Background: The effects of hepatocellular carcinoma on liver metabolism and circulating metabolites have been subjected to continuing investigation. This study compares the levels of selected metabolites in sera of hepatocellular carcinoma cases versus patients with liver cirrhosis and evaluates the influence of gender, race, and alcoholic cirrhosis on the performance of the metabolites as candidate biomarkers for hepatocellular carcinoma.

Methods: Targeted quantitation of 15 metabolites is performed by selected research monitoring in sera from 89 Egyptian subjects (40 hepatocellular carcinoma cases and 49 cirrhotic controls) and 110 U.S. subjects (56 hepatocellular carcinoma cases and 54 cirrhotic controls). Logistic regression models are used to evaluate the ability of these metabolites in distinguishing hepatocellular carcinoma cases from cirrhotic controls. The influences of gender, race, and alcoholic cirrhosis on the performance of the metabolites are analyzed by stratified logistic regression.

Results: Two metabolites are selected on the basis of their significance to both cohorts. Although both metabolites discriminate hepatocellular carcinoma cases from cirrhotic controls in males and Caucasians, they are insignificant in females and African Americans. One metabolite is significant in patients with alcoholic cirrhosis and the other in nonalcoholic cirrhosis.

Conclusions: The study demonstrates the potential of two metabolites as candidate biomarkers for hepatocellular carcinoma by combining them with α-fetoprotein (AFP) and gender. Stratified statistical analyses reveal that gender, race, and alcoholic cirrhosis affect the relative levels of small molecules in serum.

Impact: The findings of this study contribute to a better understanding of the influence of gender, race, and alcoholic cirrhosis in investigating small molecules as biomarkers for hepatocellular carcinoma. Cancer Epidemiol Biomarkers Prev; 23(1); 64–72. ©2013 AACR.

Hepatocellular carcinoma is a significant worldwide health problem with as many as 500,000 new cases diagnosed each year (1). In the United States, the rate of hepatocellular carcinoma has increased by 70% over the last two decades (2) and the increasing incidence of hepatocellular carcinoma has been associated with hepatitis C virus (HCV) infection (3–5). Chronic hepatitis develops in about 80% of those infected with HCV. Over the course of 20 years or more, 10% to 30% of HCV carriers develop cirrhosis; patients with cirrhosis have an annual risk of 1% to 2% for developing hepatocellular carcinoma (6). HCV has been reported as the main risk factor for hepatocellular carcinoma in both Egypt and the United States (7). The prognosis for patients with hepatocellular carcinoma remains extremely poor. Survival rates of patients with hepatocellular carcinoma can be improved significantly if the diagnosis were made at earlier stages, when treatment is more effective (8). Current diagnosis of hepatocellular carcinoma relies on clinical information, liver imaging, and measurement of serum α-fetoprotein (AFP). However, elevated AFP levels may be seen in patients with cirrhosis or chronic hepatitis (9). Thus, the reported sensitivity (41%–65%) and specificity (80%–94%) of AFP are not sufficient for early diagnosis and additional markers are needed (9, 10).

Characterizing metabolites in human samples can facilitate the discovery of biomarkers that may reveal early signs of cancers when the chances for cure are highest. Metabolomics provides a simultaneous assessment of numerous metabolites allowing for the quantification of individual metabolites and the characterization of phenotypic profiles based on a combination of metabolites (11–14). Metabolomic studies are typically carried out through two complementary methods: untargeted and targeted. Although untargeted metabolomics can be used to generate new hypotheses for further tests by analyzing the ion abundances of a large number of unknown metabolites, targeted approaches focus on quantitation of the levels of known metabolites to detect true differences between distinct biologic groups. Liquid chromatography coupled with mass spectrometry (LC/MS) has been applied for both untargeted and targeted quantitation of metabolites. Various LC/MS based untargeted metabolomic studies have led to the identification of significant differences of bile acids, phospholipids, fatty acids, glycolysis, urea cycle, and methionine metabolism in serum, plasma, urine, and fecal samples of patients with hepatocellular carcinoma compared with benign liver tumor or healthy subjects (9, 12–20). We previously reported results from our untargeted metabolomic study, in which we observed downregulation of metabolites of bile acids and long chain carnitines and upregulation of phospholipids and amino acids in sera of hepatocellular carcinoma cases versus cirrhotic controls (15, 16). Targeted quantitation of these relevant metabolites using isotope dilution by selected reaction monitoring (SRM) on triple quadruple (QqQ) or quadrupole linear ion trap (QLIT) mass spectrometers allows more sensitive and accurate quantitation to identify candidate biomarkers. Recent quantitative analysis of lipids in plasma by ultraperformance liquid chromatography (UPLC)–ESI–QqQ MS showed aberrant lipid metabolism of lysophosphocholines (LPC) and lysophosphatidic acid in hepatocellular carcinoma (17). Also, targeted analysis of taurocholic acid (TCA), lysophosphoethanolamine (LPE) 16:0, and LPC 22:5 by LTQ Orbitrap helped evaluate these metabolites as serum biomarkers distinguishing the different stages of hepatocarcinogenesis (18).

Gender differences have been previously observed in the onset of hepatocellular carcinoma; the risk of hepatocellular carcinoma is 2 to 7 times higher in males than in females, although this ratio varies across the world (21, 22). In addition to the speculation that men could have higher rates of environmental exposure to liver carcinogens (such as smoking or alcohol) and hepatitis virus infections, the risk difference could be attributed to estrogen and androgen effects (23, 24). Although few studies have addressed racial/ethnic differences in survival for patients with hepatocellular carcinoma, African Americans have persistently shown poor survival rate. Also, it has been reported that the sensitivity of AFP for the diagnosis of hepatocellular carcinoma in African Americans with HCV infection is less than that of patients of all other racial groups combined (25). Alcohol is one of the major risks of liver injury and has been considered as a promoter for hepatocellular carcinoma (26). It has been found that patients with alcoholic cirrhosis and bearing two copies of the RANTES G-403 variant (2G-403 genotype) had a higher rate of hepatocellular carcinoma occurrence compared with patients carrying at least one RANTES A-403 allele (27). AFP has been reported to be higher in patients with hepatocellular carcinoma arising from chronic viral conditions compared with those with alcoholic liver disease (28).

