Genetic alteration and misexpression of Polycomb group genes in hepatocellular carcinoma

Abstract

Although the abnormal expression of Polycomb-group (PcG) proteins is closely associated with carcinogenesis and the clinicopathological features of hepatocellular carcinoma (HCC), the genetic mutation profile of PcG genes has not been well established. In this study of human HCC specimens, we firstly discovered a highly conserved mutation site, G553C, in the Polycomb Repressive Complex 2 (PRC2) gene enhancer of zeste homolog 2 (EZH2). This site also harbors a single nucleotide polymorphism (SNP), rs2302427, which plays an important antagonistic role in HCC. Kaplan-Meier survival curves showed that the tumor-free and overall survival of patients with EZH2 G553C were superior to those without the mutation. The G allele frequencies in patients and healthy subjects were 0.2% and 0.122%, respectively, with significant differences in distribution. The individuals carrying the GG and the GC genotypes at rs2302427 showed 3.083-fold and 1.827-fold higher risks of HCC, respectively, compared with individuals carrying the wild-type allele. Furthermore, Immunohistochemical staining revealed that the expression levels of CBX8 (in 53/123 samples) and BMI1 (in 60/130 samples) were markedly increased in human HCC specimens. Importantly, the overall and tumor-free survival rates were significantly reduced in the group of patients who simultaneously expressed PRC1 and PRC2. These results argue that a combination of PRC1 and PRC2 expression has a significant predictive/prognostic value for HCC patients. Taken together, our results indicate the abnormal expression and genetic mutation of PcG members are two independent events; cumulative genetic and epigenetic alterations act synergistically in liver carcinogenesis.

Introduction

Hepatocellular carcinoma (HCC) is the most common form of liver cancer. Numerous studies have greatly advanced our understanding of the pivotal role of specific gene mutations, including p53, CDKN2A, CTNNB1, AXIN1, HNF1, and IRF-2, in controlling HCC development [1]. Recently, emerging evidence has demonstrated an essential role for epigenetic regulators, such as microRNA or DNA methylation, in HCC [2-5], suggesting that HCC is frequently governed by cumulative genetic and epigenetic alterations [6].

Site-specific histone modifications are the major epigenetic mechanism for controlling a stable gene transcription state, but the importance of histone methylation in HCC development is not well established [7,8]. One of the best-studied histone modifications required for the maintenance of gene silencing in HCC is the trimethylation of histone 3 lysine 27 (H3K27me3), which is mediated by Polycomb-group (PcG) proteins [4,8-13]. The PcG comprises at least two distinct complexes. First, the initiation complex, the Polycomb Repressive Complex 2 (PRC2), has a core that in humans consists of the proteins enhancer of zeste homolog 2 (EZH2) and suppressor of zeste 12 (SUZ12). Second, the maintenance complex, PRC1, includes the proteins B-lymphoma Moloney murine leukemia virus insertion region-1 (BMI1), Chromobox homolog 8 (CBX8), and others [14]. The EZH2 SET domain specifically catalyzes the trimethylation of H3K27; H3K27me3 compresses the chromatin structure and leads to the transcriptional repression of genes such as E-cadherin, RUNX3, and cyclin-dependent kinase inhibitors [15-17]. Recently, we also demonstrated the tumor-promoting activity of PcG, which occurs by direct and indirect regulation in HCC [9]. We found EZH2 occupancy at chromatin coincides with H3K27me3 at promoters and directly silences the transcription of certain tumor suppressors in HCC. The H3K27me3-related target gene network of PcG contains well-established genes (e.g., CDKN2A) and previously undescribed genes (e.g., FOXO3, E2F1, and NOTCH2). PcG also represses the expression of the TP53 tumor suppressor in HCC independently of H3K27me3 [9]. These data strongly demonstrate the functional and mechanistic significance of certain gene regulatory networks that are regulated by PcG in HCC.

