Serum proteomic analysis revealed diagnostic value of hemoglobin for nonalcoholic fatty liver disease

Transcription

1 Serum proteomic analysis revealed diagnostic value of hemoglobin for nonalcoholic fatty liver disease Chaohui Yu 1,, Chengfu Xu 1,, Lei Xu 1, Jiekai Yu 2, Min Miao 3, Youming Li 1, 1 Department of Gastroenterology, The First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou , China; 2 Cancer Institute, The Second Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou , China; 3 Department of Internal Medicine, Zhenhai Lianhua Hospital, Ningbo , China Background & Aims: Nonalcoholic fatty liver disease (NAFLD) is one of the most common chronic liver diseases worldwide. The two linked studies presented herein aimed to identify and verify new biomarkers for NAFLD. Methods: First, 70 serum samples were analyzed using proteomics approaches to identify potential biomarkers for NAFLD. Second, a total of 6944 initial NAFLD-free subjects were followed up for 3 years to evaluate the predictive value of hemoglobin for NAFLD. Results: In the first study, 20 differentially expressed protein peaks (11 up-regulated and nine down-regulated) were observed in NAFLD patients upon comparison to the controls. With the aid of bioinformatic tools, we established a biomarker pattern for NAFLD with a sensitivity of 89% and a specificity of 83%. Further analysis suggested a protein peak to be hemoglobin subunit alpha. In the second study, prospective analysis showed that subjects with higher baseline hemoglobin levels were associated with higher incidence of NAFLD. Cox proportional hazards regression analyses showed that the age, gender, and body mass index adjusted hazard ratio (95% CI) for subjects with baseline hemoglobin level in quintile 2, 3, 4, and 5 vs. quintile 1 was 1.36 ( ), 1.66 ( ), 1.76 ( ), and 1.83 ( ), respectively. Conclusions: Our study showed that serum hemoglobin may have significant predictive value for NAFLD. Ó 2011 European Association for the Study of the Liver. Published by Elsevier B.V. All rights reserved. Keywords: Nonalcoholic fatty liver disease; Biomarker; Hemoglobin. Received 30 July 2010; received in revised form 8 May 2011; accepted 11 May 2011; available online 12 July 2011 Corresponding author. Address: Department of Gastroenterology, The First Affiliated Hospital, College of Medicine, Zhejiang University, 79 Qingchun Road, Hangzhou , China. Tel.: ; fax: address: xiaofu@zju.edu.cn (Y. Li). These authors contributed equally to this work. Abbreviations: ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI, body mass index; HDL-C, high-density lipoprotein cholesterol; FPG, fasting plasma glucose; LDL-C, low-density lipoprotein cholesterol; NAFLD, nonalcoholic fatty liver disease; NASH, nonalcoholic steatohepatitis; SELDI-TOF-MS, surface enhanced laser desorption/ionization-time of flight mass spectrometry; SUA, serum uric acid; ZUCI-PDAS, Cancer Institute of Zhejiang University-ProteinChip Data Analysis System. Introduction Nonalcoholic fatty liver disease (NAFLD), which includes a range of conditions from nonalcoholic steatosis to nonalcoholic steatohepatitis (NASH) and cirrhosis, is the most common liver disease identified in Western countries [1,2]. NAFLD has also become a significant form of chronic liver disease in developing countries [3,4]. The development of NAFLD is closely associated with central obesity, type 2 diabetes, hypertension, and dyslipidemia, which form a cluster of metabolic disorders that is now recognized as the metabolic syndrome [5,6]. For this reason, NAFLD has been regarded as a hepatic manifestation of the metabolic syndrome [5]. The diagnosis of NAFLD is suspected in patients with elevated serum aminotransferases and, in many cases, it is suspected in patients with evidence of the metabolic syndrome [7]. Ultrasonography is widely used in clinical diagnosis of NAFLD. However, the sensitivity of ultrasonography is relatively low in patients with mild steatosis. In addition, inter-observer variability is another factor that may limit the usefulness of ultrasonography [8]. Computed tomography and magnetic resonance imaging may help in determining the presence and amount of fatty infiltration in the liver, whereas these techniques cannot be used to accurately determine the severity of liver damage [8,9]. To date, liver biopsy remains the gold standard for NAFLD diagnosis. However, its invasiveness, the observations of significant side effect profile, and susceptibility of this technique to sampling error ultimately raise the need for finding out reliable diagnostic biomarkers to replace liver biopsy [10,11]. Recent advances in the field of proteomics have provided us with powerful tools for studying disease diagnosis and for discovering biomarkers [12]. Among the different technologies available, surface enhanced laser desorption/ionization-time of flight mass spectrometry (SELDI-TOF-MS) is an attractive technique since it provides protein profiles in an easy and rapid manner from small amounts of protein, obtained from various biologically complex samples [13]. In addition to this, with the aid of bioinformatics tools, the data gleaned from SELDI- TOF-MS can be further analyzed and successfully used for identification of novel diagnostic biomarkers [14,15]. Recently, Trak-Smayra et al. [16] firstly applied SELDI-TOF-MS to identify serum markers of steatosis and NASH in extremely obese patients who were candidates for bariatric surgery. However, to date, the Journal of Hepatology 2012 vol. 56 j

