Molecular and Immune Landscape of Recurrent and/or Distant Metastatic Squamous Cell Carcinoma of the Head and Neck: An EORTC/IMMUCAN Project

Abstract

Purpose:

Recurrent and/or metastatic (R/M) squamous cell carcinoma of the head and neck (SCCHN) is a heterogeneous clinical entity with a poor prognosis. The molecular and immune landscape of R/M SCCHN is underexplored. To offer a comprehensive view of the tumor microenvironment and molecular profile of R/M SCCHN, we performed an in-depth molecular and immune characterization, evaluating the impact of human papillomavirus (HPV) status, tobacco and alcohol history, primary tumor site, relapse pattern, and treatment history at the genomic, transcriptomic, and immune levels.

Experimental Design:

We analyzed 253 R/M SCCHN fresh tumor biopsies from the IMMUcan project using RNA sequencing (RNA-seq), whole-exome sequencing, and multiplex immunofluorescence (mIF).

Results:

The primary clinical factor affecting the immune microenvironment was the number of treatment lines, with significant declines in T cells and B cells observed via mIF and RNA-seq as the number of R/M treatment lines progressed. IL6, IL13, IL15, and NRF2 pathways were enriched in HPV-negative R/M SCCHN compared with HPV-positive tumors, whereas no immune differences were detected between these two clinical groups. Specific genomic alterations were observed in laryngeal cancer (DDR2, FOXP1, KLF5, and ROBO2), whereas nonsmokers/nondrinkers exhibited alterations in SPEN, PBRM1, and CYLD. 11q13.3 amplification was linked to HPV-negative metastatic tumors and hypopharyngeal cancer. HPV-negative SCCHN with locoregional recurrence showed elevated EGFR and CXCL12 pathway activity. Partial epithelial–mesenchymal transition transcriptomic signatures correlated with poor survival, whereas lymphocyte infiltration, especially in the context of tertiary lymphoid structures, was associated with improved survival.

Conclusions:

Our study highlights key molecular and immune differences across R/M SCCHN subgroups, identifies potential biomarkers, and suggests biological rationales for tailored therapeutic strategies.

Translational Relevance.

Recurrent and/or metastatic (R/M) squamous cell carcinoma of the head and neck (SCCHN) remains poorly characterized at the molecular and immune levels. Most current knowledge derives from primary tumor studies, with limited data on R/M disease. Using the IMMUcan multimodal-omics workflow, we analyzed 253 patients with R/M SCCHN, integrating genomic, transcriptomic, and immune profiling. Our findings indicate that human papillomavirus status, primary tumor site, substance abuse, relapse pattern, and systemic treatments significantly influence tumor biology. This study highlights the heterogeneity of R/M SCCHN and identifies potential biomarkers for innovative therapeutic strategies. In addition, our data suggest that lymphocyte infiltration, especially in the context of tertiary lymphoid structures, and the tumor partial epithelial–mesenchymal transition state show prognostic significance at this advanced stage. These findings expand the current understanding of R/M SCCHN, providing a foundation for precision medicine approaches tailored to distinct molecular-immune profiles.

Introduction

Squamous cell carcinoma of the head and neck (SCCHN) is the eighth most common cancer worldwide (1). It is a heterogeneous disease with different primary disease locations, etiologies, treatment approaches, and relapse patterns. SCCHN arises from the epithelium of the oral cavity, oropharynx, hypopharynx, and larynx. Its carcinogenesis is multifactorial, with smoking and alcohol as primary risk factors. Human papillomavirus (HPV) is another key risk factor, particularly for oropharyngeal cancers (2). Despite distinct clinical and molecular features, tobacco/alcohol-induced and HPV-related SCCHN share the same standard treatment. Most patients present with locally advanced disease. Despite multimodal treatment, including surgery and/or (chemo)radiation, fewer than 60% remain cancer-free at 3 years. Among those who relapse, 35% to 45% have only local/regional recurrence, whereas 55% to 65% develop distant metastases (DM; ref. 3). Fewer than 10% of patients present with DM at diagnosis (4). Platinum-based chemotherapy, programmed cell death-1 (PD-1) inhibitors, and cetuximab improve the survival of recurrent and/or metastatic (R/M) SCCHN, but prognosis remains dismal, with a median overall survival (OS) of 10 to 17 months (5). No standard of care exists for patients who progress after platinum therapy and PD-1 inhibitors (2).

Understanding the tumor microenvironment (TME) and its biology is crucial for improving cancer outcomes. However, previous attempts to characterize the molecular and immune landscape of R/M SCCHN were limited by single-omic approaches and small cohorts (6–9). Currently, our understanding of the molecular landscape of SCCHN largely stems from exome and gene expression studies of primary tumors (10). Thus, the impact of primary tumor site, relapse pattern (locoregional vs. metastatic), etiology, and prior therapies on the molecular and immune landscape is largely unknown.

Large-scale cancer multimodal studies, such as The Cancer Genome Atlas (TCGA) and the Clinical Proteomic Tumor Analysis Consortium (10, 11), have primarily focused on molecular data, whereas growing spatial proteomic datasets now allow for single-cell and spatial resolutions (12). Nevertheless, none of these studies have integrated both molecular and immune profiling in SCCHN, limiting a comprehensive TME view.

Here, we present results from the IMMUCAN SCCHN cohorts, integrating genomic, transcriptomic, and immune profiling from the IMMUCAN workflow to characterize R/M SCCHN according to HPV status, tobacco and alcohol history, primary tumor site, relapse pattern, and treatment history. We further report the prognostic significance of these clinical and molecular parameters.

Materials and Methods

Patient cohorts and inclusion criteria

Patients were enrolled either in the EORTC-SPECTA protocol (NCT02834884), a prospective academic translational research infrastructure for biomaterial collection, or in the EORTC-HNCG-1559 trial (NCT03088059), a prospective biomarker-driven study that included patients with R/M SCCHN after tumor biopsy (13). Both trials are approved by the ethics committees of each participating institution and by the health authorities of each participating country and are conducted in accordance with the Declaration of Helsinki (October 2000). All the patients provided written informed consent and consented to optional translational research and genetic testing. Patients selected for this work had to meet the following inclusion criteria: (i) R/M squamous cell carcinoma of the oral cavity, oropharynx, hypopharynx, or larynx, (ii) for whom whole blood and tumor biopsy performed for R/M disease were available, and (iii) for whom the clinical characteristics and disease outcomes were prospectively recorded. The number of screened patients and the reason for exclusions are detailed in Supplementary Fig. S1A.

Clinical data

A smoker and/or drinker patient was defined as a patient having a positive smoking history and/or a drinking history of 3 or more units of alcohol per day for a man and 2 or more units of alcohol per day for a woman. A nonsmoker/nondrinker patient was defined as a patient who never smoked and who had light consumption of alcohol (occasional or less than 2 units of alcohol per day for women and less than 3 units of alcohol per day for men). OS was defined as the time from biopsy sampling to death from any cause.

Recurrent SCCHN was defined according to Odense–Birmingham: same anatomic subsite or adjacent subsite within 3 cm of the primary lesion, time interval no more than 3 years, and same p16 status for oropharyngeal carcinomas. Second primary SCCHN was excluded, as well as locally advanced SCCHN, salivary gland, and sinonasal cancers.

Sample collection and processing

For each patient, a whole blood sample (EDTA vacutainer) and a fresh tumor formalin-fixed, paraffin-embedded (FFPE) tissue sample performed at the time of R/M disease were mandatory. A frozen tumor biopsy performed at the same time point as the FFPE tumor sample was optional. Samples were analyzed through the IMMUcan pipeline (14). The quality of the FFPE material was assessed centrally by a pathologist, and only samples of sufficient quality (more than 10% viable tumor cells) were analyzed.

Germline DNA (gDNA) extracted from whole blood was sequenced for all patients. Whole-exome sequencing (WES) and RNA sequencing (RNA-seq) were performed on FFPE tumor biopsies or on frozen tumor biopsies (Supplementary Fig. S1B). After quality control, a total of 253 patients with WES datasets were included, of which 223 had RNA-seq. Three multiplex immunofluorescence (mIF) panels were performed on the corresponding FFPE tumor biopsies of the 253 patients whose WES datasets passed quality control. After quality checks, mIF data of panel 1 (cell types), panel 2 (cell state), and panel 3 (T-cell subtypes) were available for 206, 209, and 208 patients, respectively, and are included in our analyses (Supplementary Figs. S1B and S2A).