Compared to PRC2, PRC1 is more complicated and functionally diverse. Rather than catalyzing H3K27me3 methylation like PRC2, the PRC1 components include BMI1 and CBX8 family proteins, which show affinity for H3K27me3. In the prevailing model, PRC2 is recruited to specific genomic loci where it catalyzes H3K27 trimethylation [14]. The trimethylated histones in turn recruit PRC1, which catalyzes H2A ubiquitination and thereby antagonizes RNA polymerase II elongation and represses the transcription of target genes [18]. CBX8 was originally characterized as a transcriptional repressor of INK4a/ARF in fibroblasts [19]. We also found that both EZH2 and CBX8 repress INK4a/ARF expression in HCC cells through H3K27me3 methylation [9]. The functional significance of PRC1 in the initiation and development of HCC has not been completely deciphered, although its mechanistic importance in mediating the response to H3K27me3 has been addressed. As with EZH2, BMI1 over-expression was detected in 60.5% of primary HCC tumor tissues. The cumulative recurrence rate was significantly higher in BMI1-overexpressing patients than in their BMI1-negative counterparts [12]. The potential biological role of CBX8 in liver cancer remains largely undefined.

Interestingly, as an important epigenetic regulation factor, mutations in EZH2 gene are also involved in certain types of tumors. For example, sporadic point mutations affecting the Y641 and A677 residues in SET domain of EZH2 have been identified in lymphoma and myeloid neoplasms [20]. Further studies have indicated that PcG loci are sensitive targets of environmental stress; emerging evidence has shown extensive and frequent somatic PRC2 alterations in early T-cell precursor acute lymphoblastic leukemia, including deletions and sequence mutations in EED, EZH2 and SUZ12 [21]. Because of the clinical importance of PcG in HCC, it is interesting to determine whether the components of PcG are mutated in HCC, and if so, whether these mutations are clinically significant. In the present study, we investigated the clinical significance of PcG members, especially EZH2 and CBX8, and examined genetic mutations of EZH2 and CBX8 in HCC.

Materials and methods

Human HCC samples and immunohistochemistry (IHC)

Informed consent to participate in this study was obtained from each patient, and the protocol conformed to the ethical guidelines of the 1975 Declaration of Helsinki as reflected in a priori approval by the Xiamen University Medical Ethics Committee. Paraffin-embedded primary HCC tissues and corresponding adjacent non-tumorous liver samples from Han Chinese patients were obtained from the chronic liver disease biological sample bank. Department of Hepatobiliary Surgery, Zhongshan Hospital of Xiamen University [8]. During the same study period, 310 ethnic group-matched individuals who had no self-reported history of cancer at any site were enrolled as controls who underwent physical examination. Personal information and characteristics were collected from the study subjects using interviewer-administered questionnaires. All male and female patients were histopathologically diagnosed with stage I-IV HCC. The demographic data and clinicopathological features are listed in Supplementary Tables 9 and 10. Sections from paraffin-embedded samples were stained with affinity-purified anti-CBX8 or anti-BMI1 antibodies for IHC [8]. The expression of CBX8 or BMI1 in liver tissue as shown by IHC was evaluated by three independent pathologists.

DNA extraction from human HCC samples

DNA from human HCC samples was extracted from fresh tissue or from tissue that was frozen at -80°C after collection by using the E-Z 96 Tissue DNA Kit (Omega) according to the manufacturer’s protocol [22].

Exon sequencing

PCR reactions were performed in a volume of 25 µl and contained 50 ng of genomic DNA, 0.175 mM each primer, 0.2 mM dNTP mix, 1.5 mM MgCl₂, 10 mM Tris-HCl, 50 mM KCl, and 0.2 µl of Taq DNA Polymerase (TaKaRa). PCR cycling conditions were as follows: 95°C for 5 min, 30 cycles of denaturation at 95°C for 30 s, specified annealing temperature for 30 s, and extension for 30 s at 72°C, and then a final extension step for 10 min at 72°C. After purifying the PCR products, sequencing reactions were performed using a BigDye Terminator v3.1 kit (Applied Biosystems). The primers used for amplification and sequencing reactions, the sizes of the amplified PCR products, and the annealing temperatures for each pair of primers are listed in Supplementary Table 11. The HCC samples were sequenced twice independently to exclude the contamination between samples.