2 combination of SELDI-TOF-MS and bioinformatics approach has not been widely applied to the field of biomarker identification for general NAFLD patients. In this study, we first aimed to identify new biomarkers for NAFLD using a combination of SELDI-TOF-MS and bioinformatics tools, and then verified the predictive value of these potential biomarkers for NAFLD by a population-based study. Patients and methods This study was designed to identify and verify new biomarkers for NAFLD in two separate studies. The first study (Study 1) was a proteomic study designed to clarify serum protein fingerprint of NAFLD patients, and identify potential biomarkers for NAFLD. Once the results of Study 1 were obtained, we then sought to determine whether the biomarker identified in Study 1 has a predictor value for NAFLD. For this question, we conducted Study 2, a population-based prospective study among 6944 initially NAFLD-free subjects. Study 1: Proteomic study Subjects included for serum proteomic analysis In order to clarify serum protein fingerprint of NAFLD patients, 70 subjects (35 NAFLD patients and 35 controls) were recruited at the First Affiliated Hospital, College of Medicine, Zhejiang University, Hangzhou Sixth People s Hospital, Ningbo Yinzhou People s Hospital, and Ningbo Medical Treatment Center Lihuili Hospital. The diagnosis of NAFLD was based on the following criteria: (i) the hepatic histology indicated the presence of steatosis affecting more than 5% of the total hepatocytes; (ii) the patient had a history of alcohol consumption of less than 140 g/week for male and less than 70 g/week for female; (iii) the patient did not have a specific disease that could lead to hepatic steatosis, such as viral hepatitis, drug-induced liver disease, total parenteral nutrition, autoimmune hepatitis, and metabolic/genetic liver diseases [17]. The subjects included as controls were these with normal anthropometry parameters, blood pressure, liver function tests, liver ultrasound examination, and normal hepatic histological findings. A total of 35 eligible NAFLD patients and 35 healthy controls were enrolled in this study. The mean age and gender of subjects in control group were comparable with those patients from the NAFLD group. Informed consent was obtained from all the subjects and the study protocol was approved by the Hospital Ethics Committee. Clinical examinations Clinical examinations were conducted in a standard manner as previously described [18]. Fasted blood samples were obtained from an antecubital vein and the samples were used for the biochemical values analysis and proteomic analysis. The biochemical values were measured using an Olympus AU640 autoanalyzer (Olympus, Kobe, Japan) using standard methods. The biochemical values included alanine aminotransferase (ALT), aspartate aminotransferase (AST), triglyceride, total cholesterol, high-density lipoprotein cholesterol (HDL-C), lowdensity lipoprotein cholesterol (LDL-C), fasting plasma glucose (FPG), and serum uric acid (SUA). Liver biopsy specimens from the subjects were read under coded identification by a liver pathologist who was unaware of the patient details or biopsy sequence. Histological features were interpreted according to Brunt et al. [19], with modification. Steatosis was graded as follows: mild (steatosis involving less than 30% of hepatocytes), moderate (30 60% of hepatocytes affected), severe (more than 60% of hepatocytes affected). Severity of lobular inflammatory, hepatocellular necrosis, portal tract inflammation, pericellular fibrosis, portal fibrosis, and bridging fibrosis were also recorded [19]. NASH was defined as either the presence of steatosis with mixed lobular inflammation and hepatocellular ballooning, as proposed during the AASLD single topic conference [2], or the presence of fibrosis with any stage of fibrosis [20]. Proteomic analysis The ProteinChip used was the CM10 (weak cation exchange). The detailed protocols for sample preparation were listed in Supplementary subjects and methods. The chips were detected on the protein biological system II (PBS-II) plus mass spectrometer reader (Ciphergen Biosystems). The bioinformatics and