Biliary atresia (BA) is a serious but rare disease affecting neonatal infants. BA occurs in approximately 1 in 15,000 children and is more common in girls than in boys. The cause of BA is unknown and treatments are only partially successful (1,2). Early disease detection and surgical intervention using the Kasai procedure correlates with a good long-term outcome. Hence, early diagnosis is particularly important for patients with BA (3,4). Presently, there are no noninvasive diagnostic methods that can clearly identify infants with BA. Present definitive diagnosis requires a cholangiogram, liver biopsy, and surgery (5). Although laparoscopy-assisted cholangiography is a simple, accurate, and safe method for the diagnosis of prolonged jaundice in infants, there are some intrinsic limitations, such as invasiveness, and need for an experienced team (6,7). Thus, there is a need for blood assays that facilitate the differentiation between BA and other neonatal cholestatic diseases.
Recent advances in proteomics offer opportunities for discovering biomarkers in biological fluids, including serum. Surface-enhanced desorption/ionization time-of-flight mass spectrometry (SELDI-TOF-MS) coupled with ProteinChip array technology provides an efficient and sensitive means for biomarker discovery. This approach enables quantitative measurement of specific peptides/proteins in a high-throughput manner, while only requiring a small amount of material for the analysis. SELDI-TOF-MS, which generates a protein fingerprint by MS, has proven to be an attractive tool for clinical applications (8,9). This technology has been successfully applied to facilitate specific biomarker discovery for various cancers, including hepatocellular carcinoma, thyroid carcinoma, and pancreatic cancer (10–12). SELDI-TOF-MS analysis has been successfully used to identify specific biomarkers of pediatric surgery diseases, including nephroblastoma and Hirschsprung disease (13–15), and to determine genetic risk in preimplantation genetic diagnosis using maternal plasma (16). A few pilot studies have reported the potential of 2-dimensional difference gel electrophoresis (2-DE) to identify serum biomarkers that differentiate BA from other types of neonatal cholestasis and healthy subjects (17,18); however, no specific protein biomarkers were identified and validated in these reports. Additionally, proteome coverage by 2-DE is experimentally limited to proteins with molecular weights in the 10- to 120-kDa range. In comparison, the SELDI-TOF-MS technique is suitable for relatively small (<20 kDa) proteins and protein fragments (19).
In the present study, we used the SELDI-TOF-MS technology to screen for potential protein patterns specific to patients with BA and then purified the candidate protein biomarkers using high-performance liquid chromatography (HPLC). The isolated target biomarkers were identified by liquid chromatography tandem mass spectrometry (LC-MS/MS) and confirmed by enzyme-linked immunosorbent assay (ELISA).
Patients, Controls, and Serum Samples
Serum samples were obtained from 42 infants with BA, 38 non-BA neonatal cholestasis (NC) infants, and 36 healthy controls (HC). The 3 groups were similar in age and sex distribution (Table 1). Infants with BA and NC were randomly selected from patients who had been identified by pathological diagnosis at the Children's Hospital of Fudan University. HC were selected from children who were healthy (as determined by physical examination of the serum microelement in normal babies; the microelement means determination of zinc, copper, manganese, and selenium in the serum of normal children) and had not been diagnosed as having any other disease. Children with BA and NC were identified based on operative cholangiogram and liver pathology. Entry criteria for the BA study included type III classification of the BA phenotype, age younger than 90 days, and serum direct or conjugated bilirubin levels >20% of total and >2 mg/dL. Children with liver failure, malignancy, hypoxia, shock, ischemic hepatopathy within the preceding 2 weeks, extracorporeal membrane oxygenation–associated cholestasis, and/or prior hepatobiliary surgery were excluded from the present study. Children with primary hemolytic disease, drug- or total parenteral nutrition–associated cholestasis, bacterial or fungal sepsis, or birth weight <1500 g were also excluded from the present study unless they were definitively diagnosed as having BA or another cholestatic disease. Serum was obtained within a few days of enrollment from untreated patients. Our studies have been reviewed and approved by the ethics committee of the Children's Hospital of Fudan University. Informed consent was given by the participant's parents before we collected blood samples. Serum was left at room temperature for 1 hour, centrifuged at 3000 rpm for 10 minutes, and then stored at −80°C until analysis.