Genomic and transcriptomic profiling experiments

DNA and RNA from FFPE and frozen tissue were isolated using the AllPrep DNA/RNA FFPE kit (QIAGEN) or MagMAX FFPE DNA/RNA Ultra Kit (Thermo Fisher Scientific). gDNA from EDTA tubes was isolated using the PAXgene Blood DNA kit (QIAGEN). Exome libraries were generated using the Twist Human Core Exome Kit supplemented with the RefSeq and Mitochondrial Panel for enrichment (Twist Bioscience) or KAPA HyperExome kit (Roche). Tumor DNA was sequenced at a coverage of 100 to 150× or higher, and gDNA was sequenced at a coverage of 30× or higher.

Depending on the quality and quantity of the available RNA input material, the total RNA libraries were prepared using the KAPA RNA HyperPrep Kit with RiboErase (H/M/R) Globin (Roche) or the SMART-Seq Stranded Kit (Takara Bio). Next-generation sequencing was performed on an Illumina NovaSeq 6000 platform (Illumina) with a 2 × 100 bp read mode. The generated sequencing data were demultiplexed using Illumina bcl2fastq (2.20, RRID: SCR_015058), and adapters were trimmed with Skewer (version 0.2.2, RRID: SCR_001151).

Genomics data processing

Somatic mutation calling

WES data have been analyzed using the Nextflow VEGAN pipeline (v2.1.1) available at https://github.com/bioinfo-pf-curie/vegan. Briefly, sequencing reads were aligned to the Human hg38 genome using BWA-MEM (RRID: SCR_022192). Reads intersecting the exome capture with a minimum mapping quality of 20 were kept for downstream analysis. Duplicates were removed using the Picard MarkDuplicates tool (RRID: SCR_006525). Somatic mutations were called using paired tumor-normal samples, following GATK good practices with the Mutect2 software (RRID: SCR_026692).

Somatic variant annotation was performed with SnpEff (RRID: SCR_005191). Only somatic variants with a variant allele frequency >5%, a sequencing depth of 20, and annotated as coding, nonsynonymous variants were used. Polymorphisms (based on gnomAD and 1000 Genomes projects) were discarded.

Variants in oncogenes and tumor suppressor genes listed in the OncoKB Cancer Gene list (downloaded on January 26, 2024, RRID: SCR_014782; ref. 15) or COSMIC Cancer Gene Census list (downloaded on January 26, 2024, RRID: SCR_002260; ref. 16) were analyzed. For oncogenes, in-frame insertions or deletions known in Cancer Hotspots and missense variants known in Cancer Hotspots were retained, as well as variants in the promoter region of TERT (c.−146, −124, −57, −138, and −139). For tumor suppressor genes, frameshift insertions or deletions, splicing variants, stop-gained variants with nonsense-mediated decay, in-frame insertions or deletions known in Cancer Hotspots, and missense variants known in Cancer Hotspots were retained.

Tumor mutational burden

Tumor mutational burden (TMB) was called using the pyTMB package included in the VEGAN pipeline. TMB-high was defined as TMB ≥10 mutations/Mb, whereas TMB-low was defined as TMB <10 mutations/Mb.

Copy-number alterations analysis

Copy-number alteration (CNA) analysis was performed using the FACETS tool (RRID: SCR_026264) in matched normal-tumor mode, included in the VEGAN pipeline. For each alteration, FACETS returns the copy number and allelic status.

Amplifications were defined as a minimum of five copies in addition to the ploidy, whereas deletions were defined as homozygous deletions (0 copies). Focal amplifications (≤10 Mb) of oncogenes and deletions of tumor suppressor genes listed in the OncoKB Cancer Gene List (downloaded on January 26, 2024; ref. 15) or COSMIC Cancer Gene Census list (downloaded on January 26, 2024; ref. 16) were analyzed. We added E2F1 to the list of oncogenes, E2F1 amplifications being one of the main CNAs found in HPV-positive SCCHN (10). Genes significantly more frequently altered in FFPE samples than in frozen samples were filtered out (Fisher exact test, P value < 0.05). Moreover, only genes whose amplifications translated into higher RNA expression levels compared with RNA levels of tumors with neutral copy numbers [Wilcoxon test, adjusted P value Benjamini–Hochberg (BH) < 0.05] were kept, as well as genes whose homozygous deletions translated into lower RNA expression levels compared with RNA expression levels of tumors with neutral copy numbers (Wilcoxon test, adjusted P value < 0.05). Oncogenes CCND1, FGF3, FGF4, and FGF19 listed in OncoKB are located on the same locus 11q13.3. This locus contains 10 genes mainly amplified together in SCCHN (17). We therefore analyzed the frequency of the locus rather than each gene individually. Supplementary Table S1 summarizes the list of CNAs analyzed.

Purity and ploidy

The tumor ploidy and purity were estimated by FACETS. Patients without a purity value (NA given by FACETS) were excluded from all analyses. The ploidy score was only reported for frozen tumor biopsies (n = 102).

Fraction genome altered

To infer the fraction of the genome altered, we implemented the script infer_pga.py from FACETS with default parameters. The fraction genome altered (FGA) was only reported for frozen tumor biopsies (n = 102).

Pathways analysis

Fourteen selected oncogenic pathways were analyzed. For each pathway, genes common to the genes listed in the reference articles and to the oncogenes or tumor suppressor genes listed in the OncoKB Cancer Gene List (downloaded on January 26, 2024; ref. 15) or COSMIC Cancer Gene Census list (downloaded on January 26, 2024; ref. 16) were analyzed (Supplementary Table S2). Per sample, a pathway was considered altered if any of the genes harbored a somatic mutation, an amplification, or a deletion as described above.

Transcriptomics data processing

RNA quantification

Paired-end sequencing reads were aligned using STAR (version 2.7.10b, RRID: SCR_004463) against the GRCh38/hg38 human reference genome. Gene-level counts were computed with LiBiNorm using the Gencode v29 assembly.

Transcriptomic feature extraction

Prior to the different feature extraction methods, gene expression was normalized using the variance-stabilizing transformation from the DESeq2 library (RRID: SCR_015687).

Data-driven modules

We filtered genes based on expression to remove low-expressed genes (threshold = 6.5), resulting in 23,563 genes. From these, we selected the 10,000 most variable genes to find data-driven modules. For ICA modules, we computed z-scores for each gene and used the stabilized independent component analysis (ICA) algorithm from Captier and colleagues (18) with 35 components. For the community detection approach, we used the leiden algorithm (19) on the correlation matrix with a resolution of 0.55. The list of genes present in each module is available in Supplementary Table S3. To annotate the modules, we used the 100 genes with the most weight in the ICA modules and all the genes in the leiden modules. We used ShinyGO for annotation, along with g:Profiler and ToppFun to confirm unclear cases.

Immune deconvolution

We applied the consensusTME method (20) on the normalized expression matrix, with the “HNSC” and “ssgsea” options. This yielded a number of immune and stromal populations, as well as a general RNA immune score computed as a combination of the gene sets of all the different immune cells. This immune score correlated almost perfectly (Spearman’s p > 0.9) with the immune score computed with ESTIMATE (https://bioinformatics.mdanderson.org/estimate/).

Cancer pathway and cytokine activity scores

We computed pathway activity using Progeny (21) and cytokine activity scores using Cytosig (22).

Literature signature calculation

We gathered signatures for cancer-associated fibroblasts and epithelial–mesenchymal transition (EMT) from the literature (23–26) and computed a score using ssgsea.

HPV status and HPV virus identification

The HPV status was primarily determined by RNA-seq. To detect HPV presence in the tumor samples from bulk RNA-seq, we adapted the pipeline designed by Khan and colleagues (27). We extracted the reads that did not map to the human genome (using HISAT2 with default parameters, RRID: SCR_015530) and aligned them using Bowtie 2 (RRID: SCR_016368) to a viral reference created with the genomes of all the different HPV types and the Epstein-Barr virus (EBV) genome. We considered a hit if there were more than 1,000 reads concordantly and uniquely aligned to one of the viral genomes.