biostatistics were described as reported previously [14], and listed in Supplementary subjects and methods. The whole processing was performed using the Cancer Institute of Zhejiang University-ProteinChip Data Analysis System (ZUCI-PDAS, designed by Yu Jiekai and based on the MATLAB Web Server (The Math Works Inc., Natick, MA). Serum protein identification was also performed by 2-DE and MS based methods as described in Supplementary subjects and methods. Study 2: Prospective study Subjects and study design To evaluate the predictive value of hemoglobin for NAFLD, a population-based prospective study was conducted among the employees of Zhenhai Refining & Chemical Company Ltd., Ningbo, China beginning in The study population was described in detail in our previous uric acid study [21]. A total of 6944 subjects (4496 male and 2448 female) with serum hemoglobin data available in the uric acid study were included in this study. The study protocol was approved by the Hospital Ethics Committee. Baseline examinations and outcome assessments Baseline examinations and outcome assessments were performed as previously described [21]. In brief, baseline examinations included a medical history and health habit inventory taken by a physician, anthropometric measurements, hepatic ultrasonic examination, and biochemical measurements. Serum hemoglobin level was analyzed by SYSMEX XT-1800 hematology autoanalyzer according to the manufacturer s instructions. The diagnosis of fatty liver was based on the results of abdominal ultrasonography using a Toshiba Nemio 20 sonography machine with a 3.5-MHz probe (Toshiba, Tokyo, Japan). The diagnosis of metabolic syndrome was based on the new International Diabetes Federation definition [22]. For a person to be defined as having the metabolic syndrome they must have: central obesity (defined as waist circumference P90 cm for Chinese men and P80 cm for Chinese women), plus any two of the following four factors: (i) raised triglyceride level, defined as triglycerides P1.7 mmol/l or specific treatment for this lipid abnormality; (ii) reduced HDL- C level, defined as HDL-C <1.03 mmol/l in males and <1.29 mmol/l in females; (iii) raised blood pressure, SBP P130 mm Hg or DBP P85 mm Hg, or treatment of previously diagnosed hypertension; (iv) raised FPG, defined as FPG P5.6 mmol/l, or previously diagnosed type 2 diabetes. Statistical analysis Continuous variables are presented as mean and standard deviation (SD) or medians and interquartile range (IQR), as appropriate. Continuous variables were compared with Wilcoxon rank-sum test, student s t-test, Kruskal Wallis H test or one-way analysis of variance, depending on the normality of the data; categorical variables were compared using the v 2 test. The Kolmogorov Smirnov test was used to assess whether continuous data were normally distributed. To explore the association between hemoglobin level and risk for incident NAFLD, subjects were stratified according to their baseline hemoglobin levels: quintile 1, 6125 g/l; quintile 2, g/l; quintile 3, g/l; quintile 4, g/l; and quintile 5, P150 g/l. The baseline characteristics of the subjects in each quintile were compared. Kaplan Meier estimator was used to estimate the cumulative incidence of NAFLD. We used Cox proportional hazards regression analyses to estimate hazard ratios for incident NAFLD for each baseline hemoglobin quintile over the whole follow-up range. The Cox proportional regression model assumes that the effects of the predictor variables are constant over time. This was assessed by tests based on the generalization by Grambsch and Therneau using scaled Schoenfeld residuals [22]. The subjects within the first quintile were used as reference group. The data were first adjusted for age and gender and then for other covariates that might confound the relationship between the hemoglobin and NAFLD. For linear trends of risk, the number of quintiles was used as a continuous variable and tested on each model. The statistical analyses were performed using the SPSS software package version 13.0 for Windows (SPSS Inc., Chicago, IL). p <0.05 (2-tailed) was considered to be statistically significant. Results Study 1 Clinical and laboratory data Results from analysis of the NAFLD patients and controls indicated that they were different in terms of BMI, waist 242 Journal of Hepatology 2012 vol. 56 j

