ITGB6 modulates resistance to anti-CD276 therapy in head and neck cancer by promoting PF4+ macrophage infiltration

Abstract

Enoblituzumab, an immunotherapeutic agent targeting CD276, shows both safety and efficacy in activating T cells and oligodendrocyte-like cells against various cancers. Preclinical studies and mouse models suggest that therapies targeting CD276 may outperform PD1/PD-L1 blockade. However, data from mouse models indicate a significant non-responsive population to anti-CD276 treatment, with the mechanisms of resistance still unclear. In this study, we evaluate the activity of anti-CD276 antibodies in a chemically-induced murine model of head and neck squamous cell carcinoma. Using models of induced and orthotopic carcinogenesis, we identify ITGB6 as a key gene mediating differential responses to anti-CD276 treatment. Through single-cell RNA sequencing and gene-knockout mouse models, we find that ITGB6 regulates the expression of the tumor-associated chemokine CX3CL1, which recruits and activates PF4⁺ macrophages that express high levels of CX3CR1. Inhibition of the CX3CL1-CX3CR1 axis suppresses the infiltration and secretion of CXCL16 by PF4⁺ macrophages, thereby reinvigorating cytotoxic CXCR6⁺ CD8⁺ T cells and enhancing sensitivity to anti-CD276 treatment. Further investigations demonstrate that inhibiting ITGB6 restores sensitivity to PD1 antibodies in mice resistant to anti-PD1 treatment. In summary, our research reveals a resistance mechanism associated with immune checkpoint inhibitor therapy and identifies potential targets to overcome resistance in cancer treatment.

Fig. 1. ITGB6 as a pivotal driver of α-CD276 resistance in mouse HNSCC Model.

A The experimental design of the HNSCC tumorigenesis model and treatment strategy. B Representative examples of head and neck magnetic resonance imaging (MRI) for different groups (IgG = 8, Resistant = 11, Sensitive = 7 mice). The dashed area is the boundary of the tumor. D, days. C Tumor growth curves of each mouse receiving Anti-CD276 antibody or control IgG treatment. Measurements were taken every 4 days (IgG = 8, Resistant = 11, Sensitive = 7 mice). D Representative images of 4NQO-induced HNSCC (upper) and corresponding HE staining(lower) (IgG = 8, Resistant = 11, Sensitive = 7 mice). Scale bar, 1 mm (upper), 100 μm (lower). E Workflow strategy diagram for the isolation of HNSCC cells and single-cell RNA sequencing (IgG = 6, Resistant = 3, Sensitive = 3 mice). F UMAP plot showing the results after unbiased clustering. Subpopulations of epithelial cells, endothelial cells, fibroblast cells, and immune cells were identified, with each cell type colored. G Copy number of epithelial cells was annotated using CopyKAT. Cells are colored by ploidy status. H UMAP plots showing the subclusters of epithelial cells. I Heatmap of characteristic genes in the subclusters of epithelial cells. Each cell cluster is represented by three specifically expressed genes. J The UMAP plots showing the Scissor-selected epithelial cells. The red and gray dots are Scissor⁺ (cluster 1, worse survival) and Scissor⁻ (cluster 0, good survival) cells. K The proportions of subclusters of epithelial cells in Scissor 0 and Scissor 1. L Volcano plot displaying the -Log10 FDR vs Log2 fold-change of genes differentially expressed between DEG1 (Scissor 0 and Scissor 1, left) and DEG2 (CD276R and CD276S, right) in Epithelial cells. R Resistant, S Sensitive, M Venn diagram displaying the overlap of DEGs in Epithelial cells. The P values in Supplementary Data 1 and 2 were calculated by the Wilcoxon rank sum test. P values are two-sided and exact. Adjustments for multiple comparisons were made using the Bonferroni correction method. Source data and exact p values are provided as a Source Data file.

Fig. 2. ITGB6 as a crucial gene that mediates divergent responses to α-CD276 treatment.

