Reassessment of a four-gene NK-cell-lineage-associated prognostic score in histologically defined grade 3 glioma.

BACKGROUND

Historical anaplastic glioma comprises molecularly heterogeneous tumors. We reassessed a four-gene score derived from an NK-cell-lineage-associated panel, focusing on selection stability, standardized validation, and incremental value beyond established molecular features.

METHODS

We analyzed histologically recorded grade III gliomas with overall survival longer than 60 days from TCGA (training; n = 248), GSE16011 (n = 80), and REMBRANDT/GSE108474 (n = 67). Of 76 historical candidates, 73 mapped to current TCGA expression data. Ten-fold cross-validated LASSO Cox regression and 100 bootstrap resamples assessed stability. The submitted four-gene score (ARL4C, ERBB2, IQGAP1, and SMAD7) was refitted in TCGA. The continuous score was primary, and one TCGA-derived cutoff was applied unchanged to validation cohorts. TCGA models adjusted for age, sex, histology, and integrated IDH/1p19q subtype.

RESULTS

The score was associated with survival in TCGA (HR per SD, 2.37; 95% CI, 1.91-2.93) and GSE16011 (HR, 1.84; 95% CI, 1.43-2.36), but not REMBRANDT (HR, 0.93; 95% CI, 0.70-1.23; P = 0.601); validation-cohort heterogeneity was significant (P = 0.00046). In 245 TCGA patients with molecular data, the adjusted HR was 1.95 (95% CI, 1.37-2.76). Adding the score increased Harrell's C from 0.810 to 0.833 and reduced AIC from 676.23 to 664.31. At lambda.1se, only ARL4C and ERBB2 among the four genes were retained; bootstrap frequencies ranged from 0.25 to 0.74.

CONCLUSIONS

The score retained an adjusted association in TCGA and replicated in GSE16011, but failed in REMBRANDT and showed incomplete gene-selection stability. No protein-level, single-cell, or spatial validation was performed.It should be considered an exploratory bulk-transcriptomic score, not a validated NK-cell-specific biomarker or a clinically validated decision tool. Prospective molecularly classified cohorts, locked cross-platform preprocessing, protein-level validation, and cell-resolved studies are required.