For patients without RNA-seq data available (n = 30), HPV-positive patients were defined as those with p16-positive oropharyngeal cancer for whom the HPV virus was detected either by DNA polymerase chain reaction (PCR) or in situ hybridization (ISH). In total, HPV status was determined to be positive for 51 patients: 49 by RNA-seq, 1 by ISH, and another by DNA PCR.

mIF analysis

Three mIF panels were performed on consecutive slides of FFPE tumor biopsies. The first panel was designed to detect immune cell populations. It contains antibodies against CD15 (cat. #301902; RRID: AB_314194), CK (cat. #M3515; RRID: AB_2132885), CD163 (cat. #MB460; RRID: AB_3714914), CD11c (cat. #111M-15; RRID: AB_3714915), CD20 (cat. #M0755; RRID: AB_2282030), and CD3 (cat. #A0452; RRID: AB_2335677; Supplementary Fig. S3A and S3B). The second panel was designed to identify the cell state of tumor cells and CD8⁺ T cells. It contains antibodies against CD8 (cat. #108R; RRID: AB_2892088), CK, and PD-1 (cat. #315M-96; RRID: AB_1160829), PD-L1 (cat. #13684; RRID: AB_2687655), Ki-67 (cat. #275R-16; RRID: AB_1158037), and Granzyme B (cat. #MON7029C; RRID: AB_2114692; Supplementary Fig. S3C and S3D). The third panel was designed to identify the different T-cell subtypes. It contains antibodies against CD3 and CD4 (cat. #104R; RRID: AB_1516770), CD8 and FoxP3 (cat. #ab99963; RRID: AB_10675258), CD56 (cat. #156R-9; RRID: AB_2941091), and CK (Supplementary Fig. S3E and S3F). Antibody references are available in Supplementary Table S4. Staining procedures are described in detail by Eling and colleagues (28).

mIF data processing

mIF image analyses were performed with the IFQuant software, as previously described (28). Briefly, tissue segmentation was performed using the tumor cell marker (CK) to assign each cell to the tumor tissue region or to the stroma region. With panel 1, we identified tumor cells and six different immune cell phenotypes: B cells, T cells, BnT cells (cells positive for CD20 and CD3), neutrophils, dendritic cells (DC), and macrophages CD163⁺ (MacCD163). Cells not assigned to any of the phenotypes were defined as “other.” With panel 3, we identified six different T-cell and natural killer (NK) cell phenotypes: CD4⁺ T cells, CD8⁺ T cells, CD4⁺CD8⁺ T cells, regulatory T cells (Treg), NK cells, and NK T cells. Marker combinations for each cell type are available in Supplementary Table S5.

Proportions of each immune cell phenotype were computed for the total region (dividing by the total number of cells), for the stroma region (dividing by the total number of cells present in the stroma region), and for the tumor region (dividing by the total number of cells present in the tumor region). With panel 2, we identified and analyzed two tumor phenotypes: tumor cells Ki-67+ (CK+ Ki-67+) and tumor cells PD-L1+ (CK+ PD-L1+). Proportions of each tumor phenotype were calculated by the sum of tumor cells positive for the marker divided by the total number of tumor cells.

Immune phenotypes were defined based on the proportion of CD8⁺ T cells present in the different regions. A tumor sample was defined as CD8⁺ T cell immune infiltrated if the sum of the proportions of CD8⁺ T cells and CD4⁺CD8⁺ T cells in the tumor region was higher than 3%, CD8⁺ T cells immune excluded if the sum of the proportions of CD8⁺ T cells and CD4⁺CD8⁺ T cells in the stroma region was higher than 3%, and CD8⁺ T cells immune desert if the sum of the proportions of CD8⁺ T cells and CD4⁺CD8⁺ T cells in the tumor and stroma regions was lower than 3% (the three categories are illustrated in Supplementary Fig. S2B).

Tertiary lymphoid structures (TLS) were defined as patches of B cells (CD20⁺ cells), with a local B cell density above 2,000 cells/mm² and at least 40 cells (28). TLS were considered present in a sample if at least one TLS was identified in the region of interest. An example of TLS can be found in Supplementary Fig. S2C.

Data quality control

Sample labeling checking

We verified samples matching among gDNA, tumor DNA, and tumor RNA, and we verified the correspondence between the clinical gender variable and RNA gender determination.

The verification that the gDNA, tumor DNA, and tumor RNA FASTQ files originated from the same donor was performed using an in silico genotyping test. A custom Perl script first extracts the allele frequency at each genomic position specified in the bed file associated with the HumanCore-12v1-0 Illumina kit and then compares the overall concordance frequency of all alleles. For two patients, gDNA did not correspond to tumor DNA and RNA. Those patients were discarded.

We determined the transcriptomic sex and compared it with clinical sex as a rapid quality measure by looking at the expression of XIST and DDX3Y (female and male specific genes, respectively). Following this procedure, two samples were identified for potential gender mislabeling. These data were later verified with the patients’ site. Both were confirmed by the sites as data entry errors and were corrected in the final clinical data version.

FFPE and frozen tumor biopsies

For 151 patients, we analyzed WES datasets from FFPE tumor biopsies, and for 102 patients, we analyzed WES datasets from frozen tumor biopsies (Supplementary Fig. S1B). No significant difference in TMB and single-nucleotide variants was observed between FFPE and frozen tumor biopsies. For CNA, significant differences were observed between FFPE and frozen tumor biopsies: FFPE tumor biopsies harbored significantly more deletions and had higher ploidy and higher FGA (Supplementary Fig. S4A). In addition to those tests, we analyzed 12 patients’ paired WES datasets coming from a frozen tumor biopsy and the corresponding FFPE tumor biopsy (numbers not included in Supplementary Fig. S1B). Paired tests were performed, and the same results were obtained (Supplementary Fig. S4B). We therefore focused our copy-number analysis on genes that were not significantly more frequently altered in FFPE tumor biopsies than in frozen tumor biopsies, and we report the ploidy and FGA only for patients with frozen tumor biopsies analyzed (n = 102).

For RNA-seq analyses, 125 patients had RNA-seq datasets from FFPE tumor biopsies, and 98 patients had RNA-seq datasets from frozen tumor biopsies (Supplementary Fig. S1B). The variable sample conservation type (FFPE or frozen tumor biopsy) was always included in the multivariate analyses.

mIF analyses were always performed on the FFPE tumor biopsies (Supplementary Fig. S1B).

Statistical analyses

Statistical analyses were performed using R version 4.1.0.

WES data analysis

For frequencies, a Fisher exact test was performed. For continuous variables with two categories, a Wilcoxon rank-sum test was performed, and for continuous variables with more than two categories, a Kruskal–Wallis test was performed. For the variable number of R/M treatment lines, a linear regression was performed. Multiple comparisons were adjusted using the BH correction. For mutated genes and CNA frequencies, adjusted P values were obtained for all genes altered in at least 3% of the studied population. To ensure that each significant alteration identified (nominal P value < 0.05) was specific to the tested variable, multivariate analyses using logistic regression models with key clinical variables (primary diagnosis location, disease extent, substance abuse, number of lines of R/M treatments, and biopsy sampling site) were performed and are available in Supplementary Tables S3–S7. Only alterations significant in univariate and multivariate models are reported in the main text of this article.

For TMB, a linear regression model was fitted, including technical variables [DNA assay library, sample conservation type (FFPE vs. frozen), and biopsy sampling site]. For chromosomal instability scores (ploidy and FGA), a linear regression model was fitted, including technical variables (DNA assay library and biopsy sampling site). For those genomic scores, if the P value was <0.05, a multivariate analysis was performed in a second step, including the main clinical variables (primary diagnosis location, disease extent, and number of lines of R/M treatments).

RNA-seq data analysis

To ensure that the results were not due to confounding technical or clinical variables, we used a two-step approach. First, we fitted a linear model, including as covariates the biopsy sampling site, the sample conservation type (FFPE or frozen), the RNA extraction method, the RNA library protocol used, and the sample size (in a binarized form, taking the median as the threshold). We corrected the obtained P values for each clinical variable of interest using the BH correction. Then, instead of adding all possible clinical variables to the multivariate model, we added the ones that showed a significant effect (adjusted P < 0.05 in the previous test). In the results, we report the transcriptomic features that passed these filters.

mIF data analysis

For continuous variables with two categories, a Wilcoxon rank-sum test was performed, and for continuous variables with more than two categories, a Kruskal–Wallis test was performed. For the variable number of treatment lines, a linear regression model was fitted. Multiple comparisons were adjusted using BH correction per panel.

Survival analyses

Survival data were available for 204 patients. The median follow-up time was 8.14 months, starting from the biopsy to death or the last known follow-up date at the time of study base lock.

