A single-cell transcriptomic atlas of CSCC
To investigate the cellular composition of TME in CSCC, we collected six tumor tissues and two adjacent normal tissues from six treatment-naïve CSCC patients for scRNA-seq (Fig. 1A). The clinical details of the patients are provided in Table 1. Following quality filtering, a total of 50,649 cells, with an average of 2,282 genes per cell, were analyzed. This dataset comprised 31,232 cells from tumor tissues and 19,437 cells from normal tissues (Additional file1: Fig. S1A and 1B, Table S2). Unsupervised clustering using uniform manifold approximation and projection (UMAP) demonstrated distinct clustering based on tissue origin, mitigating batch effects from the samples (Fig. 1B, Additional file1:Fig. S2A). Ten major cell types were identified based on canonical markers, comprising three stromal cell types (epithelial cells, fibroblasts, and endothelial cells (ECs)), and seven immune cell types (myeloid cells, mast cells, B cells, plasma cells, NK cells, CD4+ T cells, and CD8+ T cells) (Fig. 1C, Additional file1:Table S3). Cell type ratios exhibited variations between tumor and normal tissues (Fig. 1D). Normal tissues showed abundance in ECs and fibroblasts, while B cells and mast cells were prevalent in tumor tissues, indicating an immune activation state within TME (Fig. 1E). Additionally, the cellular compositions differed significantly between tumor and normal tissues. Tumor tissues displayed a higher proportion of epithelial cells followed by immune cells, whereas normal tissues had higher proportions of fibroblasts and endothelial cells (Fig. 1F). Survival analysis indicated that high infiltration of CD4+ T cells, CD8+ T cells, NK cells, plasma cells, and B cells correlated with favorable overall survival, suggesting a positive association between an activated immune response and tumor control; however, no correlation was observed between stromal cells and survival (Fig. 1G, Additional file1:Fig. S2B). Analysis of cell communication and correlations revealed complex interactions within the TME, with most cells showing positive correlations indicative of cooperation in the TME (Fig. 1H, Additional file1:Fig. S2C and 2D). In conclusion, this study provides a comprehensive overview of the major cell types in CSCC.
Table 1
Clinical information of included patients
Patients | Samples | Age | Stage | Differentiation | Pathological type |
Patient1 | PT1 | 68 | IIA | Low | CSCC |
Patient1 | PN1 | 68 | IIA | Low | CSCC |
Patient2 | PT2 | 39 | IIIA | Moderate | CSCC |
Patient3 | PT3 | 57 | IIB | Low | CSCC |
Patient4 | PT4 | 47 | IIB | High | CSCC |
Patient5 | PT5 | 39 | IB | Low | CSCC |
Patient5 | PN5 | 39 | IB | Low | CSCC |
Patient6 | PT6 | 42 | IB | Moderate | CSCC |
Intratumoral heterogeneity of epithelial cells in CSCC
To delve deeper into understanding the phenotype and function of epithelial cells within TME dominated by CSCC, we performed subsetting and re-clustering of these cells. Through unsupervised clustering using UMAP, we delineated eight distinct clusters of epithelial cells. The distribution of these clusters varied depending on the sample source and tissue origin (Fig. 2A and 2B; Additional file1:Fig. S3A).Cluster 3 (C3), defined by its high expression of carcinoembryonic antigen (CEACAM5, CEACAM6) and mucin (MUC20) genes, exhibited characteristics typically enriched in tumor cells, specifically excluding those found in normal tissues. Clusters 4 (C4) and 6 (C6), designated as C4-KRTDAP and C6-TFF3, respectively, showed some shared marker genes with C3 but displayed a lower level of carcinoembryonic antigen expression. Notably, these two clusters demonstrated activation of immune-related pathways such as antigen processing and presentation, leukocyte chemotaxis, and antimicrobial function, categorizing them as immune-associated epithelial cells (IAEpis) (Additional file1:Fig. S3B).Cluster 5 (C5), denoted as C5-PCLAF, was predominantly present in tumor tissues and exhibited high expression of cell cycle-related genes (PCNA, CLSPN), signifying tumor-associated epithelial cells (TAEpis). Conversely, clusters 7 (C7) and 8 (C8) identified as C7-CENPF and C8-NEURL1B - tended to aggregate in tumor tissues, displaying similar patterns of metabolic pathways and signal transduction as C5. This suggested a transitional state of TAEpis within these clusters. Cluster 2 (C2), labeled as C2-DST, showcased a high expression of stromal genes (CCN1, COL17A1) associated with wound healing and matrix remodeling, predominantly found in normal tissues, representing normal epithelial cells.Cluster 1 (C1) did not exhibit distinct markers, indicating a transitional state that was less well-defined (Fig. 2C and 2D, Additional file1:Table S4).
