Effect of FRG1 alone on survival in different cancer types
Kaplan-Meir survival analysis was performed, to determine the effect of FRG1 expression on the OS across the seven most frequent cancer types. In cervix, stomach and prostate cancers there was a highly significant difference in the survival probability between high and low FRG1 expression groups (Fig. 2). In liver cancer the difference in survivability was marginally significant. In breast, lung, and colorectal cancers although the trend was there yet, the difference was not significant. Overall, this data suggests that FRG1 affects the survival in cancers but the extent of the effect is tissue specific. Analysis of FRG1 expression alone may not be enough to explain the contribution of other genes, which are affected by FRG1 directly or indirectly. Therefore, we did multigene model-based analysis in breast, lung, colorectal and liver cancer to get a clear idea about the effect of FRG1 expression on OS.
High FRG1 expression is associated with a good prognosis in the multigene model
To determine the contribution of FRG1 on survival, the effect of other genes correlated with FRG1, was neutralized using the multivariate Cox regression model.
Effect of FRG1 and correlated genes on survival in breast cancer
In breast carcinoma, initially we entered the top 20 genes (Supplementary Table S2) (rs ≥ 0.353) correlated with FRG1 to generate the multivariate cox regression model in the TCGA-BRCA dataset. Sequentially, lowest correlated genes were removed from the model till the FRG1 showed a maximum level of significant association with survival (Table 1). The hazard ratio of FRG1 was 0.133 (95% CI 0.029–0.599, p = 0.009) for breast cancer patient’s death.
In order to analyze the combined effect of FRG1 and the correlated genes (genes present in the final model) on the OS, for each breast cancer patient risk score was calculated. The patients were stratified into low-risk (n = 612) and high-risk (n = 611) groups based on the median risk score value. A significant difference (p = 2.45E-13) was observed between the groups in OS (Fig. 3A). The AUC for this risk model was 0.645 (Supplementary Fig. S2). There was significantly higher (p = 0.0001) FRG1 expression in the low-risk group compared to the high-risk group (Fig. 3B).
Table 1
Covariates present in multivariate Cox regression model in breast cancer patients.
Genes | B | Sig. | Exp(B), 95.0% CI for Exp(B) |
HPF1 | 1.233 | 0.134 | 3.433 (0.683,17.259) |
ING2 | -1.865 | 0.003 | 0.155 (0.045,0.535) |
UFSP2 | 2.672 | 0 | 14.467 (3.49,59.965) |
PFDN5 | -1.501 | 0.089 | 0.223 (0.04,1.256) |
EXOSC9 | -1.376 | 0.08 | 0.253 (0.054,1.177) |
SARNP | 0.939 | 0.049 | 2.557 (1.005,6.504) |
SRP19 | -0.34 | 0.688 | 0.712 (0.135,3.743) |
RPS3A | -0.76 | 0.39 | 0.468 (0.083,2.647) |
NDUFC1 | -0.472 | 0.514 | 0.624 (0.151,2.577) |
NACA | 0.959 | 0.302 | 2.609 (0.422,16.14) |
RWDD4 | -0.247 | 0.774 | 0.781 (0.145,4.201) |
NSA2 | -0.578 | 0.389 | 0.561 (0.15,2.092) |
TBCA | 1.132 | 0.136 | 3.102 (0.7,13.75) |
MRPS18C | 0.698 | 0.401 | 2.01 (0.395,10.239) |
TRIM56 | -0.226 | 0.663 | 0.797 (0.288,2.205) |
TTC1 | -0.236 | 0.792 | 0.79 (0.137,4.549) |
PLRG1 | 0.758 | 0.368 | 2.134 (0.409,11.134) |
MRPL1 | 0.536 | 0.498 | 1.71 (0.362,8.069) |
RPL34 | 0.588 | 0.464 | 1.8 (0.373,8.688) |
FRG1 | -2.021 | 0.009 | 0.133 (0.029,0.599) |
age_months | 0.003 | 0 | 1.003 (1.002,1.004) |
Effect of FRG1 and correlated genes on survival in Lung cancer
The top 20 genes (Supplementary Table S2) were added (rs ≥ 0.535) with FGR1 to generate the multivariate cox regression model using TCGA-MESO, TCGA-LUAD and TCGA-LUSC datasets. To investigate the prognostic effect of FRG1 on lung carcinoma patients, we applied the same strategy as described above. The final model had 17 genes where the hazard ratio of FRG1 was 0.235 (95% CI 0.074–0.742, p = 0.014) for lung cancer patient’s death (Table 2).
All the patients were stratified into low-risk (n = 559) and high-risk (n = 572) groups based on the median value of the risk score. The AUC for this risk model was 0.569 (Supplementary Fig. S2). A significant difference (p = 1.0E-6) in OS was observed between the groups (Fig. 4A). There was significantly (p = 0.0007) high FRG1 expression in the low-risk group compared to the high-risk group (Fig. 4B).
Table 2
Covariates present in multivariate Cox regression model in lung cancer patients.