SELDI-TOF-MS Analysis of Serum Protein Profiles
Frozen serum samples were thawed on ice and centrifuged at 10,000 rpm for 5 minutes at 4°C. Each serum sample (10 μL) was denatured by the addition of 20 μL of U9 buffer (9 mol/L urea, 2% 3-[(3-cholanidopropyl)dimethylammonio]-1-propanesulfonate, 50 mmol/L Tris–HCl, 1% DL-dithiothreitol, pH 9.0) and vortexed at 4°C for 30 minutes. Each sample was then diluted in 108 μL of low-stringency buffer (0.1 mol/L sodium acetate (NaAc), pH 4.0). Diluted serum sample (100 μL) was then hybridized with the weak cation exchange (WCX2) protein-chip array (Ciphergen Biosystems, Fremont, CA), which was controlled by a bioprocessor and preactivated twice with 150-μL low-stringency buffer at room temperature for 5 minutes. The diluted serum sample was allowed to react with the surface of the WCX2 chip for 60 minutes at room temperature. Each spot was then washed 3 times with NaAc buffers of various pH values and ionic strengths to eliminate nonadsorbed proteins. After allowing the surface of the array to air dry, 1 μL of saturated sinapinic acid matrix in 50% acetonitrile (ACN) and 0.5% trifluoro-acetic acid (TFA) were applied and allowed to dry. Mass spectrometric (MS) analysis was performed using a PBS-II ProteinChip reader (Ciphergen Biosystems). Mass peak detection was analyzed using ProteinChip Biomarker Software version 3.1 (Ciphergen Biosystems). The mass spectra of the proteins were generated using an average of 140 laser shots at a laser intensity of 170 AU, and the detector sensitivity was set at 6. For data acquisition of low-molecular-weight proteins, the optimizing detection mass range was set from 2 to 20 kDa for all of the sample profiles examined. The instrument was calibrated by the all-in-one peptide molecular mass standard.
Bioinformatics and Biostatistics
Serum samples were split into a training set and a testing set. A total of 22 BA, 20 NC, and 20 HC samples were randomly selected for the training sample set. To evaluate the accuracy and validity of the classification model, the remaining 20 BA, 18 NC, and 16 HC samples were selected as the test set. First, the undecimated discrete wavelet transform method was used to denoise the signals. Second, the spectra were subjected to baseline correction by aligning with a monotone local minimum curve and mass calibration. The proteomic peaks were detected and quantified by an algorithm that takes the maximal height of every denoised, baseline-corrected, and calibrated mass spectrum into account. Third, the peaks were filtered to maintain a signal-to-noise ratio (S/N) of >3. The S/N of a peak is the ratio of the height of the peak above the baseline to the wavelet-defined noise. Finally, to match peaks across spectra, we pooled the detected peaks if the relative difference in their mass sizes was ≤0.3%. The minimal percentage of each peak, appearing in all of the spectra, was specified to 10. The matched peaks across spectra were defined as a peak cluster. If a spectrum did not have a peak within a given cluster, then the maximal height within the cluster was assigned to its peak value. The normalization was performed only with the identified peak clusters. To distinguish between data of different groups, we used a nonlinear support vector machine classifier, originally developed by Vapnik, with a radial-based function kernel, a parameter gamma of 0.6, and a cost of the constrain violation of 19. The leave-one-out cross-validation method was applied to estimate the accuracy of this classifier. The capability of each peak in distinguishing data of different groups was estimated by the P value obtained using the Wilcoxon t test. P values of ≤0.01 were considered to be statistically significant. The remaining samples were analyzed to test the classification model. Samples were then categorized based on their proteomic profile characteristics. The sensitivity was defined as the probability of predicting BA, whereas the specificity was defined as the probability of predicting control samples. The positive predictive value was defined as the probability of BA.
Serum samples were mixed with U9 buffer (1:2, v/v) and incubated for 30 minutes at room temperature. Samples were then diluted in 5-mL WCX binding buffer (50 mmol/L NaAc, pH 4.0) and loaded onto the CM Ceramic Hyper DWCX SPE column (Pall Life Science, Port Washington, NY). After washing with 2 mL of WCX binding buffer, the column was eluted with 5 mL of eluting buffer (2 mol/L NaCl, 50 mmol/L NaAc, pH 4.0) at a flow rate of 0.5 mL/minute. The eluted fraction was further purified using HPLC.
