Page 19 - Read Online

Page 8 of 12 Nishioka et al. Hepatoma Res 2018;4:1 I http://dx.doi.org/10.20517/2394-5079.2017.46

Table 3. HCC-related gene sets from the chemical and genetic perturbations collection enriched in the high AFP
tumor class
Name FDR
BOYAULT_LIVER_CANCER_SUBCLASS_G3_UP 0.086
HOSHIDA_LIVER_CANCER_SUBCLASS_S2 0.088
BOYAULT_LIVER_CANCER_SUBCLASS_G123_UP 0.099
CHIANG_LIVER_CANCER_SUBCLASS_PROLIFERATION_UP 0.101
YAMASHITA_LIVER_CANCER_WITH_EPCAM_UP 0.104
SAKAI_CHRONIC_HEPATITIS_VS_LIVER_CANCER_UP 0.118
LEE_LIVER_CANCER_SURVIVAL_DN 0.147
CHIANG_LIVER_CANCER_SUBCLASS_UNANNOTATED_DN 0.186
BOYAULT_LIVER_CANCER_SUBCLASS_G23_UP 0.228
BOYAULT_LIVER_CANCER_SUBCLASS_G12_UP 0.252
HCC: hepatocellular carcinoma; FDR: false discovery rate

For survival analysis based on the molecular classification of tumors, patients whose tumors were assigned
to sub-classes S1 and S2 were grouped together because of the similarity in gene expression between
sub-classes S1 and S2 relative to S3 [Figure 1] and the significant associations of sub-classes S1 and S2
with higher AFP levels (mean 2819 ng/mL) as compared to S3 (mean 72 ng/mL). A Kaplan-Meier plot
comparing the survival rates of patients with non-S3 tumors vs. patients with S3 tumors is shown in
Figure 3B. Non-S3 tumors were significantly associated with poorer overall survival (Log-Rank P = 0.024)
and a mortality hazard ratio of 3.6 (95% confidence interval 1.1-11.6, P = 0.035). Age, gender, FIB4 > 2.87,
and HBV infection were not found to be significant predictors of overall survival following liver resection.

DISCUSSION
In this study, tumoral differences were examined at the gene signature level between HCC sub-groups
categorized on the basis of AFP and other clinical parameters. Using GSEA, we found no significant
differences in gene set enrichment between tumors categorized by patient age, gender, clinical severity of
liver fibrosis, and HBV infection. However, we did find significant differences between tumors categorized by
patient serum AFP level. These differences proved to be biologically coherent across analyses involving two
distinct gene set collections from the mSigDB molecular signature repository. Specifically, using the mSigDB
Hallmarks collection of gene sets, we found serum AFP levels > 400 ng/mL to be associated with gene set
enrichments corresponding to MYC oncogene activation, enhanced DNA replication/repair, and cell cycle
progression, all of which are defining properties of highly proliferating tumors. In addition, we found tumors
from patients with high serum AFP levels to be significantly enriched for genes associated with proteostasis,
[22]
a potential mechanism for the release of AFP by tumor cells .
Using the larger CGP gene set collection comprised of 2675 gene signatures, we found that tumors associated
with high serum AFP levels were also significantly enriched for genes belonging to several existing molecular
classification signatures for HCC. Three of these signatures have already been associated with high AFP levels
[10]
by previous studies. The first signature corresponds to the S2 tumor sub-class defined by Hoshida et al. . In
addition to being associated with high serum AFP levels, this sub-class of HCC tumors is characterized by
MYC oncogene activation and enhanced cellular proliferation. Thus, these results are concordant with the
results obtained by GSEA using the Hallmarks gene set collection. Another HCC sub-classification signature
found significantly enriched in the high AFP class of tumors corresponds to a “proliferation” sub-class of
[8]
HCC described by Chiang et al. . In addition to being associated with high serum AFP levels, this sub-class
is associated with chromosomal instability and overexpression of proliferation-related genes. The third HCC
classification signature that was significant in our analysis corresponds to an EpCam signature defined by
[12]
Yamashita et al. . In their study, this signature, when combined with AFP expression, identifies four patient
sub-groups, each with their own unique sub-signature and survival pattern. Notably, the AFP-positive sub-
[12]
groups were associated with higher TNM stage and worse clinical prognosis . Altogether, these distinct