A, B Growth curves (A) and representative images (B) of HNSCC in mice transplanted in situ in different treatment groups. Data are expressed as mean ± standard deviation (SD) (n = 6 mice). C, D Representative images of ITGB6 IHC staining (C) and percentage of ITGB6⁺ cells (D) in different treatment groups. Scale bar, 50 μm. Data are presented as mean ± SD (n = 8 mice). P values are presented by two-tailed unpaired Student’s t test. E Representative image of 4NQO-induced HNSCC (upper) and display of histological characteristics of HNSCC in different groups by H&E (down) (n = 8 mice). Scale bar, 1 mm (upper) and 100 μm (lower). F Representative immunofluorescence (IF) staining images of CD8 (green) and GZMB (red) (upper). Statistical analysis of the ratio of CD8⁺ cells and CD8⁺ GZMB⁺ cells to CD8⁺ cells (CD8T killing capacity) in different treatment groups (down). Data are expressed as mean ± SD (n = 8 mice). Scale bar, 25 μm. P values were calculated by two-tailed unpaired Student’s t test. G, H Representative images of KI67 (G) and Caspase-3 (H) IHC staining (left) and quantitation of the percentage of KI67⁺ (G) and Caspase-3⁺ (H) cells (right) in different treatment groups within tumor cells. Scale bar, 100 μm. Data are shown as mean ± SD (n = 8 mice). P values were calculated by two-tailed unpaired Student’s t-test. I Quantitation of the percentage of Caspase-3⁺ cells in different treatment groups within stromal cells. Data are shown as mean ± SD (n = 8 mice). P values were calculated by two-tailed unpaired Student’s t-test. J Display of histological characteristics of HNSCC in different groups by H&E (down) and the representative image of 4NQO-induced HNSCC (upper) (n = 8 mice). Scale bar, 100 μm (down) and 1 mm (upper). K, L Quantification of tongue SCC number (J) and lesion area (K) in different treatment groups. Data are presented as mean ± SD (n = 8 mice). P values were calculated by two-tailed unpaired Student’s t-test. Source data and exact p values are provided as a Source Data file.

Fig. 3. ITGB6 knockout overcomes α-CD276 treatment resistance in mouse HNSCC.

A The tumor growth curves of different groups of mice were plotted. Measurements were taken every 4 days. Data are shown as mean ± SD (n = 8 mice). P values were calculated by one-way ANOVA and Tukey HSD test. B Display of histological characteristics of HNSCC in different groups by H&E (left) and the representative image of 4NQO-induced HNSCC (right) (n = 8 mice). Scale bar, 100 μm (left) and 1 mm (right). C Quantification of primary tumor incidence in different treatment groups (n = 8 mice). Data are presented as percentages. Statistical significance was assessed using the Pearson chi-square test. P value is exact and two-sided. D, E Quantification of tongue lesion area (D) and SCC number (E) in different treatment groups. Data are shown as mean ± SD (n = 8 mice). P values were calculated by one-way ANOVA with Tukey’s multiple comparison test. F Representative images of KI67 IHC staining (left) and quantitation of the percentage of KI67⁺cells (right) in different treatment groups. Scale bar, 50 μm. Data are shown as mean ± SD (n = 8 mice). P values were calculated one-way ANOVA with Tukey’s multiple comparison test. G Representative images of Caspase-3 IHC staining (left) and quantitation of the percentage of Caspase-3⁺cells (right) in different treatment groups. Scale bar, 50 μm. Data are shown as mean ± SD (n = 8 mice). P values were calculated by one-way ANOVA with Tukey’s multiple comparison test. H Quantitation of the percentage of Caspase-3⁺ cells in different treatment groups within stromal cells. Data are shown as mean ± SD (n = 8 mice). P values were calculated by one-way ANOVA with Tukey’s multiple comparison test. I Representative immunofluorescence (IF) staining images of CD8 (green) and GZMB (red) (left). Statistical analysis of the ratio of CD8⁺ cells (mid) and CD8⁺ GZMB⁺ cells to CD8⁺ cells (CD8T killing capacity) in different treatment groups (right). Scale bar, 25 μm. Data are expressed as mean ± SD (n = 8 mice). P values were calculated by one-way ANOVA with Tukey’s multiple comparison test. Source data and exact p values are provided as a Source Data file.

Fig. 4. ITGB6 affects the interactions between tumor cells and PF4⁺ macrophages.

