The circRNA expression profile in LSCC and control group.
The analysis of circRNA microarray indicted that the cluster diagram showed the circRNA lineages with abnormal differentially expressed circRNAs among each group, while the circRNAs with statistically significant differences among each group were presented by the volcano diagram (Fig. 1a). Due to the spatio-temporal characteristics of circRNA expression, we set a specific threshold to ensure the feasibility of subsequent validation. The expression level of circRNA in the chip must meet a clear statistical significance (P < 0.05), and the detectable rate in the chip must be greater than 75%. Under the limitation of the above criteria, we found that compared with the control group, 122 circRNA transcriptome specificity was increased in the LSCC group, and 98 circRNAs were confirmed to be down-regulated (Fig. 1b). In order to reveal the potential biomarker for LSCC, we selected the top 20 increased circRNAs in LSCC group as candidate diagnostic makers (Fig. 1c).
Validation of significantly dysregulated circRNAs in plasma
Two staged validation of candidate circRNAs was used including 20 paired samples as training set and 100 paired samples as validation set. A randomly selected 20 paired LSCC and control samples was used. The top 20 circRNAs was first validated. Among the 20 candidate circRNAs, 10 circRNAs was confirmed with significant different expression; however, 5 circRNAs was consistent with microarray results while the rest circRNAs was presented with decreased level in LSCC group.
Based on the results in training set, the rest 100 paired samples were enrolled as validation set. We next examine the expression of the five candidate circRNAs in validation set. We found that hsa_circ_0055202, hsa_circ_0074920 and hsa_circ_0043722 was confirmed with higher expression level in LSCC, while hsa_circ_0010178 presented no significant. The hsa_circ_0009760 presented a higher expression in lncRLSCC; however, the p value was < 0.05.
Two-stage validation of candidate circRNAs was performed, with 20 pairs of paired samples as the training set and 100 pairs as the validation set. As shown in Fig. 2a, 20 pairs of LSCC and control samples were randomly selected. The top 20 circRNAs were identified for the first time. Among the 20 candidate circRNAs, three circRNAs entitled circ_0019201, circ_0011773 and circ_0122790 were proved to be differentially expressed and consistent with the microarray results. However, another four circRNAs were inconsistent with the microarray results, and the remaining circRNAs presented no difference between LSCC group and control group (Fig. 2b).
Based on the results of the training set, the remaining 100 paired samples were enrolled in the verification set. We then studied the expression of the circRNAs validation set of the three candidates. As shown in Fig. 3, we found that circ_0019201, circ_0011773 and circ_0122790 had higher expression levels confirmed by LSCC. Furthermore, the expression of the three circRNA presented a remarkable decreasing level after surgical excision.
Risk score analysis
To further explore the accuracy and specificity of these three circRNAs as potential features, we used a risk scoring formula to evaluate the diagnostic value of these three circRNAs. Firstly, we divided the control and case groups in the training set according to the 95% confidence interval (95% CI) for the control group. Logistic regression analysis was used to calculate the risk score. All plasma samples were then divided into a high-risk group (possibly LSCC) and a low-risk group (expected to be a control group). We defined the threshold as the maximum of sensitivity plus specificity. The positive predictive value (PPV) and negative predictive value (NPV) calculated in the training set were 90% and 85%, respectively. We further applied the same values to calculate risk scores for the validation set sample, with PPV and NPV of 90% and 89%, respectively (Table 2). In addition, we also used ROC curve analysis to evaluate the predictive diagnostic value of circRNA for LSCC. The areas under the ROC curve of the three circRNAs were 0.933, 0.908 and 0.965, respectively, which could well distinguish the areas under the ROC curve between the LSCC patients and the control group. In the validation set of amplified samples, the areas under the ROC curve of the three circRNAs were 0.766, 0.864 and 0.908, respectively. Combined with these three circRNAs, the area under the ROC curve between LSCC patients and the control group was 0.951 (Fig. 4).
Table 2
Risk score analysis of in LSCC and cancer-free control plasma samples
Score | 0-7.08 | 7.08–11.02 | PPVa | NPVb |
Training set | | | 0.90 | 0.85 |
LSCC | 2 | 18 | | |
Control | 17 | 3 | | |
Validation set | | | 0.90 | 0.89 |
LSCC | 10 | 90 | | |
Control | 89 | 11 | | |
aPPV, positive predictive value. |
bNPV, negative predictive value. |
Stability detection of circRNAs in plasma samples
Next, we amplified these three circRNAs in five healthy controls. Human plasma obtained from three healthy controls was frozen and thawed for 5 cycles at room temperature for 0, 12, and 24 h. It was stored at -80 ° C for approximately seven days and digested with RNase. We found that the expression levels of these three circRNAs did not change, indicating that circ_0019201, circ_0011773, and circ_0122790 were stably expressed and detectable in human plasma (Fig. 5).