Statistical analysis

Deviations from Hardy-Weinberg equilibrium (HWE) were assessed using the HWSIM program. Differences in the allele and genotype frequencies of EZH2 and CBX8 between groups were evaluated using Chi-square tests. Quantitative trait tests were performed using UNPHASED 2.404, which examines the association of genetic polymorphisms with symptom severity. The effects of haplotypes on quantitative trait variation were evaluated using the sub-program QTPHASE of the UNPHASED software, which provided a chi-square value for the association between a particular haplotype and a continuous variable based on the statistical parameters of the QTPHASE sub-program. Other statistical analyses were performed using SPSS software version 17.0 (SPSS, Inc.; Chicago, IL). Survival curves were calculated by the Kaplan-Meier method, and comparisons were performed using the log-rank test. The results for parametric variables are expressed as the means ± SD or the means ± SEM. In all cases, p<0.05 was considered statistically significant.

Results

Recurrent G553C mutation in EZH2

To uncover the genetic mutation profile of EZH2 in HCC, we performed exon sequencing of the entire EZH2 locus in 110 HCC patients (Figure 1A). We focused our analysis on novel sequence changes predicted to affect protein coding. Among these variants, we found a G-to-C mutation in 39 of 110 HCC samples at nucleotide 553 (G553C) within exon 6 of the EZH2 gene; this transversion corresponds to a D185H amino acid substitution (Figure 1B). No other EZH2 mutations were detected. Homozygous mutations were found in 3 samples, while heterozygous mutations occurred in 36 samples and were thus much more common (Figure 1B). The striking recurrence of this mutation suggested that the G553C alteration in EZH2 is a common feature of HCC. Comparing the amino acid sequence of human EZH2 with those from other species revealed that the mutation site is highly conserved (Figure 1C).

Figure 1.

A screening for EZH2 mutations in HCC. A. Genomic organization of the EZH2 locus showing alternative exons and the protein domain structure. The location of the mutation affecting Asp185 in exon 6 of the EZH2 gene is shown. B. Illustration of sequencing results. Of 110 samples, 3 showed distinctly homozygous mutations and 36 showed distinctly heterozygous mutations. The amino acid substitutions in different HCC samples at codon 185 were detected by sequencing. C. A multiple alignment of EZH2 protein sequences from thirteen species; Asp185 is conserved. D and E. Kaplan-Meier curves for tumor-free and overall survival in HCC patients with EZH2 WT and D185H. F. Serum AFP levels of HCC patients with EZH2 WT and D185H. The line in each panel indicates the median.

Next, we performed a correlation analysis between the EZH2 G553C mutation and HCC malignancy in 110 cases with detailed clinical information. Kaplan-Meier survival curves were plotted according to the mutation status. Interestingly, the tumor-free survival (Figure 1D, p=0.047, log-rank test) and overall survival (Figure 1E, p=0.049, log-rank test) of patients with the EZH2 D185H mutation were superior to those of patients without the mutation. This result suggests that the somatic mutation of D185 is most likely a loss-of-function mutation that interferes with the pathogenesis of HCC. No statistically significant correlation was found between G553C mutation and the serum alpha fetoprotein (AFP) level (Figure 1F).

