Prediagnostic Levels of Serum One-Carbon Metabolites and Risk of Hepatocellular Carcinoma Free

Liver cancer in men is the fifth most frequently diagnosed cancer worldwide and the second most frequent cause of cancer-related death (1). In high-risk populations, important risk factors for hepatocellular carcinoma include chronic infection with hepatitis B virus (HBV) and dietary aflatoxin (2). In relatively lower-risk populations, hepatitis C, excessive alcohol intake, cigarette smoking, diabetes, and obesity play a prominent role in hepatocellular carcinoma development (3–5).

The one-carbon metabolism pathway comprises a network of integrated biochemical pathways that donate and regenerate the one-carbon moieties needed for two critically important functions: nucleotide synthesis and methylation (Fig. 1; ref. 6). The one-carbon metabolism pathway is hypothesized to play an important role in hepatocellular carcinogenesis (7). For example, rats fed diets deficient in choline for one year develop hepatocellular carcinoma without the administration of carcinogens (8). In addition, global hypomethylation in hepatic tumor DNA (9) and methylation patterns that result in alterations of protooncogene and tumor suppressor gene expression are common features of liver tumor cells from rodents fed methyl-deficient diets (10, 11).

As the primary methyl donor, S-adenosylmethionine (SAM), derived from dietary sources of methionine or homocysteine (Fig. 1), has tight control of the main features of hepatic one-carbon metabolism (12). Homocysteine is either remethylated to methionine by betaine-homocysteine methyltransferase (BHMT), or metabolized to cystathionine by vitamin B₆ (pyridoxal-5-phosphate, PLP)-dependent cystathionine β-synthase (CBS; ref. 13). Mice with inactive BHMT are prone to fatty livers and hepatocellular carcinoma (14). Decreased CBS activity, as a result of PLP deficiency, is associated with reduced methylation capacity in vivo (15) and is observed in humans with cirrhosis and hepatocellular carcinoma (16, 17).

Patients with cirrhosis, a precursor of hepatocellular carcinoma, had reduced plasma PLP levels (18). Prospective studies on circulating PLP and hepatocellular carcinoma risk are needed to confirm whether reduced PLP levels lead to the development of hepatocellular carcinoma or are the consequence of impaired liver function associated with cirrhosis. In the present study, we examined the relation between levels of several one-carbon metabolites in serum collected before cancer diagnosis and the risk of developing hepatocellular carcinoma in a cohort of men in Shanghai, China, after more than 20 years of follow-up. The metabolites measured are betaine, choline, cystathionine, total homocysteine, methionine, 5-methyltetrahydrofolate (5-MTHF), PLP, and SAM.

The design of the Shanghai Cohort Study has been described in detail elsewhere (19). Briefly, 18,244 men (about 80% of eligible subjects) between 45 and 64 years of age and no history of cancer at recruitment were enrolled in the study between January 1986 and September 1989. Each participant was interviewed in person using a structured questionnaire to obtain information on demographic characteristics, use of tobacco and alcohol, usual adult diet, and medical history. Nonfasting blood samples were collected from each participant and stored at −70°C before analysis. The Institutional Review Boards at the University of Pittsburgh (Pittsburgh, PA) and the Shanghai Cancer Institute (Shanghai, PR China) have approved this study.

Identification of incident cancer cases and deaths among cohort participants has been accomplished through routine reviews of reports from the population-based Shanghai Cancer Registry and from the Shanghai Municipal Vital Statistics Office, as well as by annual in-person interviews of all surviving cohort members. Cumulatively, only 985 (5.4%) cohort participants had been lost to follow-up by the end of 2008.

As of November 1, 2012, the study had accumulated 449,523 person-years of observation. Two hundred and ninety-seven cohort participants who were free of cancer at recruitment had developed hepatocellular carcinoma. Cases were diagnosed on the basis of histopathologic confirmation (n = 50), elevated serum α-fetoprotein with consistent clinical and radiologic history (n = 74), positive computerized axial tomography scan and/or ultrasonography with consistent clinical history (n = 153), or by death certificate only (n = 20).