Purification of the Candidate Protein Markers Using HPLC
HPLC separation was performed using a SCL-10AVP (Shimadzu, Tokyo, Japan) with a Sunchrom C18 column (Great Eur-Asia Sci-Tech Development, Beijing, China) and a C18 guard column (Shimadzu). The mobile phase consisted of solvent A (5% ACN, 0.1% TFA) and solvent B (90% ACN, 0.1% TFA). HPLC separation was achieved with a linear solvent gradient: 100% A (0 min) 15% B (15 min) 65% B (65 min) 100% B (100 min) at a flow rate of 0.5 mL/min. The eluate was detected at multiple wavelengths of 214, 254, and 280 nm. Each peak fraction was collected and concentrated using a SpeedVac, and then analyzed using an AXIMA-CFRTM plus MALDI-TOF mass spectrometer (Shimadzu/Kratos, Manchester, UK) operating in the linear mode to trace the candidate protein biomarkers with a-cyano-4-hydroxycinnamic acid as matrix.
Identification of the Candidate Protein Biomarkers by LC-MS/MS
In-solution digestion of each concentrated fraction, which contains 1 candidate protein biomarker, was performed using a standard protocol. Briefly, each fraction was dissolved in 25 mmol/L NH4HCO3, reduced with 10-mmol/L DTT for 1 hour, alkylated by incubating in 40-mmol/L iodacetamide in the dark for 45 minutes at room temperature, and the reaction quenched by adding 40 mmol/L DTT for 30 minutes at room temperature. Protease K (0.1 μg; Promega Corp, Madison, WI) was then added into the sample solution and incubated for 45 minutes at 37°C. The digestion was stopped by adding formic acid to a final concentration of 0.1%. The digested sample was loaded onto a C18 column packed with Sunchrom packing material (SP-120–3-ODS-A, 3 μm) and flow-through analyzed by nano-LC-ESI (electrospray ionization)-MS/MS. The LTQ mass spectrometer was operated in the data-dependent mode in which the initial MS scan recorded the mass-to-charge (m/z) ratios of ions over the mass range from 400 to 2000 Da. The 5 most abundant ions were automatically selected for subsequent collision-activated dissociation. All of the MS/MS data were searched against a human protein database downloaded from the National Center for Biotechnology Information database using the SEQUEST program (Thermo Finnigan, San Jose, CA).
ELISA Measurement of the Serum Levels of Apo C-II and Apo C-III
Duplicate serum levels of Apo C-II and Apo C-III were measured by Apo C-II and Apo C-III ELISA kits (R&D Systems, Minneapolis, MN), respectively. The absorbance was measured at 450 nm on a microplate reader within 10 minutes.
Data were presented as mean ± standard deviation. All of the statistical analyses were performed using SPSS 13.0 software (SPSS, Chicago, IL). One-way analysis of variance was applied in the MS and ELISA data analyses between the BA and NC groups, the BA and HC groups, and the NC and HC groups. The rank-sum test was used to analyze nonhomogeneous data. P < 0.05 was regarded as statistically significant.
Serum Protein Profiles and Data Processing
Serum samples from the training set were analyzed by SELDI-TOF-MS with the WCX2 chip. Baseline readings were subtracted from the MS data followed by normalization using the total ion current. Peak clusters were generated using the Biomarker Wizard software. After carrying out Wilcoxon rank-sum tests to ascertain the relative signal strength, 25 peaks with P < 0.05 were obtained. Twenty-one protein peaks were found to be upregulated and 4 peaks were found to be downregulated in the BA group. From the random combination of protein peaks with remarkable variation, support vector machine screened out the combined model with the maximum Youden index of the predicted value, identifying 2 markers positioned at 8697 and 9098 Da. The average peak intensities of the 8697-Da protein in the BA, NC, and HC groups were 39.99 ± 15.07, 29.13 ± 13.93, and 28.01 ± 11.81, respectively. Expression of the 8697-Da protein was significantly higher in the BA group when compared with the NC and HC groups (P < 0.01). Expression of the 9098-Da protein was found to be significantly lower in the BA group (9.770 ± 5.221) when compared with the HC group (13.36 ± 7.110) (P < 0.01), whereas its expression was significantly higher compared with the NC group (7.017 ± 3.810) (P < 0.01) (Fig. 1). By combining the 2 potential markers and using the leave-one-out cross-validation method, the sensitivity of discriminating 22 BA and 20 NC samples was 95% and its specificity was 93%.