The G553C mutation in EZH2 is a single-nucleotide polymorphism (SNP)

Because of the high incidence of EZH2 G553C mutations in HCC, we assumed that the mutation was a SNP. We thus recruited 311 healthy unrelated people as a control group. Using UNPHASED genetic analysis software (V2.404), we further analyzed the relationship between rs2302427 and HCC. The demographic and clinical characteristics of the selected cases and controls are shown in Supplementary Table 1. The mean ages of the 110 cases with HCC and the 311 controls were 58.3±13.36 and 53.98±12.19 years, respectively. We did not find any significant differences in G553C distribution with respect to age or gender between the HCC and control groups (P>0.05). We analyzed the distribution of rs2302427 in both HCC tissue and in adjacent tissues. In cancer tissue, the three different rs2302427 genotypes CC, CG, and GG showed differences in their distribution between HCC patients and the control group; however, the association lacked sufficient statistical significance (Table 1, p=0.071). The allelic frequencies of G in patients and healthy subjects were 0.186% and 0.122%, respectively. There were significant differences in the allelic gene frequency distribution, which indicates that the allelic frequency of G was significantly associated with HCC (Table 1, OR=1.636, p=0.024). Individuals carrying the G allele, the GC genotype or the GG genotype at rs2302427 in the cancer tissue showed a 1.636-fold (95% CI: 1.074-2.491, p=0.021), a 1.713-fold (95% CI: 1.048-2.8, p=0.0377), or a 2.313-fold (95% CI: 0.506-10.58, p=0.361), respectively, higher risk of HCC than patients carrying the wild-type allele (Table 2).

Table 1.

Distribution of EZH2 genotypes and allelic frequencies in the study population

EZH2 (HCC tissue)	Genotypes			Allelic Frequency
rs2302427	CC	CG	GG	C	G
Patients	72	35	3	0.814	0.186
Controls	222	63	4	0.877	0.122
	χ2=5.27, d.f.=2, p=0.071			χ2=5.08, d.f.=1, p=0.024
EZH2 (adjacent tissue)	Genotypes			Allelic Frequency
rs2302427	CC	CG	GG	C	G
Patients	54	28	3	0.8	0.2
Controls	222	63	4	0.877	0.122
	χ2=6.13, d.f.=2, p=0.046			χ2=6.04, d.f.=1, p=0.013

Table 2.

Odds ratio with 95% CI of the EZH2 gene in HCC patients

Genotypes		Odds Ratio (CI 95%)	p value
EZH2 (cancer tissue)	G vs C	1.636 (1.074-2.491)	0.021
rs2302427	CG vs CC	1.713 (1.048-2.8)	0.0377
	GG vs CC	2.313 (0.506-10.58)	0.361
EZH2 (adjacent tissue)	G vs C	1.785 (1.138-2.801)	0.011
rs2302427	CG vs CC	1.827 (1.07-3.121)	0.035
	GG vs CC	3.083 (0.67-14.19)	0.199

In tissues adjacent to the HCC, the numbers of patients who carried the three different genotypes CC, CG, and GG of EZH2 rs2302427 were 54, 28, and 3, respectively, a distribution that was significantly different compared with control subjects (Table 1, p=0.046). The G allele frequencies in patients and healthy subjects were 0.2% and 0.122%, respectively, with significant differences in distribution (Table 1, p=0.013). Individuals carrying the GG and GC genotypes at rs2302427 showed a 3.083-fold (95% CI: 0.67-14.19, p=0.199) and a 1.827-fold (95% CI: 1.07-3.121, p=0.035), respectively, higher risk of HCC than individuals carrying the wild-type allele (Table 2). The G allele in adjacent tissues was also highly associated with HCC (Table 2, OR=1.785, p=0.011).

We further analyzed the distribution of rs2302427 in cancer tissue, para-cancerous tissue, and in healthy control tissue. The distributions in these three groups were in Hardy-Weinberg equilibrium. Both the genotypic and allelic distributions of rs2302427 exhibited differences between the cancer tissue and adjacent tissues (Tables 1 and 2). The allelic frequency also differed within the same group. An inconsistent genotype at rs2302427 was observed between the cancer and adjacent tissues in 8 cases (Tables 1 and 2). Based on these data, the G allele and the GC genotype act as risk factors for HCC incidence. The G-to-C mutation in both cancerous and adjacent tissues might be a self-protection mechanism for HCC patients. These data indicate that rs2302427 is not only closely associated with HCC SNPs but also a locus that is easily mutated during HCC progression.

