The POR rs10954732 polymorphism decreases susceptibility to hepatocellular carcinoma and hepsin as a prognostic biomarker correlated with immune infiltration based on proteomics

The study consisted of a discovery cohort and some validation cohorts. A schematic diagram of our workflow is presented in Fig. 1. The discovery cohort is composed of liver samples from 85 healthy subjects and 100 HCC patients undergoing hepatic surgery (Additional file 1: Table S1). The validation cohorts consisted of experimental research studies and database surveys. Experimental research was conducted with human and mouse liver tissues and paraffin specimens. Database surveys were performed from GTEx, TCGA, TIMER, GEPIA, Kaplan–Meier plotter and TISIDB databases.

Normal liver samples were obtained from patients undergoing hepatic surgery and usually diagnosed with hepatic haemangioma, and only tissue specimens with normal liver function and histological appearance were collected. Peritumor liver samples were obtained from HCC patients without receiving tumor radiotherapy, chemotherapy or targeted drug therapy before surgery; tissue specimens were taken 2 cm away from the tumor tissues. Liver specimens were stored in liquid nitrogen within 30 min after resection. Well-documented demographic information including gender, age and laboratory tests were obtained from an inpatient case record [13]. All liver samples were collected from the First Affiliated Hospital of Zhengzhou University, the Affiliated People's Hospital of Zhengzhou University and the Affiliated Cancer Hospital of Zhengzhou University. This study was carried out in accordance with the Declaration of Helsinki and approved by the Medical Ethics Committee of Zhengzhou University. Written informed consent was provided by all donors.

The genomic DNA purification kit (Beijing ComWin Biotech Co., Ltd., China) was used to isolate genomic DNA from liver tissues. Four polymorphisms for POR with frequencies more than 1% in the Chinese population based on consideration of linkage disequilibrium (LD) analysis were genotyped by polymerase chain reaction (PCR) sequencing according to our previous study [29]. Human liver microsomes (HLMs) were prepared by differential centrifugation and enzyme activity of CYP2E1 in HLMs was determined by measuring the rate of chlorzoxazone 6-hydroxylation by high-performance liquid chromatography with eight chlorzoxazone concentrations (7.8–1000 µM) as reported previously [30]. Chlorzoxazone and 6-OH-chlorzoxazone were purchased from the National Institute for the Food and Drug Control (China) and Toronto Research Chemicals, Inc. (Canada), respectively. The incubation system contained HLMs (0.3 mg/mL protein), 100 mM phosphate buffer (pH 7.4) and serial concentrations of substrate. The reaction was initiated with 1 mM NADPH with an optimal incubation time of 30 min. One mL of ethyl acetate was added to terminate the reactions and the metabolite of 6-OH-chlorzoxazone was determined by HPLC–UV. Examination on linear range, precision, relative recovery, and stability were all determined. The Michaelis–Menten affinity constant (K_m) and maximum reaction velocity (V_max) values was determined according to nonlinear regression analysis by GraphPad Prism 8.03 software. The intrinsic clearance CL_int was estimated by the Michaelis–Menten parameter estimates (CL_int = V_max / K_m).

The proteomic analysis based on normal livers and peritumor tissue from HCC patients was conducted using liquid chromatography-mass spectrometry-tandem mass spectrometry (Q-Exactive HF LC–MS/MS) by the State Key Laboratory of Proteomics, Beijing Proteome Research Center, (Beijing, China). Raw data from mass spectrometry were processed against the human UniProt protein sequence database (version 20,140,922, 20,193 sequences). Protein expression is shown as intensity-based absolute protein quantification (iBAQ) based on peak intensity. Proteomic data was normalized by log2 transformation. Proteins that were undetectable in more than 50% of samples were excluded, and missing values for each protein were replaced with half of the minimum. Further detailed methods on proteomic processing and analysis are according to a previous report [31].

The R/Bioconductor package limma v.3.24.15 34 [32] was used to identify differentially expressed proteins (DEPs) between 34 normal livers and peritumor tissues of 60 HCC patients as HCC related proteins, or between GG genotype and GA + AA genotypes in peritumor tissues of 60 HCC patients as A allele-related proteins. Proteins with both differences greater than 1.2 and P < 0.05 were considered significant. Potential biological functions of DEPs were identified by Gene ontology (GO) analysis for Enrich GO function in the R Package “clusterProfiler”. GO analysis was according to the threshold of P < 0.05 and q < 0.05.

Male BALB/c mice (18–22 g) were purchased from Beijing Vital River Laboratory Animal Technology Co. Ltd. Mice were first anesthetized with 300 mg/kg chloral hydrate by intraperitoneal injection. Each mouse was implanted in the left liver lobe with 10 μL of an H22 cell suspension (1.5 × 10⁶ cells/mL;10 μL for each mouse) or 10 μL of serum-free RPMI-1640 medium (sham operation group) by a subcapsular intrahepatic injection. The mice were kept in a Specific Pathogen Free (SPF) warm incubator until they had recovered from the anesthesia and then returned to the animal room.