A multivariate Cox proportional hazards model was independently fitted to each of the features (genomic, transcriptomic, and mIF). For the genomic features, we included only those present in at least 3% of the patients. In the multivariate Cox model, we included the following clinical variables: Eastern Cooperative Oncology Group (ECOG) performance status, substance abuse, HPV status, primary diagnosis location, number of previous treatment lines, and disease extent. P values were concatenated and corrected for multiple testing using the BH correction.

To create the four prognostic subgroups, we took the two most significant features in each direction [leiden 44 for worse prognosis, renamed partial EMT (p-EMT), and leiden 115 for better prognosis, renamed TLS-germinal center] and computed the four quartiles for each of them. We then selected Q1 as high and Q2 to Q4 as low, as this grouping yielded the best separation in prognosis.

Results

Patient characteristics

A total of 253 patients with R/M SCCHN were analyzed (Supplementary Fig. S2A). Primary diagnosis sites were the oral cavity for 57 (23%) patients, the oropharynx for 116 (46%), the hypopharynx for 46 (18%), and the larynx for 34 (13%), respectively. Fifty-one (20%) patients with SCCHN were HPV-positive. One hundred seventy-five (69%) patients had DM, whereas 78 (31%) patients recurred only locally and/or regionally. Patients were heavily pretreated, with 57% of the patients having already received two or more treatment lines for R/M disease before tumor biopsy. Clinical characteristics are summarized in Table 1.

Table 1.

Clinical characteristics.

Characteristic	Total (N = 253)	HPV-negative (N = 202)	HPV-positive (N = 51)
Sex
Female	44 (17.4%)	37 (18.3%)	7 (13.7%)
Male	209 (82.6%)	165 (81.7%)	44 (86.3%)
Age (years)
Median (min, max)	62 (21, 87)	62 (21, 87)	59 (40, 81)
ECOG
0	58 (22.9%)	42 (20.8%)	16 (31.4%)
1	178 (70.4%)	147 (72.8%)	31 (60.8%)
2	10 (4%)	8 (4%)	2 (4%)
Missing	7 (2.8%)	5 (2.5%)	2 (3.9%)
Substance abuse
Nonsmoker/nondrinker	45 (17.8%)	24 (11.9%)	21 (41.2%)
Smoker and/or drinker	207 (81.8%)	177 (87.6%)	30 (58.8%)
Missing	1 (0.4%)	1 (0.5%)	0 (0%)
Primary disease location
Oral cavity	57 (22.5%)	52 (25.7%)	5 (9.8%)
Oropharynx	116 (45.8%)	75 (37.1%)	41 (80.4%)
Hypopharynx	46 (18.2%)	44 (21.8%)	2 (3.9%)
Larynx	34 (13.4%)	31 (15.3%)	3 (5.9%)
HPV status
HPV-negative	202 (79.8%)	202 (100%)	0 (0%)
HPV-positive	51 (20.2%)	0 (0%)	51 (100%)
Primary disease stage
I–II	26 (10.3%)	20 (9.9%)	6 (11.8%)
III	35 (13.8%)	24 (11.9%)	11 (21.6%)
IVA–B	124 (49%)	105 (52%)	19 (37.3%)
IV-C	44 (17.4%)	36 (17.8%)	8 (15.7%)
Missing	24 (9.5%)	17 (8.4%)	7 (13.7%)
Primary disease grade
G1	45 (17.8%)	41 (20.3%)	4 (7.8%)
G2	87 (34.4%)	70 (34.7%)	17 (33.3%)
G3	68 (26.9%)	48 (23.8%)	20 (39.2%)
Missing	53 (20.9%)	43 (21.3%)	10 (19.6%)
Disease extent
Distant metastatic disease	175 (69.2%)	135 (66.8%)	40 (78.4%)
Locoregional only disease	78 (30.8%)	67 (33.2%)	11 (21.6%)
Number of R/M treatment lines
Prior to biopsy
M naïve	19 (7.5%)	15 (7.4%)	4 (7.8%)
0	45 (17.8%)	41 (20.3%)	4 (7.8%)
1	46 (18.2%)	33 (16.3%)	13 (25.5%)
2	71 (28.1%)	59 (29.2%)	12 (23.5%)
3	52 (20.6%)	41 (20.3%)	11 (21.6%)
>3	20 (7.9%)	13 (6.4%)	7 (13.7%)
Biopsy site
Primary site	102 (40.3%)	92 (45.5%)	10 (19.6%)
Regional lymph node	57 (22.5%)	44 (21.8%)	13 (25.5%)
Distant metastasis	94 (37.2%)	66 (32.7%)	28 (54.9%)

Abbreviation: M naïve, upfront metastatic patients naïve to any treatment.

The most frequent genomic alterations and immune cell proportions of the entire cohort are shown in Supplementary Fig. S2D and S2E.

R/M HPV-positive and HPV-negative SCCHN have distinct molecular profiles but similar immune infiltration

The molecular and immune differences of HPV-related malignancies have been largely characterized in early disease, but little is known in heavily pretreated patients. We investigated whether HPV-positive SCCHN retained distinct molecular and immune traits in the R/M setting compared with HPV-negative cases. Forty-four (88%) patients were HPV16-positive, three were HPV33-positive, and one each had HPV31, HPV35, or HPV45. HPV type was undetermined for one patient due to missing RNA-seq data.

The most frequently mutated and amplified genes in HPV-positive and HPV-negative patients are shown in Supplementary Figs. S5A, S5B, and S6A–S6D, with frequencies similar to those previously described in primary tumors (10).

HPV status did not affect TMB levels. Ploidy and FGA were significantly higher in HPV-negative SCCHN compared with HPV-positive SCCHN (median FGA: 63% vs. 26%, P = 2.4e−7, Wilcoxon ranked-sum test; median ploidy: 2.85 vs. 2.07, P = 0.001, Wilcoxon ranked-sum test). 11q13.3 amplification reached 46%, and CDKN2A deletion reached 27% in HPV-negative tumors (Supplementary Fig. S6D). In HPV-positive SCCHN, the main CNAs were 11q13.3 amplifications (10%), CYLD deletions, and AKT2 and E2F1 amplifications (8%; Supplementary Fig. S6C). The cell cycle, p53, and Hippo pathways were significantly more frequently altered in HPV-negative SCCHN (nominal P < 0.01, Fisher exact test). In HPV-positive tumors, the PI3K and ubiquitin pathways were the most frequently altered pathways (Fig. 1A; Supplementary Table S6).

Figure 1.

HPV status affects the genomic and transcriptomic profiles but not the immune infiltration. A, Oncogenic pathway alterations in HPV-positive (n = 51) and HPV-negative SCCHN (n = 202). B, Correlation network of transcriptomic features affected by HPV status (n = 223). Significant results come from a multivariate model including technical covariates (see “Materials and Methods”), tumor primary location, smoker/drinker status, disease extent, and number of systemic treatment lines received. C and D, Comparison of the RNA-seq immune score according to HPV status in all samples (J, n = 223) or only metastatic patients naïve to any treatment (K, n = 14; HPV-positive = 3). P values from the Wilcoxon test. E, Immune cell proportions in the total tumor region in HPV-positive (n = 42) and HPV-negative (n = 164) R/M SCCHN. BnT cell, B near T cells defined as CD20⁺ and CD3⁺ cells; MacCD163, macrophages CD163⁺.

HPV-positive SCCHN exhibited a distinct transcriptome from HPV-negative SCCHN (Fig. 1B). Twenty-three percent of the tested features (60/256) were significantly associated with HPV status (adjusted multivariate P < 0.05, linear model). The most differential feature, an unsupervised module (leiden 27) containing meiotic genes (MEI1, SYCP2, and SMC1B), accurately identified HPV-positive tumors (specificity = 99%, sensitivity = 92%, and adjusted multivariate P = 4e−57, linear model; Supplementary Fig. S6E). Transcriptomic modules related to the cell cycle and DNA repair/metabolism were enriched in HPV-positive tumors (leiden 50, 55, 119, and 129 and ICA 16; Fig. 1B), including one involved in DNA integrity, repair, and chromosome maintenance (XRCC2, POLQ, and C1orf112; adjusted multivariate P = 6e−12, linear model, Supplementary Fig. S6F).