Association of the rs2302427 SNP with clinical symptoms

The effect of haplotypes on quantitative trait variation was evaluated using the QTPHASE sub-program of UNPHASED software, which provided a chi-square value for the association between a special haplotype and a continuous variable. Using this software, we examined the association of rs2302427 with phenotype. The tumor-free survival period varied significantly with different genotypes in adjacent tissues as follows: survival for the CC genotype was 10.22 months, survival for the CG genotype was 17.79 months, and survival for the GG genotype was 30.67 months (Table 3, p=0.049). The survival for the C allele was 11.78 months, whereas the survival for the G allele was 20.06 months (Table 3, p=0.019), which indicated that the G allele may be a protective factor. Although in cancerous tissues there are no significant differences in the survival period with various genotypes, we observed that the survival period in patients with the GG genotype was longer than that for the other 2 genotypes. This may partially confirm the hypothesis that the G allele might be a protective factor (Table 3). The results show that the rs2302427 polymorphism of EZH2 is significantly associated with the risk of HCC.

Table 3.

Quantitative trait analysis of EZH2 genotypes and alleles

EZH2 (cancer tissue)	Genotypes				Allelic Frequency
rs2302427	CC	CG	GG	P	C	G	P
Overall survival	13.83	13.51	29.67	0.403	13.77	15.88	0.494
Tumor free survival	12.30	12.89	29.67	0.296	12.42	15.34	0.308
AFP	1.76	1.54	1.44	0.791	1.714	1.523	0.508
EZH2 (adjacent tissue)	Genotypes				Allelic Frequency
Rs2302427	CC	CG	GG	P	C	G	P
Overall survival	12.65	18.25	30.67	0.181	13.8	20.44	0.075
Tumor free survival	10.22	17.79	30.67	0.049	11.78	20.06	0.019
AFP	1.84	1.53	0.74	0.388	1.776	1.389	0.221

The expression of PRC1 is stimulated in HCC and is correlated with poor prognosis

It is unclear whether PRC1 family members are associated with liver carcinogenesis. To determine the clinical significance of PRC1 in HCC, we measured the expression of CBX8 and BMI1 in primary HCC samples from patients. Immunohistochemical (IHC) staining revealed that HCC tissues exhibited robust expression of these factors. Moreover, CBX8 staining was exclusively nuclear in HCC tissues but not in adjacent tissues (Figure 2A and Supplementary Figure 1A). The expression of CBX8 was markedly increased in 53 out of 123 HCC samples, accounting for 43.1% of the tumors examined. Similarly, the IHC detection revealed over-expression of BMI1 compared to adjacent normal tissues in 60 of 130 HCC tissue samples (Figure 2B). We further performed a correlation analysis between PRC1 expression and HCC malignancy. Kaplan-Meier survival curves showed that the 5-year overall survival and tumor-free survival were significantly lower in the CBX8 or BMI1 over-expressing HCC patients than in the under-expressing patients (Figure 2C and 2D, log-rank test). The serum level of AFP was dramatically elevated in the CBX8 over-expressing group compared with the under-expressing group (Figure 2E, p=0.005). The hyper-expression of BMI1 was also associated with increased AFP levels; however, the association lacked sufficient statistical significance (Figure 2F, p=0.303). These results point to the clinical significance of PRC1 as a biomarker for HCC diagnosis and prognostic evaluation. Further correlation analyses in both cohorts showed that a robust expression level of PRC1 was not associated with the degree of HCC aggressiveness, including differentiation, tumor multiplicity, and neoplasm staging (Supplementary Figure 1B-H).