In this study, we evaluate the levels of metabolites in sera of hepatocellular carcinoma cases and patients with liver cirrhosis. This is accomplished by targeted quantitation of selected metabolites in sera from two cohorts (Egypt and the United States) using isotope dilution on a QLIT MS coupled with an UPLC (UPLC–QLIT–MS). The metabolites are chosen from candidates discovered by our previous untargeted analysis of metabolites in the Egyptian cohort (15). To address the challenge in detecting true differences between hepatocellular carcinoma cases and cirrhotic controls due to the heterogeneity of the patients and their disease, we propose to compare differences among homogeneous subpopulations through stratification of the subjects by gender, race, alcoholic cirrhosis, and tumor stage. This offers the opportunity to identify subgroup-specific biomarkers for accurate detection of hepatocellular carcinoma. We present our observations about the influence of gender, race, and alcoholic cirrhosis on the performance of candidate biomarkers for hepatocellular carcinoma.

### Study cohorts

The samples in this study were obtained from participants recruited in Egypt and the United States. The Egyptian cohort consists of adult patients with hepatocellular carcinoma or cirrhosis recruited from the outpatient clinics and inpatient wards of the Tanta University Hospital (Tanta, Egypt). The participants consist of 89 subjects (40 hepatocellular carcinoma cases and 49 patients with liver cirrhosis). The characteristics of the patient population in this study are summarized in Table 1. All participating patients provided informed consent before taking part in the study. The study protocol was approved by the Tanta University Ethical Committee. Patients were diagnosed to have cirrhosis on the basis of established clinical, laboratory and/or imaging criteria with ultrasound examination. Cases were suspected to have hepatocellular carcinoma if they had focal lesion in ultrasound examination and/or elevated AFP above 200 ng/mL, but only diagnosed to have hepatocellular carcinoma based on well-established diagnostic imaging criteria with spiral CT scanning and/or histopathology examination of liver biopsy. Clinical stages for hepatocellular carcinoma cases were determined on the basis of the tumor–node–metastasis (TNM) staging system. Suspected cases that were not confirmed by CT or biopsy were excluded from both groups of this study.

Table 1.

Characteristics of the Egyptian study cohort

Hepatocellular carcinoma (N = 40)Cirrhosis (N = 49)P
Age Mean (SD) 53.2 (3.9) 53.8 (7.6) 0.3530
Gender Male 77.5% 67.3% 0.3474
HCV serology HCV Ab+ 100.0% 100.0% 1.0000
HBV serology HbsAg+ 0.0% 6.1% 0.2492
MELD Mean (SD) 18.6 (7.7) 18.9 (7.1) 0.1328
MELD ≤ 10 20.0% 12.2% 0.3863
AFP Median (IQR) 275.9 (1,244.3)
Hepatocellular carcinoma stage Stage I 72.5%
Stage II 15.0%
Stage III 5.0%
Unknown 7.5%
Hepatocellular carcinoma (N = 40)Cirrhosis (N = 49)P
Age Mean (SD) 53.2 (3.9) 53.8 (7.6) 0.3530
Gender Male 77.5% 67.3% 0.3474
HCV serology HCV Ab+ 100.0% 100.0% 1.0000
HBV serology HbsAg+ 0.0% 6.1% 0.2492
MELD Mean (SD) 18.6 (7.7) 18.9 (7.1) 0.1328
MELD ≤ 10 20.0% 12.2% 0.3863
AFP Median (IQR) 275.9 (1,244.3)
Hepatocellular carcinoma stage Stage I 72.5%
Stage II 15.0%
Stage III 5.0%
Unknown 7.5%

Abbreviations: MELD, model for end-stage liver disease; IQR, interquartile range.

The U.S. participants in this study comprised of adult patients, recruited from the hepatology clinics at MedStar Georgetown University Hospital (Washington, DC). The cohort consists of 110 participants (56 hepatocellular carcinoma cases and 54 patients with liver cirrhosis). The characteristics of the cohort are summarized in Table 2. All patients provided informed consent and the study was approved by the Institutional Review Board at Georgetown University (Washington, DC). All patients were diagnosed to have liver cirrhosis on the basis of established clinical, laboratory, and/or imaging criteria. Cases were diagnosed to have hepatocellular carcinoma based on well-established diagnostic imaging criteria and/or histology. Clinical stages for hepatocellular carcinoma cases are determined on the basis of the TNM staging system. Controls were required to be hepatocellular carcinoma free for at least 6 months from the time of study entry.

Table 2.

Characteristics of the U.S. study cohort

Hepatocellular carcinoma (N = 56)Cirrhosis (N = 54)P
Age Mean (SD) 59.7 (7.4) 58.2 (8.4) 0.3489
Gender Male 75.0% 63.0% 0.2168
Ethnicity Caucasian 55.4% 59.3%
African American 30.4% 25.9% 0.8926
Others 14.3% 13.0%
HCV serology HCV Ab+ 51.8% 35.2% 0.0392
HCV RNA+ 55.4% 33.3% 0.6356
HBV serology Anti-HBC+ 14.3% 7.4% 0.2893
HbsAg+ 28.6% 20.4% 0.6777
Alcoholic cirrhosis Alcoholic cirrhosis+ 28.6% 33.3% 0.6811
MELD Mean (SD) 11.0 (6.4) 14.0 (6.0) 0.0002
MELD ≤ 10 44.6% 14.8% 0.0008
AFP Median (IQR) 14.5 (103.6) 4.8 (13.6) 0.0002
Hepatocellular carcinoma stage Stage I 53.6%
Stage II 23.2%
Stage III 7.1%
Stage IV 1.8%
Unknown 14.3%
Hepatocellular carcinoma (N = 56)Cirrhosis (N = 54)P
Age Mean (SD) 59.7 (7.4) 58.2 (8.4) 0.3489
Gender Male 75.0% 63.0% 0.2168
Ethnicity Caucasian 55.4% 59.3%
African American 30.4% 25.9% 0.8926
Others 14.3% 13.0%
HCV serology HCV Ab+ 51.8% 35.2% 0.0392
HCV RNA+ 55.4% 33.3% 0.6356
HBV serology Anti-HBC+ 14.3% 7.4% 0.2893
HbsAg+ 28.6% 20.4% 0.6777
Alcoholic cirrhosis Alcoholic cirrhosis+ 28.6% 33.3% 0.6811
MELD Mean (SD) 11.0 (6.4) 14.0 (6.0) 0.0002
MELD ≤ 10 44.6% 14.8% 0.0008
AFP Median (IQR) 14.5 (103.6) 4.8 (13.6) 0.0002
Hepatocellular carcinoma stage Stage I 53.6%
Stage II 23.2%
Stage III 7.1%
Stage IV 1.8%
Unknown 14.3%

