Tumor immune microenvironment delineates progression trajectories of distinct nasopharyngeal carcinoma phenotypes

Abstract

We investigate the molecular landscape of locally advanced nasopharyngeal carcinoma (LA-NPC) subtypes: limited (L), ascending (A), descending (D), and ascending-descending (AD). Using a cohort of 994 patients, we perform germline and somatic whole-exome sequencing (WES), transcriptomic profiling, multiplex immunohistochemistry (mIHC), and spatial histopathological analyses of tumor whole-slide images (WSIs). Germline WES reveals the most variants in AD subtypes, but somatic WES shows no subtype-specific mutations. Transcriptomics reveals higher extracellular matrix (ECM) gene expression in A and AD subtypes and higher immune gene expression in D and AD subtypes, agreeing with deconvolution and mIHC. Tumor immune microenvironment (TIME) of node-negative (N0) and node-positive (N+) L subtypes, considered early nasopharyngeal carcinoma (NPC), resembles A and D subtypes, respectively, suggesting distinct evolutionary trajectories. Spatial WSI analyses identify the most immune-dense tumors among D subtypes and association of TIME with disease-free survival in AD subtypes. These findings highlight the TIME's role in LA-NPC progression and its potential impact on treatment strategies.

Keywords: ascending; descending; genomics; nasopharyngeal carcinoma; tumor immune microenvironment.

Graphical abstract

Figure 1

Clinical characteristics and germline variants associated with NPC subtypes (n = 303 [L], 201 [A], 246 [D], and 244 [AD]) (A) Representative MRI scans of each NPC subtype (red, bulky tumor; blue, small tumor). (B) CONSORT diagram of patient selection for the NCCS cohort. (C) Association of NPC subtypes with OS and DFS. (D) Manhattan plot showing p values of germline variants associated with L, A, D, or AD subtype relative to healthy controls. Points above dashed line (p = 8.9 × 10⁻⁷) are significant variants post-Bonferroni adjustment. (E) Variants significantly associated with NPC subtypes identified in both the comparison against healthy controls and the comparison against all controls. (F and G) Tumor expression of FADS1 and DFS in A-subtype patients with CC vs. TC genotype of rs72643557, respectively. The boxplots represent the IQR, with the horizontal line indicating the median. Mann-Whitney U test and log-rank test was used to compare between CC vs. TC in (F) and (G), respectively. Abbreviations: NPC, nasopharyngeal carcinoma; CONSORT, Consolidated Standards of Reporting Trials; QC, quality control; OS, overall survival; DFS, disease-free survival; L, limited; A, ascending; D, descending; AD, ascending-descending.

Figure 2

Somatic mutational landscape across NPC subtypes (n = 51 [L], 29 [A], 30 [D], and 43 [AD]) (A) Top 20 genes (top) and selected immune and NF-κB pathways (bottom) affected by somatic mutations across all NPC subtypes. (purple, genes involved in defense response, NIK-NF-κB signaling, and negative regulation of NF-κB transcription factor activity GO gene sets; ∗, genes involved in NIK-NF-κB signaling; ^#, genes involved in the negative regulation of NF-κB transcription factor activity). (B) Top 10 COSMIC SBS mutational signatures enriched across all NPC subtypes. (C) Comparison of total mutations across the NPC subtypes. (D and E) Association of mutation status with expression of CYLD and NFKBIA, respectively. The boxplots represent the IQR, with the horizontal line indicating the median. Mann-Whitney U test was used to compare between NPC subtypes unless otherwise indicated. Abbreviations: COSMIC, Catalogue of Somatic Mutations in Cancer; SBS, single base substitution; GO, Gene Ontology; KW test, Kruskal-Wallis test.

Figure 3

ECM and immune gene expression differences in NPC subtypes (n = 40 [L], 35 [A], 33 [D], and 48 [AD]) (A) Top 20 GO gene sets enriched in A-subtype versus D-subtype tumors based on whole transcriptome profiles. (B and C) Differential expression of genes in A versus D subtypes in GO ECM-related and immune-related gene sets. (D) Expression heatmaps across all NPC subtypes for the top 15 ECM- and top 15 immune-related gene sets identified from GSEA, respectively. (E and F) Proportion of each NPC subtype within GSVA score clusters as defined in (D), with comparisons between subtypes made using chi-squared test. Numbers annotated on the bar plots represent number of samples and percentage for each subtype. Abbreviations: A, ascending; D, descending; GO, Gene Ontology; ECM, extracellular matrix.

Figure 4

Immune profiling across NPC subtypes using RNA-seq deconvolution (n = 40 [L], 35 [A], 33 [D], and 48 [AD]) (A) Comparison of immune scores among NPC subtypes computed using ESTIMATE, quanTIseq, and CIBERSORT algorithms, respectively. (B) Immune scores of specific immune subsets based on quanTIseq deconvolution. The boxplots represent the IQR, with the horizontal line indicating the median. Mann-Whitney U test was used to compare between NPC subtypes.

Figure 5

Distribution of tumor and immune subsets in NPC subtypes analyzed by mIHC (n = 3 [L-N0], 4 [L-N+], 8 [A], 8 [D], and 8 [AD]) (A) Representative mIHC images of CD20⁺ B cells and CK+ tumor cells (top), CD8⁺ T cells and FOXP3+ Tregs (middle), and CLDN1+ tumor cells (bottom) across NPC subtypes. (B–G) Distribution of various cell subpopulations in mIHC images across NPC subtypes. The boxplots represent the IQR, with the horizontal line indicating the median. Mann-Whitney U test was used to compare between NPC subtypes. Scale bar, 50 μm.

Figure 6

Analysis of TIME in NPC subtypes using whole slide imaging (n = 39 [L], 35 [A], 33 [D], and 48 [AD]) (A) Workflow of spatial analyses leading to immune classification based on spatial distribution and density of immune-rich tiles within the WSI of the tumor. (B) Proportion of immune-rich tiles in WSIs for the four subtypes. (C) Mean tumor-immune tile distances in WSIs for the four subtypes. (D) Distribution of the three immune classes in WSIs for the four subtypes. (E) Association of immune classes with DFS for both NCCS and JXCH cohorts. No hazard ratio was computed for JXCH cohort due to absence of DFS events in the immune-dense class. Log-rank test was used to compare between immune classes. The boxplots represent the IQR, with the horizontal line indicating the median. Mann-Whitney U test was used to compare between NPC subtypes unless otherwise indicated. Abbreviations: WSIs, whole-slide images; DFS, disease-free survival.