Figure 2.

Increased expression of CBX8 or BMI1 is associated with poor prognosis of HCC. A and B. Representative micrographs of high-level CBX8 or BMI1 expression in primary HCC paraffin sections as visualized by IHC staining. The dotted lines indicate junctures between the tumor (T) and adjacent normal tissues (N). The original magnification values are 100× and 400×, respectively. C and D. Kaplan-Meier curves for overall and tumor-free survival in HCC patients with CBX8 or BMI1 hyper-expression (+) or under-expression (-). E and F. Serum AFP levels of HCC patients with CBX8 or BMI1 hyper-expression (+) or under-expression (-). The line in each panel indicates the median.

Simultaneous detection of PRC1 and PRC2 expression is a better predictor of HCC prognosis

To further evaluate the clinical relationship between PRC1 and PRC2 in HCC, we compared PRC1 IHC results to previous EZH2 expression data [8]. A trend toward lower 5-year overall and tumor-free survival rates was observed in groups with stronger expression of either EZH2 or CBX8 compared to under-expressing patients, but the association lacked sufficient statistical significance (Figure 3A). Significantly reduced 5-year overall (Figure 3A, p=0.000, log-rank test) and tumor-free survival rates (Figure 3B, p=0.001, log-rank test) were observed in patients who simultaneously expressed CBX8 and EZH2 compared to CBX8/EZH2 under-expressing patients, especially at the early post-operative stage. Additionally, the serum level of AFP was significantly elevated in the EZH2 and CBX8 co-expressing group compared to the EZH2/CBX8 under-expressing group (Figure 3C, p=0.000). Similarly, the overall survival (Figure 3D, p=0.015, log-rank test) and tumor-free survival (Figure 3E, p=0.009, log-rank test) in the group with elevated expression of both EZH2 and BMI1 was significantly lower than those of patients whose cancerous tissue showed lower expression of EZH2 or BMI1. The expression of AFP with high expression of both EZH2 and BMI1 in cancerous tissue was significantly higher than that with high expression of either EZH2 or BMI1 alone (Figure 3F, p=0.029). However, the overall and tumor-free survival showed no significant difference between the group with both BMI1 and CBX8 hyper-expression and the BMI1 and CBX8 hypo-expressing group (Supplementary Figure 2A and 2B). The expression of AFP in these two groups also showed no statistically significant difference (Supplementary Figure 2C). Collectively, our results argue that the combination of PRC1 and PRC2, but not the combination of CBX8 and BMI1, has a significant predictive/prognostic value for HCC patients.

Figure 3.

Simultaneous expression of EZH2 and CBX8 is useful for predicting poor HCC prognosis. A and B. Kaplan-Meier curves for overall and tumor-free survival in HCC patients with EZH2 and/or CBX8 hyper-expression or under-expression. C. Serum AFP levels of HCC patients with EZH2 and/or CBX8 hyper-expression (+) or under-expression (-). D and E. Kaplan-Meier curves for overall and tumor-free survival in HCC patients with EZH2 and/or BMI1 hyper-expression or under-expression. F. Serum AFP levels of HCC patients with EZH2 and/or BMI1 hyper-expression (+) or under-expression (-). The line in each panel indicates the median.

The mutational profile of CBX8 in HCC

As for EZH2, we performed a mutation screening of the entire CBX8 locus in HCC as well by exon-sequencing (Figure 4A). Among the 110 examined liver cancer specimens, 1 sample showed a homozygous C-to-T mutation of base 767 in the 5th exon of the CBX8 gene (Figure 4B). In addition, in 13 cases, the 5th exon of CBX8 showed a homozygous G-to-T mutation in base 950 (corresponding to G317V); 56 additional cases were heterozygous for the G950T mutation (Figure 4C), which is the functional SNP rs4889891. Comparing the CBX8 amino acid sequence with those of other species revealed that this amino acid is highly conserved (Figure 4D). Correlation analysis of the survival curve and of AFP expression with CBX8 mutation was performed for 110 liver cancer patients. No correlation with tumor-free survival or overall survival was found with CBX8 mutation in the cancer tissue of the patients (Figure 4E and 4F). AFP expression was clearly higher in patients with CBX8 mutation than in patients without CBX8 mutation (Figure 4G). This result showed that CBX8 mutation was not clearly associated with the prognosis of HCC patients but that it had some significance for the diagnosis.