HPN expression in the livers of 6 normal subjects and 6 peritumor tissues of HCC patients was determined by western blotting (WB) with primary anti-HPN antibody (ab189246, optimal dilutions at 1:1500). Expression of CD68, CD163 and IL-10 in liver samples for the HCC mouse model was determined by immunohistochemistry (IHC). Validation of HPN expression in various cancers including HCC was obtained by exploring the GTEx, TCGA and TIMER databases.

The correlation of HPN mRNA level with clinical prognosis was evaluated by analyzing the TCGA database, the Kaplan–Meier plotter database and the GEPIA database. Subsequently, the correlation of HPN with the tumor immune cell infiltration level was assessed by exploring the TIMER platform and TISIDB website. The prognostic value of HPN expression in predicting prognosis based on immune cells was determined by analyzing the Kaplan–Meier plotter database.

Statistical analyses were carried out using SPSS version 24.0 software (SPSS Inc. Chicago, IL, USA). Data are expressed as mean ± SE. The Wilcoxon Mann–Whitney test was used for pairwise comparison between two independent groups. Receiver operating characteristic (ROC) curves were constructed to evaluate the models' predictive values in terms of sensitivity, specificity, and respective areas under the curves (AUCs). P values are 2-sided and a P value of less than 0.05 is defined as statistically significant.

A Hardy–Weinberg equilibrium test indicated that the genotype distribution of the four SNPS (rs10954732, rs2286822, rs1135612 and rs1057868) for POR included in the study were in Hardy–Weinberg equilibrium (Additional file 1: Table S2). The rs10954732 (G > A) polymorphism was associated with significantly lowered susceptibility to HCC, demonstrated by a 68.8% decreased risk of HCC in individuals with GA + AA genotype (A allele carriers) ((Odds Ratio (OR) = 0.312, 95% confidence interval (CI), 0.116–0.838, P = 0.021)) in comparison with GG wild-type (Fig. 2a), suggesting that the rs10954732 A allele carrier is a protective factor for HCC. For the remaining polymorphisms (rs2286822, rs1135612 and rs1057868) no significant relationship was detected between these three polymorphisms and HCC risks (Additional file 1: Table S2). The protective factor of the rs10954732 A allele for HCC was further substantiated by examining the prognosis of HCC patients. As depicted in the Kaplan–Meier survival curve, patients with the A allele were more likely to have longer overall survival (OS) compared with the GG wild-type (median OS, 437 vs 220 days, P = 0.0150, Fig. 2b). Taken together, these results suggest an important role for the rs10954732 GA + AA genotype in decreased HCC susceptibility and longer OS.

Moreover, serum biochemical measurements revealed higher levels of aspartate transaminase (AST) and alanine aminotransferase (ALT) in patients than in the normal group, while lower levels in patients with the A allele when compared with individuals carrying the GG genotype (Fig. 2c), results which further corroborated the protective role of the A allele. To explore the underlying mechanism, the effect of the G > A mutation on activity of cytochrome P450 2E1 (CYP2E1), an enzyme involved in hepatocarcinogenesis, was evaluated. In line with AST and ALT, significantly increased CYP2E1 activity was found in the HCC group (Fig. 2d), whereas decreased CYP2E1 activity was found in A allele carriers (Fig. 2e). These results suggest that the decreased HCC susceptibility associated with the rs10954732 A allele might be related to the lowered metabolic activity of CYP2E1.

To explore the mechanism underlying the reduced HCC susceptibility associated with the rs10954732 polymorphism, proteomic analysis of liver tissues from 16 GG genotypes and 44 GA + AA genotypes derived from the 60 HCC patients was performed to identify rs10954732-related DEPs. There were 222 DEPs identified with the heat-map-based screening criteria of |log2-fold change (FC)|≥ 0.20 and P value < 0.05 (Fig. 3a). Subsequently, 117 statistically significant DEPs including 86 up-regulated and 31 down-regulated proteins in the GA + AA genotypes were selected after adjusting for P < 0.05, as shown in the volcano plot (Fig. 3b).

Further functional analysis of the DEPs by gene ontology (GO) enrichment analysis shows that dysregulated proteins were significantly enriched in immune, inflammation and metabolism, as demonstrated by neutrophil and T cell activation, immune and inflammatory response, energy derivation, ATP, oxidoreduction coenzyme and glycogen metabolic processes, regulation of endocytosis, and electron transport chain (Fig. 3c). These results suggest an altered tumor-extrinsic microenvironment in GA + AA genotypes by regulation of the immune response, inflammatory response and altered metabolic reprogramming, which in part provides an underlying mechanism by which the A allele confers a lower risk of HCC.

