Expression pattern of p53-pathway correlate with survival
We retrieved 67 genes involved in p53-pathway from the KEGG pathway database (pathway: map04115). Quantile method [22] was used to remove genes with low expression level and 62 genes remained. Finally, the expression level of the 62 genes were log transformed for further analysis (Fig. 1A).
Performing consensus clustering analysis in the discovery cohort, we found that all patients were clustered into 3 subgroups (Fig. 1A). The clinic characteristics of patients in 3 subgroups are listed in Table 1. The results showed that the age, TP53 mutation and histological diagnosis were significantly different between 3 clusters. The frequency of TP53 mutation in cluster 2 (55.4%) is much higher than that in cluster 1 (13.2%) and cluster 2 (31.8%). Comparing the survival curve between 3 clusters, we found that cluster 1 has the best prognosis outcome, while cluster 2 and cluster 3 showed comparable prognosis (Fig. 1B). There is a significant difference in prognosis outcome between cluster 1 and cluster 3 (log rank p=0.031) (Fig. 1C).
Comparing the expression level of genes between cluster 1 and cluster 3, we identified a total of 12 deferentially expressed genes (DEGs) (Fig. 1D), 10 of which was over-expressed in cluster 3. GO enrichment analysis showed that these DEGs were mainly related to cell cycle arrest (Fig. 1E).
Table 1
The clinical characteristics of each cluster
Variables
|
Cluster 1 (n=106)
|
Cluster 2 (n=177)
|
Cluster 3 (n=132)
|
p.value
|
Gender:
|
|
|
|
0.083
|
Female
|
47 (44.3%)
|
58 (32.8%)
|
42 (31.8%)
|
|
Male
|
59 (55.7%)
|
119 (67.2%)
|
90 (68.2%)
|
|
Age:
|
|
|
|
<0.001
|
<65
|
31 (29.5%)
|
69 (40.4%)
|
71 (54.6%)
|
|
>=65
|
74 (70.5%)
|
102 (59.6%)
|
59 (45.4%)
|
|
Histological diagnosis
|
|
|
|
<0.001
|
Intestinal type
|
16 (15.1%)
|
44 (24.9%)
|
17 (12.9%)
|
|
NOS type
|
46 (43.4%)
|
64 (36.2%)
|
46 (34.8%)
|
|
Diffuse type
|
15 (14.2%)
|
15 (8.47%)
|
37 (28.0%)
|
|
Tubular type
|
23 (21.7%)
|
40 (22.6%)
|
10 (7.58%)
|
|
Else
|
6 (5.66%)
|
14 (7.91%)
|
22 (16.7%)
|
|
TP53 mutation:
|
|
|
|
<0.001
|
No
|
92 (86.8%)
|
79 (44.6%)
|
97 (78.2%)
|
|
Yes
|
14 (13.2%)
|
98 (55.4%)
|
27 (21.8%)
|
|
TNM Stage:
|
|
|
|
0.666
|
I
|
14 (14.4%)
|
30 (17.6%)
|
13 (10.6%)
|
|
II
|
30 (30.9%)
|
53 (31.2%)
|
40 (32.5%)
|
|
III
|
44 (45.4%)
|
67 (39.4%)
|
58 (47.2%)
|
|
IV
|
9 (9.28%)
|
20 (11.8%)
|
12 (9.76%)
|
|
T stage:
|
|
|
|
0.041
|
T1
|
8 (7.92%)
|
10 (5.71%)
|
4 (3.08%)
|
|
T2
|
16 (15.8%)
|
46 (26.3%)
|
26 (20.0%)
|
|
T3
|
42 (41.6%)
|
83 (47.4%)
|
56 (43.1%)
|
|
T4
|
35 (34.7%)
|
36 (20.6%)
|
44 (33.8%)
|
|
N stage:
|
|
|
|
0.817
|
N0
|
33 (32.7%)
|
54 (32.0%)
|
36 (28.6%)
|
|
N1
|
32 (31.7%)
|
43 (25.4%)
|
37 (29.4%)
|
|
N2
|
16 (15.8%)
|
38 (22.5%)
|
25 (19.8%)
|
|
N3
|
20 (19.8%)
|
34 (20.1%)
|
28 (22.2%)
|
|
M stage
|
|
|
|
0.840
|
M0
|
94 (93.1%)
|
155 (93.9%)
|
118 (92.2%)
|
|
M1
|
7 (6.93%)
|
10 (6.06%)
|
10 (7.81%)
|
|
Construction and validation of p53-pathway-related prognostic signature
To evaluate the prognostic value of 12 p53-pathway-related DEGs, we first performed univariate Cox regression analysis and found that 3 of 12 genes were significant associated with the survival outcome (Fig. 2A), including SERPINE1 (p<0.001), THBS1 (p=0.001) and GADD45B (p=0.004). Then, the multivariate Cox regression model was employed to construct a prognostic signature as following: The distribution of risk score, survival status and expression profile of signature gene were showed in Fig. 2E.
Using the median risk score as the critical value, all patients were classified into high and low risk group. As shown in Fig. 2B, the patients in the low-risk group tends to live longer than those in the high-risk group (log rank test, p<0.001). The prognosis signature was further validated within two patients’ subgroups with early stage (stage I/II) and late stage (stage III/IV) cancer, respectively. In both subgroups, patients were divided into the low-risk group with better survival and the high-risk group with worse outcomes (Fig. 2C and D). The prognosis signature was also validated by 2 external cohort, including GSE84437 (Fig. 3D) and GSE66229 (Fig. 3E). From the figures, we observed that two external cohort were consistently classified into the high-risk group with worse survivals and low-risk group with better outcome. Multivariate analysis also identified the prognostic signature as an independent prognostic factor (adjusted p<0.001), which is independent of the tumor stage and age (Table 2).
Table 2
multivariable analysis of prognostic signature in the prediction of gastric cancer survival
Variables
|
HR (95% CI)
|
P value
|
Age (Year) (>=65 vs <65)
|
1.030 (1.013,1.047)
|
<0.001
|
TNM stage
|
|
|
Stage II vs. Stage I
|
1.276 (0.659, 2.470)
|
0.047
|
Stage III vs. Stage I
|
1.978 (1.054, 3.712)
|
0.034
|
Stage IV vs. Stage I
|
3.427 (1.778, 6.607)
|
<0.001
|
Risk score (high vs. Low)
|
2.172 (1.544, 3.056)
|
<0.001
|
Finally, we constructed a nomogram by integrating the p53-pathway-based risk score and the well-known prognostic factors including age and TNM stage (Fig. 3A). The calibration curve showed that the nomogram performed well, compared with an ideal model (Fig. 3B). The AUC of the ROC curve of the nomogram for 5 years reached 0.71, higher than that of TNM stage (AUC = 0.66), indicating better performance of our prognosis signature (Fig. 3C).
Functional analysis of the p53-pathway-related gene
To explore the potential target gene that regulated by three genes of the prognostic signature, we construct a co-expression network of the p53-pathway using the WGCNA package in R. The co-expression network showed that the three genes may regulated 10 genes involved in p53-pathway (Fig. 4A), including CDK1, CCNE2, CHEK1 and so on. GO (Fig. 4B) and KEGG analysis (Fig. 4C) showed that the differential expressed genes identified by comparing high risk and low risk group were enriched in several important pathways, including Extracellular matrix organization, ECM-receptor interaction, PI3K-Akt signaling pathway and so on.