The present study was approved by the Ethics Review Committees of Hospitals involved in this study. This study was a hospital-based two-phase retrospective study (Supplementary Fig. 1), including the first stage for screening positive SNPs in CBX4 gene affecting HCC survival and the second stage for validating clinical significance of positive CBX4 SNPs. In this study, patients were from the Southwest Guangxi Zhuang Autonomous Regin, a high incident area of HCC. For screen stage, the inclusion criterion cases are as follows: (1) cases with pathologically diagnosed HCC; (2) cases featuring cancer in I-II Tumor, Node, and Metastasis (TNM) stage and receiving curative resection treatment; (3) cases with available biopsy samples and clinic-pathological data for further analyses; and (4) cases understanding the objective of the study and providing informed consent. The exclusion criteria consisted of: (1) cases receiving chemotherapy or radiotherapy treatment before liver tumor resection; and (2) cases rejected, dropped out, or lost information. According to our inclusion and exclusion criterion of cases, a total of 598 patients were selected for the first-phase analyses in the Affiliated Hospitals of Youjiang Medical University for Nationalities and Guangxi Medical University during January 2016 and December 2018. Meanwhile, a total of 328 HCC cases receiving non-curative resection as their primary therapy were selected for the second-phase analyses in the same hospitals according to our previously reported inclusion and exclusion criterion of cases [13]. These cases were divided into two groups: transarterial chemoembolization (TACE) treatment group (n = 163) and non-TACE control group (n = 165), according to whether they accepted postoperative adjuvant TACE (pa-TACE) as their initial treatment.

In this study, each patient provided written informed consent and donated 5 mL peripheral venous blood for the analyses of CBX4 genotypes. Surgical removed samples with HCC were obtained from each patient for analyzing the amount of CBX4 protein expression and aflatoxin B1 (AFB1)-DNA adducts. At the same time, the corresponding demographic data and clinicopathological information was also collected using a standard interviewer-administered questionnaire and/or medical records by a Youjiang Cancer Institution staff member. Follow-up was conducted according to our previously described method [14]. In brief, all patients received a series of monitoring for detecting any tumor recurrence. In this study, the last follow-up time was set on May 31, 2021, and final survival status was ascertained using telephone contact and medical records.

AFB1 exposure level was elucidated using the amount of AFB1-DNA adducts (AADA) in tissue samples with tumor as previously described [15]. To analyze, AFB1 exposure level was divided into two groups: high exposure (AADA $>$2.87 $\mu$mol/mol DNA) and low exposure (AADA $\le$2.87 $\mu$mol/mol DNA), according to the average AADA among all patients with HCC. More detailed AFB1-exposure analyses can be found in the Supporting Materials and Methods.

A total of 1376 SNPs were obtained in the CBX4 gene from the ENSEMBL database using the Programming Interface Tools. Based on the criteria of minor allele frequency (MAF) more than 0.001, thirty-three of these SNPs were finally chosen for further analysis in this study (Supplementary Table 1). Genotyping was performed by using SNaPshot method (Applied Biosystems [ABI], Foster City, CA, USA), according to the manufacturers’ instructions. For quality control, controls were included in each run, and repeated genotyping and sequencing of a random 5% subset yielded 100% identical genotypes. More detailed SNP selection and genotyping information analyses can be found in the Supporting Materials and Methods.

The expression levels of CBX4 protein and the elucidation of microvessel density (MVD) were analyzed by immunohistochemistry in tissue slides. In this study, the amount of CBX4 protein-expressing was calculated according to the value of IRS systems [16]. More detailed corresponding information can be found in the Supporting Materials and Methods.

