Prognostic Signature of NRLs in PC Patients
Flow diagram of the study is presented in Fig. 1. Based on the correlation analysis between lncRNAs and NRGs expression, 161 NRLs were identified, and the lncRNA-mRNA co-expression network revealed their interconnections (Fig. 2A). Clinical characteristics were comparable between the training set and the validation set (Table 1). In the training set, 20 NRLs associated with prognosis were screened out from the above 161 NRLs through UniCox analysis, including LINC00857, AC012213.4, AL672291.1, AC093675.1, AC080013.4, LINC02245, PDCD4-AS1, PTPRN2-AS1, AC141930.2, LINC02593, MEG9, AL078600.1, CCNT2-AS1, LINC00852, AL590787.1, AC005332.6, TRAF3IP2-AS1, AL008729.2, AC002091.1 and WAKMAR2 (Fig. 2B). LASSO and MultiCox regression analysis were then performed, and a total of 5 NRLs were ultimately used to construct the risk model, including LINC00857, AL672291.1, PTPRN2-AS1, AC141930.2, and MEG9 (Fig. 2C-E). The coefficients of each lncRNA in the prognostic risk formula were obtained based on MultiCox regression analysis. Therefore, the RS of PC patients were calculated according to the following formula: RS = LINC00857 × 0.395 + AL672291.1 × (-1.869) + PTPRN2-AS1 × (-0.489) + AC141930.2 × (-1.636) + MEG9 × (-0.735). Subsequently, the training and validation sets were divided into low- and high- risk subgroups, using the median RS of the training set as the cut-off value to assess the predictive power of the prognostic signature. The Kaplan-Meier curve demonstrated that the OS of the high-risk group was significantly lower than that of the low-risk group (P < 0.001, Fig. 3A). As shown in Fig. 3B, patients in the validation set had a similar result (P = 0. 014). These results suggested that the 5-NRLs signature had a good prognostic predictive power for PC patients.
Table 1
Clinical characteristics of 178 patients with pancreatic cancer
Character
|
Whole
cohort
|
Training
set
|
Validation set
|
P-value
|
n = 178
|
n = 89
|
n = 89
|
|
Age
|
|
|
|
0.881
|
<=65
|
94(52.81%)
|
48(53.93%)
|
46(51.69%)
|
|
> 65
|
84(47.19%)
|
41(46.07%)
|
43(48.31%)
|
|
Gender
|
|
|
|
0.880
|
Female
|
80(44.94%)
|
41(46.07%)
|
39(43.82%)
|
|
Male
|
98(55.06%)
|
48(53.93%)
|
50(56.18%)
|
|
Grade
|
|
|
|
0.113
|
G1
|
31(17.42%)
|
20(22.47%)
|
11(12.36%)
|
|
G2
|
95(53.37%)
|
43(48.31%)
|
52(58.43%)
|
|
G3
|
48(26.97%)
|
24(26.97%)
|
24(26.97%)
|
|
G4
|
2(1.12%)
|
2(2.25%)
|
0(0%)
|
|
unkonw
|
2(1.12%)
|
0(0%)
|
2(2.25%)
|
|
Stage
|
|
|
|
0.614
|
Stage I
|
21(11.8%)
|
13(14.61%)
|
8(8.99%)
|
|
Stage II
|
147(82.58%)
|
70(78.65%)
|
77(86.52%)
|
|
Stage III
|
3(1.69%)
|
1(1.12%)
|
2(2.25%)
|
|
Stage IV
|
4(2.25%)
|
2(2.25%)
|
2(2.25%)
|
|
unknow
|
3(1.69%)
|
3(3.37%)
|
0(0%)
|
|
T
|
|
|
|
0.745
|
T1
|
7(3.93%)
|
3(3.37%)
|
4(4.49%)
|
|
T2
|
24(13.48%)
|
14(15.73%)
|
10(11.24%)
|
|
T3
|
142(79.78%)
|
69(77.53%)
|
73(82.02%)
|
|
T4
|
3(1.69%)
|
1(1.12%)
|
2(2.25%)
|
|
unknow
|
2(1.12%)
|
2(2.25%)
|
0(0%)
|
|
M
|
|
|
|
1.000
|
M0
|
80(44.94%)
|
39(43.82%)
|
41(46.07%)
|
|
M1
|
4(2.25%)
|
2(2.25%)
|
2(2.25%)
|
|
unknow
|
94(52.81%)
|
48(53.93%)
|
46(51.69%)
|
|
N
|
|
|
|
0.595
|
N0
|
49(27.53%)
|
22(24.72%)
|
27(30.34%)
|
|
N1
|
124(69.66%)
|
63(70.79%)
|
61(68.54%)
|
|
unknow
|
5(2.81%)
|
4(4.49%)
|
1(1.12%)
|
|
Predictive Power of Prognostic Signature Based on 5-NRLs
