Head and neck cancer pEMT scores split in two: the malignant arm tracks long progression-free survival on cetuximab, the stromal arm does not
Synopsis
In head and neck squamous cell carcinoma the partial EMT gene list is expressed by both malignant cells and fibroblasts, yet it has been scored only in bulk tumours. Dividing its 100 genes by the cells that express them shows bulk co-expression is dominated by the stromal arm, so standard pEMT scores mainly measure stroma; a single-cell pseudobulk classifier and a 12-gene malignant core recover the malignant arm in any bulk transcriptome. Among 40 patients given platinum plus cetuximab, both malignant measures were higher in those with long progression-free survival (area under the curve 0.78 and 0.82) and the stromal core was not, but the malignant arm could not be separated from the Basal subtype, which discriminated better (0.93).
Interpretation
One pEMT gene list belongs to two cell types in single cells but reports fibroblasts in bulk tumours. The pEMT programme had been scored only in bulk transcriptomes, to stratify patients by survival and radiotherapy response, and gene covariation in bulk tissue largely follows cell-type proportions. In the Puram dataset where the list was defined, COL17A1 drew 99.3%, LAMB3 99.1%, LAMC2 98.3% and LAMA3 96.0% of its expression from malignant cells, whereas COL1A1 drew 96% and TAGLN 98% from fibroblasts; in a signed co-expression network of 520 TCGA head and neck primaries, 54 of the 76 pEMT genes present fell into one 1,411-gene module whose eigengene correlated with Hallmark EMT at +0.96 and with the stromal core at +0.91.
The malignant arm can be scored on its own from bulk data without relying on bulk co-expression. Earlier stroma corrections adjusted whole bulk profiles or excluded stroma-rich tumours, and excluded tumours received no score. A three-class elastic-net classifier trained on pseudobulks from two single-cell datasets yields pEMT specificity; in the 40 GSE65021 patients with recurrent or metastatic head and neck cancer, pEMT specificity had a median of +0.27 against −0.50 (p = 0.004), an area under the curve of 0.78 (95% confidence interval 0.62 to 0.92), the malignant core 0.82 (0.68 to 0.94) and the stromal core 0.42 (0.24 to 0.61).
The malignant arm cannot be separated from the Basal subtype in the patient cohort, and the Basal score discriminates better. pEMT specificity was highest in Basal tumours in TCGA, whereas KS, Hallmark EMT and the Puram signature peaked in the Mesenchymal subtype. In GSE65021, 10 of the 12 patients called Basal by the Walter centroids had long progression-free survival against 4 of the other 28 (Fisher p = 6.0 × 10⁻⁵), and the Basal centroid score gave an area under the curve of 0.93 (0.83 to 1.00), exceeding the malignant core by 0.11 (0.02 to 0.23) and correlating with it at 0.80; in the GSE84713 and GSE183881 xenograft panels the Basal score gave 0.79 and 0.95, with a pooled odds ratio of 3.55 (1.23 to 10.2).
The malignant arm does not mark favourable prognosis in cohorts not selected for cetuximab. A marker of good prognosis in general could have explained the cetuximab association. Across TCGA-HNSC, GSE41613, CPTAC-3 and GSE65858, comprising 869 tumours and 350 deaths, pEMT specificity gave a pooled hazard ratio of 1.25 (DerSimonian-Laird 1.03 to 1.52, Hartung-Knapp 0.88 to 1.78, I² of 64%), the malignant core 1.19 (Hartung-Knapp 0.90 to 1.56) and the stromal core 1.11 (0.87 to 1.43).
Perspective
The analysis addresses gene programmes defined in single cells and scored in bulk tissue in head and neck squamous cell carcinoma, for readers working on tumour transcriptomes and biomarkers. The malignant core and the Basal score kept their association when each patient was scored alone, suggesting a targeted assay is feasible, whereas pEMT specificity would need a reference for each platform. The authors propose scoring pre-treatment biopsies for the malignant core, pEMT specificity, the Basal subtype and epithelial differentiation in a larger cetuximab-treated cohort with unselected outcomes, ideally with a comparator arm, and testing the reversal prediction on paired specimens from the trametinib window study.
The cetuximab cohort has 40 patients in two extreme groups, which inflates discrimination, and the confidence interval for pEMT specificity is wide (0.62 to 0.92); all patients received platinum, so the association with outcome cannot be separated from chemosensitivity or from prognosis within recurrent disease, and progression-free survival is available only as a dichotomy. The malignant arm could not be separated from the Basal subtype in the patient cohort, and in the xenografts the Basal score did as well or better. Several arm definitions were examined, three of them added after the first GSE65021 results. The xenograft panels are small, cannot separate the two arms, and define response in the mouse. The prognostic value of pEMT specificity varied between cohorts (I² of 64%), and the subtype result rests on TCGA alone. Both malignant measures fall as non-malignant cells dilute a tumour, so they partly track purity, although adjusting for stromal and immune content did not weaken the cetuximab association.