3 JOURNAL OF HEPATOLOGY Table 1. Data comparison between NAFLD patients and controls. Variable Controls NAFLD patients p value Age (years) 48.3 (9.6) 49.1 (13.8) Gender (male/female, n) 14/21 19/ a Type 2 diabetes (yes/no, n) 5/30 6/ a Body mass index (kg/m 2 ) (2.44) (3.21) Waist circumference (cm) 78.0 (5.5) 93.7 (9.1) <0.001 Systolic blood pressure (mmhg) (2.7) (14.4) Diastolic blood pressure (mmhg) 78.7 (9.2) 81.7 (10.1) Alanine aminotransferase (U/L) 21.5 ( ) 57.0 ( ) b Aspartate aminotransferase (U/L) 22.5 ( ) 32.0 ( ) b Triglyceride (mmol/l) 4.28 ( ) 4.97 ( ) b Total cholesterol (mmol/l) 4.22 (0.78) 5.01 (1.37) HDL cholesterol (mmol/l) 1.09 ( ) 1.18 ( ) b LDL cholesterol (mmol/l) 2.20 ( ) 2.97 ( ) <0.001 b Fasting plasma glucose (mmol/l) 4.86 ( ) 5.39 ( ) b Serum uric acid (mmol/l) (92.3) (75.7) Data are expressed as mean (SD) or median (IQR). a v 2 test; b Wilcoxon rank-sum test; HDL, high-density lipoprotein; LDL, low-density lipoprotein. circumference, ALT, AST, triglyceride, total cholesterol, LDL cholesterol, FPG, and SUA, while there was no difference in terms of age, gender, type 2 diabetes, systolic blood pressure, diastolic blood pressure, or HDL cholesterol between the two groups (Table 1). Among 35 NAFLD patients, mild steatosis, moderate steatosis, and severe steatosis was observed in 23, 7, and 5 patients, respectively. Among the 12 patients with moderate and severe steatosis, NASH was observed in three patients. No abnormal hepatic histological finding was observed in control subjects. Typical hepatic histological findings were illustrated in Supplementary Fig. 1. Serum biomarker identification for NAFLD In order to clarify serum protein fingerprint of NAFLD patients, 70 serum samples from 35 histologically confirmed NAFLD patients and 35 controls were analyzed by ProteinChip based SELDI-TOF- MS. We first evaluated the reproducibility of the ProteinChip by using eight serum samples from one single subject, which were subsequently applied to each chip in a random fashion. The coefficient of variance for the selected peaks after being normalized by intensity was 14.5% and the coefficient of variance for selected peaks mass was 0.05%. For proteins profiles comparison, a total of 507 protein peaks were analyzed, and the m/z of these protein peaks ranged from 1006 to 16,113 Da. Bonferroni adjustment was applied in multiple comparisons, the adjusted p value is 0.05/507 = , which is an upper bound for the correct p value. The comparison of proteins profiles obtained from NAFLD patients and controls showed a total of 20 differentially expressed proteins peaks. Eleven of these peaks were highly expressed in NAFLD patients whereas nine of these peaks were low expressed. The top 10 differentially expressed protein peaks were selected, randomly combined, and input into the support vector machine. Among these, the peaks with m/z of 2760, 2957, 2967, 3269, 3403, 5814, and 6306 Da were significantly low expressed in NAFLD patients, while the peaks with m/z of 8567, 13,756, and 15,124 Da were significantly high expressed in NAFLD patients (Supplementary Fig. 2). The accuracy of all models was calculated and the diagnostic model that achieved the highest Youden s index was selected as the final model to distinguish the two groups. The diagnostic model to differentiate NAFLD patients from controls was comprised of six biomarkers with m/z of 2760, 2957, 2967, 5814, 6306, and 15,124 Da. The peak with m/ z of 15,124 Da, which was supported to be hemoglobin subunit alpha, was significantly up-regulated in NAFLD group (Supplementary Methods and Results); while the other five peaks were significantly down-regulated. The diagnostic model had a sensitivity of 89% and a specificity of 83% as evaluated by leave-one cross-validation. Study 2 Baseline characteristics At baseline, 6944 subjects (4496 male and 2448 female) were included, whereas 466 subjects (317 male and 149 female) did not complete the 3-year follow-up examination. All subjects did not have competing events including death during the 3-year follow up. Baseline characteristics were not significantly different between the subjects lost to follow-up and those with successful follow-up (Supplementary Table 1). The baseline characteristics of subjects in each hemoglobin quintile are shown in Table 2. Age, male gender, BMI, waist circumference, blood pressure, serum liver enzymes, serum lipids, FPG, and SUA all tended to increase at higher hemoglobin levels (all p <0.001). In contrast, HDL-C decreased with as hemoglobin increased (p <0.001). Association of hemoglobin with incident NAFLD To explore the association of hemoglobin with incident NAFLD, subjects were stratified into quartiles according to their baseline Journal of Hepatology 2012 vol. 56 j

