Identification of DElncRNAs between PCa and non-carcinoma tissue samples
Based on thresholds of |log2FC| > 1.0 and false discovery rate (FDR) q < 0.05, altogether 451 lncRNAs, among which, 307 were up-regulated whereas 144 were down-regulated, were discovered to show different expression between PCa and the non-carcinoma tissue samples, and they were applied in later stepwise survival analysis (Supplementary:TableS1). The expression profiles were intuitively reflected by volcano plots (Figure 2). According to the DElncRNAs profiles, the unsupervised hierarchical cluster analysis was carried out, which suggested the possibility to distinguish between PCa and non-carcinoma samples (SupplementaryFigure S1). To identify prognosis-related lncRNAs which are associated with patients' OS in PCa, the lncRNA expression profiles were evaluated by univariate Cox regression analysisdata. 36 lncRNAs related to OS were selected at the threshold of 0.05 (Supplementary2: Table S2), which were then applied for stepwise multivariate Cox regression analysis, and four lncRNAs therein (HOXB-AS3, LINC01679, PRRT3-AS1, YEATS2-AS1) (as shown in Figure 3, Table2) were ultimately screened out from the 451 lncRNAs identified before to establish a predictive model.
Expression profiles of those 4 lncRNAs were integrated with the related regression coefficients to construct a prognosis nomogram. Based on the above-mentioned analysis, the riskscore formula was the sum of 4 lncRNAs expression levels weighted by the corresponding relative regression coefficients obtained upon multivariate Cox regression, as shown below: survival risk score = (0.524052278 × HOXB-AS3 expression) + (1.092940489 × YEATS2-AS1 expression) + (-0.663887578 × LINC01679 expression) + (-0.504710478 × PRRT3-AS1 expression). Of these four lncRNAs, two had positive coefficients upon multivariate Cox regression analysis, associated with high risk because the up-regulated level indicated the reduced patient OS (HOXB-AS3 and YEATS2-AS1, Coef > 0) and the remaining two lncRNAs (LINC01679 and PRRT3-AS1, Coef < 0) were shown negative coefficients upon Cox regression analysis, which indicated that such lncRNAs were protective, because cases having up-regulate lncRNAs levels tended to show extended OS relative to patients having decreased expression (Figure3).
Favorable performance for the4-lncRNA prognostic model in the prediction ofOSfor PCa cases
The 493 cases were classified as high- (n=246) or low-risk (n=247) group in line with the median riskscore (0.943) obtained based on those 4 lncRNAs expression levels (also defined as the survival risk score, SRS) (Figure 4 and Supplementary : Table S3). Difference in survival was determined by log-rank test. The K-M method was used for survival analysis. It was illustrated from Figure 5 that, the KM OS curves for both groups on the basis of 4 lncRNAs showed notable difference (p=3.3e-03). Typically, the low-risk group showed significant correlation with favorable prognosis compared with high-risk group. The AUC values of time-dependent ROC curves were calculated to evaluate our constructed nomogram prognostic ability. The greater AUC value is indicative of the higher nomogram performance, and that AUC of >0.90 has excellent performance. Based on the analysis results of ROC curves, the AUC values at 1, 3 and 5 years were 0.997, 0.929 and 0.928, separately (Figure 6), revealing the high sensitivity and specificity of our 4 lncRNAs-based signature in the prediction of OS risk for PCa cases. Additionally, the model C-index was calculated, (C-index=0.9203, CI: 0.8482-0.9924, p-value=3.13447e-30), exhibiting good model performance.
4 lncRNAs-basednomogram’s predictive abilitywas not dependent onother clinicopathological factors
For investigating the distinguishing ability of our constructed 4 lncRNAs-based signature of the survival risk for PCa cases after considering additional possible traditional prognostic factors, univariate as well as multivariate analysis was conducted for evaluating the model independent prognostic significance. The multivariate analysis results demonstrated that our 4 lncRNAs-based signature might be used to be the potent factor to independently predict PCa OS rate from other clinical factors (hazard ratio(HR)=1.014, 95% CI 1.005-1.023, p=0.003), shown in Table 3, and Figure 7, compared with conventional clinicopathological factors such as age, Gleason score and TNM classification.
Contrast of the four-lncRNA signature with Gleason score
Hierarchical analysis showed that PCa with Gleason score>7 could be further stratified as 2 groups that had different survival by the nomogram (log-rank test, p<0.05, Figure 8). Combining the 4 lncRNAs-based signature and the Gleason score remarkably increased the model prognostic capacity compared with the Gleason score alone (AUC, 0.949 vs. 0.896 vs. 0.633) (Figure 9).
Identification of the prognostic lncRNAs signature associated biological functional characteristic
After determining the associations between those 4 lncRNAs and the PCGs, this study selected co-expression between 1 of those 4 lncRNAs and 4080 PCGs (|Pearson correlation coefficient| > 0.4, P-value < 0.05, and q-value <0.05). Afterwards, GO and KEGG analyses were conducted for the lncRNAs-related PCGs to display the possible functions for those 4 prognostic lncRNAs. The 4080 PCGs were most significantly enriched into 28 GO terms and they were located in the extracellular matrix (ECM) and cell membrane, and their functions are mostly associated with the adhesion, activation and transport of substances across the extracellular matrix and cell membrane(such as GO: 0003779: actin binding, GO: 0022803: passive transmembrane transporter activity, GO: 0015267: channel activity, GO: 0005216: ion channel activity, GO: 0005178: integrin binding, GO: 0005539: glycosaminoglycan binding, GO: 0022836: gated channel activity, GO: 0019955: cytokine binding, GO: 0098631: cell adhesion mediator activity, GO: 0046873: metal ion transmembrane transporter activity) (Figure 10A and Supplementary Figure S2 and Table S4). Three KEGG pathways were enriched which are mainly concerned with the interaction and binding between cells or between cells and ECM, and cell proliferation (including hsa04390: Hippo signaling pathway, hsa04514: Cell adhesion molecules (CAMs), hsa04510: Focal adhesion) (Figure 10B, and Supplementary: Figure S3, Figure S4 and Table S5 ).