In examining the functions of epithelial cells, gene set enrichment analysis (GSVA) was employed to quantify the activity levels of metabolic and signaling pathways. The findings revealed heightened metabolic activity in epithelial cells within tumor tissues, particularly evident in the increased activity of sphingolipid metabolism, the pentose phosphate pathway, and glycerophospholipid metabolism. Conversely, tryptophan metabolism, taurine and hypotaurine metabolism, and fatty acid biosynthesis exhibited higher scores in normal epithelial cells (Fig. 2E; Additional file1:Fig. S3C).Further analysis showed that intratumoral epithelial cells were characterized by the activation of signaling pathways such as HIF-1, mTOR, notch, and VEGF, while displaying a reduction in immune-related signaling (Fig. 2F). Examination of specific clusters highlighted that C5-PCLAF and C8-NEURL1B demonstrated similar metabolic profiles, with heightened activation in TME. In contrast, C2-DST appeared to be in a state of metabolic equilibrium (Fig. 2G; Additional file1:Fig. S3D).Interestingly, distinct signaling patterns were observed between C2 and C8, with C2 predominately activating calcium and cytokine-cytokine receptor signaling pathways (Fig. 2H). Transcription factor (TF) analysis using SCENIC identified specific TFs associated with each cluster, although C1 did not exhibit any unique TFs (Fig. 2I).Survival analysis indicated that a greater abundance of C8 correlated with a more favorable overall prognosis, suggestive of an “immune-hot” TME in this subtype of CSCC due to the activation of the NF-κB and TNF signaling pathways (Fig. 2J; Additional file1:Fig. S3E). In summary, these results elucidate the intratumoral diversity and plasticity of epithelial cells in CSCC.
The interconversion of inflammatory and cancer-associated fibroblasts correlates with the survival outcome of CSCC patients
Reclustering of fibroblasts identified five distinct clusters, with these subsets displaying universal distribution across samples but showing disparities between tumor and adjacent normal tissues (Fig. 3A and 3B, Additional file1:Fig. S4A). Cluster 1 (C1), labeled as C1-SPER4, exhibited elevated expression of SPER4 along with several chemokines such as CXCL1, CXCL14, and PDGFRA. This fibroblast cluster, prevalent in normal tissues, was classified as inflammatory-associated fibroblasts (IAFs). Cluster 2 (C2), known as C2-MMP11, showed high expression of matrix metallopeptidase genes like MMP11, MMP2, and MMP14, a range of collagens, and other extracellular matrix components including COL1A1, COL3A1, COL1A2, COL5A2, and COL12A1. Moreover, the presence of FAP, a characteristic gene of cancer-associated fibroblasts (CAFs), designated this cluster as CAFs. Functional analysis indicated that CAFs were associated with the activation of extracellular matrix organization and collagen fibril organization (Additional file1:Fig. S4B).Clusters 3 (C3) and 5 (C5) were identified as subsets of myofibroblasts, both expressing typical myofibroblast marker genes such as ACTA2 and genes involved in myogenesis like MYH11, MUSTN1, and DES. However, while C3 exhibited moderate expression of these genes, suggesting an immature state, C5-MYH11 was predominant in normal tissues, whereas C3-MUSTN1 showed higher levels in tumor tissues. Cluster 4 (C4), identified as C4-RGS5, comprised pericytes characterized by the expression of the pericyte marker RGS5 (Fig. 3C and 3D, Additional file1:Table S5).