### Chemicals and reagents

Vacutainer sterile vacuum tubes were purchased from Becton Dickinson and Company. We purchased from Sigma-Aldrich LC check solutions (mixture of dodecanamide, 11-deoxycoricosterone, cortisone, thyroxine, epitestosterone, and N-benzoyl-D-phenylalanine), ammonium acetate (NH4OAC), ammonium bicarbonate (NH4HCO3), glycholic acid (GCA), taurochenodeoxycholic acid (TCDCA), glycodeoxycholic acid (GDCA), glycodeoxycholic-2,2,4,4-d4 acid (D4_GDCA), glycodeoxycholic-2,2,4,4-d4 acid (D4_GCA), Phe-Phe, 2C13_Phe-Phe, 3β,6β-dihydroxy-5β-cholan-24-oic acid, and oleoyl carnitine. LPC 16:0, LPC 17:0, D35_LPC 18:0, LPE 18:0, D54_PE 14:0 were purchased from Avanti Polar Lipids. Glycochenodeoxycholic acid 3-sulfate (3sulfo-GCDCA) was synthesized by PharmAgra Labs. D5_sulfo-GCDCA was synthesized by Toronto Research Chemicals. Acetonitrile, methanol, isopropanol, and chloroform were from Fisher Scientific. Metabolomics MS System Test Mix (theophylline, caffeine, hippuric acid, 4-nitrobenzoic acid, nortriptyline HCl) was purchased from Waters. All other common chemicals of analytic grade were purchased from Sigma-Aldrich.

### Blood collection

Human blood was drawn from the participants in the Egyptian and U.S. cohorts and collected by venipuncture into 10 mL BD Vacutainer sterile vacuum tubes without the presence of anticoagulant. The blood was immediately centrifuged at 1,000 × g for 10 minutes at room temperature. The serum supernatant was carefully collected and centrifuged at 2,500 × g for 10 minutes at room temperature. After aliquoting, serum was kept frozen at −80°C until use. Primary tubes and serum aliquots were labeled using anonymous confidential code numbers with no personal identifiers. Identification codes were cross-referenced with clinical information in a passcode protected computer system.

### Metabolite extraction

Metabolites were extracted using a 96-well filter plate that allows high throughput serum sample preparation. Briefly, serum samples from 96 hepatocellular carcinoma cases and 103 cirrhotic controls representing the two cohorts (Egypt and the United States) were thawed and vortexed in separate storage boxes. Six samples randomly selected from the hepatocellular carcinoma groups and 6 from the cirrhosis group (50 μL each) were simultaneously pipetted into odd numbered wells and even numbered wells, respectively, in each row of the 96-well filter plate through a multiple channel pipettor. This was followed by in-well protein precipitation by adding 200 μL chilled methanol containing 1% NH4HCO3 and a mixture of isotope-labeled internal standards. After in-well agitation on an oscillating table, vacuum was applied to filtrate and remove the precipitated proteins. The methanol filtrate of approximate 200 μL was then ready for LC/MS analysis.

### Internal standards and QC runs for quality assessment

To evaluate the quality of metabolite extraction and the impact of matrix effects, a mixture of internal standards (0.1 pmoL/μL), including 2C13_Phe–phe, D4_GCA, D4_GDCA, D5_3sulfo-GCDCA, D35_LPC 18:0, and D54_PE 14:0, was spiked into serum. In addition, we ran six serum quality control (QC) samples along with the hepatocellular carcinoma cases and cirrhotic controls to evaluate the reproducibility of the analytical methods. The QC samples were pooled from both the hepatocellular carcinoma cases and cirrhotic controls.

### Target metabolites

Target metabolites were chosen from a list of candidate metabolite biomarkers discovered by our previous hepatocellular carcinoma biomarker discovery study using the serum samples from Egyptian cohort (15). The list includes bile acids (GDCA, GCA, TCDCA, 3sulfo-GCDCA, 3β,6β-dihydroxy-5β-cholan-24-oic acid), phospholipids (LPC 17:0, LPC 16:0, and LPC 18:0 and LPE 18:0, LPE 16:0, LPE 20:4, LPE 22:6, and Phe–Phe), long chain carnitines of oleoyl carnitine and linoleaidyl carnitine. SRM-based quantitation of these 15 metabolites was done by UPLC–QqQLT in sera from hepatocellular carcinoma cases and cirrhotic controls in the Egyptian and the U.S. cohorts as outlined in the next section.

### Quantitation of target metabolites by UPLC–QqQLIT

From each sample, 10 μL was injected onto a reverse-phase 50 × 2.1 mm ACQUITY 1.7 μm CSH C18 column using a Waters ACQUITY UPLC system with a gradient mobile phase consisting of 40% acetonitrile solution containing 10 mmol/L NH4OAC and 0.1% formic acid (FA) (A) and 10% acetonitrile in isopropanol containing 0.1% FA and 10 mmol/L NH4OAC (B). The whole run lasted 11 minutes. The gradient consisted of 100% A for 1.0 minutes at a flow rate of 0.25 mL/min then a ramp to 50% B from 1.0 to 5.0 minutes at a flow rate of 0.25 mL/min, then ramp to flow rate of 0.4 mL/min and 100% B from 5.1 to 9.1 minutes, return to 0.25 mL/min flow rate and 100% A from at 9.2 minutes, followed by a hold at 100% A until 11 minutes. The column temperature was set to 50°C. UPLC–QqQLIT data were collected in the SRM mode on the AB Sciex QTRAP 4000 MS operating in the positive mode with a capillary voltage of 4,500 V and source temperature of 450°C. The curtain gas flow rate was set to 20, sheath gas flow rate GS1 was set to 40.00, sheath gas GS2 was set to 20.00 and exit potential was 10 V. Each target metabolite was monitored by two SRM transitions: one transition works for identification and the other transition for quantitation shown in Table 3. Quantitation of each metabolite was performed by calculating the ratio of the MS signal of the metabolite to its corresponding isotope-labeled internal standard. To obtain accurate and reproducible SRM quantitation data, dwell time and scheduled SRM detection were customized for each target metabolite, allowing us to acquire enough data points (>15 points) across the chromatographic elution profile.