Five to 10 control subjects among cohort participants without diagnosis of hepatocellular carcinoma were individually matched to an index case by date of birth (within 2 years), date of blood draw (within 1 month), and neighborhood of residence at recruitment in our previous studies (20–22). For the present study, matched pairs included the first 2 control subjects that were originally matched to each case. In addition, for the present study, an additional 39 control subjects were included who tested positive for hepatitis B surface antigen (HBsAg) in previous studies (20–22) to increase the sample size within this group (i.e., positive HBsAg controls).

All serum samples of a given matched set (containing the samples from the case and the matched controls) of all 297 case–control sets were tested in the same batch for all laboratory measurements. Serum total homocysteine was determined by liquid chromatography–electrospray ionization (ESI)/tandem mass spectrometry (LC–ESI/MS-MS) as previously described (23). Serum betaine, choline, methionine, cystathionine, PLP, and SAM were quantified by the methods described previously (24, 25). Serum 5-MTHF was determined by LC–ESI/MS-MS. The samples were analyzed following injection of 20 μL of extract on a Synergi Hydro 4 μ 150 × 3 mm maintained at 30°C (Phenomenex) and eluted in a gradient with buffer A (100% water with 0.1% formic acid) and buffer B (100% methanol with 0.1% formic acid). The flow rate was 0.5 mL per minute, with a step-wise gradient over a total run time of 10 minutes: 0.0 to 5.5 minutes, 5% B; 5.6 minutes, 50% B (linear gradient 2); 7.5 minutes, 75% B (linear gradient); 7.6 minutes, 5% B (linear gradient). The observed (m/z) values of the fragment ions were 5-MTHF (m/z 460 → 313) and ¹³C₅5-MTHF (m/z 465 → 313). The interassay precision measures of all analytes included in the present study were below 10% coefficient of variation (Supplementary Table S1). All data were collected and processed using Analyst Software Version 1.4.2 (Applied Biosystems). The one-carbon metabolites were measured in the Bottiglieri Laboratory (Baylor University, Waco, TX).

Seropositivity for the HBsAg was determined using a standard radioimmunoassay (AUSRIA; Abbott Laboratories) in the Govindarajan Laboratory. The serologic status of the antibodies to hepatitis C virus (anti-HCV) were determined on the first 76 hepatocellular carcinoma cases and their 410 matched controls using ELISA Version 2.0 (Ortho). Only one case and one control were positive for anti-HCV (26). We did not measure anti-HCV on the remaining cases and controls in the present study given, its negligible role in hepatocellular carcinoma in this Chinese population and thus, to reserve precious serum specimens for future research.

Statistical testing was conducted on logarithmically transformed values of metabolites due to their skewed distributions. The analysis of covariance method was used to examine the difference in the mean concentration levels of serum metabolites between subjects with and those without risk factors for hepatocellular carcinoma among controls only. In these analyses, the base 2 logarithm (log₂) of serum concentrations of a given metabolite was the dependent variable. We present the back transformed values, i.e., the ratios of serum concentrations calculated as 2 raised to the power of the beta estimates (Supplementary Table S2).

We conducted the same statistical analyses on matched and unmatched case–control sets with similar results (Supplementary Table S3). All results presented in this report were based on unmatched analyses with maximized sample size. In the unmatched analyses, we broke originally matched case–control sets and used unconditional logistic regression models including all matching factors to calculate ORs and their corresponding 95% confidence intervals (CI) and P values. Study subjects were grouped into quintile categories based on the distributions of serum micronutrients. To compute the linear trend tests we used ordinal variables for serum metabolites where each category was assigned the median value within each quintile of the corresponding metabolite. To adjust for potential confounding effects of established risk factors for hepatocellular carcinoma, the multivariate logistic regression models also included the following variables: cigarette smoking (ever vs. never smokers), heavy alcohol consumption (≥4 drinks/day vs. nondrinkers or <4 drinks/day), self-reported history of physician-diagnosed liver cirrhosis (yes vs. no), and the presence of HBsAg in serum (yes vs. no). Given the extremely low prevalence of anti-HCV in the study population (e.g., 1.3% in hepatocellular carcinoma cases and 0.2% in control subjects), serologic status of anti-HCV was not determined on all study subjects included in the present study and thus, not included in the multivariate regression models. Statistical computing was conducted using the SAS version 9.2 statistical software package (SAS Institute Inc.). All P values quoted are two-sided and considered statistically significant if less than 0.05.