Metabolic analysis unveiled heightened activity in intratumoral fibroblasts, showcasing increased metabolic pathways like glycolysis, gluconeogenesis, and pyrimidine metabolism, alongside reduced cholesterol metabolism and steroid biosynthesis levels (Fig. 3E, Additional file1:Fig. S4C). Observation of the fibroblasts within TME demonstrated concurrent activation of oncogenic and immune-related signaling, emphasizing the diverse nature of fibroblasts within the TME (Fig. 3F).Further exploration of each cluster revealed that C2-MMP11-CAFs exhibited the most vigorous metabolic processes, in contrast to a shared metabolic pattern between C3-MUST1 and C5-DES, while C1-SPER4 displayed a more quiescent metabolic profile (Fig. 3G, Additional file1:Fig. S4D). Consistent with these findings, C2-MMP11 showed activation across various oncogenic signaling pathways, C3 activated the calcium signaling pathway, and C5 activated the phosphatidylinositol signaling system (Fig. 3H).TF analysis identified distinct TFs specific to each cluster. The similar TF expression between the two subsets of cancer-associated fibroblasts (CAFs) suggested the potential for mutual transformation (Fig. 3I). Trajectory analysis revealed a differentiation pathway from C1-SPER4-IAFs to C2-MMP11-CAFs (Fig. 3J). Moreover, survival analysis indicated that a high infiltration of IAFs correlated with favorable survival outcomes in CSCC patients, while the presence of CAFs displayed an opposing trend. Conversely, the other subsets showed no significant correlation with patient survival (Fig. 3K, Additional file1:Fig. S4E).In summary, these findings illuminate the intricate landscape of fibroblasts in CSCC.
Myeloid cell heterogeneity in CSCC correlates with tumor progression
Myeloid cells encompass significant immune cell populations within TME that exert anti-tumor effects. Sub-clustering of myeloid cells post-batch correction revealed eight distinctive subsets (Fig. 4A). While non-unique subsets were evident across samples, notable differences were observed between tumor and adjacent normal tissues (Fig. 4B, Additional file1:Fig. S5A). Specifically, C1-C1QA was characterized as macrophages (C1QA, C1QB), C4-CD163 identified as tumor-associated tumor-associated macrophages (TAMs) expressing key metabolic genes like SLC40A1 and FOLR2, linked to TAMs proliferation and polarization. Dendritic cell subsets were also identified, with C5-CD1C representing conventional type 2 dendritic cells (cDC2) expressing CD1C, FCER1A, and HLA-DQB1, and C8-LAMP3 denoted as mature dendritic cells, or cDC3, with high expression of LAMP3 and FCN1 enabling migration to tumors via elevated CCR7 expression.Four subsets of neutrophils were clustered in CSCC, with C2-S100A8 exhibiting high expression of pro-inflammatory genes like IL1B, indicative of conventional neutrophils. On the other hand, C3-CXCR4 and C6-CXCL8 displayed high expression of CXCL8 and CSF3R predominantly in tumor tissues, representing a subset of tumor-associated neutrophils (TANs). C7-ISG15 exhibited high expression of interferon-induced and stimulated genes (IFIT2, IFIT3, ISG15, and ISG20), characterizing a type I interferon-producing neutrophil phenotype, both displaying high expression of the neutrophil-specific antigen CD16B (encoded by FCGR3B) (Fig. 4C and 4D, Additional file1:Fig. S5B, Table S6).Further characterization of the C6 and C7 neutrophil subsets involved differential gene expression and pathway enrichment analysis. Results showed distinct transcription profiles between the two clusters, with C6-CXCL8 activating oncogenic signaling pathways like MAPK and NF-κB, while C7-ISG15 enhancing immune-related pathways such as chemokine signaling, Toll-like receptor signaling, and antigen processing and presentation (Additional file1:Fig. S5C and 5D).
Unlike stromal cells, the metabolism and signal transduction of myeloid cells in tumor tissues are significantly suppressed (Fig. 4E and 4F, Additional file1:Fig. S5E). Upon detailed analysis of each cluster, it was evident that macrophages and dendritic cells displayed enhanced metabolic activity and signaling capabilities compared to neutrophils, underscoring their crucial role in TME (Fig. 4G and 4H, Additional file1:Fig.S5F). While unique TFs were identified in each cluster, C6-CXCL8-TANs and C7-ISG15-Neutrophils exhibited similar TF expression levels, suggesting a potential interconversion between these two neutrophil subsets (Fig. 4I). Trajectory analysis of the four neutrophil subsets revealed three distinct differentiation paths originating from C2-S100A8 conventional neutrophils (Fig. 4J). Consistent with the aforementioned findings, survival analysis of these clusters indicated that increased infiltration of C6-CXCL8-TANs was associated with unfavorable overall survival, while the trend was reversed for C7. Additionally, high infiltration of the two dendritic cell subsets predicted a more favorable overall survival outcome (Fig. 4K, Additional file1:Fig.S5G).