Table 3.

Experimental parameters for SRM quantitation of target metabolites

Compound[M+H] + m/zSRMInternal standardCE (V)RT (min)DPCXP
GDCA 450.5 450.5 → 90.9 D4-GCDCA 109 5.0 91
450.5 → 414.3  23  91 26
GCA 466.4 466.4 → 412.3 D4-GCA 27 3.15 101 26
466.4 → 337.2  33  101 20
TCDCA 500.4 500.4 → 464.2 D5-TCA 23 3.20 106
500.4 → 126.1  55  106
3Sulfo-GCDCA 547.3 547.3 → 414.2 D5-3sulfo GCDCA 31 2.30 76 24
[M+NH4+]+ 547.3 → 432.3  19  76 26
3β,6β-Dihydroxy-5β-cholan-24-oic acid 393.3 393.3 → 357.3 D4-GCDCA 13 5.5 66
393.3 → 81.1  63  66
LPC 17:0 510.5 510.5 → 184.2 D35-LPC18 37 6.90 101 10
510.5 → 86.2  69  101
LPC 16:0 496.5 496.5 → 184.2 D35-LPC18 37 6.63 101 10
496.5 → 86.2  69  101
LPC 18:0 524.4 524.4 → 184.2 D35-LPC18 37 7.04 101 10
524.4 → 86.2  69  101
LPE 18:0 482.3 482.3 → 341.3 D54 PE14 29 7.44 86 20
482.3 → 62.2  35  86 10
LPE 16:0 454.28 454.3 → 313.3 D54 PE14 29 6.67 86 20
454.3 → 62.2  35  86 10
LPE 20:4 502.3 502.3 → 361.3 D54 PE14 29 6.36 86 20
502.3 → 62.2  35  86 10
LPE 22:6 526.3 526.3 → 385.3 D54 PE14 29 6.30 86 20
526.3 → 62.2  35  86 10
Phe–Phe 313.0 313.0 → 120.1 2C13Phe–phe 53 1.32 120 10
313.0 → 103.1  53  120 10
Oleoyl carnitine 426.3 426.3 → 85.0 D3-Palmitoyl carnitine 55 6.60 96
426.3 → 62.4  63  96
Linoleaidyl carnitine 424.3 424.3 → 85.0 D3-Palmitoyl carnitine 55 6.30 96
424.3 → 62.4  63  96
Compound[M+H] + m/zSRMInternal standardCE (V)RT (min)DPCXP
GDCA 450.5 450.5 → 90.9 D4-GCDCA 109 5.0 91
450.5 → 414.3  23  91 26
GCA 466.4 466.4 → 412.3 D4-GCA 27 3.15 101 26
466.4 → 337.2  33  101 20
TCDCA 500.4 500.4 → 464.2 D5-TCA 23 3.20 106
500.4 → 126.1  55  106
3Sulfo-GCDCA 547.3 547.3 → 414.2 D5-3sulfo GCDCA 31 2.30 76 24
[M+NH4+]+ 547.3 → 432.3  19  76 26
3β,6β-Dihydroxy-5β-cholan-24-oic acid 393.3 393.3 → 357.3 D4-GCDCA 13 5.5 66
393.3 → 81.1  63  66
LPC 17:0 510.5 510.5 → 184.2 D35-LPC18 37 6.90 101 10
510.5 → 86.2  69  101
LPC 16:0 496.5 496.5 → 184.2 D35-LPC18 37 6.63 101 10
496.5 → 86.2  69  101
LPC 18:0 524.4 524.4 → 184.2 D35-LPC18 37 7.04 101 10
524.4 → 86.2  69  101
LPE 18:0 482.3 482.3 → 341.3 D54 PE14 29 7.44 86 20
482.3 → 62.2  35  86 10
LPE 16:0 454.28 454.3 → 313.3 D54 PE14 29 6.67 86 20
454.3 → 62.2  35  86 10
LPE 20:4 502.3 502.3 → 361.3 D54 PE14 29 6.36 86 20
502.3 → 62.2  35  86 10
LPE 22:6 526.3 526.3 → 385.3 D54 PE14 29 6.30 86 20
526.3 → 62.2  35  86 10
Phe–Phe 313.0 313.0 → 120.1 2C13Phe–phe 53 1.32 120 10
313.0 → 103.1  53  120 10
Oleoyl carnitine 426.3 426.3 → 85.0 D3-Palmitoyl carnitine 55 6.60 96
426.3 → 62.4  63  96
Linoleaidyl carnitine 424.3 424.3 → 85.0 D3-Palmitoyl carnitine 55 6.30 96
424.3 → 62.4  63  96

Abbreviations: CE, collision energy; DP, de-clustering potential; CXP, collision cell exit potential.

### Quality assessment of the quantitation data

The quantitation performance was evaluated on the basis of the linearity of the standard curves [e.g., coefficient of determination (R2)], signal reproducibility (e.g., SD; coefficient of variation, CV), dynamic range of detection, analyte recovery (percentage of extraction efficiency), accuracy, and ESI matrix effects. Each standard curve was created by at least five points of standard solutions and each point had triplicate injections at the beginning, in the middle and at the end of run. The majority of the target metabolites achieved more than 80% recovery except Phe–Phe (67%). This could be due to the hydrophilic property of Phe–Phe that causes the sample loss during extraction and the poor retention on the C18 column. The coeluting interferences from serum sample cause ion enhancement on Phe–Phe and GCA and ion suppression on the rest of target metabolites during the ionization process, of which LPE 18:0 has the worst ion suppression effect of 31%. However, the presence of spiking isotope-labeled internal standards enables us to correct the above detection variations with remarkable linearity (R2 > 0.999) and reproducibility (CV < 10% and accuracy > 85%). The six serum QC samples that were analyzed along with the other serum samples yielded a relative SD < 15%.