The mean age (± SD) of cases at diagnosis of hepatocellular carcinoma was 66.3 (±7.3) years, whereas the corresponding age of control subjects at the time of case's diagnosis was 65.9 (±6.9) years. The average time interval between blood draw and cancer diagnosis among cases was 9.8 (±5.8) years (range, 1 month to 22 years). Table 1 shows the distributions of selected baseline characteristics and risk factors for hepatocellular carcinoma in cases and control subjects. The prevalence of ever smokers and positive test for HBsAg were higher among hepatocellular carcinoma cases than control subjects. Heavy alcohol consumption was slightly more prevalent in cases than controls. There were no differences by body mass index or prevalence of self-reported history of physician-diagnosed type II diabetes between hepatocellular carcinoma cases and controls. Ten percent of hepatocellular carcinoma cases and 1% of controls reported a history of physician-diagnosed liver cirrhosis. Approximately, 60% of hepatocellular carcinoma cases and 11% of controls tested positive for HBsAg in serum (Table 1). The prevalence of liver cirrhosis between HBsAg-positive and -negative cases was 12.5% and 7.5%, respectively. The corresponding values among controls were 1.9% and 1.1%, respectively.

Among control subjects, serum levels of cystathionine, 5-MTHF, and PLP were lower in current smokers than never smokers (Supplementary Table S2). Regular alcohol drinkers had elevated serum choline level but reduced serum levels of cystathionine and SAM. Compared with HBsAg-negative control subjects, HBsAg-positive individuals had elevated serum level of choline, but had comparable levels of all other analytes. Serum cystathionine level was approximately 50% higher in controls with a history of cirrhosis than in controls without a history of cirrhosis. Most of the measured analytes were weakly correlated with each other (Supplementary Table S4).

Table 2 shows geometric means of serum one-carbon metabolites among all hepatocellular carcinoma cases and control subjects as well as among subgroups stratified by HBsAg status. Among total and HBsAg-negative subjects, cases had statistically significant higher levels of cystathionine than controls. Among HBsAg-positive subjects, cases had statistically significant lower levels of choline and PLP and higher levels of methionine and SAM than controls. There were no statistically significant differences in serum levels of betaine, homocysteine, and 5-MTHF between cases and controls among total subjects or subgroups stratified by HBsAg status.

Table 3 shows the ORs for hepatocellular carcinoma in relation to quintile levels of serum one-carbon metabolites. Before adjustment for potential confounders, increased serum levels of homocysteine and PLP were associated with reduced risk of hepatocellular carcinoma, whereas increased serum levels of betaine, cystathionine, methionine, and SAM were associated with increased risk of hepatocellular carcinoma (all P_trend < 0.05). After adjustment for potential confounders including cigarette smoking, heavy consumption of alcohol, HBsAg status, and history of cirrhosis, the inverse relation for hepatocellular carcinoma risk with serum choline levels strengthened and became statistically significant, whereas the positive association with betaine and homocysteine weakened considerably and became statistically non-significant. The statistical significance of the inverse association with PLP and the positive associations with cystathionine, methionine, and SAM remained (all P_trend < 0.05).

Given the interrelationship of the measured one-carbon metabolites, we examined whether the metabolite-hepatocellular carcinoma risk associations were independent of each other (Table 3). The additional adjustment considerably attenuated the association for hepatocellular carcinoma risk with levels of cystathionine and SAM (both P_trend ≥ 0.49). Although the trend test for the association between methionine and hepatocellular carcinoma risk was statistically significant (P_trend = 0.03), the ORs for individual quintile levels were not statistically significant. The inverse associations with PLP and choline remained statistically significant after adjustment for all other one-carbon metabolites (Table 3).