A Strategy for the treatment of HNSCC mice and single-cell RNA sequencing (n = 2 mice for each group). B UMAP plot showing four cell clusters from mice HNSCC tissues, colored by cell cluster. D, days. C Heatmap of signature genes for four cell clusters. Each cell cluster is represented by three specifically expressed genes. D Copy number of epithelial cells was estimated using CopyKAT. E UMAP embedding of the inferred CopyKAT diploid and aneuploid copy number profiles (all epithelial cells). D, days. F The UMAP diagram of the immune cell subpopulation (left) and the cell ratio between the two groups (right). G Heatmap showing the ligand-receptor interaction between tumor cells and immune cell subpopulations based on CellPhoneDB. H UMAP plot of macrophage subpopulation, colored by clustering. I Circle plot showing the differences in the number and strength of intercellular communications between the two groups by CellChat. Both the strength and number of intercellular communications between PF4⁺ macro and tumor cells were significantly reduced. (The blue line represents intercellular communication downregulation, red represents up-regulation, and the width of the line represents the number of up- or downregulation). J, K Dot plot showing the expression of receptor-ligand pairs between PF4⁺ macrophages and tumor cells in WT and cKO HNSCC tissues by CellChat (J) and CellPhoneDB (K) analysis. Data are presented as communication probabilities. The size of the dots represents the communication probability, and the color scale represents the communication strength. P values were calculated using a permutation test, assessing the significance of cell–cell communication by comparing observed mean expression with a null distribution generated by random permutations. P values are two-sided and exact, with significance indicated (0.01 < P < 0.05 and P < 0.01) (J). Data are presented as mean expression levels. The size of the dots represents the -log10(P value), and the color scale represents the mean expression levels. P values were calculated using a permutation test, which generates a null distribution by randomly shuffling cell labels to determine the specificity of interactions. P values are two-sided and exact, with significance indicated (0.01 < P < 0.05 and P < 0.01) (K). Source data and exact p values are provided as a Source Data file.

Fig. 5. PF4 macrophages are accompanied by worse survival and less CD8⁺T-cell infiltration and killing capacity.

A Kaplan–Meier curve depicting the overall survival (OS) of patients with HNSCC (TCGA-Cohort) in the model group based on PF4 signature (n = 259 in PF4_low and n = 260 in PF4_high). P values were calculated by log-rank test. B Kaplan–Meier curve depicting the overall survival (OS) of patients with Pan-Cancer (TCGA-Cohort) in the model group based on PF4 signature. P values were calculated by log-rank test (n = 4892 in PF4_low and n = 4892 in PF4_high). C Kaplan–Meier curve depicting the overall survival (OS) of patients with Immunotherapy-treated renal cancer (GEO-Cohort) in the model group based on PF4 signature (n = 155 in PF4_low and n = 156 in PF4_high). P values were calculated by log-rank test. D Representative images of PF4 IHC staining in PF4_low (upper) and PF4_high (lower) (left) groups with HNSCC (FAH-SYSU-Cohort) (n = 40 in PF4_low and n = 41 in PF4_high). Scale bar, 100 μm. Kaplan–Meier curve depicting the overall survival (OS) of patients with HNSCC (FAH-SYSU-Cohort) in the model group based on PF4 expression (right). P values were calculated by log-rank test. E The experimental design of the HNSCC tumorigenesis model and treatment strategy (left) and representative image of PF4 IHC staining in CTL (upper) and DTR (down) groups (right). F Representative image of 4NQO-induced HNSCC in different groups (n = 6 mice). Scale bar, 1 mm. G Display of histological characteristics of HNSCC in different groups by H&E (n = 6 mice). Scale bar, 100 μm. H, I Quantification of tongue lesion area (H) and SCC number (I) in different treatment groups. Data are shown as mean ± SD (n = 6 mice). P values were calculated by one-way ANOVA with Tukey’s multiple comparison test. J Representative images of ITGB6 IHC staining in ITGB6_low (left) and ITGB6_high (right) groups with HNSCC (FAH-SYSU-Cohort) (n = 81 biological replicates). Scale bar, 100 μm. K Representative images of CD276 IHC staining in CD276_low (upper) and CD276_high (down) groups with HNSCC (FAH-SYSU-Cohort) (n = 81 biological replicates). Scale bar, 100 μm. L–N Correlation of PF4 IHC score with ITGB6 IHC score (L) and CD276 IHC score (M) (n = 81 for each group). And the correlation of ITGB6 IHC score with CD276 IHC score (N) (n = 81 for each group). Data are presented as scatter plots with a linear regression line. The shaded area represents the 95% confidence interval. Pearson correlation coefficients (R values) were calculated to assess the linear relationship between the variables. P values are exact and two-sided. Source data and exact p values are provided as a Source Data file.