HPV-negative associated features were FGF2 pathway activation and several transcriptomic modules (leiden 28, 32, and 99 and ICA 24) without specific biological enrichment but with a strong presence of long noncoding RNAs (lncRNA; Supplementary Fig. S6G). Six of the eleven (LINC01980, LINC00858, etc.) lncRNAs in these modules have been implicated in oncogenesis (29, 30). MAGE-type antigens (SAGE1, MAGEA10, POTEE, CT45A1, etc.) formed an unsupervised module (ICA 12) that was more expressed in HPV-negative tumors, as were the genes located on 11q13.3 (leiden 60). Several other pathways, including NRF2, EGFR, IL6, IL13, and IL15, had higher expression in HPV-negative tumors (Fig. 1B).

We did not observe any differences in immune score (Fig. 1C) or deconvoluted immune cells (data not shown) about HPV status. However, in the small subgroup of metastatic patients naïve to any treatment with RNA-seq data available (n = 14), we retrieved higher immune scores among HPV-positive tumors (n = 3) compared with HPV-negative tumors (n = 11; P = 0.02; Wilcoxon rank-sum test; Fig. 1D). Immunofluorescence analyses confirmed the transcriptomic data (Fig. 1E). Again, in metastatic treatment-naïve patients, we observed a trend toward higher immune infiltration in HPV-positive tumors (Supplementary Fig. S5C). As 59% (30/51) of HPV-positive SCCHN in our cohort were also smokers and/or drinkers, we hypothesized that this may contribute to the absence of differences in immune infiltration. We therefore compared immune infiltration in HPV-negative SCCHN (n = 174) to nonsmoker/nondrinker HPV-positive SCCHN (n = 21; Supplementary Fig. S5D), yet no significant differences were observed.

Altogether, these findings indicate that HPV-positive tumors retain their characteristic genomic and transcriptomic identity even at this advanced R/M stage, but they lose their enhanced immune infiltrate.

Genomic alterations in R/M HPV-negative SCCHN vary according to tobacco and alcohol exposure and primary tumor location

The effects of substance abuse and anatomic subsite on the molecular and immune landscape have been described in early localized disease, but they are unexplored in R/M SCCHN. We defined a nonsmoker/nondrinker patient as a patient who never smoked and had light consumption of alcohol (occasional or less than 2 units of alcohol per day for women and less than 3 units of alcohol per day for men). Among 202 HPV-negative patients, 24 (12%) were nonsmoker/nondrinker patients. Their tumors harbored significantly more mutations in SPEN (17% vs. 1%), PBRM1 (12% vs. 1%), and MTOR (8% vs. 0%), more deletions of CYLD (8% vs. 0%), and more splicing pathway alterations (29% vs. 7%; nominal P < 0.05, Fisher exact test, Supplementary Table S4). Conversely, these tumors had fewer 11q13.3 amplifications (25% vs. 47%; nominal P < 0.05; Fisher exact test; Supplementary Fig. S7A; Supplementary Table S7).

With respect to primary tumor location (Supplementary Table S8), hypopharynx tumors harbored significantly more 11q13.3 amplifications (70% vs. 25%–47%; nominal P < 1e−4; Fisher exact test, Supplementary Fig. S7B) and CDC73 mutations (7% vs. 0%–1%; nominal P < 0.05; Fisher exact test, Supplementary Fig. S7B). Laryngeal cancers showed a specific genomic profile: DDR2, FOXP1, KLF5, and ROBO2 mutations were exclusive to laryngeal cancers (6% each), and they showed higher mutation rates and increased amplifications in several genes (nominal P < 0.05; Fisher exact test; Supplementary Fig. S7B).

In contrast, smoking status and primary cancer location had limited impacts on immune and transcriptomic profiles (data not shown).

The molecular landscape of locally and/or regionally recurring SCCHN differs from that of SCCHN with DM

As some patients developed only locoregional recurrence (LRR), we postulated that comparing them with patients with DM could reveal different immune or molecular landscapes that may help to tailor therapy.

HPV-negative LRR SCCHN (n = 67) harbored significantly more NOTCH1 (21% vs. 6%), NFE2L2 (18% vs. 7%), and TGFBR1 (5% vs. 0%) mutations and more alterations in the splicing pathway (18% vs. 7%) than DM SCCHN (n = 135; nominal P < 0.05; Fisher exact test). HPV-negative DM SCCHN had significantly more amplifications of 11q13.3 (56% vs. 25%, nominal P = 4e−5; Fisher exact test; Fig. 2A; Supplementary Table S9). In HPV-positive patients, TP53 mutations were more frequent in LRR (3/11, 27%) than in DM (1/40, 2.5%; nominal P < 0.05; Fisher exact test; Fig. 2B).

Figure 2.

The molecular landscape of SCCHN that recurs only locally and/or regionally is different from that of SCCHN with DM. A, Significant differences in genomic alterations in locoregional-only versus distant metastatic HPV-negative SCCHN (n = 202). P value from the Fisher exact test. B, Significant differences in genomic alterations in locoregional-only versus distant metastatic HPV-positive SCCHN (n = 51). P value from the Fisher exact test. C, Transcriptomic features significantly associated with disease extent (n = 223). Significant results come from a multivariate model including technical covariates (see “Materials and Methods”), tumor primary location, smoker/drinker status, disease extent, and number of systemic treatment lines received. D, Correlation between the EGFR activity score and two unsupervised modules: leiden 12 (left) and ICA 27 (right; n = 223). E, Biological annotation of the ICA 27 module. The annotation was done with ShinyGO using the Gene Ontology Biological Process (GO BP) database. F, Effect of EGFR ligands, receptor expression, and EGFR copy number in the activation of the EGFR pathway (n = 223). P values from the linear model including the variables are represented; * < 0.05; *** < 0.001. The x-axis (estimates) corresponds to the estimated effect of the parameter. Error bars correspond to the 95% confidence interval (CI). G, Left: Effect of CXCL12, ACKR3, and CXCR4 (CXCL12 receptors) expression, as well as HRAS expression and mutation, in the activation of the CXCL12 pathway (n = 223). P values from the linear model including the variables are represented; * < 0.05; *** < 0.001. The x-axis (estimates) corresponds to the estimated effect of the parameter. Error bars correspond to the 95% CI. Right: CXCL12 pathway activity according to the mutation status of HRAS (n = 223). P value from the Wilcoxon test. CNV, copy number variation; SNV, single-nucleotide variation; WT, wild-type.

Disease extent strongly influenced the transcriptome: 28% of tested features (72/256) differed significantly between LRR and DM patients (multivariate adjusted P < 0.05, linear model; Fig. 2C). LRR tumors exhibited higher activity of EGFR/VEGF/MAPK pathways, hypoxia signature, and CXCL12, IL22, OSM, IL10, and CSF3 signaling. DM SCCHN showed higher expression of the aforementioned lncRNA modules (leiden 28, 32, and 99) and modules linked to cell projection morphogenesis (ICA 33; Supplementary Fig. S8A).

EGFR pathway activity exhibited higher expression in LRR samples (Fig. 2D) and correlated with some unsupervised modules (leiden 12, ICA 27, and leiden 44). ICA 27 was enriched in genes involved in morphogenesis, cell migration, and cell adhesion (Fig. 2E). Leiden 44 and ICA 27 showed a high overlap with the p-EMT signature described by Puram and colleagues (31), indicating that the signature is replicable and relevant in our dataset (Supplementary Fig. S8B). Furthermore, published SCCHN EMT signatures (25) did not exhibit differences between LRR and DM samples, suggesting limited applicability to clinical samples. Interestingly, the p-EMT program expression was higher in LRR samples, and as indicated above, it was correlated to EGFR pathway activity. Leiden 12 contained EGFR ligands, including AREG, EREG, and HBEGF (Supplementary Table S3), and their expression was tightly linked to pathway activation (Supplementary Fig. S8C). To investigate if genomic alterations contributed to pathway activation, we fitted a linear model including these ligands, EGFR expression, and EGFR copy number. No EGFR mutations were detected in our cohort. A significant effect was only observed for HBEGF and AREG expression (Fig. 2F).

CXCL12 signaling also had higher expression in LRR disease and has been proposed to be involved in the mechanism of action of tipifarnib, a farnesyl transferase inhibitor, active in HRAS-mutated R/M SCCHN (32). We therefore investigated whether HRAS mutations and expression were associated with CXCL12 activity. HRAS mutations were present in six of our patients (2%): four LRR and two DM. We fitted a linear model including HRAS mutations and expression, CXCL12 expression, and the expression of two receptors, CXCR4 and ACKR3. HRAS mutations and expression were independently associated with higher CXCL12 pathway activity (Fig. 2G).