We further examined the inverse associations between serum levels of all one-carbon metabolites and hepatocellular carcinoma risk in subgroups defined by HBsAg status and history of cirrhosis. Among HBsAg-negative individuals who had no history of cirrhosis at recruitment, the inverse associations for hepatocellular carcinoma risk with serum choline and PLP remained and the magnitude of the ORs were comparable with those observed in total subjects although the test for trend for choline was statistically borderline significant (P_trend = 0.06; Table 4). On the other hand, a statistically significant positive association between serum methionine levels and hepatocellular carcinoma risk was only present in men with positive HBsAg and/or liver cirrhosis (Table 4). Cigarette smoking or alcohol consumption did not significantly modify the association for hepatocellular carcinoma risk with serum levels of choline, PLP, or methionine (Supplementary Table S5). Similar to the results from all subjects, there were no statistically significant associations between serum levels of betaine, cystathionine, homocysteine, 5-MTHF and SAM, and risk of hepatocellular carcinoma in subgroups stratified by HBsAg status and history of cirrhosis as well as by smoking and alcohol consumption (data not shown).

We conducted sensitivity analyses for choline and PLP in relation to the risk of developing hepatocellular carcinoma after excluding individuals with diagnosis of hepatocellular carcinoma within 5 years after blood draw. The independent associations with choline and PLP remained. Compared with the lowest quintiles, the multivariate-adjusted ORs (95% CIs) of hepatocellular carcinoma for the second, third, fourth, and fifth quintiles of choline were 0.78 (0.44–1.37; P = 0.38), 0.75 (0.43–1.33; P = 0.33), 0.46 (0.23–0.90; P = 0.02), and 0.31 (0.13–0.77; P = 0.01) and for PLP they were 0.60 (0.34–1.04; P = 0.07), 0.64 (0.36–1.12; P = 0.12), 0.73 (0.42–1.27; P = 0.26), and 0.45 (0.24–0.84; P = 0.01; both P_trend < 0.05). Additional analyses among hepatocellular carcinoma cases that occurred more than 10 years after blood collection showed similar results; the corresponding ORs (95% CIs) for choline were 1.09 (0.53–2.24; P = 0.81), 0.98 (0.47–2.03; P = 0.95), 0.78 (0.34–1.79; P = 0.55), and 0.42 (0.14–1.29; P = 0.13) and for PLP they were 0.52 (0.25–1.05; P = 0.07), 0.72 (0.36–1.41; P = 0.34), 0.81 (0.40–1.62; P = 0.55), and 0.44 (0.20–1.00; P = 0.049). We conducted similar sensitivity analyses for methionine among individuals with positive HBsAg and/or history of liver cirrhosis, because these individuals were at increased hepatocellular carcinoma risk with increasing methionine levels. The positive association between serum methionine levels and hepatocellular carcinoma risk was stronger when the analyses were restricted to cases diagnosed within the first 5 years after blood collection than cases diagnosed after 5 years after blood collection. Compared with the lowest quintile of methionine, the ORs (95% CIs) for the second, third, fourth, and fifth quintiles were 3.28 (0.66–16.33; P = 0.15), 7.09 (1.42–35.38; P = 0.02), 5.89 (0.95–36.43; P = 0.06), and 10.83 (1.88–62.49; P = 0.008), respectively, for developing hepatocellular carcinoma within the first 5 years after blood draw (P_trend = 0.008). The corresponding ORs (95% CIs) for the development of hepatocellular carcinoma after 5 or more years after blood draw were 1.51 (0.55–4.15; P = 0.43), 1.04 (0.36–3.00; P = 0.94), 3.27 (1.18–9.05; P = 0.02), and 2.87 (1.01–8.13; P = 0.048; P_trend = 0.02). We repeated all analyses described above after excluding hepatocellular carcinoma cases identified through death certificates only (n = 20). The results did not change materially (data not shown).

Using prospective data from a cohort of Chinese men living in a hepatocellular carcinoma high-risk region, we evaluated the relationship between prediagnostic serum levels of one-carbon metabolites and hepatocellular carcinoma risk. Our main finding was an inverse association with higher serum levels of PLP, the bioactive form of vitamin B₆, and choline on hepatocellular carcinoma risk. High serum choline and PLP were associated with a statistically significant 65% and 56% decrease in hepatocellular carcinoma risk, respectively. These novel data strongly support an important role for these one-carbon metabolites in hepatocarcinogenesis. Vitamin B₆ and choline should be further evaluated for their potential to protect against the development of hepatocellular carcinoma.