4 Table 2. Baseline characteristics of study subjects according to hemoglobin quintiles. Baseline hemoglobin quintile p value a 1 (n = 1454) 2 (n = 1417) 3 (n = 1384) 4 (n = 1376) 5 (n = 1313) Age (yrs) 45.2 (12.1) 47.1 (13.7) 44.4 (13.0) 43.0 (12.7) 42.4 (12.3) <0.001 Gender (male/female, n) 151/ / / / /96 <0.001 Type 2 diabetes (yes/no, n) 33/ / / / / Body mass index (kg/m 2 ) (2.74) (2.82) (2.56) (2.40) (2.63) <0.001 Waist circumference (cm) 71.7 (7.2) 74.6 (7.7) 78.1 (7.4) 80.5 (7.0) 80.9 (7.4) <0.001 Systolic blood (15.6) (15.3) (14.0) (13.7) (13.4) <0.001 pressure (mmhg) Diastolic blood 72.4 (8.8) 75.1 (9.0) 75.4 (8.8) 77.3 (9.0) 78.7 (8.9) <0.001 pressure (mmhg) Alanine 16.0 ( ) 18.0 ( ) 21.0 ( ) 24.0 ( ) 26.0 ( ) <0.001 aminotransferase (U/L) Aspartate 18.0 ( ) 19.0 ( ) 19.0 ( ) 20.0 ( ) 20.0 ( ) <0.001 aminotransferase (U/L) Triglyceride (mmol/l) 0.92 ( ) 1.09 ( ) 1.17 ( ) 1.30 ( ) 1.37 ( ) <0.001 Total cholesterol (mmol/l) 4.70 (0.96) 4.84 (0.93) 4.70 (0.93) 4.76 (0.90) 4.82 (0.93) <0.001 HDL cholesterol (mmol/l) 1.41 ( ) 1.36 ( ) 1.23 ( ) 1.22 ( ) 1.21 ( ) <0.001 LDL cholesterol (mmol/l) 2.57 (0.76) 2.69 (0.77) 2.68 (0.75) 2.72 (0.71) 2.78 (0.76) <0.001 Fasting plasma 4.46 ( ) 4.53 ( ) 4.44 ( ) 4.43 ( ) 4.39 ( ) <0.001 glucose (mmol/l) Serum uric acid (mmol/l) (72.8) (74.3) (71.4) (68.3) (71.9) <0.001 Data are presented as mean (SD) or median (IQR). The subjects were grouped according to baseline hemoglobin quintiles: quintile 1, 6125 g/l; quintile 2, g/l; quintile 3, g/l; quintile 4, g/l; and quintile 5, P150 g/l. HDL, high-density lipoprotein; LDL, low-density lipoprotein. a p values are computed from v 2 test for categorical data and on Kruskal Wallis H test or one-way analysis of variance for continuous data, depending on the normality of the data. hemoglobin levels. The cumulative incidence of NAFLD was calculated by dividing the number of cases by the numbers of subjects in each hemoglobin quintile. A total of 773 subjects developed NAFLD during the follow up study. Kaplan Meier estimator showed that the cumulative incidence of NAFLD increased progressively with increased baseline hemoglobin level (Fig. 1). This observation indicated that the subjects with higher baseline hemoglobin levels were more likely to develop NAFLD than those with lower levels. High hemoglobin predicts increase risk of NAFLD To estimate hazard ratios of each hemoglobin quartile for incident NAFLD, Cox proportional hazards regression analyses were applied. As shown in Table 3, with comparison to the subjects in quintile 1, baseline hemoglobin levels were positively correlated with the hazard ratios for incident NAFLD. After adjustment for age and gender, with comparison to subjects in quintile 1, the hazard ratios (95% CI) for subjects in quintile 2, quintile 3, quintile 4, and quintile 5 were 1.43 ( ), 1.88 ( ), 2.21 ( ), and 2.42 ( ), respectively (p for trend <0.001; Table 3). The relationship of hemoglobin with incident NAFLD remained significant even when further adjusted for five components of metabolic syndrome, which are all major cofactors that impact the development of NAFLD (Table 3). When mean hemoglobin increases, the prevalence of anemia should decrease accordingly. In the second study, at total of 212 subjects fulfilled the diagnostic criteria of anemia (Hb <120 g/l for male and Hb <110 g/l for female) at baseline evaluation. With comparison to non-anemic subjects, the 3-year cumulative incidence of NAFLD was significantly lower for anemic subjects (4.2% vs %; p <0.001). In age and gender adjusted Cox analysis, anemia was showed to be a protective factor that reduced the development of NAFLD, the hazard ratios (95% CI) was 0.50 ( ). These results indirectly indicated that hemoglobin level is an important factor that predicts the development of NAFLD and the risk increases with increase in baseline hemoglobin levels. Discussion In this study, we first analyzed serum samples using SELDI-TOF- MS and bioinformatics tools, and successfully established a serum protein fingerprint diagnostic model for NAFLD. A further population-based prospective study showed that serum hemoglobin level independently predicts the development of NAFLD and may serve as a biomarker for the disease. The field of biomarker is an area of fast growing interest in the setting of NAFLD [23]. As it has been well established, the serum contains thousands of proteins or peptides that regulate a large number of physiologic functions [24]. The alteration of specific proteins or peptides may be related to a specific disease [25]. Therefore, the identification of the protein patterns in the serum could provide a valid clinical diagnosis of many diseases. Traditional knowledge-based approaches such as enzyme-linked immunosorbent assay based assays are classical and reliable methods to achieve biomarker identification [26]. However, these 244 Journal of Hepatology 2012 vol. 56 j