Genes | B | Sig. | Exp(B), 95.0% CI for Exp(B) |
HPF1 | 0.801 | 0.09 | 2.227 (0.882,5.621) |
MRPS18C | 0.449 | 0.373 | 1.566 (0.584,4.2) |
ANAPC10 | -1.668 | 0.008 | 0.189 (0.055,0.652) |
LSM6 | 0.725 | 0.149 | 2.064 (0.771,5.525) |
ATP5PO | -0.004 | 0.993 | 0.996 (0.425,2.338) |
UBE2B | 0.018 | 0.966 | 1.019 (0.432,2.403) |
THOC7 | -0.65 | 0.229 | 0.522 (0.181,1.504) |
NDUFS4 | 0.347 | 0.362 | 1.416 (0.671,2.988) |
RWDD4 | 0.331 | 0.53 | 1.393 (0.495,3.917) |
COX7B | -0.295 | 0.477 | 0.744 (0.33,1.681) |
GSTO1 | 0.368 | 0.191 | 1.444 (0.833,2.505) |
RPL24 | 0.143 | 0.691 | 1.154 (0.57,2.335) |
UBE2D3 | 0.951 | 0.113 | 2.587 (0.799,8.377) |
UXT | -0.246 | 0.617 | 0.782 (0.299,2.048) |
LSM3 | 0.791 | 0.124 | 2.206 (0.805,6.041) |
RPL34 | -0.508 | 0.157 | 0.602 (0.298,1.216) |
SHANK2 | 0.221 | 0.048 | 1.247 (1.002,1.552) |
FRG1 | -1.448 | 0.014 | 0.235 (0.074,0.742) |
age (Months) | 0.001 | 0.025 | 1.001 (1,1.002) |
FRG1 and correlated genes do not predict survival in the colorectal cancer
To investigate the prognostic effect of FRG1 using colorectal cancer TCGA-READ and TCGA-COAD datasets, the top 20 correlated genes (Supplementary Table S2) were added (rs ≥ 0.964) in the multivariate cox regression model (Supplementary Table S3). Models with all the 20 genes, as well as any other combination of genes didn’t show significant effect of FRG1 on the OS of the colorectal cancer patients.
Effect of FRG1 and correlated genes on survival in liver cancer
The top 20 genes correlated (Supplementary Table S2) with FRG1, with rs cutoff ≥ 0.539, were used to generate the multivariate cox regression model using TCGA-LIHC and TCGA-CHOL datasets. The final model had 16 genes (Table 3) where the hazard ratio of FRG1 was 0.18 (95% CI 0.034–0.948, p = 0.043) for liver cancer patient’s death.
Next, to determine the effect in the multigene model, the patients were divided into the low-risk group (n = 231) and high-risk group (n = 231) based on the median risk score. The AUC for this risk model was 0.616 (S2 Fig.). A significant (p = 0.0001) difference in OS was observed between the two groups (Fig. 5A). Comparison of FRG1 expression between the high-risk group and low-risk group (Fig. 5B) showed significantly (p < 0.0001) higher expression in the low-risk group.
Table 3
Covariates present in multivariate Cox regression model in liver cancer patients.
Genes | B | Sig. | Exp(B), 95.0% CI for Exp(B) |
HPF1 | 0.791 | 0.238 | 2.206 (0.593,8.209) |
POMP | 0.885 | 0.174 | 2.424 (0.676,8.691) |
UXT | -0.222 | 0.772 | 0.801 (0.179,3.591) |
RREB1 | -1.001 | 0.089 | 0.367 (0.116,1.164) |
LMTK2 | -0.551 | 0.289 | 0.576 (0.208,1.596) |
NDUFC1 | -1.594 | 0.009 | 0.203 (0.061,0.676) |
EP300 | 0.786 | 0.335 | 2.195 (0.445,10.844) |
NCOA2 | -0.613 | 0.146 | 0.542 (0.237,1.239) |
MRPL54 | -1.53 | 0.009 | 0.217 (0.069,0.683) |
KMT2C | -0.615 | 0.259 | 0.541 (0.186,1.574) |
PRR14L | -0.073 | 0.914 | 0.93 (0.251,3.449) |
UFSP2 | 1.583 | 0.022 | 4.871 (1.254,18.923) |
HACD2 | 0.919 | 0.033 | 2.507 (1.075,5.846) |
CELF1 | 0.741 | 0.421 | 2.097 (0.346,12.73) |
NCOA6 | -0.044 | 0.954 | 0.957 (0.216,4.245) |
NDUFS5 | 1.581 | 0.004 | 4.86 (1.681,14.049) |
FRG1 | -1.716 | 0.043 | 0.18 (0.034,0.948) |
age (Months) | 0.001 | 0.055 | 1.001 (1,1.002) |
FRG1 knockdown in HEK293T reduces expression of HPF1, RPL34 and EXOSC9
From the top 20 genes correlated with FRG1 across cancer types, we found that three genes (HPF1, RPL34 and EXOSC9) were common. We hypothesized that these genes could be part of pathway/pathways in which FRG1 has a role and could affect their expression. To validate this, the expression level of these three genes was analyzed in response to FRG1 depletion in the HEK293 cell line by quantitative real-time PCR. We observed that knockdown of FRG1 led to a significant decrease in expression of HPF1 (0.68-fold, p-value = 0.011), RPL34 (0.65-fold, p-value = 0.025) and EXOSC9 (0.54-fold, p-value = 0.012) (Fig. 6). These findings confirm the effect of FRG1 in transcriptional regulation of HPF1, RPL34, and EXOSC9, which could be direct or indirect.
To figure out the pathway/s where FRG1 may have a role, we used genes that show correlation with FRG1 expression and the genes, which interact with FRG1 (HIPPIE database) as input in the STRING database. Individual networks for each cancer type are shown in Fig. 7. Thereafter all the networks were merged and the intersection was obtained using the Merge tool of Cytoscape, giving us the most common pathway (Fig. 7). The merged pathway had 17 nodes (MEPCE, LARP7, SUMO2, UBE2O, HECW2, RBPMS, JUN, ESR2, SART3, EXOSC8, FRG1, PARP2, C4orf27 (HPF1), EFTUD2, SNRPD3, CWC22 and AQR) and 21 edges. All the networks formed are statistically significant with protein-protein interaction (PPI) enrichment p-value < 0.05.