Characterization of T cells in CSCC
T cells play a pivotal role in immune responses. Re-clustering of T and NK cells unveiled four clusters of CD8+ T cells, five clusters of CD4+ T cells, one cluster of double-positive cells, and one cluster of NK cells (Fig. 5A). While no specific clusters were consistent across patients, distinct tissue origins were evident in some clusters (Fig. 5B, Additional file1:Fig.S6A). Noteworthy CD8 clusters included CD8-C1-ZNF683, characterized by high expression of ZNF683 and T cell-related chemokines and cytokines like CCL4, CCL5, GZMB, and IFNG, labeled as tissue-resident memory T cells (Trm). CD8-C3-GZMK displayed heightened GZMK expression and TNFSF9 co-stimulatory molecules, being newly recognized as a transitional state denoted as effector memory T cells (Tem). CD8-C6-CXCL13, with a pronounced expression of inhibitory genes like HAVCR2, LAG3, and PDCD1, denoted a state of exhaustion (Tex). Notably, this cluster showed high levels of cytokines GZMB, PRF1, and IFNG, underscoring its anti-tumor function, identified as tumor-specific T cells expressing CD39 (ENTPD1-encoded) and CD103 (ITGAE-encoded). Consistently, CD8-C6-CXCL13 T cells were exclusively observed in tumor tissues (Additional file1:Fig. S6B). CD8-C8-GPR183 were classified as central memory T cells (Tcm). Regulatory T cell (Treg) subsets, CD4-C4 and CD4-C9, expressed Treg markers FOXP3 and IL2RA (CD25), with CD4-C4 exhibiting higher gene expression and a tumor-exclusive presence, indicating an enhanced suppressive nature. CD4-C2-IL7R, notably expressing homing receptor CCR7, was categorized as central memory T cells (Tcm). CD4-C7-CD40LG notably expressed TNF, defining it as effector T cells (Teff). CD4-C10-CXCL13 exhibited elevated CXCL13 and BHLHE40 expression, recently classified as Th1-like cells (Th1). NK cells were distinguished by the expression of NKG7. The CD4 and CD8 double-positive T cell subset, DP-C11-HIST1H1B, displayed high expression of cell-cycle-related genes like MKI67, STMN1, and TOP2A, indicating proliferative potential (Fig. 5C and 5D, Additional file1:Table S7).
Metabolic analysis unveiled a metabolic response of T cells to hypoxia in TME. Both CD4+ and CD8+ T cells upregulated oxidative phosphorylation, glycolysis, gluconeogenesis, and fatty acid metabolism (Fig. 5E). Furthermore, T cells within the tumor tissue exhibited heightened activity in seven metabolic processes (Additional file1:Fig. S6C). Consistent with these findings, activation of HIF-1 signaling in T cells was observed (Fig. 5F). Subsequent analysis of each cluster revealed that CD4-C9-FOXP3low, CD4-C10-CXCL13, and CD8-C6-CXCL13 displayed robust metabolic activity (Fig. 5G, Additional file1:Fig. S6D). Interestingly, both CXCL13+ CD8+ and CD4+ T cells displayed activation of apoptosis signaling pathways, including ferroptosis, necroptosis, and general apoptosis signaling pathways (Fig. 5H).SCENIC analysis revealed a novel TF, nuclear factor interleukin-3-regulated (NFIL3), was specific to CD8-C6-CXCL13 and is recognized as a pivotal immune regulator. NFIL3 overexpression inhibits Tregs function and regulates cytokine expression in Th2 cells (21, 22). Conversely, the role of NFIL3 in the formation or function of the CD8-C6-CXCL13 cluster remains largely unexplored (Fig. 5I). Trajectory analysis indicated that CD8-C1-ZNF683 and CD4-C2-IL7R represent the initial states of CD8+ and CD4+ T cells, respectively. Furthermore, CD8-C3-GZMK, CD8-C6-CXCL13, and CD8-C8-GPR183 delineate distinct differentiation pathways from CD8-C1-ZNF683. Additionally, a potential association between CD4-C10-CXCL13 and Tregs was observed (Fig. 5J). Survival analysis revealed a positive correlation between high T cell infiltration levels and improved survival outcomes (Fig. 5K, Additional file1:Fig. S6E). In conclusion, our findings suggest an enhanced activity of T cells within the TME.