### Statistical analysis

A univariate logistic regression model (Eq. 1) was used to evaluate the log-transformed quantitation data for the 15 metabolites in terms of distinguishing hepatocellular carcinoma cases from cirrhotic controls.

where Y is the binary response variable with |$Y\, = \,1$| indicating that the sample is from the hepatocellular carcinoma group, X is the metabolite expression level, the exponential of β is the OR of a subject coming from the hepatocellular carcinoma group by increasing one unit of the metabolite expression level. Our interest is in testing the null hypothesis |${\rm H}_0:\beta \, = \,0$| against the alternative |${\rm H}_1:\beta \, = \,\beta ^*$|⁠, where |$\beta ^* \, \ne \,0$|⁠, that the metabolite is related to the binary response variable. In addition, a proportional odds model (Eq. 2) was applied to evaluate the metabolite effects on two subgroups based on hepatocellular carcinoma stage: early (stage I) and late (stages II, III, and IV).

where Y is the multinomial distributed response variable with |$Y\, = \,0$| indicating that the sample is from the cirrhotic control group, |$Y\, = \,1$| denotes that the sample is from the early-stage hepatocellular carcinoma group, and |$Y\, = \,2$| indicates that the sample is from the late-stage hepatocellular carcinoma group. Finally, the logistic regression model in Eq. 1 was applied separately to subcohorts stratified by gender, race, or alcoholic cirrhosis to investigate the influence of these factors. This analysis helped us to discover gender, race, or alcoholic cirrhosis-specific biomarkers. For the U.S. cohort, the AFP values were available for both hepatocellular carcinoma cases and cirrhotic controls. So, we considered three logistic regression models with the following predictors: (i) gender and AFP; (ii) gender and selected metabolites; and (iii) gender, selected metabolites, and AFP. Area under the receiver operating characteristic (AuROC) curve was used to evaluate the performance of each model. Model cross-validation was performed 500 times by randomly choosing 35% of the samples from each group. The remaining 65% are used for model training.

Supplementary Table S1 presents the results of the statistical analysis based on the log-transformed metabolite levels measured by SRM in 40 hepatocellular carcinoma cases and 49 cirrhotic controls in the Egyptian cohort. The hepatocellular carcinoma cases are stratified by stage into two groups. The first group consist of 29 stage I cases. The second group has 8 cases of stages II and III combined. For each metabolite, the statistical significance levels (P values) are presented on the basis of the univariate logistic regression model in distinguishing hepatocellular carcinoma from cirrhosis and the univariate proportional odds model for the subgroups stratified by tumor stage. We observe that the metabolites with significant and consistent discrimination between hepatocellular carcinoma cases and cirrhotic controls are primarily involved in bile acid biosynthesis, long-chain carnitine, and small peptide metabolism. Deregulation of bile acids (GCDCA, GDCA, GCA, and TCDCA etc.) in serum has been reported in association with liver diseases, for example, liver cirrhosis and hepatitis (11, 19, 20). Previous hepatocellular carcinoma studies also indicated the downregulation of those bile acids metabolites in serum compared with liver cirrhosis (15, 16, 20, 29). Meanwhile the abnormal expression of conjugated bile acids and sulfation product was investigated in injured liver during hepatocellular carcinoma carcinogenesis that could lead to deteriorating detoxification of endogenous and exogenous lipophilic compounds (20, 30, 31). As shown in Supplementary Table S1, GCA, GDCA, and TCDCA significantly discriminate stage II and III hepatocellular carcinoma cases from cirrhotic controls (P < 0.05). Downregulation of these three metabolites was observed in the progression of the disease from cirrhosis to early-stage and to late-stage, indicating their metabolic deregulation during the process of hepatocellular carcinoma tumorigenesis. Although GCA, TCDCA, and 3sulfo-GCDCA do not significantly distinguish early-stage hepatocellular carcinoma cases from cirrhotic controls, we observe consistent pattern and statistically significant discrimination of stage I cases and stage II and III cases from cirrhosis for GDCA.

Supplementary Table S2 presents statistical analysis results for the U.S. cohort with 56 hepatocellular carcinoma cases and 54 cirrhotic controls. Two metabolites (3sulfo-GCDCA and 3β,6β-dihydroxy-5β-cholan-24-oic acid) significantly distinguish hepatocellular carcinoma cases from cirrhotic controls. The early-stage group consists of 30 stage I hepatocellular carcinoma cases. The late-stage group has 18 hepatocellular carcinoma cases of stages II, III, and IV combined. Consistent downregulation patterns and statistically significant discrimination of early-stage hepatocellular carcinoma from cirrhosis are observed for these two metabolites. LPC 17:0 can significantly discriminate early-stage from cirrhotic controls with an upregulation pattern. These three metabolites may reveal early signs of hepatocellular carcinoma in patients with liver cirrhosis. Of note, 3β,6β-dihydroxy-5β-cholan-24-oic acid can distinguish late-stage hepatocellular carcinoma from cirrhotic controls as well.

