NFIX in neural stem cell biology and muscle development58,59, its specific role in the tumor microenvironment of HNSCC and particularly in HNSCC CAFs remains unexplored. In HNSCC patients, we found that NFIX is predominantly expressed in SFRP2_CAFs and exhibits a significant correlation with the expression of CCL2 and siNFIX leads to alterations in CCL2 expression. Importantly, within the gene regulatory network we constructed, NFIX-CCL2 axis demonstrated a remarkable increase in activity specifically within SFRP2_CAFs. Furthermore, through silico perturbation of NFIX, we observed alterations in the fate of SFRP2_CAFs cells. These findings further emphasize the specificity of NFIX in SFRP2_CAFs and highlight its crucial role in regulating the identity of SFRP2_CAFs.

Despite the promising findings displayed in our research, one limitation is that the intricate interactions observed may not be fully replicated in in vitro experiments. In summary, our study successfully identified a novel phenotype of CAFs in HNSCC at both the bulk and single-cell levels. Remarkably, these CAFs have been identified as originating from smooth muscle cells. This distinct cellular subset not only carries prognostic implications but also assumes a crucial role in tumor promoting. Additionally, our research has elucidated key transcription factors that exert influence on the development of this specific CAFs phenotype. We have summarized this study procedure (Fig. 8). The insights gained from this study hold substantial potential for the identification of therapeutic targets. Future treatment modalities for HNSCC could be directed towards reversing the phenotype of SFRP2_CAFs, with a focus on developing efficacious therapeutic strategies that specifically target the invasive, metastatic and immunosuppressive capabilities of this cell type.

Fig. 8: Workflow of this study.
figure 8

We crafted Fig. 8 utilizing Adobe Illustrator 2023 and Servier Medical Art (https://smart.servier.com). The picture unveiled the presence of SFRP2_CAFs in HNSCC, which secrete CCL2 to recruit SPP1_TAMs. These interactions are mediated by MIF-CD74, thereby facilitating tumor invasion, metastasis, apoptosis resistance and immune evasion. The transcription factor NFIX is identified as a key regulatory factor for SFRP2_CAFs.

Full size image