OC is a heterogeneous multifactorial disease resulting from genetic and epigenetic alterations. OC is challenging to manage due to its aggressive nature, high metastatic rate, and late diagnosis leading to poor 5-year overall survival rates. In addition, a considerable proportion of patients (3–50%) with OPMD are at high risk of transforming into invasive carcinoma 18. In particular, patients with high-grade dysplasia (around 12%) are at risk for malignant transformation 19. Early detection and timely targeted therapy are well-known strategies for improving patient outcomes. As such, recent research has focused on identifying biomarkers for the diagnosis and prognosis of OC using a liquid biopsy approach. An ideal biomarker should have a high sensitivity and specificity through non-invasive, simple, and cost-effective methods, such as human saliva.
Our unique approach of combining TCGA tumour tissue miRNA sequencing data and salivary small RNA sequencing data to identify miRNAs led to the development of a robust miRNA-based panel. Furthermore, choosing the differentially expressed miRNAs in TCGA data that overlapped with saliva sequencing data and the inclusion of most up and down-regulated miRNAs further increase the efficiency of our miRNA panel. More importantly, including over-expressed and under-expressed miRNAs further strengthened our biomarker discovery approach. In contrast, most previously published studies have considered only upregulated miRNAs 15. Furthermore, our systemic approach of selecting five miRNA reference genes to normalize RT-qPCR data ensures the reliability of the miRNA expression analysis. As a result, a combination of three highly stable miRNAs was considered as reference genes for normalizing miRNA RT-qPCR quantification. Also, using LNA-based technology for the RT-qPCR further ensures the reliability of the expressions of miRNAs.
Regarding the non-invasive diagnosis of OC, the 8-miRNA signature achieved a diagnostic efficiency with an AUC, sensitivity, and specificity of 0.954, 86%, and 90%, respectively. Further validation of our candidate miRNAs in tissue samples confirmed that most of our candidate miRNAs exhibited similar expression patterns as those observed in saliva. However, due to the limited sample size, a few miRNAs did not show significant and similar patterns as those of saliva. For instance, miR-7-5p and miR-486-3p demonstrated similar patterns observed in saliva, indicating their association with the tumour. Moreover, the localization of miR-7-5p in FFPE tissue samples further confirms its association with the tumour.
Furthermore, the most clinically relevant finding was the development of a risk probability score to detect and stratify patients at high risk of developing OC. Notably, patients with high-grade dysplasia are more prone to malignant transformation. Several studies have demonstrated that early detection and timely targeted therapy could be the best strategy to improve patient outcomes. Concurrently, our risk probability score predicted the development of OC in two OPMD at least four weeks prior, when comparing to the gold standard biopsy-proven diagnosis. This finding could be a game changer in the management of high-risk OPMD patients, as this test could be used as a screening test to predict the risk of OC cancer development in high-risk OPMD patients. However, we acknowledge that there are four patients with risk probability scores higher than the cut-off who have not yet developed OC. This may be attributed to the fact that we only collected patient outcomes during the study period. Nevertheless, we are continuously monitoring these patients to track their progress. Another clinically relevant finding was the development of a four-miRNA signature to discriminate OPMD patients from OC. The discriminative efficiency of the four-miRNA signature based on AUC, sensitivity, and specificity was 0.9115, 90%, and 82.7%, respectively. miR-4707-3p demonstrated the highest discriminative efficiency, followed by miR-7-5p, miR-215-5p and miR-10b-5p.