In addition to A allele-related proteins, a further proteomic analysis with 60 HCC patients and 34 normal subjects identified 1342 DEPs, which are considered as HCC-related proteins. The intersection between the 222 A allele-related DEPs and the 1342 HCC-related DEPs identified 46 proteins (Fig. 3d). Changing trends of expression for these 46 A allele-related DEPs in HCC compared with normal group and in the rs10954732 GA + AA genotype compared with GG genotype was next investigated. Surprisingly, an opposite trend was observed in HCC patients when grouped by rs10954732 mutation. Specifically, we found that 27/36 (75.0%) proteins up-regulated in the A allele carriers showed significant down-regulation, and 7/10 (70.0%) proteins down-regulated in individuals carrying A allele were up-regulated in HCC patients (Fig. 3e). The opposite direction for protein changes in the HCC group and A allele carriers once again confirmed the protective role of the A allele mutation.

We speculate that the above 34 proteins with opposite expression trends in HCC patients when grouped by the rs10954732 mutation might be involved in lowered susceptibility to HCC. Therefore, we focused the 34 A allele-related DEPs with regard to already-known cancer-related proteins based on a thorough literature review, and receiver operating characteristic (ROC) curves was constructed. Nine out of 34 proteins showed the most notable changes in expression in HCC patients with a high AUC (> 0.70) (Fig. 3f), indicating high power to distinguish HCC patients from healthy subjects. The significant top DEPs included 8 down-regulated proteins (HPN, CLEC4G, ABCB10, CTSZ, PACSIN3, ENPP1, C14orf1 and DHRS3) and one up-regulated protein (MYLK). Among them, hepsin (HPN) showed the largest AUC at 0.802.

Hepsin was the only DEP with an AUC > 0.8 (Fig. 4a). In contrast to the significant down-regulation in the HCC group, expression of HPN was higher in the A allele carriers compared with wild-type (P = 0.014). Moreover, there was no significant difference in HPN expression when grouped by G > A mutation in normal subjects (Fig. 4b). That’s to say, lowered HPN expression was detected in HCC patients, while A allele carriers were more likely to have higher expression of HPN, findings further validating that the decreased HCC susceptibility underlying the rs10954732 A allele might be related to increased HPN expression.

To further elucidate the prognostic value of HPN, a Kaplan–Meier survival analysis was performed with our discovery cohort. As shown in Fig. 4c, a lower level of HPN was associated with worse prognosis in HCC patients, evidenced by a prominently decreased (61.6%) OS in the HPN low-expressing group (290 vs 756 days, P = 0.0057). However, this association was only in patients genotyped with the A allele, and no difference in OS was found between HPN high- and low-expressing patients in the GG wild-type group (Fig. 4d). Similarly, patients with higher HPN expression were more likely to have lower levels of AST, ALT,γ-glutamyl transpeptidase (GGT) and globulin (GLO) (Fig. 4e), indicating that lower HPN expression is an indicator for poorer prognosis of HCC.

Altogether, results from the discovery cohort strongly indicate that lower expression of HPN in A allele carriers is a risk factor for HCC occurrence and is correlated with poorer prognosis.

Lowered expression of HPN in HCC peritumoral tissues as compared with normal tissues was first confirmed by the validation set obtained from the GEx and TCGA databases (Fig. 4f), with a AUC at 0.736. Next, the analysis of RNA-seq data from the TIMER database also showed deregulated expression in tumor tissues from various cancers, including significantly lower expression in HCC (Fig. 4g). The lower expression of HPN in peritumor tissues and tumor tissues of HCC patients indicate a protective role for HPN in HCC occurrence.

The prognostic value of lower HPN expression was substantiated by a remarkably shortened OS of HCC patients (by 34.5%) from the TCGA database (1397 vs 2131 days, P = 0.0056) (Fig. 5a) and poor prognosis from the GEPIA database (HR = 0.68, P = 0.03) (Fig. 5b). Furthermore, this finding was again corroborated by analysis of the Kaplan–Meier plotter database. Low HPN expression was associated with worse prognosis in HCC (OS: HR = 0.54, 95% CI = 0.38 to 0.76, P = 0.00044; PFS: HR = 0.68, 95% CI = 0.51 to 0.91, P = 0.0094; Fig. 5c, d).

Subsequently, association between HPN expression and prognosis of HCC patients with different clinical characteristics was investigated. Notably, the expression of HPN was not always associated with prognosis of HCC. Specifically, a low HPN level correlated well with both worse OS (Fig. 5e) and PFS (Fig. 5f) in males (OS: HR = 0.47, P = 0.001; PFS: HR = 0.66, P = 0.022), Asians (OS: HR = 0.33, P < 0.001; PFS: HR = 0.51, P = 0.002) and patients without alcohol use (OS: HR = 0.53, P = 0.007; PFS: HR = 0.621, P = 0.021), without hepatitis viral infection (OS: HR = 0.59, P = 0.021; PFS: HR = 0.54, P = 0.005). In other cases, low expression of HPN was weakly or uncorrelated with poor prognosis, suggesting that the prognostic value of HPN expression was influenced by the clinical characteristics of HCC patients.