Immune infiltrates did not differ between LRR and DM SCCHN (Supplementary Fig. S8D). In summary, we identified several differences between tumors that remain locoregional and those that acquire metastatic ability, highlighting potential biomarkers and therapeutic targets.

Treatment history has an impact on the molecular and immune landscape of R/M SCCHN

The majority of patients in our cohort were heavily treated before biopsy sampling (Table 1). A small subgroup of patients was treatment-naïve (n = 19); these patients were metastatic at diagnosis. We investigated the influence of treatment history on the molecular and immune landscape.

In HPV-negative SCCHN, TMB and ploidy increased with the number of treatment lines administered in the R/M setting (nominal P < 0.05, linear model; Fig. 3A; Supplementary Fig. S9A). CDKN2B deletions, KIT and MYCL amplifications, and alterations in the cell cycle and MYC pathways significantly increased with the number of treatment lines (nominal P < 0.05; linear model; Fig. 3B). Conversely, NOTCH1 mutations were more frequent in SCCHN not yet treated for R/M disease (nominal P < 0.05, linear model; Fig. 3B; Supplementary Table S10). Multivariate analyses integrating the different systemic treatments revealed that CDKN2B deletions, KIT amplifications, and cell-cycle alterations were not related to any specific treatment, whereas MYC pathway alterations were significantly more frequent after exposure to anti–PD-(L)1 treatment (multivariate P < 0.01, linear model).

Figure 3.

Treatment history has an impact on the molecular and immune landscape of R/M SCCHN. A, TMB (n = 202: M naïve n = 15, 0 n = 41, 1 n = 33, 2 n = 59, 3 n = 41, and >3 n = 13), according to the number of treatment lines in HPV-negative tumors. P values from a linear model taking the number of treatment lines as a continuous numerical parameter. B, Genomic alterations according to the number of treatment lines in HPV-negative tumors (n = 202: M naïve n = 15, 0 n = 41, 1 n = 33, 2 n = 59, 3 n = 41, and >3 n = 13). From left to right: mutations, copy-number variations (CNV), and pathway alterations. P values from a linear model taking the number of treatment lines as a continuous numerical parameter. C, Representative transcriptomic features decreasing with the number of systemic treatment lines. Left: IFNγ activity score. Right: plasma cells obtained through deconvolution. P values from the Kruskal–Wallis test on features corrected for other clinical and technical variables using a linear model. D–F, Cell proportions according to the number of previous systemic treatment lines. P values from the linear model taking the number of treatment lines as a continuous numerical parameter. D, General immune cell proportions (first mIF panel) in the total region (n = 206: M naïve n = 15, 0 n = 40, 1 n = 39, 2 n = 55, 3 n = 41, and >3 n = 16). E, T-cell subtypes (second mIF panel) in the total region (n = 208: M naïve n = 17, 0 n = 40, 1 n = 37, 2 n = 57, 3 n = 40, and >3 n = 17). F, Percentage of PD-L1+ tumor cells (n = 209: M naïve n = 19, 0 n = 40, 1 n = 37, 2 n = 59, 3 n = 41, and >3 n = 16). G, Proportion of SCCHN with TLS present in metastatic SCCHN naïve to any treatment (n = 15), in R/M relapsing after curative intent treatment (n = 40), and in R/M refractory to one or more R/M treatment line(s) (n = 155). P value from the Fisher exact test. BnT cell, B near T cells defined as CD20⁺ and CD3⁺ cells; M naïve, untreated metastatic patients; PD-L1, programmed cell death ligand 1.

In HPV-positive SCCHN, TMB also increased with the number of systemic treatments given in the R/M setting (nominal P < 0.01, linear model). Additionally, we observed an increase in B2M mutations and in DNA repair pathway alterations as R/M treatment lines increased (nominal P < 0.05, linear model; Supplementary Fig. S9B; Supplementary Table S10).

Interestingly, 2 (10%) of the 19 upfront metastatic patients had NFKBIA truncating mutations, which belong to the NF-κB inhibitor family and play a tumor suppressive role (33). This gene was never mutated in the other subgroups and was described in the SCCHN TCGA cohort, but at a very low frequency (0.2%; ref. 10).

We selected the transcriptomic features affected by the number of systemic treatment lines (57/256; multivariate adjusted P < 0.05; Kruskal–Wallis test) and clustered them to study the different patterns of expression (Supplementary Fig. S9C). Three main patterns stood out. First, 27 features showed high expression in patients relapsing after curative-intent treatment but not yet treated for recurring disease. Among these features, we found high TGFβ and CXCL12 activity, several enriched fibroblast populations, and high activity scores for some cytokines (IL3, IL21, IL22, and OSM; Supplementary Fig. S9D). Second, 22 features decreased linearly with the number of treatment lines. Those features were immune-related: IFNγ pathway activity scores, deconvoluted lymphocytes and DCs, and several unsupervised modules containing immune genes (Fig. 3C). Finally, eight features exhibited a first peak in metastatic naïve patients and a second peak in patients with three or more systemic therapy lines, most of these features being unsupervised modules (Supplementary Fig. S9E).

Similar to transcriptomic findings, mIF analyses showed a significant decrease in T-cell and B-cell proportions as R/M treatment lines increased (adjusted P < 0.001, linear model; Fig. 3D). With respect to T-cell subtypes, mainly CD4⁺ T cells (adjusted P < 1e−4, linear model) and Tregs decreased (adjusted P < 0.01, linear model; Fig. 3E). In parallel, a decrease in PDL1+ tumor cell proportion was observed in patients exposed to more treatment lines (nominal P < 0.01, linear model; Fig. 3F). TLS presence also differed according to patient treatment history. TLS were present in 60% of treatment-naïve patients, 40% of patients relapsing after curative intent treatment but not yet treated for recurring disease, and 15% of patients refractory to R/M treatment(s) (P = 1e−5; Fisher exact test; Fig. 3G).

These results underscore the impact of systemic treatment lines in the TME, especially in the T-cell compartment, and the unique transcriptomic character of relapsing tumors not yet treated for recurring disease.

Coamplified genes of the 11q13.3 locus are not homogeneously expressed

Given that 11q13.3 amplification was the most common CNA, we investigated its transcriptomic and clinical correlates. One transcriptomic unsupervised module (leiden 60) correlated strongly with 11q13.3 CNA (Fig. 4A; Spearman’s p = 0.86; P < 2e−16). It was formed by six genes, all of them located in the 11q13.3 locus (CCND1, CTTN, FADD, MIR548K, PPFA1, and LTO1). However, not all genes from this region were present in this transcriptomic module. We inspected the genomic–transcriptomic correlation for all 11q13.3 genes (Fig. 4B). Three genes (FGF3, FGF4, and FGF19) showed lower correlation between CNA and transcript expression (FGF3: Spearman’s p = 0.34; FGF4: Spearman’s p = 0.35; FGF19: Spearman’s p = 0.47; P < 0.001). When further inspecting these correlations, we observed that not all tumors harboring high copy numbers of FGF had high FGF expression, and inversely, high FGF expression was observed in some tumors without CNA (Fig. 4C). Furthermore, these three genes were coexpressed, forming a distinct expression cluster (Fig. 4D). Taken together, these results suggest that elevated expression of CCND1, CTTN, FADD, MIR548K, and LTO1 is almost entirely induced by high copy numbers of 11q13.3, whereas other mechanisms are involved in the regulation of FGF3, FGF4, and FGF19 expression. As growth factors are known for their autocrine signaling loops in cancer (34), we studied whether the FGF receptors correlated with the FGF cluster expression, and FGFR1 showed a moderate correlation (Spearman’s p = 0.34–0.42; P < 0.001; Fig. 4E). Furthermore, they tightly correlated with an unsupervised module (ICA 26), which showed an enrichment in genes linked to neural development (Fig. 4F). FGFs have been largely implicated in the development of the nervous system (35), providing a biological link between these components in our dataset and a hint toward a real biological effect of the FGF genes, possibly through FGFR1.

Figure 4.