Figure 4.

A screening for CBX8 mutations in HCC. (A) CBX8 locus showing exons structure. The mutations in exon 5 are shown. Illustration of sequencing results. Of 110 HCC samples, 1 showed a distinctly homozygous mutation in codon 256 causing an amino acid replacement (B) and 13 showed distinct homozygous mutations and 56 showed distinct heterozygous mutations at codon 317, causing an amino acid replacement (C). (D) A multiple alignment of CBX8 from eleven species. (E-G) Kaplan-Meier curves for tumor-free (E) and overall survival (F) and Serum AFP levels (G) in HCC patients with CBX8 WT and G317V. The line in each panel indicates the median.

We further analyzed the distribution of the rs4889891 SNP in the 110 HCC cases and in 355 controls (Supplementary Tables 2, 3, 4 and 5). The alleles with the highest distribution frequency at CBX8 rs4889891 in cancer tissue were homozygous C/C. Individuals carrying the C/A genotype at rs4889891 in the cancer tissue showed a 1.553-fold (95% CI: 0.949-2.539) higher risk of HCC than individuals carrying the wild-type alleles. Quantitative trait analysis showed no significant associations between CBX8 rs48-89891 and symptom severity, including overall and tumor-free survival (P>0.05).

Discussion

As evidence of human genome evolution, SNPs represent a type of genetic mutation that has become the third generation of genetic markers. In complex diseases such as cancer, some minor alleles in SNPs act as risk factors, whereas others are protective. Unlike SNPs, characterizing somatic mutations and their functional consequences remains a formidable challenge due to their low individual abundance.

Because of the importance of the PcG family in epigenetic regulation, its misexpression is closely associated with liver carcinogenesis [8]. However, the genetic mutation profiles of PcG genes in HCC remain unclear. Yu et al. observed that one polymorphic C allele at SNPs rs6950683 and rs3757441 was strongly associated with a lower risk of HCC [10]. In the present study, we discovered another highly conserved mutation site in HCC, G553C, which is also the SNP site rs2302427. Strikingly, we found that HCC patients carrying the minor G553C allele gain a significant survival advantage compared with those carrying the wild-type allele. The rs2302427 variant was found in both liver cancer and in paracancerous tissues; it was significantly associated with overall and tumor-free survival when found in paracancerous tissues but not when found in liver cancer tissue. This unexpected finding might be related to the distribution of rs2302427. In fact, the distributions of this SNP in cancer tissue and paracancerous tissues were different. It seems the SNP may convert into a mutation site depending on the cellular context. An interesting finding is that the tumor survival rate in G carriers is higher than in wild-type C carriers. Notably, the above results clearly confirm that the G553C point mutation of EZH2 in liver cancer is a meaningful and protective genetic change that has significant clinical diagnostic and prognostic value. Basically, each single SNP is uniform within an individual. Unexpectedly, we observed a different distribution of rs2302427 in cancerous tissues and adjacent tissues, suggesting that the G553C mutation may be both a SNP and a somatic mutation, a phenomenon that has not been reported in previous studies. It further suggested that the G allele plays an important role in the transition from benign hyperplasia to cancer. Our results show that the rs2302427 site might be a protective SNP that plays an important antagonistic role in the process of liver cancer development; it may thus provide a potential genotype and allele marker to classify patients for different treatment strategies. Whether this mutation initiates consecutive protective mechanisms against tumorigenesis, such as apoptosis, autophagy, and/or senescence, requires further exploration. Furthermore, how the G553C mutation affects the function of the EZH2 protein remains an open question. Our immunoprecipitation (IP) data showed that the G553C mutation did not affect the interaction between EZH2 and DNMT1 despite the mutation site within the DNMT1-interacting region (Supplementary Figure 3). In addition to the PRC2, we also found frequent mutations encoding the G317V (rs4889891) substitution in the CBX8 protein of PRC1 in HCC. Although there was no association between rs4889891 and HCC, and although this mutation had no statistically significant association with prognosis indicators such as tumor-free survival, it did show a strong correlation with the AFP level, which is generally considered as a diagnostic marker of HCC.