All 178 PC patients were divided into high-risk (n = 103) and low-risk (n = 75) subgroups according to the median RS calculated from the 5-NRLs signature. According to the Kaplan-Meier analysis, the OS of patients in the high-risk group was significantly shorter than that in the low-risk group (P < 0.001, Fig. 4A). The heatmap of the expression of 5-NRLs in different risk subgroups, distribution of RS and survival status of PC patients were shown in Fig. 4B-D, respectively. In order to further confirm the signature's validity, the RS and relevant clinicopathological characteristics (age, gender, tumor histopathological grade, clinical stage, T stage, M stage and N stage) were subjected to UniCox analysis. The results found that tumor histopathological grade (HR = 1. 327, 95% CI = 1. 032 ~ 1.706, P = 0. 027) and RS (HR = 1.356, 95% CI = 1.204 ~ 1.527, P < 0.001) could be relevant risk factors for prognosis of PC patients (Fig. 5A). Subsequently, a MultiCox analysis was performed, and the results demonstrated that only RS could be considered as an independent prognostic risk factor (HR = 1.362, 95% C1 1.197 ~ 1.550, P < 0.001, Fig. 5B). These findings suggested that the 5-NRLs signature had advantages over clinicopathological parameters in predicting the survival of PC patients.
Verifying the predictive ability of 5-NRLs signature
To further validation of the predictive ability of the 5-NRLs signature, the ROC curves were plotted for RS and clinicopathological features, meanwhile, the AUC was separately calculated for each index. In the whole cohort, the predictive ability of the prognostic signature was found to be higher than those of the common clinicopathological features, and the AUC of RS for 1, 3, and 5-years survival rates were 0.730, 0.791 and 0.799, respectively (Fig. 6A-B). The 1-year C-index of RS was 0.680 (Fig. 6C). In the training set, the AUCs of RS for 1, 3, and 5-years survival rates were 0.806, 0.748 and 0.831, respectively (Fig. 6D-E). In the validation set, the AUCs of RS for 1, 3, and 5-years survival rates were 0.663, 0.779 and 0.732, respectively (Fig. 6F-G). These results suggested that the 5-NRLs signature could be effective in predicting the prognosis of PC patients. In addition, a Sankey diagram was constructed in this study, which not only adequately shows the interaction between these 5 NRLs and NRGs, but also further illustrates the association between the 5-NRLs risk signature and OS of PC patients (Fig. 6H).
Nomogram and Calibration Curve
In order to investigate the clinical applicability of the model, a nomogram for predicting OS at 1, 3 and 5 years in patients with PC was constructed on the basis of RS and clinicopathological factors (Fig. 7A). The calibration curve indicated that the OS of PC patients predicted based on the nomogram was in excellent agreement with the actual OS (Fig. 7B-D). It further confirmed the accuracy and generalization ability of this nomogram model.
PCA and Biological Pathways Analyses
To further determine the clustering of the signature on the distribution of patients with different risks, the 3D scatter plots of PCA showed the distribution of patients based on different patterns, respectively. It was found that the samples grouped based on our signature had obvious clustering characteristics (Fig. 8A-C), and patients were divided into different quadrants based on their risk scores, suggesting that the prognostic signature has a high sensitivity for distinguishing different risk of PC patients.