We conducted stratified analyses to evaluate the impact of gender on the ability of the metabolites in distinguishing hepatocellular carcinoma cases and cirrhotic controls in both the Egyptian and the U.S. cohorts, as well as the impact of race and alcoholic cirrhosis in the U.S. cohort because it has been known that such cohort compositions can affect the relative concentration of small-molecule metabolites in the serum. Supplementary Table S1 shows that all except GCA, 3sulfo-GCDCA, and Phe–Phe are significant in the male Egyptian group (31 hepatocellular carcinoma cases and 33 cirrhotic controls), whereas only Phe–Phe shows significance in the female Egyptian group (9 hepatocellular carcinoma cases and 16 cirrhotic controls). Supplementary Table S2 indicates that for the U.S. cohort, the two significant metabolite candidates are significant in the male group (42 hepatocellular carcinoma cases and 34 cirrhotic controls) but insignificant in the female group (14 hepatocellular carcinoma cases and 20 cirrhotic controls). Oleoyl carnitine is significant in the female group but not in the male group. The U.S. cohort includes 63 Caucasians (31 hepatocellular carcinoma cases and 32 cirrhotic controls) and 31 African Americans (17 hepatocellular carcinoma cases and 14 cirrhotic controls). We observe that none of the two-metabolite candidates can differentiate hepatocellular carcinoma cases from cirrhotic controls among African Americans, but can differentiate hepatocellular carcinoma cases in the Caucasian group. Though oleoyl carnitine does not seem to be significant in considering all participants in the U.S. cohort, it shows marginal significance in discriminating hepatocellular carcinoma cases from cirrhotic controls among African American patients. As female participants account for only about 30% in both cohorts and African Americans account for less than 30% in the U.S. cohort, the observations about gender and race could be merely due to lack of adequate statistical power. Also, we observed that in patients with alcohol-induced cirrhosis (16 hepatocellular carcinoma cases and 18 cirrhotic controls), 3sulfo-GCDCA distinguishes hepatocellular carcinoma cases from cirrhotic controls, whereas in patients with nonalcoholic cirrhosis (40 hepatocellular carcinoma cases and 36 cirrhotic controls), 3β,6β-dihydroxy-5β-cholan-24-oic acid distinguishes the two groups.

Because 3sulfo-GCDCA and 3β,6β-dihydroxy-5β-cholan-24-oic acid significantly discriminate hepatocellular carcinoma cases from cirrhotic controls in the U.S. cohort and are marginally significant in the Egyptian cohort, we evaluated the diagnostic capability of these two metabolites combined with gender in a panel. Table 4 presents the evaluation of the model through AuROC curve. A model trained using the U.S. cohort with gender, 3sulfo-GCDCA, and 3β,6β-dihydroxy-5β-cholan-24-oic acid as predictors yielded an AuROC of 0.74 in distinguishing the hepatocellular carcinoma cases from cirrhotic controls. An AuROC of 0.70 was obtained in evaluating the model by cross-validation. The model resulted in an AuROC of 0.66 when the Egyptian cohort is used as an independent testing set. Another model trained using the Egyptian cohort yielded an AuROC of 0.69. The model resulted in an AuROC of 0.63 in cross-validation and 0.70 when using the U.S. cohort as a testing set. A model built using the combination of the U.S. and Egyptian cohort as a training set yielded an AuROC of 0.71 and an AuROC of 0.69 in cross-validation. Because AFP measurements were unavailable for the cirrhotic controls in the Egyptian cohort, we compared the performance of these metabolites together with AFP for the U.S. cohort only. Gender and AFP alone resulted in an AuROC of 0.69 using the U.S. cohort as a training set and an AuROC of 0.68 in cross-validation. The combination of gender, AFP and the two metabolites increased the AuROC to 0.79 during training, and increased the AuROC to 0.74 in cross-validation. For further evaluation, we split the U.S. cohort into two groups based on the time of sample collection that is the samples that were collected before March 2012 (US1) and those collected after March 2012 (US2). US1 includes 20 hepatocellular carcinoma cases and 20 cirrhotic controls, whereas US2 includes 36 hepatocellular carcinoma cases and 34 cirrhotic controls. These two sets are used as training and testing sets interchangeably to evaluate the performance of the metabolites with and without AFP. The results indicate that improved diagnosis of hepatocellular carcinoma in high-risk population of cirrhotic patients can be achieved by combining the metabolites with AFP and gender information.

Table 4.

Performance evaluation of three logistic regression models using different predictors: (i) AFP and gender; (ii) two metabolites and gender; (iii) AFP, two metabolites, and gender

Egypt (TRN) and US (TES)
Egypt and US (TRN)TESUS (TRN) and Egypt (TES)US1 (TRN) and US2 (TES)US2 (TRN) and US1 (TES)
TRNCRVTRNCRVUSUS1US2TRNCRVTESTRNCRVTESTRNCRVTES
AFP AuROC — — — — — — — 0.69 0.68 — 0.75 0.75 0.70 0.70 0.67 0.70
95% CI — — — — — — — (0.59, 0.80) (0.50, 0.86) — (0.59, 0.90) (0.48, 0.98) (0.56, 0.83) (0.56, 0.83) (0.43, 0.91) (0.53, 0.86)
Metabolite AuROC 0.71 0.69 0.69 0.63 0.70 0.78 0.65 0.74 0.70 0.66 0.81 0.74 0.63 0.71 0.66 0.73
95% CI (0.63, 0.78) (0.57, 0.82) (0.58, 0.80) (0.43, 0.83) (0.60, 0.80) (0.64, 0.93) (0.51, 0.78) (0.64, 0.83) (0.53, 0.87) (0.54, 0.77) (0.67, 0.94) (0.47, 0.98) (0.49, 0.76) (0.58, 0.84) (0.43, 0.89) (0.57, 0.90)
AFP and Metabolite AuROC — — — — — — — 0.79 0.74  0.89 0.81 0.68 0.74 0.67 0.75
95% CI — — — — — — — (0.69, 0.88) (0.58, 0.91) — (0.79, 0.99) (0.58, 0.99) (0.54, 0.82) (0.60, 0.88) (0.42, 0.92) (0.60, 0.91)
Egypt (TRN) and US (TES)
Egypt and US (TRN)TESUS (TRN) and Egypt (TES)US1 (TRN) and US2 (TES)US2 (TRN) and US1 (TES)
TRNCRVTRNCRVUSUS1US2TRNCRVTESTRNCRVTESTRNCRVTES
AFP AuROC — — — — — — — 0.69 0.68 — 0.75 0.75 0.70 0.70 0.67 0.70
95% CI — — — — — — — (0.59, 0.80) (0.50, 0.86) — (0.59, 0.90) (0.48, 0.98) (0.56, 0.83) (0.56, 0.83) (0.43, 0.91) (0.53, 0.86)
Metabolite AuROC 0.71 0.69 0.69 0.63 0.70 0.78 0.65 0.74 0.70 0.66 0.81 0.74 0.63 0.71 0.66 0.73
95% CI (0.63, 0.78) (0.57, 0.82) (0.58, 0.80) (0.43, 0.83) (0.60, 0.80) (0.64, 0.93) (0.51, 0.78) (0.64, 0.83) (0.53, 0.87) (0.54, 0.77) (0.67, 0.94) (0.47, 0.98) (0.49, 0.76) (0.58, 0.84) (0.43, 0.89) (0.57, 0.90)
AFP and Metabolite AuROC — — — — — — — 0.79 0.74  0.89 0.81 0.68 0.74 0.67 0.75
95% CI — — — — — — — (0.69, 0.88) (0.58, 0.91) — (0.79, 0.99) (0.58, 0.99) (0.54, 0.82) (0.60, 0.88) (0.42, 0.92) (0.60, 0.91)

NOTE: The metabolites are 3sulfo-GCDCA and 3β,6β-dihydroxy-5β-cholan-24-oic acid.