The presence of PCLAF + TAEpis showed a negative correlation with the abundance of CXCL13+ CD8+ T cells.
To investigate the cellular interactions within TME of CSCC, we conducted Cellchat analysis of the identified clusters. Our findings demonstrated extensive interactions among most clusters, particularly between epithelial cells and fibroblasts (Fig. 6A, Additional file1:Table S8). Specifically, we observed that the collagen signaling pathway played a significant role in mediating these interactions (Fig. 6B). Further scrutiny revealed fibroblasts and epithelial cells as the primary cells engaged in interactions (Fig. 6C). Subsequent receptor-ligand analysis unveiled interactions between collagen-related genes like COL1A1 with ITGA1 or ITGB1, predominantly occurring in T cells and stromal cells (Additional file1:Fig S7A and 7B). Correlation analysis further supported the interrelationships among these cell types. Moreover, a negative correlation was noted between C5-PCLAF TAEpis and C6-CD8 CXCL13 T cells, identified as novel tumor-reactive T cells (Fig. 6D) (23).To characterize C5-PCLAF TAEpis, we initially conducted an intersection analysis of the marker genes of C5-PCLAF TAEpis with epithelial cells, identifying CD24 as a specific membrane marker for C5-PCLAF TAEpis suitable for immunofluorescence labeling (Additional file1: Figure S7C, Table S9). Subsequently, utilizing multi-immunofluorescence, we labeled CD24, CD8A, and PD-1 for two distinct cell types. The findings revealed a negative correlation between CD24 expression and CD8+ PDCD1+ T cells, implying the formation of a physical barrier (Fig. 6E). Consistent with these observations, Immunohistochemistry (IHC) analysis indicated a negative correlation between CD24 and PD-1 expression (Additional file1:Fig. S7D and 7E). Survival analysis elucidated that increased CD24 expression correlated with poorer survival outcomes, while elevated PD-1 expression was associated with a more favorable prognosis (Fig. 6F). Furthermore, high PD-1 expression was linked to greater tumor cell differentiation, while CD24 showed no significant correlation (Tables 2 and 3). Subsequently, we assessed the expression of CD24 in response to anti-PD-1 therapy among CSCC patients who underwent radiotherapy combined with anti-PD-1 blockade, a detailed listing of patient characteristics in Table 4. The analysis revealed higher CD24 expression levels in tumor tissues of non-responsive patients (Fig. 6G and 6H). Additionally, CD24 expression was identified as a significant predictor of the response to anti-PD-1 therapy with enhanced specificity and sensitivity (AUC:0.768). These collective findings suggest a potential pro-tumor role of PCLAF+ TAEpis in suppressing tumor-specific CD8+ T cells.
