Fig. 4

Deregulation of LNM gene signatures in lymph node metastasis (LNM) positive primary HNCs and lymph node metastases. Differential expression of LNM gene signatures in A lymph node-positive primary HNC (LNM +) relative to LNM- primary HNC (LNM0) and B lymph node metastases (i.e., metastatic tumors) relative to patient-matched primary HNCs. The LNM gene signatures shown consist of genes negatively (anti-LNM) and positively (pro-LNM) associated with LNM status in our meta-analysis, as well as genes with six unsupervised LNM gene clusters (L1-6). Expression scores were calculated for each LNM gene signature (i.e., set of LNM-associated genes) as the mean of expression (normalized counts) of all genes within the signature. A Deregulation of LNM gene signatures in LNM + primary HNCs. i Box plots of LNM gene signature scores in LNM0 (Cyan, N = 3) and LNM + (Red, N = 6) primary HNCs, within the Puram scRNA-Seq dataset, with cells (points) stratified by LNM status and cell type. ii Heatmap of scaled mean differences between LNM + and LNM0 primary HNCs of LNM gene signature scores, within each cell type, in the Puram scRNA-Seq dataset. The point color gradient indicates the scaled mean differences of gene signature scores between LNM + and LNM0 tumors. Point sizes indicates negative log10 p-values (Wilcoxon rank sum test). B Box plots of LNM-associated gene signatures in primary HNCs and patient-matched LNMs in i bulk and ii single-cell RNA-Seq datasets. i Box plots of LNM gene signature scores in primary HNCs (gray, N = 29) and patient-matched LNMs (red, N = 72) of the Huang bulk RNA-Seq dataset (EGAD00001004489) [98]. Points represent patient samples. LNM gene signature scores are calculated as mean expression (normalized counts) of genes within each LNM-associated gene signature (i.e., set of genes). Asterisks indicate Wilcoxon rank sum test p-values: **p < 0.01, ***p < 0.001. ii Box plots of LNM gene signature scores in malignant cells (points) of primary HNCs (gray) and patient-matched LNMs (red) in two scRNA-Seq study datasets, including the Puram and Stanford datasets. Cells are stratified by tumor type (primary tumor or LNM), with groups consisting of malignant cells of combined patient samples of each tumor type. Asterisks indicate linear regression p-values for association of the LNM gene signatures with tumor type, adjusted for covariates known to correlate with these genes. These include cell cycle phase and TP53 somatic mutation status (In the Puram dataset for which TP53 mutation data was available). ***p < 0.001, **p < 0.01. Forest plots to the right of each box plot indicate regression coefficients (estimates) derived from the multiple linear regression models. These models estimate associations of LNM gene signature scores (dependent variable) with sample type (independent variable) adjusted for cell cycle phase (covariate) and TP53 mutation status (covariate)