A 4-gene signature from histologically normal surgical margins predicts local recurrence in patients with oral carcinoma: clinical validation

Abstract

Prognostic biomarkers for recurrence of Oral Squamous Cell Carcinoma (OSCC) are urgently needed. We aimed to independently validate a 4-gene expression signature (MMP1, COL4A1, P4HA2, THBS2) predictive of OSCC recurrence risk. Gene expression was measured using Nanostring nCounter^® in 245 histologically normal surgical resection margins from 62 patients. Association between risk scores for individual patients and recurrence was assessed by Kaplan-Meier analysis. Signature performance was quantified by concordance index (CI), hazard ratio (HR) and the area under receiver operating characteristics (AUC). Risk scores for recurrence were significantly higher than recurrence-free patients (p = 9.58e-7, Welch's t-test). A solid performance of the 4-gene signature was determined: CI = 0.64, HR = 3.38 (p = 1.4E-4; log-rank test), AUC = 0.71. We showed that three margins per patient are sufficient to preserve predictive performance (CI = 0.65; HR = 2.92; p = 2.94e-3; AUC = 0.71). Association between the predicted risk scores and recurrence was assessed and showed HR = 2.44 (p = 9.6E-3; log-rank test, N = 62). Signature performance analysis was repeated using an optimized threshold (70^th percentile of risks), resulting in HR = 3.38 (p = 1.4E-4; log-rank test, N = 62). The 4-gene signature was validated as predictive of recurrence risk in an independent cohort of patients with resected OSCC and histologically negative margins, and is potentially applicable for clinical decision making on adjuvant treatment and disease monitoring.

Figure 1

Boxplot depicting distribution of risks derived from margins samples of patients without (n = 28; 45%) and with locally recurrent tumors (n = 34; 55%).

Figure 2

Receiver operating characteristics depicting patient classification performance of the signature. Horizontal and vertical lines indicate sensitivity and false positive rate (1 – specificity) achieved when using median risk as the threshold. Associated statistics, including area under the curve and accuracy, are listed in the bottom right corner.

Figure 3

Kaplan-Meier plot depicting survival of the patients with low risk (LR) and high risk (HR) as estimated by applying the signature, stratifying patients using the median of the obtained risks (A) and optimized threshold, equal to 70^th percentile of the obtained risks. (B) Colour-shaded areas depict 95% confidence intervals of the survival curves. Obtained statistics are listed in the bottom left corner.

Figure 4

Violin plots depicting distributions of the major indicators of the signature performance, including concordance index (CI), area under ROC (AUC) and hazard ratio (HR), obtained using one, two and three randomly chosen samples per patient to assess patient risk. Annotations show minimum, 2.5^th percentile, median, 97.5^th percentile and maximum of the given distributions.