Overall survival (OS) time was calculated from the date of patients receiving curative therapy to the date of death or last follow-up, whereas tumor recurrence-free survival (RFS) time was done from the date of patients accepting curative therapy to the date of tumor recurrence or last follow-up. The association between patient survival, demographic and clinicopathological features, and CBX4 SNPs were estimated using Kaplan–Meier survival model (with the log-rank test). In this model, median OS time (MST), median RFS time (MRT) and corresponding 95% confidence interval (95% CI) was calculated. Mean survival time or ND (not determined) would be shown if MST or MRT could not be computed. Death risk and tumor recurrence risk (also hazard ratio, HR) for interesting variables was evaluated using univariate or multivariate Cox regression models. In multivariable Cox regression models, significant variables were selected using the stepwise forward selection based on a likelihood ratio test (p-value less than 0.05 for entering and more than 0.10 for removal in models). The relationship between CBX4 SNPs and the clinicopathological features of patients (including tumor size, grade, stage, MVD, and the amount of CBX4 protein expression) was analyzed by Spearman’s correlation method. All statistical analyses were finished utilizing the SPSS (Version 18, SPSS Institute, IBM Corp., Chicago, IL, USA).

A total of 598 patients with HCC were included in the screening stage of our study and all cases’ demographic and clinicopathological information was shown in Table 1. To explore whether these factors modified the outcome of HCC patients, we accomplished a series of Kaplan–Meier survival and univariable Cox regression survival analyses. AFB1 exposure level, tumor size, tumor grade, TNM stage, and tumor MVD were found to significantly affect patients’ survival. For example, these patients featuring high AFB1 exposure had a shorten MST and MRT compared with those with low AFB1 exposure (MST, 24.00 vs. 41.00 months; MRT, 28.00 vs. 52.27 months). Furthermore, results from Cox regression analyses also displayed that the patients having high AFB1 exposure featured a higher death risk (HR = 2.46, 95% CI = 1.99–3.04) and tumor recurrence risk (HR = 5.15, 95% CI = 3.56–7.45). However, other factors (including age, gender, race, hepatitis B virus (HBV) and hepatitis C virus (HCV) infection status, drinking, smoking, and liver cirrhosis) were not significantly associated with the survival of patients with HCC (Table 1).

One hundred and sixty-three patients receiving pa-TACE treatment (TACE group) and 165 cases not doing this type of treatment (non-TACE group) were included in the validating stage. The baseline demographic and clinicopathological features were well matched between two groups (Supplementary Table 2). Results from survival analyses displayed that the pa-TACE treatment can significantly prolong the OS and RFS time of patients with HCC (Supplementary Fig. 3A,B, left plots) and decrease death risk and tumor reoccurrence risk of cases (Supplementary Fig. 3A,B, right plots).

A total of 33 SNPs in the CBX4 genes were genotyped in this study. The linkage disequilibrium (LD) of these SNPs was analyzed among different individuals with HCC and a weak LD for some SNPs was observed (Supplementary Fig. 2). Table 2 summarizes the association between the genotypes of 33 selected SNPs in CBX4 gene and the outcome of HCC patients. Among of these SNPs, rs528507291 (SNP15) and rs77447679 (SNP20) significantly modified the OS and RFS of HCC patients. Given that number of cases having CBX4 genotypes with rs528507291 G allele is very small, only rs77447679 was further investigated in this study. As shown in Fig. 1, HCC patients carrying genotypes with rs77447679 A alleles (also rs77447679-CA and -AA), compared to those without A alleles (rs77447679-CC), had a lower MST (35 months, 21 months, and 12 months for CC, CA and AA genotype respectively) (Fig. 1A, left plot) and MRT (42 months for CC, 30 months for CA and 12 months for AA respectively) (Fig. 1B, left plot). Results from univariable Cox regression analyses further exhibited that compared to those without A alleles (HR = 1), these patients having rs77447679 A alleles featured an increasing death risk [HR (95% CI/p), 2.56 (2.05–3.19/6.67 $\times$ 10${}^{-17}$) for rs77447679-CA and 7.09 (5.61–8.97/4.59 $\times$ 10${}^{-60}$) for rs77447679-AA respectively] (Fig. 1A, right plot). This is indicative of the death risk of HCC patients gradually increasing with number of A alleles. A similar trend was also shown in the analysis of RFS [corresponding tumor recurrence risk value: 1.73 (1.27–2.37/0.001) and 7.53 (5.64–10.06/9.98 $\times$ 10${}^{-43}$)] (Fig. 1B, right plot).