5 Cumulative incidence rate of NAFLD No at risk Quintile: Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 p trend < Follow-up (yrs) Fig. 1. Kaplan Meier curves reflecting cumulative incidence rate of NAFLD according to quintiles of baseline hemoglobin level. Subjects were stratified into quartiles according to their baseline hemoglobin levels; subjects with higher hemoglobin levels had an increased cumulative incidence rate of NAFLD. p value for trend is computed from Cox analysis, the number of quintiles was used as a continuous variable and tested on the model. traditional methods are time-consuming and may be not the best choice for identification of new biomarkers [27]. Recent advance in the field of proteomics provide a high throughput unbiased omic approach for the biomarker identification. Among these techniques, SELDI-TOF-MS coupled with ProteinChip technology offers a sensitive, high throughput, and rapid method for analyzing a complex mixture of protein and peptide [13]. Using this method, we identified 11 up-regulated and nine down-regulated JOURNAL OF HEPATOLOGY serum proteins peaks in the NAFLD patients as compared to controls. The generation of the serum protein fingerprint for NAFLD patients has provided valuable information for further biomarkers investigations. The development of NAFLD is a complex multifactor process involving the perturbation of multiple gene and protein regulations [9]. In light of this multifactorial natural of NAFLD, it is plausible that a combination of multiple biomarkers is necessary to improve the diagnosis of NAFLD. Bioinformatics tools are especially effective for the analysis of complex data from genomic, proteomic, and metabomic investigations. In this study, we applied sophisticated bioinformatics tools to analyze the complex data obtained from SELDI-TOF-MS analyses. We successfully established a new diagnostic model for NAFLD that was characterized by high sensitivity and specificity. This suggests that it is possible to establish diagnostic model for NAFLD by combining SELDI-TOF-MS and bioinformatics analyses. Our prospective study showed that serum hemoglobin is significantly associated with risk for incident NAFLD. A significant association between hemoglobin and atherosclerosis has also been reported in recent studies [22,28,29]. Huang et al. [30] observed that serum free hemoglobin may serve as a novel biomarker for acute ischemic stroke with a sensitivity of 70.2% and a specificity of is 85.3%. Moreover, serum hemoglobin levels have also been observed to be significantly elevated in extremely obese NAFLD patients by Trak-Smayra et al. [16], and our recent cross-sectional study demonstrated that the serum hemoglobin level was significantly associated with prevalence of NAFLD [31]. In this prospective study, we showed that baseline serum hemoglobin level was positively associated with risk for development of NAFLD, and this association was independent of body mass index, type 2 diabetes, and other metabolic diseases. These results indicated that hemoglobin may also serve as a reliable biomarker for NAFLD, which is closely related with atherosclerosis and stroke. Serum hemoglobin level is influenced by dietary habit. Usually, low serum hemoglobin level rarely occurs among subjects frequently eating animal foods. With comparison to Western foods, the Chinese food pattern is unique in that it is characterized Table 3. Risk of development of NAFLD according to baseline hemoglobin quintiles in unadjusted and adjusted models. Models Baseline hemoglobin quintile χ 2 test p a Hazard ratio (95% confidence interval) 1 (n = 1454) 2 (n = 1417) 3 (n = 1384) 4 (n = 1376) 5 (n = 1313) Unadjusted 1 [reference] 1.54 ( ) 2.11 ( ) 2.53 ( ) 2.76 ( ) <0.001 Adjusted for age and 1 [reference] 1.43 ( ) 1.88 ( ) 2.21 ( ) 2.42 ( ) <0.001 gender Adjusted for age, 1 [reference] 1.36 ( ) 1.66 ( ) 1.76 ( ) 1.83 ( ) gender, and BMI Adjusted for age, 1 [reference] 1.43 ( ) 1.88 ( ) 2.20 ( ) 2.42 ( ) <0.001 gender, and T2DM Adjusted for age, gender, and metabolic diseases b 1 [reference] 1.37 ( ) 1.70 ( ) 1.79 ( ) 1.74 ( ) The subjects were grouped according to baseline hemoglobin quintiles: quintile 1, 6125 g/l; quintile 2, g/l; quintile 3, g/l; quintile 4, g/l; and quintile 5, P150 g/l. BMI, body mass index; MS, metabolic syndrome; T2DM, type 2 diabetes. a p values for trend are computed from Cox analysis, the number of quintiles was used as a continuous variable and tested on each model. b Metabolic diseases, including central obesity, raised triglyceride level, reduced HDL-C level, raised blood pressure and raised FPG, were defined according to the definition recommended by the new International Diabetes Federation definition. Journal of Hepatology 2012 vol. 56 j