In addition to genomic mutations, abnormal expression of PcG proteins is often seen in different types of cancer. In our previous findings, the expression level of the EZH2 protein was markedly increased in HCC, and there was significantly lower overall and tumor-free survival in EZH2-over-expressing HCC patients [8]. In the present study, we found that PRC1 expression levels are also involved in the clinical progression and outcome of HCC patients. As with EZH2, the expression of CBX8 and BMI1 was stimulated in 43.1% and 46.2% of HCC tissues, respectively. A trend toward lower rates of overall and tumor-free survival was observed in cases with the over-expression of CBX8 or BMI1 compared with patients with lower expression levels. Notably, patients with high levels of PRC1 and PRC2 co-expression showed markedly decreased overall survival and tumor-free survival rates, especially at the early post-operative stage, compared to patients with hyper-expression of only one of these proteins. In contrast, over-expression of EZH2 or CBX8 was associated with a very positive outcome for HCC patients. The overall and tumor-free survival rates in such patients were maintained at 90% and 80%, respectively, at 60 months after operation. These data strongly suggest that a combination of PRC1 and PRC2 expression might have a significant prognostic value for HCC patients. It appears that both PRC1 and PRC2 play important roles in promoting HCC but that their target gene networks are relatively independent. According to our previous results, PRC1 (CBX8 and BMI1) and PRC2 (EZH2 and SUZ12) displayed very different gene regulation profiles in HCC [9]. Microarray results indicated that numerous genes are exclusively targeted by either PRC2 or PRC1. Only 220 genes in HCC cells were significantly co-regulated by EZH2, SUZ12, BMI1, or CBX8 [9]. In addition, the PcG family regulates gene expression in both an H3K27me3-dependent and an H3K27me3-independent manner [9]. It appears that both PRC1 and PRC2 play important roles in promoting HCC and that their target gene networks are relatively independent. Multivariate Cox regression analysis also indicated that PRC1 and PRC2 are independent prognostic markers for HCC (Supplementary Tables 5 and 6). Because of their relatively independent pathways, we propose that the combination of PRC1 and PRC2 could act as an even more precise biomarker for predicting HCC outcomes. The detailed mechanism must be dissected to identify cooperative mechanisms between PRC1 and PRC2 in HCC. Conversely, the combination of CBX8 and BMI1 did not display a significant predictive/prognostic value for HCC. Our previous findings indicated that CBX8 and BMI1 share the vast majority of target genes in HCC [9]. Here, we found that the protein expression levels of CBX8 and BMI1 are significantly correlated in HCC specimens (Supplementary Table 7). Therefore, we confirmed that the PRC1 members CBX8 and BMI1 are mutually overlapping diagnostic targets.

We further wondered whether robust expression of EZH2 could be affected by EZH2 mutation. However, the results showed that EZH2 overexpression was not significantly associated with EZH2 mutation (Supplementary Table 8), indicating that abnormal EZH2 expression is not affected by mutation. Combining all these data, altered protein expression and genetic mutation of PcG are two independent events that occur simultaneously in liver carcinogenesis. Therefore, to verify the significance of PcG in cancer progression, not only protein expression but also genetic mutation should be considered.

Disclosure of conflict of interest

None.