Fig. 1.

Association between survival and CBX4 rs77447679 polymorphism in 598 patients with HCC. (A,B) The effects of CBX4 rs77447679 genotypes on OS (A) and RFS (B) of HCC. (C–D) The joint effects of aflatoxin B1 exposure combining with CBX4 rs77447679 genotypes on OS (C) and RFS (D). Cumulative hazard function was plotted by the Kaplan–Meier methodology and the p value was calculated with two-sided log-rank tests. MST and MRT was also indicated. Right plots show hazard ratio (dot) and corresponding 95% confidential interval (vertical line segment). Abbreviations: AC1, patients with low AFB1 exposure and rs77447679-CC; AC2, patients with low AFB1 exposure and rs77447679-CA; AC3, patients with low AFB1 exposure and rs77447679-AA; AC4, patients with high AFB1 exposure and rs77447679-CC; AC5, patients with high AFB1 exposure and rs77447679-CA; AC6, patients with high AFB1 exposure and rs77447679-AA; mo, months; MRT, median tumor recurrence-free survival time; MST, median overall survival time; OS, overall survival; RFS, tumor recurrence-free survival.

To explore whether the effects of CBX4 rs77447679 SNP on survival were modified by the clinicopathological features of HCC patients, we accomplished a succession of stratified analyses on basis of the clinicopathological feature variables (Table 3). In these analyses, risk genotypes (HR $>$1) of CBX4rs77447679 SNP were combined into a variable. Similar MST and MRT were observed in the analyses of some stratified variables, including age, gender, race, smoking and drinking history, HBV and HCV status, and liver cirrhosis. However, other stratified variables such as AFB1 exposure level, tumor size, MVD, and tumor grade and stage were found different MST and MRT (Table 3).

In view of AFB1 exposure as a main environmental cause of HCC among Guangxi population, we questioned whether there was a genotypes-environment interactive role in HCC progression. Therefore, we finished a joint analysis. In this analysis, the combination of AFB1 exposure levels and CBX4 rs77447679 genotypes were into six groups: AC1, AC2, AC3, AC4, AC5 and AC6 (Fig. 1C,D). Results showed that the effects of CBX4 rs77447679 on HCC survival would increase under the conditions of increasing AFB1 exposure level. For example, MST (95% CI) for patients with rs77447679-AA would shorten from 20.00 (9.20–30.80) months to 11.00 (10.15–11.85) months when AFB1 exposure level was from low to high (Fig. 1C, left plot). Corresponding death risk was from 9.94 (3.93–25.14) to 39.14 (23.14–67.03) (Fig. 1C, right plot). Further analyses prove this joint effect between CBX4 rs77447679 genotypes and AFB1 exposure levels was not a kind of multiplicative but additive interaction.

Next, we investigated the association between these different stratified clinicopathological features and CBX4 rs77447679 SNP (Table 4). Results from Spearman correlation analyses displayed that CBX4 rs77447679 SNP was significantly positively linked with tumor size (r = 0.17), tumor grade (r = 0.23), tumor stage (r = 0.23), and tumor MVD (r = 0.42).

Because clinicopathological features such as bigger tumor size, higher MVD, and lower tumor grade are correlated with poor HCC survival, we aimed to pinpoint whether reduced survival among patients with risk genotypes of rs77447679 was an indirect reflection of relationship between the CBX4 rs77447679 SNP and these clinicopathological features or, alternately, whether CBX4 rs77447679 SNP could act as an independent biomarker for HCC survival. To answer it, we finished a multivariable analysis based on Cox regression model (Table 5). Results exhibited that CBX4 rs77447679 SNP was identified as an independent biomarker for OS [HR (95% CI) = 2.06 (1.65–2.59) and 4.64 (3.63–5.93) for rs77447679-CA and -AA, respectively], as well as tumor size, grade and MVD. For RFS, HR (95% CI) adjusted by AFB1 exposure, tumor size, tumor grade and MVD was 1.73 (1.27–2.37) and 7.53 (5.64–10.06) for rs77447679-CA and -AA, respectively (Table 5).