6 by a high intake of cereals and vegetables and a low intake of animal foods. This may partially explain why the prevalence of NAFLD is significantly lower in China than in Western countries. Future research should thus examine the association between hemoglobin, serum ferritin level, iron intake, and NAFLD. In addition, due to ethnic differences in dietary habit, the homogeneous ethnicity of the sample should also be born in mind for future related studies [32]. The relationship between serum hemoglobin and NAFLD may be partially modulated by haptoglobin level. As we know, haptoglobin is an acute-phase protein and its secretion increases in response to injury or inflammatory insults. Importantly, haptoglobin acts mainly as an antioxidant, binding to free hemoglobin and thereby inhibiting the hemoglobin-induced oxidative damage to tissues. Because both inflammation and oxidative injury are thought to contribute to the development of NAFLD, haptoglobin has the potential to be a determinant of NAFLD risk and can modulate the risk linked to free hemoglobin levels [33]. A limitation of this study is that serum insulin levels have not been analyzed. Nevertheless, we adjusted for the metabolic syndrome, instead of insulin resistance, in the Cox analysis. As we know, insulin resistance has been repeatedly observed to be significantly associated with metabolic syndrome in many large sample size studies [34 36]. The second limitation is that, due to sample limitation, we could not establish a diagnostic model to distinguish simple steatosis from NASH. Yet despite this, the diagnosis of simple steatosis does have significant clinic importance. By obtaining an early diagnosis for hepatic steatosis, we can start intervention at the very beginning of the disease, thereby decreasing the negative effects caused by the disease. In addition, we can also monitor the effects of different therapeutic interventions in a simple manner that is independent of a liver biopsy. The third limitation is that several important confounders, including urban residence, income, occupation, high serum ferritin, iron intake, energy, protein, and fat have not been examined in this study; further studies are needed to clarify this point. Moreover, further studies are also needed to investigate the diagnostic value of the five remaining biomarkers for NAFLD. In summary, our results showed that the serum protein fingerprint analysis using the SELDI-TOF-MS technique combined with bioinformatic approaches can help to discover novel biomarkers for NAFLD. Hemoglobin may have a significant predictive value for NAFLD. Further studies on the biomarkers identified herein may eventually help establish a reliable noninvasive tool for diagnosis of NAFLD. Conflict of interest The authors who have taken part in this study declared that they do not have anything to disclose regarding funding or conflict of interest with respect to this manuscript. Acknowledgments This study was supported by Chinese State Key Project for Hightech (No. 2006AA02A308), National Science & Technology Pillar Program (No. 2008BAI52B03), National Natural Science Foundation of China (Nos , , ), Science & Technology Foundation of Zhejiang Province (No. 2008C ), and Zhejiang Provincial Natural Science Foundation of China (Nos. Y , Y ). The authors thank Prof. Junping Shi, Dr. Mingzhu Li, and Dr. Hong Li for their help in sample collection. The authors also thank Dr. Tong Huang and Dr. Jingyi Yuan at Zhenhai Lianhua Hospital, Ningbo, China for their valuable work in data collection. Supplementary data Supplementary data associated with this article can be found, in the online version, at doi: /j.jhep References [1] Jimba S, Nakagami T, Takahashi M, Wakamatsu T, Hirota Y, Iwamoto Y, et al. Prevalence of non-alcoholic fatty liver disease and its association with impaired glucose metabolism in Japanese adults. Diabet Med 2005;22: [2] Neuschwander-Tetri BA, Caldwell SH. Nonalcoholic steatohepatitis: summary of an AASLD single topic conference. Hepatology 2003;37: [3] Fan JG, Zhu J, Li XJ, Chen L, Li L, Dai F, et al. Prevalence of and risk factors for fatty liver in a general population of Shanghai, China. J Hepatol 2005;43: [4] Amarapurkar DN, Hashimoto E, Lesmana LA, Sollano JD, Chen PJ, Goh KL. How common is non-alcoholic fatty liver disease in the Asia-Pacific region and are there local differences. J Gastroenterol Hepatol 2007;22: [5] Marchesini G, Brizi M, Bianchi G, Tomassetti S, Bugianesi E, Lenzi M, et al. Nonalcoholic fatty liver disease: a feature of the metabolic syndrome. Diabetes 2001;50: [6] Marchesini G, Bugianesi E, Forlani G, Cerrelli F, Lenzi M, Manini R, et al. Nonalcoholic fatty liver, steatohepatitis, and the metabolic syndrome. Hepatology 2003;37: [7] Oh MK, Winn J, Poordad F. Review article: diagnosis and treatment of nonalcoholic fatty liver disease. Aliment Pharmacol Ther 2008;28: [8] Wieckowska A, Feldstein AE. Diagnosis of nonalcoholic fatty liver disease: invasive versus noninvasive. Semin Liver Dis 2008;28: [9] Angulo P. Nonalcoholic fatty liver disease. N Engl J Med 2002;346: [10] Shackel NA, McCaughan GW. Liver biopsy: is it still relevant. Intern Med J 2006;36: [11] Emanuele E. Is biopsy always necessary? Toward a clinico-laboratory approach for diagnosing nonalcoholic steatohepatitis in obesity. Hepatology 2008;48: [12] Ward DG, Suggett N, Cheng Y, Wei W, Johnson H, Billingham LJ, et al. Identification of serum