Since the important role of HPN in the immune system and the quantity and activity status of tumor-infiltrating lymphocytes can influence the prognosis of HCC patients, the correlation of HPN expression with markers of different subsets of immune cells in our data set was next analyzed. Our results revealed that HPN abundance correlated positively with immune cell gene marker levels, such as CD14, CD36, CD40, CD46, CD276 and VSIG4 and correlated negatively with the markers CD59, HLA-DRA, HLA-DAP1, NRP1 and STAT1 (Fig. 6a). These significant correlations between HPN expression and marker genes from immune cells like tumor-infiltrating monocytes, M2 macrophages, dendritic cells and T-helper cells suggests that HPN in A allele-related HCC might be involved in the regulation of tumor immune infiltration in HCC.

Consequently, this assumption was further tested by investigating the correlation between HPN expression and tumor-infiltrating immune cell subtypes in HCC patients in the TIMER database. We determined that the HPN level in tumor tissues was negatively associated with the infiltration levels of B cells (r = − 0.222, P = 3.27e−05), CD8 + T cells (r = − 0.26, P = 1.05e-06), macrophages (r = − 0.257, P = 1.53e−06), neutrophils (r = − 0.221, P = 3.39e-05) and dendritic cells (r = − 0.29, P = 5.01e-08) (Fig. 6b). No significant association was observed with CD4 + T cells. These results strongly suggest that HPN can modulate the infiltration of immune cells into tumor tissues in HCC.

To further support this inference, correlation between HPN expression in HCC and the levels of immune marker gene expression, which can represent the status of tumor-infiltrating immune cells, was assessed by analysis of the TIMER and GEPIA databases. The results indicate that HPN expression was in strongly negative correlation with 35 marker genes of infiltrating immune cells regardless of whether the correlation analysis was adjusted for purity or not (Additional file 1: Table S3), such as tumor-associated macrophages (TAM), M2 macrophages, monocytes, and neutrophils. Moreover, to further explore the regulation of immune molecules by HPN, correlation between HPN expression and 28 tumor-infiltrating lymphocytes (TILs) was analyzed across human heterogeneous cancers in the TISIDB database. The results show that HPN is associated with 13 TILs in HCC (Fig. 6c). Especially, monocytes (r = 0.486, P < 2.2e−16) and CD56dim natural killer cells (CD56dim) (r = 0.296, P = 6.42e−09) displayed the greatest positive correlations, while activated CD4 T cells (Act-CD4) (r = − 0.42, P < 2.2e−16) and Type 2T helper cells (Th2) (r = − 0.38, P = 3.54e−14) displayed the greatest negative correlations. These findings provide a potential mechanism underlying the oncogenic mechanisms of HPN, which might be by regulating the tumor microenvironment through tumor immunity and recruiting and regulating immune cells in HCC.

Kaplan–Meier plot analyses of HPN expression in HCC in B cells, CD4 + memory T cells, CD8 + T cells, macrophages, NK T cells, Treg T cells, Th1 cells, Th2 cells was next performed to assess the predictive potential of HPN in HCC based on immune cells. We found that HCC patients with lower HPN levels in enriched B cells (P = 2.1e−006), CD4 + memory T cells (P = 8.8e−03), CD8 + T cells (P = 8e−03), natural killer T cells (p = 5.8e−03), Treg T cells (p = 3.5e−03), type 1 T helper cells (P = 4.7e−03), type 2 T-helper cells (P = 3.5e−03) cohort were more likely to have a worse prognosis (Fig. 7a–h). A similar trend was found in macrophages, though it was not statistically significant. The above analysis suggests that low expression of HPN affects the prognosis of HCC patients through immune infiltration.

Finally, to validate the proteomic results and database surveys, expression of HPN was measured by a WB method in another set of human liver samples from HCC patients and in a mouse HCC model. In agreement with proteomic findings, HPN levels were dramatically lower in peritumoral specimens of HCC patients than in normal livers (P < 0.001, Fig. 8a). Consistently, significantly lower HPN expression also was detected in a mouse HCC model (P < 0.01, Fig. 8b–d). The results further verified that lower HPN expression conferred increased HCC occurrence. Furthermore, expression of marker of tumor-infiltrating immune cells like CD68, CD163 and IL-10 significantly increased in a mouse HCC model group by paraffin specimen analysis (P < 0.001, Fig. 8e), which further demonstrates HPN-related tumorigenesis and progression related to tumor immune cell infiltration in HCC.