Two patterns of regulation at the chr11q13.3 locus. A, Correlation between genomic copy number of the chr11q13.3 locus and the transcriptomic module leiden 60 (n = 223). P value from Spearman correlation test. B, Correlation between genomic copy number and gene expression of the genes located in the chr11q13.3 locus (n = 223). C, Correlation between FGF19 genomic copy number and its expression (n = 223). D, Left: correlation between the expression of the genes located in the chr11q13.3 locus (n = 223). Right: correlation between the genomic copy number of the genes located in the chr11q13.3 locus (n = 253). E, Correlation between the three FGF genes of the chr11q13.3 locus, the possible FGF receptors, and the module ICA 26 (n = 223). Dot size and color correspond to the Spearman correlation coefficient. F, Biological annotation of the ICA 26 module. The annotation was done with ShinyGO using the Gene Ontology Biological Process (GO BP) database. G, Clinical factors affecting the genomic amplification of chr11q13.3 (n = 253). P values from a logistic regression model including the four represented clinical variables. H, Expression of FGF19 (VST values) along the number of received systemic treatment lines (n = 223). OR, odds ratio; VST, variance stabilizing transformation.

When inspecting the clinical variables associated with each process, we observed that the 11q13.3 CNA was linked to HPV-negative metastatic tumors, originating mainly from the hypopharynx and smokers and/or drinkers (Fig. 4G). However, the clinical variable that most affected FGF cluster overexpression was the number of treatment lines received (Fig. 4H).

Our results indicate that the 11q13.3 locus harbors two relevant and functional clusters of genes that contribute to metastatic progression in an independent manner, each in response to different clinical parameters.

p-EMT and TLS signatures are prognostic factors in our R/M SCCHN cohort

Prognostic factors in heavily pretreated patients with R/M SCCHN remain poorly characterized; therefore, we investigated the clinical and omic factors affecting survival at this disease stage. To this end, using patients with available survival data (n = 204), we fitted multivariate Cox models for each omic feature independently, including the most relevant clinical variables (ECOG status, HPV status, substance abuse, primary diagnosis location, number of prior treatment lines, and disease extent). Most strikingly, known prognostic factors in early disease, such as HPV status and TP53 mutations, did not show any significant effects on survival. Given that most of our patients were heavily pretreated, we speculated that some systemic treatments may attenuate these described prognostic effects. The two most common systemic treatments that patients had received before sampling were platinum salts (81%) and anti–PD-1 therapy (65%). When selecting platinum-naïve patients, no effects on survival were detected for TP53 mutations or HPV status using a log-rank test (Supplementary Fig. S10A); however, in the anti-PD-1–naïve subcohort (patients who had not received anti–PD-1 treatment nor would receive it after sampling), HPV status and TP53 had the expected positive prognostic value (Supplementary Fig. S10B). In our whole population with survival data available, 10% of the tested omic features (36/374) showed effects on survival (adjusted P < 0.05, multivariate Cox model; Fig. 5A). Only features from RNA-seq and mIF were present among the significant ones. The most significant features corresponded to the data-driven p-EMT signatures (leiden 44 and ICA 27), which showed a strong negative prognostic effect in our cohort of R/M SCCHN. The most significant feature associated with a better prognosis was leiden 115. In an atlas of all human single-cell types, genes from leiden 115 (CXCR5, CD22, and FCER2) were most expressed in germinal-center B cells (Supplementary Fig. S10C), and in our dataset, leiden 115 was significantly more expressed in samples with a detected TLS through mIF (Supplementary Fig. S10D; P value = 1e−7; Wilcoxon rank-sum test), suggesting that leiden 115 represents a TLS-associated germinal-center B-cell signature. Using these two features (see “Materials and Methods”), the p-EMT signature (leiden 44 and ICA 27) and the TLS-germinal center signature (leiden 115), we could create four groups with distinct prognoses, ranging from a median OS of more than 18 months for TLS germinal center-high/pEMT-low patients to 6 months in TLS germinal center-low/pEMT-high patients (Fig. 5B).

Figure 5.

p-EMT and TLS signatures are prognostic factors in R/M SCCHN. A, Features significantly associated with OS in the whole cohort (n = 178 for RNA-seq, n = 168 for IF). P values from a multivariate Cox model including ECOG status, HPV status, tumor primary location, smoker/drinker status, disease extent, and number of systemic treatment lines received, corrected with BH correction. B, Kaplan–Meier plots showing differences in survival among the patients stratified by p-EMT (leiden_44) and TLS/germinal-center (leiden_115) signatures. The P value corresponds to a log-rank test.

Our results indicate that known clinical prognostic factors might not be relevant in heavily pretreated R/M SCCHN and that p-EMT and immune infiltrates, particularly TLS and lymphocytes, affect survival at this advanced stage in our cohort. Further validation studies are needed in independent cohorts to assert the predictive capacity of these features.

Discussion

Addressing critical knowledge gaps is essential for developing new therapies. We provide an in-depth molecular and immune characterization of R/M SCCHN, evaluating the impact of HPV status, tobacco and alcohol history, primary tumor site, relapse pattern, and treatment history at the genomic, transcriptomic, and immune levels.

The primary clinical factor affecting the immune microenvironment was the number of treatment lines, with significant declines in T cells observed via mIF and RNA-seq as the number of R/M treatment lines progressed. HPV presence and tobacco or alcohol abuse are known to increase and decrease the abundance and activation of intratumor T cells in early-stage treatment-naïve SCCHN, respectively (36–38). However, this was not observed in R/M SCCHN, possibly due to the extensive treatment history. The loss of an enhanced immune infiltrate in HPV-positive patients could be due to several factors. One hypothesis is that HPV-positive tumors initially present viral antigens (e.g., E6/E7) that promote an adaptive immune response. However, the elimination of immunogenic clones by therapy or immune editing may result in the outgrowth of less immunogenic or antigen-poor subclones in recurrent/metastatic lesions. In any case, this observation should be considered when interpreting immunotherapy clinical trial outcomes. Nonetheless, it is important to note that B- and T-cell infiltrates remain independent prognostic factors in R/M SCCHN, as described in locally advanced SCCHN (36, 38).

Our data further support that HPV-positive and -negative SCCHN progress and develop therapeutic resistance through different signaling pathways. IL6, IL13, IL15, and NRF2 pathways were enriched in HPV-negative R/M SCCHN compared with HPV-positive tumors. Increased IL6 levels were shown to confer radioresistance and chemoresistance to oral squamous cell carcinoma through activation of the NRF2 pathway (39) and to facilitate SCCHN tumor progression by inducing resistance to ferroptosis (40). IL13 and IL15 pathways are also associated with SCCHN progression (41, 42). The main molecular characteristics of HPV-positive tumors in our cohort, such as cell-cycle deregulation, altered DNA repair mechanisms, and mutations in the PI3K pathway, have already been described in early-stage SCCHN (43), confirming that these tumors retain a specific identity throughout the R/M spectrum.

In our cohort, 11q13.3 amplifications were independently associated with hypopharyngeal cancers, alcohol and tobacco consumption, and DM. Gene expression analysis identified a gene cluster (CTTN, CCND1, FADD, MIR548K, PPFIA1, and LTO) strongly correlated with copy number and linked to cell migration, proliferation, and invasion (Fig. 4A and B), whereas FGF genes showed no such correlation, making them unlikely drivers as previously suggested (17). This may partly explain the high metastatic potential of hypopharyngeal SCCHN. Previous studies in primary tumors have also linked this site to 11q13.3 amplification (44). On the other hand, the FGF genes in that locus showed independent regulation, and they were associated with a neurodevelopmental signature and high FGFR1 expression. These results indicate that a subgroup of metastatic patients could potentially benefit from FGFR inhibitors. The link of this transcriptomic axis to metastatic potential should be explored and confirmed in experimental studies. We found higher EGFR pathway activity in patients with HPV-negative locoregionally recurrent SCCHN. We also observed that EGFR pathway activation was linked to its ligand expression (AREG, EREG, and HBEGF), supporting their potential as predictive biomarkers of anti-EGFR activity, as suggested by Huang and colleagues (11). These findings align with LUX-H&N1, in which afatinib improved progression-free survival (PFS) in SCCHN with LRR and p16-negative disease (45), and with SPECTRUM, in which chemotherapy plus panitumumab improved PFS and OS in p16-negative patients (46). The association between CXCL12 and HRAS mutations also warrants further investigation for the use of CXCL12 activity as a biomarker for tipifarnib response.

