Aberrant expression of miRNAs is one of the potential biomarkers for the detection of cervical cancer and its precursor lesions11,12. We previously described differences between normal and cervical neoplasia by the fold-change of levels of certain miRNAs. The previous setting was just suited for the comparison between normal and disease groups. Thus, it remained necessary to establish -ΔCt values calculating Ct values of each target relative to the internal reference control to estimate in the external cohort. For the ideal internal reference control, candidate miRNAs would need to be expressed at a constant level in healthy women and through all disease stages. Hence, we preferred to use a novel internal reference control, namely, the average Ct of miR-3180 and miR-7109-5p instead of RNU48 which we previously employed as an internal control (Table S1).
Our previous work had shown that fold-differences of 4 miRNAs in AD were higher than in SCC in the discovery cohort. Here we added miR-155-5p because its fold-difference was also high, especially for SCC (10-fold). At least because this score was higher than miR-20b-5p (4-fold)15, this was also the reason that miR-155-5p had been selected as a candidate biomarker for cervical cancer screening by others16–18. We had estimated the accuracy of screening with 4 miRNAs separately, but combining candidate miRNAs offers the possibility of further increasing accuracy 13,19. Considering that the majority of cervical cancer derives from SCC (80%)20, we adopted a group of miRNAs including miR-20b-5p, -126-3p, -451a, and -144-3p, -155-5p for the validation cohort. Although the accuracy of miR-155-5b alone was inferior to the others in Figure 1b, a combination of the 5 miRNAs showed the minimum AIC value and the maximum AUC compared with other combinations (Table S2). This indicated that the combination of 5 miRNAs was best for accurate cervical cancer screening (Table 1).
Nomograms are utilized for predicting the risk of cervical neoplasia or recurrence21. MiRNA was one of the variables in a published nomogram with prognostic value for cervical cancer22. To the best of our knowledge, there are no published reports regarding the use of nomograms including miRNAs for cervical cancer screening. Hence, we constructed nomograms using 5 miRNAs for cervical cancer screening with the validation cohort and calculated the predicted probability for an external cohort. We also integrated HPV genotypes and patient age into the nomogram. HPV16 and 18 imbued the highest risk for cervical cancer, followed by HPV31 and 3323,24. The rate of the predicted probability was consistent with clinical staging. One case (SC-1244 in Table S3 and S4) at stage IIIC1r showed low predicted probability in nomograms 1, 3 and 4. In this patient, we noticed necrosis on the surface of the cervix during metastasis assessment by colposcopy (Figure S3). This condition could have resulted in our inability to acquire a suitable specimen for miRNA analysis. Another case (SC-1170) at stage IA1 had only a 2 mm depth stromal invasion. The low score of predicted probability in cervical cancer was associated with good prognosis during the 2-year observation period (Figure 3), and a low score was consistent with early clinical stage. Despite some analyses linking the aberrant expression of a combination of miR-216b,-585b and -764125 and others 22 with significant differences of patient survival, there are no reports regarding the association between prognosis and the expression of the miRNAs described in the present paper.
In the clinical setting, it may happen that collected samples are left at room temperature for some time. It would be practically important to know that the results of predicted probability analyses would still be reliable when using such material rather than specimens frozen immediately after acquisition. As shown in Table S4, we found no difference in the yield of RNA irrespective of storage conditions and importantly, there was no difference in predicted probability estimates. Thus, it might not be necessary to freeze specimens immediately on collection. This is an advantage of a screening test using miRNAs from cervical mucus. Therefore, specimens derived from liquid-based cytology26 or self-sampling27 may be an alternative approach in future.
This paper has some limitation. The samples were collected in a single University hospital. Therefore, cohorts may differ from the general population. Conversely, characteristics of each specimen were analyzed in more detail including histology and HPV genotyping (Table S5). Compared to our previous report, the performance of the ROC with the novel internal control reference was little different from our previous report. However, the novelty of the present paper is to show the predicted probability for patients with cervical cancer and its precursor lesions in an external cohort. As we know that persistent high-risk HPV infections worsen the risk of cervical cancer, nomograms including HPV genotypes could be more promising for predicting full-blown cervical cancer in individuals with precursor lesions. We showed that miRNAs from mucus are stable when the samples were left at room temperature overnight. This is approximately 20-25°C throughout the year, but we have not tested higher temperatures that might apply elsewhere, and we also did not test a longer period of time. Stability of miRNAs was reported to be dependent on their specific sequence, storage or freeze-thaw conditions of either sera or plasma 28–32. More information regarding the different conditions for storage of larger numbers of specimens could address this.
In summary, here we identified miRNAs highly expressed in cervical mucus of women with a range of lesions from normal through to cervical neoplasms. A combination of 5 miRNAs and an internal reference control was best suited for use a diagnostic marker of cervical neoplasia. Highly reliable performance of miRNA prediction was shown using several different nomograms on the external cohort. We further showed that the predicted probability calculated from analysing specimens either immediately frozen or left at room temperature overnight was identical. We conclude that the development of tests to detect aberrant expression of miRNAs could offer useful candidate ancillary screens for cervical neoplasia in a clinical setting.