Similar studies have reported the utilization of salivary miRNAs for the diagnosis of OC. Koopaie et al. reported that both miR-15a and miR-16-1 were downregulated in saliva samples of OSCC patients (n = 15) compared to healthy controls (n = 15). miR-15a showed a sensitivity and specificity of 93.3% and 86.67%, respectively. In contrast, miR-16-1 showed a sensitivity and specificity of 86.67% and 92.33%, respectively. miR-15a shows more sensitivity than our miRNA signature in discriminating OSCC from controls, which may be due to the low sample number recruited in their study 20. Similarly, Romani et al. reported that a panel of miR-106-5p, miR-423-5p, and miR-193b was able to distinguish OC (n = 55) from healthy controls (n = 39) with an AUC of 0.98, sensitivity of 97.4% and specificity of 94.2%. Their study considered only the upregulated miRNAs, whereas we included both upregulated and down-regulated ones to eliminate bias in selecting biomarkers. Duz et al. reported salivary miR-139-5p as a biomarker for the early detection of tongue squamous cell carcinoma 21. However, the discriminative efficiency of their study is lower than our study. Furthermore, Yap et al., reported a risk score combining miR-21-5p, miR-100-5p, and let-7-5p that could be used to assess the risk of OSCC 22. However, their AUC (0.868) and specificity (81.5%) were less than in our present study.
Overall, the results of the present study were consistent with other recent studies indicating the possibilities of using miRNAs as diagnostic and predictive biomarkers. However, except for some miRNAs reported in the present study, others have not been previously reported in OC, and we experimentally highlight their discriminative potential for the first time. For instance, Chou et al. found that miR-486-3p acted as a tumour-suppressive miRNA in OC by targeting the well-known oncogene DDR1 23. They reported that miR-486-3p is down-regulated in tumours compared to their matched normal adjacent tissues. Our results also show a downregulation in saliva samples which is in concordance with the previous results. Similarly, Li et al. reported that miR-182-5p promoted the growth of OC by targeting CAMK2N1, thus functioning as an oncogenic miRNA 24. Even though we could not find a significant difference in the levels of miR-182-5p in saliva samples between OC and controls in this study, there is a notable upregulation in the OC cohort. Since miRNAs are secreted from cancer cells into saliva, when miRNAs are downregulated in the tumour, their secretion into saliva can also be reduced, resulting in the downregulation in saliva and vice versa. In contrast, miR-431-5p was reported to be downregulated in OC tissue when compared to adjacent normal tissue and act as a tumour suppressor in tongue squamous cell carcinoma 25. However, our results demonstrate that miR-431-5p is upregulated in OC tissues and downregulated in saliva samples of OC patients. This discrepancy may be due to the limited tissue samples employed in this study.
Other miRNAs from our panel have not been reported previously in oral cancer, but some have been reported as potential biomarkers or regulators in other cancer types. miR-7-5p was reported as a tumour-suppressor in non-small cell lung cancer, hepatocellular carcinoma, and pancreatic ductal adenocarcinoma and was involved in enhancing temozolomide sensitivity in drug-resistant glioblastoma cells 26–29. Similarly, miR-215-5p was reported as a tumour-suppressor multiple myeloma and as a biomarker of diagnostic importance in osteosarcoma 30–32. Concurrently, miR-3614-5p was reported as an oncogene in hepatocellular carcinoma, non-small cell lung cancer. In contrast, it was identified as a tumour suppressor cadmium-induced breast cancer 33–35 whereas miR-4707-3p was reported in oesophageal carcinoma 36. miR-10b-5p was reported as a regulator of PIEZO1 in breast cancer and as a tumour-suppressor in primary hepatic carcinoma 37,38.
Although our miRNA signatures were robust in diagnosing and predicting OC, we acknowledge the discrepancies in the expression levels between small RNA sequencing data and validation phase using RT-qPCR. This may be due to the small cohort of patients used in the discovery phase. Furthermore, we could not trace the clinicopathological details of some patients as they were unavailable in the healthcare provider’s record, and we collected outcomes data for OPMD patients only during the study period.
To conclude, we have discovered and validated a salivary miRNA panel to diagnose and predict the development of OC. Furthermore, we have developed a risk probability score to stratify the risk of developing OC in patients with high-grade dysplasia and Stage I OC, thus, for the first time developing an algorithm to predict the risk of malignant transformation in OPMD patients. Nevertheless, further validation of the reported salivary miRNA signatures in multi-centred cohort trial is warranted prior to clinical uptake.