We confirmed that the most frequently altered genes in R/M SCCHN were similar to those reported in primary localized, untreated tumors (10), indicating no shifts in major oncogenic alterations. However, none of these genomic alterations were found to be prognostic for survival. Surprisingly, HPV status was also not associated with improved survival. This contrasts with the findings reported by Fakhry and colleagues (47) in a prospective clinical trial evaluating locally advanced oropharyngeal cancers, as well as with other retrospective studies on R/M SCCHN cohorts (48, 49). Of note, patients in these studies had not received any systemic treatments in the R/M setting. As we have shown, treatments greatly affect SCCHN, and most of the patients in our cohort were heavily pretreated. The lack of an increased immune infiltrate or activity in our HPV-positive patients might be related to the observed absence of a beneficial HPV prognostic effect. Another important factor might be the paradigm shift in the standard of care. All previous studies referred to anti-PD-1–naïve patients, but given the change in treatment paradigm, most of our patients had received and progressed on anti–PD-1 treatment, which could affect the prognostic factors. On the other hand, p-EMT has been shown to organize the TME in SCCHN, but it did not show prognostic effects in early disease (50). Our report is the first to indicate that this transcriptomic program is indeed a differential factor between locoregional and metastatic patients and that it dominates patient prognosis in R/M SCCHN. Understanding this tumor state could pave the way for therapeutic options in patients with R/M SCCHN.

Our study has several limitations. First, the cohort is highly heterogeneous, reflecting the inherent diversity of recurrent and metastatic SCCHN. Patients with varying primary tumor sites, etiologies, disease extents, and treatment histories were included, and differences in sample preservation protocols and nucleic acid library preparation methods were also present. To address this, key clinical and technical variables were incorporated as covariates in the statistical analyses, thereby minimizing the likelihood that observed associations resulted from unintended sources of variability. Second, although patient recruitment was conducted prospectively within EORTC clinical trials, the analyses presented here were performed retrospectively. Third, the limited cohort size affects statistical power. For instance, we could not address the effects of specific treatments on the TME to distinguish which of them were most tightly affecting the immune compartment. Finally, our results are highly descriptive, given that the goal was to characterize the heterogeneity of R/M SCCHN, but further validation at the protein level or in new cohorts would be required to confirm our findings.

spite these limitations, this work extends the understanding of the immune and molecular landscape of R/M SCCHN, unravels its heterogeneity, and raises hypotheses to design new investigations to study new biomarkers and treatments in biologically different subgroups.

Supplementary Material

Supplementary Figure S1

Figure S1. Study flow diagram

Supplementary Figure S2

Figure S2. Study workflow and multi-omics approach

Supplementary Figure S3

Figure S3. Three multiplex immunofluorescent staining of SCCHN samples with 3 different 6-plex panels and their deconvolution

Supplementary Figure S4

Figure S4. FFPE vs frozen tumor biopsies comparisons

Supplementary Figure S5

Figure S5. HPV status affects the genomic and transcriptomic profiles but not the immune infiltration

Supplementary Figure S6

Figure S6. Genomic and transcriptomic alterations linked to HPV status

Supplementary Figure S7

Figure S7. Genomic alterations associated with substance abuse and primary disease location.

Supplementary Figure S8

Figure S8. No differences in immune proportions according to disease extent

Supplementary Figure S9

Figure S9. The treatment history has an impact on the molecular and immune landscape of R/M SCCHN, related to Figure 3.

Supplementary Figure S10

Figure S10. p-EMT and TLS signatures are prognostic factors in R/M SCCHN

Supplementary Supplementary Tables S1-S10

Table S1. List of copy number alterations analyzed. Table S2. List of oncogenic pathways analyzed. Table S3. List of genes present in each module. Table S4. mIF panels. Table S5. mIF phenotypes. Table S6. Frequencies of genes alterations in HPV-positive R/M SCCHN and in HPV-negative R/M SCCHN, related to Figure 1A-E. Table S7. Frequencies of genes alterations in smoker and/or drinker HPV-negative patients and in non-smoker and non-drinker HPV-negative patients, related to Figure S5A. Table S8. Frequencies of genes alterations in HPV-negative R/M SCCHN according to primary disease location related to Figure S5B. Table S9. Frequencies of genes alterations in locoregional recurrence only and in distant metastatic disease SCCHN, related to Figure 2A-B. Table S10. Frequencies of genes alterations according to number of R/M treatment lines prior biopsy, related to Figure 3B and Figure S7A-B.

Data Availability

WES data (FASTQ and VCF files) can be accessed at European Genome-phenome Archive (EGA; accession numbers: EGAD50000002205 for EORTC-SPECTA and EGAD50000002178 for EORTC-HNCG-1559). RNA-seq data (FASTQ and count files) can be accessed at EGA (accession numbers: EGAD50000002206 for EORTC-SPECTA and EGAD50000002179 for EORTC-HNCG-1559). mIF data (tsv files with proportions of cell types) are available at https://doi.org/10.5281/zenodo.17940577. Code used for analysis and figure generation can be found at https://github.com/ImmucanWP7/immucan_scchn_characterization. Additional resources for each study can be found on ClinicalTrials.gov under their respective IDs: NCT03088059 (EORTC-HNCG-1559) and NCT02834884 (EORTC-SPECTA protocol). Further information and requests for resources should be directed to and will be fulfilled by the lead contact, J.-P. Machiels (jean-pascal.machiels@uclouvain.be).

Authors’ Disclosures

A. van der Elst reports grants from the Belgian National Research Fund (FNRS/Télévie), Fondation Saint-Luc (Cliniques universitaires Saint-Luc, UCLouvain, Brussels), and the Keeping Me Alive Foundation (ASBL) during the conduct of the study, as well as nonfinancial support from AstraZeneca outside the submitted work; in addition, the IMMUcan project has received funding from the Innovative Medicines Initiative 2 Joint Undertaking under grant agreement No. 821558, supported by the European Union’s Horizon 2020 and European Federation of Pharmaceutical Industries and Associations (EFPIA). The SPECTA Platform is supported by Alliance Healthcare, which will become Censora. The UPSTREAM trial (EORTC-HNCG-1559, NCT03088059) has received funding from Boehringer Ingelheim, Innate Pharma/AstraZeneca, Pfizer, GSK, and Incyte. B. Bodenmiller reports nonfinancial support from Navignostics during the conduct of the study. H.S. Hong reports employment with the Healthcare business of Merck KGaA. R. Galot reports other support from Bristol Myers Squibb, Merck Sharp & Dohme (MSD), Merck, and Daiichi Sankyo outside the submitted work. P. Bossi reports grants and personal fees from Merck, MSD, and BeiGene and personal fees from AstraZeneca, Daiichi Sankyo, GSK, LEO Pharma, and Bicara Therapeutics outside the submitted work. C. Even reports personal fees from MSD, LEO Pharma, and Merck Serono outside the submitted work. P. Saintigny reports grants from Roche, ADMIR, SmartCatch, and DeepLife during the conduct of the study. C. Lefebvre reports other support from Servier outside the submitted work. J.-P. Machiels reports other support from Pfizer, Roche, Bayer, Merck Serono, Boehringer Ingelheim, Bristol Myers Squibb, Novartis, Incyte, Cue Biopharma, ALX Oncology, ITeos, Etherna, Nektar Therapeutics, F-star, Seagen, Genmab, Astellas Pharma, CureVac, MSD, GSK, Merus, Ipsen, and ANAVEON, as well as other support from Amgen, Bristol Myers Squibb, Pfizer, MSD, Gilead Sciences, and Sanofi outside the submitted work. No disclosures were reported by the other authors.

Authors’ Contributions

A. van der Elst: Data curation, software, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. D. Herrero-Saboya: Software, formal analysis, visualization, methodology, writing–original draft, writing–review and editing. L. Michon: Software, formal analysis. M. Morfouace: Conceptualization, funding acquisition, methodology, project administration. R. Liechti: Software. P. Devanand: Resources. D. Schulz: Resources. M. Persoons: Data curation. S. Rusakiewicz: Resources. N. Eling: Resources. P.-A. Nicolas: Software. M.-S. Robert: Project administration. S. Tissot: Resources. S. Déglise: Resources. B. Palau Fernandez: Resources. B. Bodenmiller: Resources. H.S. Hong: Funding acquisition, project administration. R. Galot: Resources. P. Bossi: Resources. J. Oliveira: Resources. M. Pracht: Resources. C. Even: Resources. P. Saintigny: Supervision. C. Lefebvre: Supervision. L. Martignetti: Supervision. J.-P. Machiels: Conceptualization, supervision, methodology, writing–original draft, writing–review and editing.