We further conducted GO and GSEA-KEGG enrichment analyses for the differential genes between the different risk subgroups in order to investigate their potential biological functions and mechanisms. The results of GO analysis suggested that they were primarily enriched in immunoglobulin-related pathways in the cellular component (CC) group, immune response activation signaling and immune response activation cell surface receptor signaling pathways in the biological process (BP) group, and antigen binding-related functions in the molecular function (MF) group (Fig. 8D-E). According to the GSEA-KEGG analysis's findings, the high-risk subgroup mostly participated in necroptosis-related pathways (proteasome, oxidative phosphorylation, RNA degradation); cancer-related pathways (P53 signaling pathway, pancreatic cancer); immune-related pathways (Pathogenic Escherichia coli infection, systemic lupus erythematosus) (Fig. 8F).
The Importance of the 5-NRLs Signature in TME
The internal and external environments in which tumor cells are located have a close relationship to the development of tumors, and the TME typically comprises tumor cells, mesenchymal cells, and immune cells [23]. The stromal cell and immune cell infiltration levels of the different risk subgroups were compared using the ESTIMATE algorithm. According to the findings, the immune cell scores and stromal cell scores were significantly higher in the low-risk subgroup (Fig. 9A-C), while the tumor purity scores were markedly higher in the high-risk subgroup (Fig. 9D). By employing the 'CIBERSORT' R package, we further compared 22 immune infiltrating cells in the high- and low-risk subgroups. As shown in Fig. 9E, M0 macrophages and regulatory T cells (Tregs) were considerably upregulated in the high-risk subgroup, suggesting that these abnormally infiltrated immune cells may be linked to the progression of PC. However, the low-risk subgroup had considerably higher levels of naive B cells, memory B cells, plasma cells, CD8+ T cells, activated CD4+ T memory cells, and monocytes (Fig. 9E). These results indicated that the 5-NRLs signature might play an important role in regulating the TME of PC.
The Importance of the 5-NRLs Signature in Immunotherapy and Chemotherapy
With regard to the therapeutic treatment of numerous cancers, immunotherapy has produced impressive outcomes, and pertinent biomarkers that forecast patient response to immunotherapy, such as TMB and immunological checkpoints, have shown tremendous promise [24, 25]. In the current study, we compared the somatic mutational variants between the different risk subgroups. The results demonstrated that the TMB was significantly lower in the low-risk subgroup (Fig. 10A). Moreover, PC patients with high-risk and high-TMB had the worst prognosis (Fig. 10B-C).
According to a further analysis of the expression levels of common immunological checkpoints, the majority of immune checkpoints demonstrated stronger activation in the low-risk subgroup (Fig. 11A). Also, the TIDE scores were substantially lower in the high-risk subgroup (Fig. 11B). Based on the IPS of PC patients obtained from the TCIA database, we only discovered significant differences between high- and low-risk subgroups when both PD-1 and CTLA4 were negative responses (Fig. 11E-F).
Drug sensitivity analysis was performed by evaluating the IC50 of 138 conventional chemotherapeutic agents included in the “pRRophetic” R package between the different risk subgroups. Sixteen chemotherapeutic agents that were significantly different and had application in PC patients were selected for visualization (Supplement Table S2). Patients in the high-risk subgroup seemed to be more responsive to Bryostatin.1, Epothilone.B, Erlotinib, Lapatinib, and Paclitaxel, which suggests that these agents may be more suitable for PC patients with high RS (Fig. 12A-E). Nevertheless, AICAR, Axitinib, Bosutinib, DMOG, Lenalidomide, Methotrexate, Rapamycin, Shikonin, Sunitinib, Temsirolimus and Vorinostat had lower IC50 values in the low-risk subgroup, which indicates these agents might be better suited to PC patients with lower RS (Fig. 12F-P). These results suggested that the 5-NRLs signature is helpful in selecting immunotherapy and chemotherapy agents for PC.