Abbreviations: CI, confidence interval; CRV, cross-validation; TES, the cohort used for testing; TRN, the cohort used for training.

In this study, we evaluated the levels of targeted metabolites in sera of hepatocellular carcinoma cases and cirrhotic controls from two cohorts (Egypt and the United States) by SRM on UPLC–ESI–QLITMS. We found that hepatocellular carcinoma in Egyptian cohort is associated with dysregulated levels of bile acids, LPCs, LPEs, sphingophospholipid, long-chain carnitines, and amino acids. Hepatocellular carcinoma in the U.S. cohort is associated with abnormal bile acids metabolism. We found that two metabolites (3sulfo-GCDCA and 3β,6β-dihydroxy-5β-cholan-24-oic acid) are relevant for both the Egyptian and the U.S. cohorts. Together with gender and AFP, these two metabolites led to significant improvement in AuROC compared with gender and AFP only (AuROC 0.66–0.81 vs. 0.66–0.75 based on cross-validation and independent testing). This indicates the potential of these two candidate biomarkers in combination with AFP and gender information for improved diagnosis of hepatocellular carcinoma in high-risk population of cirrhotic patients. Thus, following appropriate clinical validation, the metabolic perturbations of these candidates can be used for early detection of hepatocellular carcinoma or to monitor the progression of cirrhosis to hepatocellular carcinoma. The proportional odds model discovered candidate biomarkers that are specific to the tumor stage level. This improves the ability to detect early-stage hepatocellular carcinoma in high-risk population of liver cirrhosis when treatment is more effective. Our stratified statistical analysis reveals that gender, race, and alcoholic cirrhosis play a decisive role in affecting the relative concentration of small molecules in the serum. These findings will not only contribute to a greater understanding of the implications of health disparities in hepatocellular carcinoma, but also improve the ability to accurately diagnose hepatocellular carcinoma through gender, race, and alcoholic cirrhosis-specific biomarkers.

K. Shetty has received commercial research grant support from Bayer Pharmaceuticals and has honoraria from speakers' bureau from Onyx. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Xiao, C. Di Poto, D.H. Ziada, H.W. Ressom

Development of methodology: J. Xiao, D.H. Ziada, K. Shetty, H.W. Ressom

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J. Xiao, C. Di Poto, L. Zhang, D.H. Ziada, K. Shetty

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Xiao, Y. Zhao, R.S. Varghese, B. Zhou, M.G. Tadesse, H.W. Ressom

Writing, review, and/or revision of the manuscript: J. Xiao, Y. Zhao, R.S. Varghese, B. Zhou, M.G. Tadesse, D.H. Ziada, K. Shetty, H.W. Ressom

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R.S. Varghese, K. Shetty

Study supervision: H.W. Ressom

This work was supported by NCI grant R21CA153176 (to H.W. Ressom). The quantitation data were generated at Georgetown Lombardi Proteomics and Metabolomics Shared Resource supported by NIH/NCI grant P30-CA051008.

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.