Because our previous studies have shown that increasing CBX4 expression progressed HCC angiogenesis and shorten HCC survival [8, 9], we explored whether CBX4 rs77447679 SNP is associated with CBX4 protein expression by immunohistochemistry. Results from Spearman correlation analyses showed that this SNP was positively related to the amount of CBX4 protein in tissues with HCC (r = 0.57, p = 3.62 $\times$ 10${}^{-52}$) (Fig. 2A).

Fig. 2.

CBX4 SNPs correlating with CBX4 expression. CBX4 protein expression was evaluated using the immunohistochemistry scores of IRS system (to see Material and methods) and expression scores are shown as box plots, with horizontal lines representing the median, the bottom and the top of the boxes representing the 25th and 75th percentiles, respectively, and vertical bars representing the range of data. Rs77447679 SNP is positively associated with CBX4 expression (A), but Rs2289728 SNP not (B).

Because a recent molecular epidemiological study has shown that CBX4 rs2289728 SNP (G to A mutation) may increase HCC risk [11] and results from eQTL database (Blood eQTL) (https://www.genenetwork.nl/bloodeqtlbrowser) display that this SNP is associated with decreased expression of CBX4 in blood cells, we also analyzed the association between CBX4 rs2289728 SNP and CBX4 expression in tissue samples with HCC. Results showed that this SNP was not significantly correlated with CBX4 expression (r = 0.07, p = 0.10) (Fig. 2B).

To further validate the clinic significance of CBX4 rs77447679 SNP, we investigated the effects of this SNP on pa-TACE treatment improving HCC prognosis in the second stage analyses. In this stage, the association between CBX4 rs77447679 SNP and HCC prognosis was first analyzed and results like screening stage were observed (Supplementary Fig. 3C,D). Next, we analyzed possible different effects of pa-TACE on HCC patients with different genotypes of CBX4 rs77447679 SNP (Fig. 3A–F). Among these patients having rs77447679-CA or -AA genotypes, pa-TACE treatment significantly prolonged their OS (Fig. 3C,E, left plots) and RFS time (Fig. 3D,F, left plots). Also, this kind of treatment significantly decreased death risk [HR (95% CI), 0.26 (0.18–0.38) for patients with rs77447679-CA and 0.18 (0.06–0.48) for patients with rs77447679-AA, respectively] and tumor reoccurrence risk [HR (95% CI), 0.33 (0.23–0.47) for patients with rs77447679-CA and 0.18 (0.07–0.49) for patients with rs77447679-AA, respectively] of patients with HCC (Fig. 3C–F, right plots). However, similar effects were not observed among those cases having rs77447679-CC genotype. Altogether, these results imply that different genotypes of CBX4 rs77447679 SNP may differentially change therapeutic role of pa-TACE treatment on HCC.

Fig. 3.

CBX4 rs77447679 polymorphism modifying the therapeutic effects of pa-TACE treatment on the survival of patients with HCC. (A,B) The effects of pa-TACE treatment on OS (A) and RFS (B) of HCC cases featuring rs77447679-CC genotype. (C,D) The effects of pa-TACE treatment on OS (A) and RFS (B) of HCC cases featuring rs77447679-CA genotype. (E,F) The effects of pa-TACE treatment on OS (A) and RFS (B) of HCC cases featuring rs77447679-AA genotype. Cumulative hazard function was plotted by the Kaplan–Meier methodology and the p value was calculated with two-sided log-rank tests. MST and MRT was also indicated. Right plots show hazard ratio (dot) and corresponding 95% confidential interval (vertical line segment). Abbreviations: mo, months; MRT, median tumor recurrence-free survival time; MST, median overall survival time; OS, overall survival; RFS, tumor recurrence-free survival; pa-TACE, post-operative adjuvant transarterial chemoembolization; TACE, transarterial chemoembolization.