Fig. 6. CX3CL1-CX3CR1 axis is required for PF4⁺ macrophages recruitment and activation.

A Representative images of 4NQO-induced HNSCC (upper) and corresponding HE staining (lower) in different treatment groups (n = 8 mice). Scale bar, 1 mm (upper), 100 μm (lower). B, C Quantification of SCC number (B) and tongue lesion area (C) in different treatment groups. Data are presented as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. D–G. Representative images of KI67 (D) and Caspase-3 (F) IHC staining and quantitation of the percentage of KI67⁺ (E) and Caspase-3⁺ (G) cells in different treatment groups. Scale bar, 50 μm. Data are shown as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. H–J Representative immunofluorescence (IF) staining images of CD8 (green) and GZMB (red) (H). Statistical analysis of the ratio of CD8⁺ cells (I) and CD8⁺ GZMB⁺ cells to CD8⁺ cells (CD8T killing capacity) (J) in different treatment groups. Scale bar, 25 μm. Data are expressed as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. K, L Representative flow plots show the frequency of the F4/80⁺PF4⁺ macrophages (K) and quantification of the proportions of F4/80⁺PF4⁺cells (L) in different treatment groups. Data are shown as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. M Representative flow plots show the frequency of the F4/80⁺ macrophages in different treatment groups. Data are shown as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. Source data and exact p values are provided as a Source Data file.

Fig. 7. PF4⁺ macrophages suppress the cytotoxicity and induce the exhaustion of CXCR6⁺ CD8⁺ T cells.

A Circle plot showing the differences in the number and strength of intercellular communications between the two groups by CellChat (n = 2 mice for each group). Both the strength and number of intercellular communications between PF4⁺ macrophages and CD8⁺ T cells were significantly reduced. (The blue line represents intercellular communication downregulation, red represents up-regulation, and the width of the line represents the number of up- or downregulation). B Dot plot showing the expression of receptor-ligand pairs between PF4⁺ macrophages and CD8⁺ T cells in CTL and cKO HNSCC tissues by CellChat analysis. C Data are presented as communication probabilities. The size of the dots represents the communication probability, and the color scale represents the communication strength. P values were calculated using a permutation test, assessing the significance of cell–cell communication by comparing observed mean expression with a null distribution generated by random permutations. P values are two-sided and exact, with significance indicated (0.01 < P < 0.05 and P < 0.01) (J). D UMAP plots showing the subclusters of epithelial cells. E Pseudotime evolution trajectories inferred by Monocle3 after reclustering of CD8⁺ T cells. F Proportion of three CD8⁺ T-cell subclusters. G UMAP (left) and pseudotime evolution trajectories inferred by Monocle3 (right) of three CD8⁺ T-cell subclusters with HNSCC (GEO-Cohort1) (n = 18 biological replicates). H UMAP (left) and pseudotime evolution trajectories inferred by Monocle3 (right) of three CD8⁺ T-cell subclusters with HNSCC (GEO-Cohort2) (n = 63 biological replicates). I Representative immunofluorescence (IF) staining images of CD8 (red) and GZMB (green) (left). Statistical analysis of the ratio of CD8⁺ cells (mid) and CD8⁺ GZMB⁺ cells to CD8⁺ cells (CD8T killing capacity) (right) in different groups with HNSCC (FAH-SYSU-Cohort) (n = 40 in PF4_low and n = 41 in PF4_high). Scale bar, 25 μm. Data are expressed as mean ± SD. P values were calculated by two-tailed unpaired Student’s t-test. J Representative immunofluorescence (IF) staining images of CD8 (red) and PDCD1 (green) (left). Statistical analysis of the ratio of CD8⁺ cells (mid) and CD8⁺ PDCD1⁺ cells to CD8⁺ cells (CD8T exhaustion) (right) in different groups with HNSCC (FAH-SYSU-Cohort) (n = 40 in PF4_low and n = 41 in PF4_high). Scale bar, 25 μm. Data are expressed as mean ± SD. P values were calculated by two-tailed unpaired Student’s t-test. J, K. Correlation of PF4 IHC score with the ratio of CD8⁺GZMB⁺ cells (J) and CD8⁺PDCD1⁺ cells (K) (n = 81 biological replicates). Data are presented as scatter plots with a linear regression line. The shaded area represents the 95% confidence interval. Pearson correlation coefficients (R values) were calculated to assess the linear relationship between the variables. P values are exact and two-sided. Source data and exact p values are provided as a Source Data file.