biomarkers for colon cancer by proteomic analysis. Br J Cancer 2006;94: [13] Chapman K. The ProteinChip Biomarker System from Ciphergen Biosystems: a novel proteomics platform for rapid biomarker discovery and validation. Biochem Soc Trans 2002;30: [14] Wang JX, Yu JK, Wang L, Liu QL, Zhang J, Zheng S. Application of serum protein fingerprint in diagnosis of papillary thyroid carcinoma. Proteomics 2006;6: [15] Chen YD, Zheng S, Yu JK, Hu X. Artificial neural networks analysis of surfaceenhanced laser desorption/ionization mass spectra of serum protein pattern distinguishes colorectal cancer from healthy population. Clin Cancer Res 2004;10: [16] Trak-Smayra V, Dargere D, Noun R, Albuquerque M, Yaghi C, Gannage-Yared MH, et al. Serum proteomic profiling of obese patients: correlation with pathology and evolution after bariatric surgery. Gut 2009;58: [17] Zeng MD, Fan JG, Lu LG, Li YM, Chen CW, Wang BY, et al. Guidelines for the diagnosis and treatment of nonalcoholic fatty liver diseases. J Dig Dis 2008;9: [18] Li Y, Xu C, Yu C, Xu L, Miao M. Association of serum uric acid level with nonalcoholic fatty liver disease: a cross-sectional study. J Hepatol 2009;50: Journal of Hepatology 2012 vol. 56 j

7 JOURNAL OF HEPATOLOGY [19] Brunt EM, Janney CG, Di BAM, Neuschwander-Tetri BA, Bacon BR. Nonalcoholic steatohepatitis: a proposal for grading and staging the histological lesions. Am J Gastroenterol 1999;94: [20] Adams LA, Sanderson S, Lindor KD, Angulo P. The histological course of nonalcoholic fatty liver disease: a longitudinal study of 103 patients with sequential liver biopsies. J Hepatol 2005;42: [21] Xu C, Yu C, Xu L, Miao M, Li Y. High serum uric acid increases the risk for nonalcoholic fatty liver disease: a prospective observational study. PLoS One 2010;5:e [22] Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika 1994;81: [23] Yilmaz Y, Ulukaya E, Dolar E. Serum M30 levels: a potential biomarker of severe liver disease in nonalcoholic fatty liver disease and normal aminotransferase levels. Hepatology 2009;49:697. [24] Pieper R, Gatlin CL, Makusky AJ, Russo PS, Schatz CR, Miller SS, et al. The human serum proteome: display of nearly 3700 chromatographically separated protein spots on two-dimensional electrophoresis gels and identification of 325 distinct proteins. Proteomics 2003;3: [25] Rosty C, Christa L, Kuzdzal S, Baldwin WM, Zahurak ML, Carnot F, et al. Identification of hepatocarcinoma intestine pancreas/pancreatitis-associated protein I as a biomarker for pancreatic ductal adenocarcinoma by protein biochip technology. Cancer Res 2002;62: [26] Younossi ZM, Jarrar M, Nugent C, Randhawa M, Afendy M, Stepanova M, et al. A novel diagnostic biomarker panel for obesity-related nonalcoholic steatohepatitis (NASH). Obes Surg 2008;18: [27] Yilmaz Y, Dolar E. Biomarkers for noninvasive biochemical diagnosis of nonalcoholic steatohepatitis: tools or decorations. World J Gastroenterol 2009;15: [28] Buttari B, Profumo E, Petrone L, Pietraforte D, Siracusano A, Margutti P, et al. Free hemoglobin: a dangerous signal for the immune system in patients with carotid atherosclerosis. Ann NY Acad Sci 2007;1107: [29] Rother RP, Bell L, Hillmen P, Gladwin MT. The clinical sequelae of intravascular hemolysis and extracellular plasma hemoglobin: a novel mechanism of human disease. JAMA 2005;293: [30] Huang P, Lo LH, Chen YC, Lin RT, Shiea J, Liu CK. Serum free hemoglobin as a novel potential biomarker for acute ischemic stroke. J Neurol 2009;256: [31] Xu L, Xu CF, Yu CH, Miao M, Li YM. Haemoglobin and non-alcoholic fatty liver disease: further evidence from a population-based study. Gut 2009;58: [32] Takeuchi M, Okamoto K, Takagi T, Ishii H. Ethnic difference in inter-east Asian subjects with normal glucose tolerance and impaired glucose regulation: a systematic review and meta-analysis focusing on fasting serum insulin. Diabetes Res Clin Pract 2008;82: [33] Nakagawa T, Muramoto Y, Hori M, Mihara S, Marubayashi T, Nakagawa K. A preliminary investigation of the association between haptoglobin polymorphism, serum ferritin concentration and fatty liver disease. Clin Chim Acta 2008;398: [34] Hanley AJ, Karter AJ, Festa A, D Agostino Jr R, Wagenknecht LE, Savage P, et al. Factor analysis of metabolic syndrome using directly measured insulin sensitivity: the Insulin Resistance Atherosclerosis Study. Diabetes 2002;51: [35] Resnick HE, Jones K, Ruotolo G, Jain AK, Henderson J, Lu W, et al. Insulin resistance, the metabolic syndrome, and risk of incident cardiovascular disease in nondiabetic American Indians: the Strong Heart Study. Diabetes Care 2003;26: [36] Jeppesen J, Hansen TW, Rasmussen S, Ibsen H, Torp-Pedersen C, Madsbad S. Insulin resistance, the metabolic syndrome, and risk of incident cardiovascular disease: a population-based study. J Am Coll Cardiol 2007;49: Journal of Hepatology 2012 vol. 56 j