Integrated single-cell and spatial analysis identifies context-dependent myeloid-T cell interactions in head and neck cancer immune checkpoint blockade response

Abstract

Background: Approximately 15-20% of head and neck cancer squamous cell carcinoma (HNSCC) patients respond favorably to immune checkpoint blockade (ICB). Previous single-cell RNA-Seq (scRNA-Seq) studies identified immune features, including macrophage subset ratios and T-cell subtypes, in HNSCC ICB response. However, the spatial features of HNSCC-infiltrated immune cells in response to ICB treatment need to be better characterized.

Methods: Here, we perform a systematic evaluation of cell interactions between immune cell types within the tumor microenvironment using spatial omics data using complementary techniques from both 10X Visium spot-based spatial transcriptomics and Nanostring CosMx single-cell spatial omics with RNA gene panel including 435 ligands and receptors. In this study, we used integrated bioinformatics analyses to identify cellular neighborhoods of co-localizing cell types in single-cell spatial transcriptomics and proteomics data. In addition, we used both publicly available scRNA-Seq and in-house spatial RNA-Seq data to identify spatially constrained Ligand-Receptor interactions in Responder patients.

Results: With 522,399 single cells profiled with both RNA and protein from 26 patients, in addition to spot-resolved spatial RNA-Seq from 8 patients treated with ICB together with bioinformatics analysis of publicly available single-cell and bulk RNA-Seq, we have identified a spatial and cell-type specific context-dependency of myeloid and T cell interaction difference between Responders and Non-Responders. We defined further cellular neighborhood and the sources of chemokine CXCL9/10-CXCR3 interactions in Responders, emerging targets in ICB, as well as CXCL16-CXCR6, CCL4/5-CCR5, and other underappreciated and potential markers and targets for ICB response in HNSCC. In addition, we have contributed a rich data resource of cell-cell Ligand Receptor interactions for the immunotherapy and HNSCC research community.

Discussion: Our work provides a comprehensive single-cell and spatial atlas of immune cell interactions that correlate with response to ICB in HNSCC. We showcase how integrating multiple technologies and bioinformatics approaches can provide new insights into potential immune-based biomarkers of ICB response. Our results suggested refining future studies using preclinical animal models in a more context-specific manner to elucidate potential underlying mechanisms that lead to improved ICB responses.

Keywords: cell-cell interactions; cellular neighborhoods; head and neck squamous cell carcinoma; immune checkpoint blockade; ligand-receptor interactions; myeloid cells; scRNA-seq; spatial transcriptomics.

Figure 1.. Overview of spatial and single-cell immune landscape in HNSCC treated with Immune Checkpoint Blockade (ICB)

a. Uniform Manifold Approximation and Projection (UMAP) visualization, average percentage by cell type (star annotated the cell type abundance difference between Responders (R) and Non-Responders (NR), with p<0.05, Wilcox test, also see Suppl 1A for detailed comparison), cell type gene signatures defined from differentially expressed gene (DEG) list, of HNSCC R and NR patients treated with ICB identified from Nanostring CosMx single-cell spatial protein profiling data. b. UMAP visualization, average percentage by cell type (star annotated the cell type abundance difference between R and NR, with p<0.05, Wilcox test, also see Suppl 1C for detailed comparison), cell type gene signatures defined from DEG list, of HNSCC R and NR patients treated with ICB identified from Nanostring CosMx single-cell spatial RNA profiling data. c. Major cell type compartment percentage (using computational deconvolution) by spot for representative NR and R samples sequenced with 10X Visium spatial transcriptomics, and representative cell type markers. Distribution of cell type compartment proportion in 8 R and NR samples from 10X Visium. d. Gene signatures from DEGs of different myeloid subsets, UMAP visualization from reanalyzed treatment-naïve HNSCC scRNA-seq. e. UMAP visualization of reanalyzed myeloid subsets split by PBMC and tumor samples. f. Label-transferred defined myeloid cell subsets and established T cell subsets in IBC-treated HNSCC R and NR using computational label transfer.

Figure 2.. Single-cell imaging-based spatial omics suggests enrichment of a stromal-immune mixed cellular neighborhood in Responders

a. Normalized enrichment score of cell type proportion in cellular neighborhood from protein data identified using a window-based neighborhood identification method. b. Representative fields of view (FOVs) from immune-CAF mixed neighborhood inferred from protein data: the left plot shows representative immune and stromal marker fluorescence and the right plot includes fibroblast and immune cell types within the neighborhood (one R FOV representative above, one NR FOV representative below). Cellular neighborhood frequency comparison between R and NR in protein FOVs. c. Normalized enrichment score of cell type proportion within each cellular neighborhood from RNA data identified using a window-based neighborhood identification method. d. Representative FOVs from immune-CAF neighborhood inferred from RNA data: the left plot shows representative immune and stromal marker fluorescence, and the right plot includes CAF and immune cell types within the neighborhood (one R FOV representative above, one NR FOV representative below). Cellular neighborhood frequency comparison between R and NR in protein FOVs. e. DEGs of DCs in CD45+ neighborhoods between R and NR. f. DotPlot of CXCL9/10 ligand and their receptor (CXCR3) in myeloid and T cell subsets between R and NR in Myeloid_K17+_Tumor_Mixed and ap_iCAF_Immune_Mixed neighborhoods, showing opposite frequency differences.