Fig. 8. CXCL16 as a crucial regulator of CXCR6⁺CD8⁺ T-Cell activation and induction of exhaustion upon overstimulation.

A Representative flow plots show the frequency of the CXCR6⁺ CD8⁺ GZMB⁺ T cells (left) and quantification of the proportions of CXCR6⁺ CD8⁺ GZMB⁺ T cells (right) in groups with different PF4⁺ Macrophage and CD8⁺ T-cell proportions. Data are shown as mean ± SD (n = 3 biological replicates). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. B Representative flow plots show the frequency of the CXCR6⁺ CD8⁺ PDCD1⁺ T cells (left) and quantification of the proportions of CXCR6⁺ CD8⁺ PDCD1⁺ T cells (right) in groups with different PF4⁺ Macrophage and CD8⁺ T-cell proportions. Data are shown as mean ± SD (n = 3 biological replicates). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. C, D Representative flow plots show frequency of the CXCR6⁺ CD8⁺ GZMB⁺ T cells (C) and quantification of the proportions of CXCR6⁺ CD8⁺ GZMB⁺ T cells (D) in groups with different concentrations of CXCL16 recombinant protein. Data are shown as mean ± SD (n = 3 biological replicates). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. E, F Representative flow plots show the frequency of the CXCR6⁺ CD8⁺ PDCD1⁺ T cells (E) and quantification of the proportions of CXCR6⁺ CD8⁺ PDCD1⁺ T cells (F) in groups with different concentrations of CXCL16 recombinant protein. Data are shown as mean ± SD (n = 3 biological replicates). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. G Representative images of 4NQO-induced HNSCC (upper) and corresponding HE staining(lower) in groups with different concentrations of CXCL16 recombinant protein. Scale bar, 1 mm (upper), 100 μm (lower). H–K Representative images of KI67 (H) and Caspase-3 (J) IHC staining and quantitation of the percentage of KI67⁺ (I) and Caspase-3⁺ (K) cells in groups with different concentrations of CXCL16 recombinant protein. Scale bar, 50 μm. Data are shown as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. Source data and exact p values are provided as a Source Data file.

Fig. 9. ITGB6 additionally facilitated tolerance to α-PD1 treatment in mice.

A The experimental design of the HNSCC tumorigenesis model and treatment strategy. w, weeks. B Representative examples of head and neck magnetic resonance imaging (MRI) for different groups (n = 8 mice). The dashed area is the boundary of the tumor. D, days. C Representative images of 4NQO-induced HNSCC (upper) and corresponding HE staining (lower) (left). Quantification of SCC number (upper) and tongue lesion area (lower) (right) in different treatment groups. Scale bar, 1 mm (upper), 100 μm (lower). Data are shown as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. D, E Representative images of ITGB6 (D), PF4 (E) IHC staining (left) and quantification of the percentage of ITGB6⁺ (D), PF4⁺ (E) (right) in different treatment groups. Scale bar, 50 μm. Data are presented as mean ± SD (n = 8 mice). P values are presented by one-way ANOVA with Tukey’s multiple comparison test. F Tumor growth curves for each mouse treated with either the combination of α-PD1 and α-ITGB6 or the control α-PD1 treatment, with measurements recorded every four days (n = 8 mice). G Representative images of 4NQO-induced HNSCC (upper) and corresponding HE staining(lower). Scale bar, 1 mm (upper), 100 μm (lower). H, I Quantification of tongue lesion area (left) and SCC number (right) in different groups. Data are shown as mean ± SD (n = 8 mice). P values were calculated by two-tailed unpaired Student’s t-test. Source data and exact p values are provided as a Source Data file.