Data on PLP in relation to hepatocellular carcinoma risk from prior prospective studies have been lacking. Previous cross-sectional studies have consistently shown lower serum PLP levels in patients with liver cirrhosis, a precursor of hepatocellular carcinoma, than healthy individuals (27). The major concern of these prior studies was that the lower PLP levels observed among cirrhotic patients might have been the consequence of compromised liver function. In the present study, the inverse association between serum PLP and hepatocellular carcinoma risk among individuals without hepatitis B (i.e., HBsAg-negative) and without a history of cirrhosis ruled out the possibility that the association was confounded by compromised liver function. Furthermore, the observed inverse PLP-hepatocellular carcinoma risk association for the development of hepatocellular carcinoma more than 10 years after blood draw further supports the notion that the association is not likely to be due to compromised liver function or other underlying conditions related to hepatocarcinogenesis.

Our finding for an inverse association with higher serum PLP levels on hepatocellular carcinoma development is consistent with recent data from large longitudinal cohort studies that showed statistically significant strong inverse associations between circulating PLP and incidence of lung cancer (28), colorectal cancer (29), and breast cancer (30). The mechanism(s) by which vitamin B₆ protects against development of these cancers and hepatocellular carcinoma in particular, is unknown. PLP-dependent enzymes are involved in over 150 biologic enzyme reactions (31) and thus the role of PLP as a cofactor likely affects numerous cellular mechanisms that may be directly or indirectly related to carcinogenesis (32). For example, PLP is a cofactor for CBS and cystathionine γ-lyase (CGL) in the two initial steps in the synthesis of glutathione (33), a major antioxidant and redox regulator (34). In addition, as a cofactor for serine hydroxymethyltransferase (15), PLP deficiencies may cause aberrant DNA methylation such as global hypomethylation (35) and promoter hypermethylation (36). Experimental studies are warranted to elucidate the biologic mechanism(s) underlying the possible protective role of vitamin B₆ in hepatocarcinogenesis.

The present study, for the first time, showed a statistically significant inverse association between serum choline levels and hepatocellular carcinoma risk in humans. It is well established that rodents fed choline-deficient diets spontaneously developed liver tumors (8). Methyl-deficient diets can lead to rapid fat accumulation in the liver, increased lipid peroxidation, necrotic and apoptotic cell death, increased cell proliferation, depletion of intracellular methyl pool that could lead to uracil misincorporation into DNA, and DNA strand breakage (37). In addition, feeding rats a methyl-deficient diet results in global hypomethylation of liver DNA (9) and promoter hypermethylation of tumor suppressor genes (38). Sixty-eight percent of 57 healthy adults fed a diet low in choline for up to 42 days developed fatty liver and/or liver damage (39). The findings of the present study are consistent with previous animal and human studies.

The present study showed a statistically significant positive association between serum levels of methionine and hepatocellular carcinoma risk among individuals with chronic HBV infection and/or cirrhosis. Methionine is a nutritionally indispensable amino acid with major roles in human metabolism as a substrate for protein synthesis, a methyl donor for SAM synthesis, and as a source of sulfur for the formation of cysteine and glutathione via the trans-sulfuration pathway (40). Rats fed methionine-deficient diets, either with or without choline spontaneously developed liver tumors (41, 42), and methionine metabolism flux is associated with development of hepatocellular carcinoma and liver disease in humans (12). Patients with cirrhosis have reduced SAM synthetase activity, thus leading to the accumulation of methionine (43). Among men with chronic HBV and/or cirrhosis, the observed association between serum methionine and risk of hepatocellular carcinoma within a relatively short time interval after blood draw (e.g., <5 years) was stronger than the association observed with risk of hepatocellular carcinoma within a longer time interval after blood draw (e.g., ≥5 years). This result further supports the hypothesis that the liver with compromised function or underlying conditions related to hepatocellular carcinoma development could result in elevated levels of methionine. Experimental studies in humans are warranted to confirm this hypothesis.