Table 2
Correlation of PCNA expression and clinical parameters
| Total (n = 50) | High (n = 25) | Low (n = 25) | p |
Age | 50.08 ± 11.32 | 52.16 ± 9.84 | 46.68 ± 12.97 | 0.124 |
Stage, n (%) | | | | 0.919 |
I | 24(48) | 14 (45) | 10 (53) | |
II | 20(40) | 13 (42) | 7 (37) | |
III | 6(12) | 4 (13) | 2 (11) | |
Stage2,n(%) | | | | 0.825 |
I | 24(48) | 14 (45) | 10 (53) | |
II + III | 26(52) | 17 (55) | 9 (47) | |
Differentiation,n(%) | | | | 0.273 |
L | 17 (34) | 8 (26) | 9 (47) | |
M | 18 (36) | 12 (39) | 6 (32) | |
H | 15 (30) | 11 (35) | 4 (21) | |
Differentiation2,n(%) | | | | 0.445 |
L + M | 35 (70) | 20 (65) | 15 (79) | |
H | 15 (30) | 11 (35) | 4 (21) | |
Table 3
Correlation of PD-1 expression and clinical parameters
| Total (n = 50) | High (n = 25) | Low (n = 25) | p |
Age | 50.08 ± 11.32 | 49.4 ± 10.36 | 50.76 ± 12.39 | 0.676 |
Stage, n (%) | | | | 0.571 |
I | 24(48) | 14(56) | 10(40) | |
II | 20(40) | 9(36) | 11(44) | |
III | 6(12) | 2(8) | 4(16) | |
Stage2,n(%) | | | | 0.396 |
I | 24(48) | 14(56) | 10(40) | |
II + III | 26(52) | 11(44) | 15(60) | |
Differentiation,n(%) | | | | < 0.001 |
L | 17 (34) | 10 (40) | 7 (28) | |
M | 18 (36) | 0 (0) | 18 (72) | |
H | 15 (30) | 15 (60) | 0 (0) | |
Differentiation2,n(%) | | | | < 0.001 |
L + M | 35 (70) | 10 (40) | 25 (100) | |
H | 15 (30) | 15 (60) | 0 (0) | |
Table 4
characteristics of CESC patients treated with radiotherapy and anti-PD-1 blockade
| Response(CR + PR) (n = 25) | Non-response(PD + SD) (n = 12) |
Age | 50.04 ± 14.19 | 48.08 ± 9.811 |
Histological grade, n (%) | | |
Well or moderately differentiated | 18(72%) | 2(17%) |
Poorly differentiated | 4(16%) | 8(66%) |
Unknown | 3(12%) | 2(17%) |
Expression of PD-L1 | | |
< 1% | 7(28%) | 5(42%) |
≥ 1–49% | 8(32%) | 3(25%) |
≥ 50% | 8(32%) | 3(25%) |
Unknown | 2(8%) | 1(8%) |
Surgery | | |
Yes | 9(36%) | 2(17%) |
No | 16(64%) | 10(83%) |
Tumor(T) | | |
0 | 1(4%) | 0 |
1 | 2(8%) | 0 |
2 | 9(36%) | 1(8%) |
3 | 11(44%) | 7(58%) |
4 | 2(8%) | 4(34%) |
Node(N) | | |
0 | 7(28%) | 2(16%) |
1 | 12(48%) | 5(42%) |
2 | 6(24%) | 5(42%) |
Metastasis(M) | | |
0 | 20(80%) | 8(66%) |
1 | 5(20%) | 4(34%) |
PCLAF+TAEpis inhibit infiltration and function of T cells
To elucidate the role of PCLAF+ TAEpis, we initially isolated these cell types from tumor and adjacent tissues in tumor-bearing mice based on the membrane markers EpCAM and CD24 using flow cytometry, defining them as PCLAF+ TAEpis (TAEpis). Concurrently, we isolated EpCAM+ epithelial cells from cervical tissue of tumor-free mice, referred to as normal epithelial cells (NAEpis). Subsequently, U14 mouse cervical tumor cells were co-implanted with TAEpis or NAEpis into mice (Fig. 7A). The baseline tumor volumes were similar among the groups before anti-PD-1 treatment (Fig. 7B). Notably, TAEpis promoted tumor growth and attenuated the efficacy of anti-PD-1 treatment, while NAEpis had no impact on tumor progression (Fig. 7C and 7D). Consistent with these findings, TAEpis reduced the survival of tumor-bearing mice following anti-PD-1 treatment compared to NAEpis (Fig. 7E). Furthermore, we examined the tumor-infiltrating immune subsets within five groups (Additional file1:Fig. S8), revealing a significant decline in T cells, including CD3, CD4, and CD8+ T cells, in the TAEpis plus U14 group. Although PD-1 treatment partially restored the T cell ratios, a sustained decrease was observed in the TAEpis plus U14 group compared to the NAEpis plus U14 group under PD-1 treatment (Fig. 7F-K). Notably, the ratios of CD4+ and CD8+ T cells remained consistent across these groups (Fig. 7L-M). Additionally, functional analysis indicated that co-implantation with TAEpis resulted in reduced secretion of IFN-γ in both CD4 and CD8+ T cells, while TNF-α levels exhibited a slight decrease in the TAEpis group (Fig. 7N-Q).