1.
Montalto
G
,
Cervello
M
,
Giannitrapani
L
,
Dantona
F
,
Terranova
A
,
Castagnetta
LA
.
Epidemiology, risk factors, and natural history of hepatocellular carcinoma
.
2002
;
963
:
13
20
.
2.
Yu
MC
,
Yuan
JM
,
Govindarajan
S
,
Ross
RK
.
Epidemiology of hepatocellular carcinoma
.
Can J Gastroenterol
2000
;
14
:
703
9
.
3.
El-Serag
HB
.
Hepatocellular carcinoma and hepatitis C in the united states
.
Hepatology
2002
;
36
:
s74
83
.
4.
Wong
JB
,
McQuillan
GM
,
McHutchison
JG
,
Poynard
T
.
Estimating future hepatitis C morbidity, mortality, and costs in the united states
.
Am J Public Health
2000
;
90
:
1562
9
.
5.
El-Serag
HB
.
Hepatocellular carcinoma
.
N Engl J Med.
2011
;
365
:
1118
27
.
6.
Ikeda
K
,
Saitoh
S
,
Suzuki
Y
,
Kobayashi
M
,
Tsubota
A
,
Koida
I
, et al
Disease progression and hepatocellular carcinogenesis in patients with chronic viral hepatitis: a prospective observation of 2215 patients
.
J Hepatol
1998
;
28
:
930
8
.
7.
Poustchi
H
,
Ghajarieh Sepanlou
S
,
Esmaeili
S
,
Mehrabi
M
,
A
.
Hepatocellular carcinoma in the world and the middle east
.
Middle East J Dig Dis.
2011
;
2
:
31
41
.
8.
Altekruse
SF
,
McGlynn
KA
,
Reichman
ME
.
Hepatocellular carcinoma incidence, mortality, and survival trends in the united states from 1975 to 2005
.
J Clin Oncol
2009
;
27
:
1485
91
.
9.
Schwegler
EE
,
Cazares
L
,
Steel
LF
,
BL
,
Johnson
DA
,
Semmes
OJ
, et al
SELDI-TOF MS profiling of serum for detection of the progression of chronic hepatitis C to hepatocellular carcinoma
.
Hepatology
2005
;
41
:
634
42
.
10.
Barbie
DA
,
Tamayo
P
,
Boehm
JS
,
Kim
SY
,
Moody
SE
,
Dunn
IF
, et al
Systematic RNA interference reveals that oncogenic KRAS-driven cancers require TBK1
.
Nature
2009
;
462
:
108
12
.
11.
Chen
T
,
Xie
G
,
Wang
X
,
Fan
J
,
Qiu
Y
,
Zheng
X
, et al
Serum and urine metabolite profiling reveals potential biomarkers of human hepatocellular carcinoma
.
Mol Cell Proteomics
2011
;
10
:
M110.004945
.
12.
Fernie
AR
,
Trethewey
RN
,
Krotzky
AJ
,
Willmitzer
L
.
Metabolite profiling: from diagnostics to systems biology
.
Nat Rev Mol Cell Biol
2004
;
5
:
763
9
.
13.
Kind
T
,
Tolstikov
V
,
Fiehn
O
,
Weiss
RH
.
A comprehensive urinary metabolomic approach for identifying kidney cancerr
.
Anal Biochem
2007
;
363
:
185
95
.
14.
Nicholson
JK
,
Wilson
ID
.
Opinion: understanding ‘global’ systems biology: metabonomics and the continuum of metabolism
.
Nat Rev Drug Discov
2003
;
2
:
668
76
.
15.
Xiao
JF
,
Varghese
RS
,
Zhou
B
,
Nezami Ranjbar
MR
,
Zhao
Y
,
Tsai
TH
, et al
LC-MS based serum metabolomics for identification of hepatocellular carcinoma biomarkers in egyptian cohort
.
J Proteome Res
2012
;
11
:
5914
23
.
16.
Ressom
HW
,
Xiao
JF
,
Tuli
L
,
Varghese
RS
,
Zhou
B
,
Tsai
TH
, et al
Utilization of metabolomics to identify serum biomarkers for hepatocellular carcinoma in patients with liver cirrhosis
.
Anal Chim Acta
2012
;
743
:
90
100
.
17.
Patterson
,
Maurhofer
O
,
Beyoglu
D
,
Lanz
C
,
Krausz
KW
,
Pabst
T
, et al
Aberrant lipid metabolism in hepatocellular carcinoma revealed by plasma metabolomics and lipid profiling
.
Cancer Res
2011
;
71
:
6590
600
.
18.
Tan
Y
,
Yin
P
,
Tang
L
,
Xing
W
,
Huang
Q
,
Cao
D
, et al
Metabolomics study of stepwise hepatocarcinogenesis from the model rats to patients: potential biomarkers effective for small hepatocellular carcinoma diagnosis
.
Mol Cell Proteomics
2012
;
11
:
M111.010694
.
19.
Yang
J
,
Zhao
X
,
Liu
X
,
Wang
C
,
Gao
P
,
Wang
J
, et al
High performance liquid chromatography mass spectrometry for metabonomics: potential biomarkers for acute deterioration of liver function in chronic hepatitis B
.
J Proteome Res
2006
;
5
:
554
61
.
20.
Yin
P
,
Wan
D
,
Zhao
C
,
Chen
J
,
Zhao
X
,
Wang
W
, et al
A metabonomic study of hepatitis B-induced liver cirrhosis and hepatocellular carcinoma by using RP-LC and HILIC coupled with mass spectrometry
.
Mol Biosyst
2009
;
5
:
868
76
.
21.
McGlynn
KA
,
Tsao
L
,
Hsing
AW
,
Devesa
SS
,
Fraumeni
JF
Jr
.
International trends and patterns of primary liver cancer
.
Int J Cancer
2001
;
94
:
290
6
.
22.
Parkin
DM
,
Bray
F
,
Ferlay
J
,
Pisani
P
.
Global cancer statistics, 2002
.
CA Cancer J Clin
2005
;
55
:
74
108
.
23.
Naugler
WE
,
Sakurai
T
,
Kim
S
,
Maeda
S
,
Kim
K
,
Elsharkawy
AM
, et al
Gender disparity in liver cancer due to sex differences in MyD88-dependent IL-6 production
.
Science
2007
;
317
:
121
4
.
24.
Yu
MW
,
Chen
CJ
.
Elevated serum testosterone levels and risk of hepatocellular carcinoma
.
Cancer Res
1993
;
53
:
790
4
.
25.
Nguyen
MH
,
Garcia
RT
,
Simpson
PW
,
Wright
TL
,
Keeffe
EB
.
Racial differences in effectiveness of alpha-fetoprotein for diagnosis of hepatocellular carcinoma in hepatitis C virus cirrhosis
.
Hepatology
2002
;
36
:
410
7
.
26.
Li
S
,
Liu
H
,
Jin
Y
,
Lin
S
,
Cai
Z
,
Jiang
Y
.
Metabolomics study of alcohol-induced liver injury and hepatocellular carcinoma xenografts in mice
.
J Chromatogr B Analyt Technol Biomed Life Sci
2011
;
879
:
2369
75
.
27.
Charni
F
,
Sutton
A
,
Rufat
P
,
Laguillier
C
,
Mansouri
A
,
Moreau
R
, et al
Chemokine RANTES promoter dimorphisms and hepatocellular carcinoma occurrence in patients with alcoholic or hepatitis C virus-related cirrhosis
.
Cancer Epidemiol Biomarkers Prev
2011
;
20
:
1439
46
.
28.
Lee
HS
,
Chung
YH
,
Kim
CY
.
Specificities of serum alpha-fetoprotein in HBsAg+ and HBsAg− patients in the diagnosis of hepatocellular carcinoma
.
Hepatology
1991
;
14
:
68
72
.
29.
Wang
B
,
Chen
D
,
Chen
Y
,
Hu
Z
,
Cao
M
,
Xie
Q
, et al
Metabonomic profiles discriminate hepatocellular carcinoma from liver cirrhosis by ultraperformance liquid chromatography-mass spectrometry
.
J Proteome Res
2012
;
11
:
1217
27
.
30.
Goto
T
,
Myint
KT
,
Sato
K
,
O
,
Kakiyama
G
,
Iida
T
, et al
LC/ESI-tandem mass spectrometric determination of bile acid 3-sulfates in human urine 3beta-sulfooxy-12alpha-hydroxy-5beta-cholanoic acid is an abundant nonamidated sulfate
.
J Chromatogr B Analyt Technol Biomed Life Sci
2007
;
846
:
69
77
.
31.
Gowda
GA
.
Human bile as a rich source of biomarkers for hepatopancreatobiliary cancers
.
Biomark Med
2010
;
4
:
299
314
.