In the present study, we showed a positive association between serum cystathionine or SAM and hepatocellular carcinoma risk. Clinical studies have shown that patients with cirrhosis had elevated serum level of SAM (44). Given the attenuation of these associations with hepatocellular carcinoma risk after adjustment for the other one-carbon metabolites, the role of cystathionine and SAM in the development of hepatocellular carcinoma may depend on other one-carbon metabolites.

Serum, 5-MTHF levels did not differ between hepatocellular carcinoma cases and controls in our study, nor did we observe an association between serum total homocysteine and hepatocellular carcinoma risk. Our results of a null association between serum homocysteine and hepatocellular carcinoma risk is consistent with the findings of previous studies on any cancer (45) and specific cancers including the lung (28), prostate (46), breast (47), and pancreas (48).

During subject recruitment at baseline, blood samples were kept on ice (at ∼4°C) for approximately 3 to 4 hours before serum was separated. To evaluate the potential impact of the blood collection procedures on the levels of the metabolites quantified in the present study, we conducted a study of 10 healthy individuals who each donated multiple vials of a blood sample that were each kept at room temperature and 4°C for 0, 2, 4, 6, 24, and 48 hours before their plasma was separated. PLP levels were remarkably stable, with negligible differences between measurements of plasma that was separated immediately after blood draw and those kept in either room temperature or 4°C for 4 hours. Similar results were observed for betaine, cystathionine, and methionine (all changes were <5%). SAM levels were relatively stable with approximately 12% increase in samples with a 4-hour delay in separation. Less stable metabolites were observed for 5-MTHF with 21% decrease, homocysteine with 25% increase, and choline with 26% increase. The least stable metabolite was S-adenosylhomocysteine (SAH; 53% increase). For this reason, we did not include SAH in our analysis. The weak or null associations for SAM, 5-MTHF, and homocysteine with hepatocellular carcinoma risk may be due in part to these variations that would have occurred to a similar degree in hepatocellular carcinoma cases and control subjects. Similarly, these variations may have also contributed to an observed association between choline levels and hepatocellular carcinoma risk that was an underestimate of the true association.

There are notable strengths and limitations of our study. The strengths of the present study included the prospective study design, allowing for the measurement of one-carbon metabolites in serum specimens collected up to 22 years before the clinical diagnosis of hepatocellular carcinoma. Although we cannot completely rule out the possibility that the disease progress influenced circulating one-carbon metabolite levels, our main findings for PLP and choline remained after restricting our analyses to men without chronic HBV infection or history of cirrhosis. A limitation of our study is that we only had single time-point serum specimens. If any changes in diet or lifestyle during follow-up had influenced circulating one-carbon metabolite levels, these changes would result in an observed exposure-disease risk association that was biased toward the null.

In summary, we have shown that higher prediagnostic serum levels of PLP, the bioactive form of vitamin B₆ and serum choline were associated with statistically significant reduced risk of hepatocellular carcinoma. These results suggest a protective role of these one-carbon metabolites in the development of hepatocellular carcinoma in humans. Future studies are warranted to confirm these novel findings in other study populations and to evaluate their potential chemopreventive effect on hepatocarcinogenesis.

No potential conflicts of interest were disclosed.

Conception and design: J.-M. Yuan

Development of methodology: E. Arning, T. Bottiglieri, J.-M. Yuan

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Arning, T. Bottiglieri, Y.-T. Gao, J.-M. Yuan

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L.M. Butler, R. Wang, J.-M. Yuan

Writing, review, and/or revision of the manuscript: L.M. Butler, E. Arning, T. Bottiglieri, Y.-T. Gao, J.-M. Yuan

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y.-T. Gao, J.-M. Yuan

Study supervision: Y.-T. Gao, J.-M. Yuan

The authors thank Ms. Xue-Li Wang of the Shanghai Cancer Institute for her assistance in data collection and management and the staff of the Shanghai Cancer Registry for their assistance in verifying cancer diagnoses in study participants.

This work was supported by the United States National Cancer Institute at the NIH (grant numbers R01 CA43092 and R01 CA144034).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.