Figure 3.. CD14+ monocytes sharing a transcriptional signature with PBMC monocytes have increased outgoing signaling in Non-Responders

a. Schematic of in silico cell-cell interaction (CCI) analysis. Comparison analyses were plotted with interaction rank between groups (with R-enriched interactions at the lower end) on the x-axis and log2 fold change of CellChat interaction count on the y-axis. Enriched interactions are visualized with the left box representing senders, and the right box representing receiver cell types. b. Rank plot of CCIs enriched by Response for pre- and post-ICB. c. UpSet plots representing the degree of overlap of LR interactions to receiving T cell types sent from PBMC-like CD14 monocytes both pre- and post-ICB. d. Alluvial plots representing sent signal to T cell subsets from PBMC-like CD14 monocytes, pre-ICB, and the specificity and receiving cell type of these same interactions post-ICB. The left strata and links represent pre-ICB receiving cell type, and the right strata represent post-ICB specificity by Response. Grey stratum represents interactions either gained or lost following ICB. Left: post-ICB specificity of NR-specific interactions from pre-ICB. Right: post-ICB specificity of R-specific interactions from pre-ICB.

Figure 4.. Spatial and cell-type context-specific identification of myeloid-T cell interactions in HNSCC ICB Response

a. Circos plots representing CCI interaction probability from myeloid to T cell compartments by pathway for pre- and post-treatment using previously published pathways. Chord color represents pathway. Interactions from each sender/receiver pair are categorized by group specificity: specific to Responders (R), Non-Responders (NR), or shared (Sh), and converted to percentage of total probability contribution to all interactions within that pair and group specificity. b. Rank plot of CCIs within the chemokine pathway pre- and post-ICB. Interactions on the top left are enriched in R, and interactions at the bottom right are enriched in NR. Individual interactions are represented by a 2-segment box, the left box indicating sender cell type, the right box indicating receiver cell type, ligand is written in the left box, and receptor written in the right box. c. Unsupervised hierarchical clustering identified context-specific myeloid-T LR interaction differences between HNSCC ICB R and NR. (Left) Heatmap represents 7 Ligand-Receptor interaction likelihood probability patterns, merged into 3 major patterns: R-enriched, NR-enriched and Mixed. (Right) LR gene expression in representative neighborhood and cell type specificity identified for R and NR. d. DotPlot for representative LR gene expression in myeloid and T cells within select immune-rich neighborhoods.

Figure 5.. CXCL9/10-CXCR3 and CXCL16-CXCR6 are elevated in immune-rich regions of Responder tumors in Visium tissues

a. Schematic of LR bidirectional co-expression score calculation in CD45+ spots (as defined by >5% immune cells per spot from SCDC deconvolution). b. Rank plot of the difference in R vs. NR mean CD45+ spot proportion for a given LR pair. The left side of the plot highlights interactions enriched in R, and the right side highlights interactions enriched in NR. Select chemokine interactions are highlighted in red. c. Top: Spatial feature plot of CXCL9, CXCL10, CXCR3, CXCL16, and CXCR6 expression per spot for one example R (top) and NR (bottom) tumor. Bottom: Spatial feature plot and boxplots of select bidirectional co-expression LR scores per spot in CD45+ spots for R and NR tumors. The frequency of LR+ CD45+ spots among CD45+ spots was split by response for CXCL9-CXCR3, CXCL10-CXCR3, and CXCL16-CXCR6.

Figure 6.. Publicly available independent cohort validation of elevated chemokine interactions in ICB Responder patients

a. Volcano plots show DEGs of CD68+ and CD45+ RNA-Seq from Nanostring GeoMX for NR vs R, highlighting immune LR pairs inferred from Fig 4C. b. Comparison of three top LRs between Responders (n=20) vs. Non-Responders (n=76) with Ligands in CD68+ samples and Receptors in CD45+ samples. c. Comparison of CXCL9 and CD274(PD-L1) expression by Response for publicly available clinical trial data from ICB-treated HNSCC patients^,. d. Summary of main differences between myeloid-T cell interactions between HNSCC Responders vs Non-Responders.