Identification and validation of seven RNA binding protein genes as a prognostic signature in oral cavity squamous cell carcinoma

Figures & data

Table 1. Primers for qRT-PCR

Gene	Forward	Reverse
β-ACTIN	GCCGCCAGCTCACCAT	TCGTCGCCCACATAGGAATC
ERMP1	CTCTACCTGATCGCGCTGC	CTGTAGTCCTGGGGCCAATG
RNASE3	CCCACAGTTTACGAGGGCTC	ACCCGGAATCTACTCCGATGA
ARL4D	GCGGCTCACGAGAGATAAC	GGTCTTTCCAGCAGAGTCCA
CSRP2	TGGGAGGACCGTGTACCAC	CCGTAGCCTTTTGGCCCATA
ULK1	AGCACGATTTGGAGGTCGC	GCCACGATGTTTTCATGTTTCA
ZC3H12D	GCTGACACCCCTATCAGAGAG	GGTCGTCGTAGCAGACCAG
RPS28	CCGTCTGCAGCCTATCAAG	CTCGCTCTGACTCCAAAAGG

Figure 1. Flow chart for analyzing RBPs in OCSCC

Figure 2. Heat map (a) and volcano plot (b) to show the differently expressed RBPs in OCSCC

Table 2. RBPs associated with the prognosis of OCSCC in univariate Cox regression analysis

RBPs	Hazard ratio (95% CI)	P value
ANLN	1.362 (1.018 − 1.821)	0.037
ENO1	1.664 (1.037 − 2.671)	0.035
ERMP1	1.320 (1.061 − 1.644)	0.013
PGK1	1.978(1.311 − 2.985)	0.001
KPNA3	1.757 (1.013 − 3.045)	0.045
GPD2	1.796 (1.166 − 2.766)	0.008
P4HA1	1.374 (1.020 − 1.850)	0.037
EGLN3	1.228 (1.017 − 1.483)	0.033
PLOD2	1.301(1.027 − 1.648)	0.029
RACGAP1	1.629(1.023 − 2.595)	0.040
RNASE2	1.512(1.012 − 2.258)	0.044
RNASE3	13.051(1.977 − 86.151)	0.008
CSRP2	0.672(0.519 − 0.871)	0.003
ARL4D	1.368(1.015 − 1.846)	0.040
ULK1	0.620(0.419 − 0.917)	0.017
ZC3H12D	0.495(0.251 − 0.976)	0.042
RPS28	0.636 (0.406 − 0.997)	0.048

Figure 3. Construction of the prognostic model. (a, b) Lasso analysis identified 11 RBPs closely related to the prognosis of OCSCC. (c) Forest plot of the multiple Cox regression. Seven RBPs were determined as independent predictors for the prognosis of OCSCC

Figure 4. Survival analysis of TCGA training dataset. A. Dividing high and low risk group by median risk score; B. Survival time and status of OCSCC patients. C. Heatmap of the 7 significant RBPs expression. D. T-SNE analysis demonstrated that the patients in low and high risk groups were distributed in two directions. E. Survival curves showed noticeable stratification of high and low risk group

Figure 5. Predictive performance of the prognostic model in training dataset. A. ROC curves of the prognostic model in 1-, 3-, 5-year. B. ROC curves of the clinical risk factors and RBPs-related risk score for 1-year survival. C, D. Univariable and multiple Cox regression analysis involving clinical risk factors and the risk score

Figure 6. Internal and external validation of the prognostic model. A, B. The distribution of survival time, life status and the expression patterns of the 7 prognostic RBPs in TCGA internal testing dataset and GSE41613 testing dataset. C, D. Survival curves of high and low risk group showed noticeable stratification in two testing datasets. E, F. ROC curve of TCGA testing group and GSE41613 testing group in 1-, 3-, 5-year survival

Figure 7. Nomogram to predict 1-, 3- and 5-year survival rate of OCSCC patients. A. Prognostic nomogram to predict survival of OCSCC patients based on TCGA dataset. B, C. The calibration plots for 1-, 3- and 5-year survival in TCGA training dataset (b) and testing dataset (c) demonstrated that the nomogram-predicted survival fit well with the actual survival

Figure 8. The mRNA expression level of seven prognostic genes in the normal oral keratinocyte cells (NOK) and OCSCC cells (CAL33). The expression of ERMP1, RNASE3, ARL4D were significantly upregulated in CAL33 than the control (NOK), while the expression of CSRP2, ULK, ZC3H12D, RPS28 were significantly downregulated in CAL33 compared with the NOK cells. *P < 0.05, **P < 0.01, ***P < 0.001

Figure 9. GO functional, KEGG pathway enrichment analysis and correlation network of the 7 prognostic RBPs. A, B. GO functional, KEGG pathway enrichment analysis. C. Correlation network of the 7 hub RBPs. The correlation coefficients were denoted by different colors