Molecular Signatures of Tumour and Its Microenvironment for Precise Quantitative Diagnosis of Oral Squamous Cell Carcinoma: An International Multi-Cohort Diagnostic Validation Study

Abstract

Background: Heterogeneity in oral potentially malignant disorder (OPMD) poses a problem for accurate prognosis that impacts on treatment strategy and patient outcome. A holistic assessment based on gene expression signatures from both the tumour cells and their microenvironment is necessary to provide a more precise prognostic assessment than just tumour cell signatures alone.

Methods: We reformulated our previously established multigene qPCR test, quantitative Malignancy Index Diagnostic System (qMIDS) with new genes involved in matrix/stroma and immune modulation of the tumour microenvironment. An algorithm calculates and converts a panel of 16 gene mRNA expression levels into a qMIDS index to quantify risk of malignancy for each sample.

Results: The new qMIDS^V2 assay was validated in a UK oral squamous cell carcinoma (OSCC) cohort (n = 282) of margin and tumour core samples demonstrating significantly better diagnostic performance (AUC = 0.945) compared to previous qMIDS^V1 (AUC = 0.759). Performance of qMIDS^V2 were independently validated in Chinese (n = 35; AUC = 0.928) and Indian (n = 95; AUC = 0.932) OSCC cohorts. Further, 5-year retrospective analysis on an Indian dysplastic lesion cohort (n = 30) showed that qMIDS^V2 was able to significantly differentiate between lesions without transformation and those with malignant transformation.

Conclusions: This study validated a novel multi-gene qPCR test on a total of 535 tissue specimens from UK, China and India, demonstrating a rapid minimally invasive method that has a potential application for dysplasia risk stratification. Further study is required to establish if qMIDS^V2 could be used to improve OPMD patient management, guide treatment strategy and reduce oral cancer burden.

Keywords: FOXM1 diagnostic biomarkers; clinical translation; dysplasia; early detection; early oral cancer biomarkers; microenvironment; molecular diagnostics; oral premalignant disorders; personalised medicine; prognostic biomarkers; qMIDS; squamous cell carcinoma.

Figure 1

Summary of patient demographic data (age, sex, ethnicity, substance use habits and tissue anatomical sites) associated with all tissue samples used in this study from UK, China and India, respectively. The number of samples are indicated within the pie-chart with % in parenthesis under each data labels. In the UK cohort, each patient contributed either single or multiple samples (paired margins, core and neck metastasis), hence the total number of UK patients were 170 contributing 282 specimens. In China and India cohorts, each patient provided only a single tissue specimen.

Figure 2

Comparisons between qMIDS^V1 and qMIDS^V2. (A) Biomarker gene panel and their respective functional groups in qMIDS^V1 compared to qMIDS^V2. Diagrams indicate the removal of less influential genes from qMIDS^V1 and replacement of new genes with functional regulation of stroma matrix and immune modulation in qMIDS^V2. The qMIDS algorithm [24] was used to compute 16 gene expression levels into a qMIDS malignancy index (MI) for each sample. (B), Case study using a single OSCC tumour core tissue biopsy for qMIDS^V1 and qMIDS^V2 comparison. Photograph showing the cut site of a strip of tissue across the tumour sample which was subsequently dissected into 10 pieces of 1 mm³ tissue fragments. Each fragment was subjected to qMIDS^V1 and qMIDS^V2 assays simultaneously and their corresponding qMIDS indexes were shown adjacent to the data points. Of note, some fragments (e.g., S1, S2 and S10) showed much larger differences between qMIDS^V1 and qMIDS^V2 which may reflect the molecular heterogeneity across the tumour tissue. (C), Data from (B) were plotted as box-whisker dot plots (box horizontal lines represent median and 25–75% percentiles, whiskers represent lowest and highest values, outliers are beyond the whiskers), t-test were performed. p-values were indicated in the panel above. (D), Similar to methods in (B,C), each sample was cut into 9–24 fragments for qMIDS^V1 and qMIDS^V2 comparison whereby paired and unpaired margin and tumour core samples were analysed. A total of n = 498 sub-fragments (from paired samples of 7 patients) and n = 204 sub-fragments (unpaired samples of 10 patients) were independently analysed. Top panels show box-whisker dot plots of individual fragments for each patient (x-axis showed individual patients’ sample IDs). Panels below show box-whisker dot plots of average values of all fragments from each sample and corresponding statistical t-test p-values.

Figure 3

Independent validation in UK cohort on margin and OSCC tumour core samples showing data comparisons between qMIDS^V1 and qMIDS^V2. Due to insufficient tissue sample left for qMIDS^V1 assays, only a subset of samples (n = 102) was analysed with qMIDS^V1 compared to qMIDS^V2 (n = 282). (A), Box-whisker dot plots (box horizontal lines represent median and 25–75% percentiles, whiskers represent lowest and highest values, outliers are beyond the whiskers) showing the segregation of data and t-test analysis p-values for qMIDS^V1 and qMIDS^V2. (B), Diagnostic test performances for qMIDS^V1 and qMIDS^V2 were calculated based on the cut-off value at 4.0 (dotted line as shown in (A). (C), Diagnostic test performance results for qMIDS^V1 and qMIDS^V2. TN, true negative; FN, false negative; FP, false positive; TP, true positive. (D), Data from (A) were separately subjected to ROC analysis showing the comparison between qMIDS^V1 and qMIDS^V2 with respective AUC values as shown within the panel.

Figure 4

Multi-cohort qMIDS^V2 diagnostic performance comparisons across geographically and ethnically distinct OSCC cohorts. (A,B), China cohort samples (snapped frozen samples): (A), normal oral mucosa (NOM; grey) and oral squamous cell carcinomas (OSCC; red). Student’s t-test p-value is indicated within the panel. (B,C), Indian cohort samples (FFPE): (B), Samples were grouped according to histopathology grading (WHO 2017): NOM, Mild (yellow)/Moderate (pink) Dysplasia (Dysp), Severe (orange) Dysplasia and OSCC (red). C, Oral lichen planus (OLP), submucous fibrosis (OSF) and dysplastic oral leukoplakia (OL) were compared. Outliers are indicated by black outlined symbols and t-test p-values are indicated above each chart. (D), Diagnostic test performance were compared between China and India OSCC cohort data obtained from (A,B). (E), Diagnostic test performance table for OSCC comparing between UK (extracted from Figure 3), China and India. TN, true negative; FN, false negative; FP, false positive; TP, true positive.

Figure 5

Comparison between pathological dysplasia grading method (WHO OED 2017) and qMIDS^V2 for predicting malignant transformation in an Indian dysplasia cohort. (A), Dysplasia patients with 5-year outcome data (n = 30 from Figure 4B) were re-grouped according to their dysplasia transformation status: non-transformed or transformed into OSCC within 5 years. Student’s t-test p < 0.004 and Mann–Whitney U-test (p < 2 × 10⁻⁶) were performed due to skewed data distribution. Outliers are indicated by black outlined symbols and t-test p-values are indicated above the chart. (B), Prognostic performance of dysplasia grading (cut-off between mild/moderate and severe dysplasia grades) and qMIDS^V2 (at cut-off 4.0 and 2.2) were analysed on the dysplasia cohort (n = 30) from (A) and their respective prognostic efficiencies are tabulated in (C). TN, true negative; FN, false negative; FP, false positive; TP, true positive.