Protein Peak Validation
The remaining 20 BA, 18 NC, and 16 HC serum samples, representing the test set, were analyzed to validate the accuracy and validity of the classification model derived from the training set. The classification model differentiated the BA samples from the NC samples with 94.1% sensitivity and 91.8% specificity. The positive predictive value was found to be 87.5%. The area under the receiver operating characteristics curve of this model was 0.984.
Purification and Identification of the Candidate Protein Biomarkers
Serum samples from HC were used for the purification of the 9098-Da protein, and serum samples from BA were used for the purification of the 8697-Da protein using WCX solid-phase extraction and HPLC. Figure 2 shows the results of MALDI-TOF-MS analysis of the 2 purified candidate protein biomarkers. After digestion with modified trypsin, the peptide mixture was analyzed by nano-LC-MS/MS. Table 2 shows the identification results of the 2 candidate protein biomarkers, Apo C-II (GI: 757915) and Apo C-III (GI: 757913). The high sequence coverage and accurate molecular weight measurement by MALDI-TOF-MS resulted in the whole sequence of the 2 candidate protein markers.
ELISA Analysis of the 2 Candidate Protein Biomarkers
The expression levels of Apo C-II in the BA, NC, and HC groups were 67.55 ± 3.528, 59.83 ± 4.964, and 99.32 ± 9.069 ng/mL, respectively. The levels of Apo C-III in the BA, NC, and HC groups were 1113 ± 87.71, 703.4 ± 67.82, and 786.5 ± 75.0 ng/mL, respectively. Statistical analysis showed that the Apo C-II expression was significantly lower in the BA group and significantly higher in the NC group compared with the HC group (P < 0.05). Apo C-III expression was significantly elevated in the BA group compared with both the NC and HC groups (P < 0.05).
BA is a destructive inflammatory obliterative cholangiopathy affecting some neonates that differentially affects both intrahepatic and extrahepatic bile ducts. The progression of biliary tract obstruction diseases is associated with proteomic changes, which may be partially reflected by the changes in the blood composition of patients. For instance, the lipopolysaccharide-binding protein mediates lipopolysaccharide-induced liver injury and mortality in the setting of biliary obstruction (20). Proteins are sometimes secreted by cholangiocytes (biliary epithelium cells) or hepatocytes in response to the presence of a biliary obstruction disease or the hallmark of predisposing factors to BA occurrence (21–23). Because whole blood is in contact with essentially all of the tissues in the human body and provides a dynamic reflection of the physiological and pathological status, serum often changes before the detection of other clinical symptoms and has diagnostic value for early disease detection (24). For these reasons, the use of serum biomarkers has an advantage over other available screening methods for early diagnosis of diseases. Importantly, the experimental procedure described in the present study did not involve prefractionation and albumin depletion, which are methods adopted for studying serum and inevitably may eliminate low-abundance and low-molecular-weight disease markers. Use of serum without pretreatment also affords high-throughput and a cost-effective approach.
In the present study, Apo C-II was found to be downregulated in serum from patients with BA compared with healthy infants, which is consistent with previously reported data (18); however, Apo C-II was upregulated in serum from patients with BA compared with patients with non-BA NC. Consistent with our data, previous reports demonstrated that Apo C-II expression was elevated in individuals with primary biliary cirrhosis and BA compared with non-BA NC (17,25). Apo C-II is a protein component of very-low-density lipoproteins and chylomicrons. Apo C-II plays a crucial role in lipoprotein metabolism as an activator of lipoprotein lipase, which hydrolyzes triglycerides and thus provides free fatty acids for cells. Mutations in this gene cause hyperlipoproteinemia type IB, characterized by hypertriglyceridemia, xanthomas, and increased risk of pancreatitis and early atherosclerosis. Low Apo C-II levels in infants with BA-associated liver cirrhosis indicate deficient triglyceride metabolism, which is consistent with a previous report (26). A 2007 study also suggested that Apo C-II is a novel substrate for matrix metalloproteinases; Apo C-II deficiency correlates with the pathophysiology of liver fibrosis associated with BA (18).
Apo C-III was upregulated in serum from patients with BA compared with patients with non-BA NC and healthy infants. Apo C-III is a protein component of extremely-low-density lipoprotein. Apo C-III also plays an important role in the catabolism of triglyceride-rich lipoproteins because it is a potent inhibitor of lipoprotein lipase. Mutations in this gene cause coronary heart disease in normal as well as noninsulin diabetes mellitus. In addition, it should be noted that the use of serum lipoprotein-X (LP-X) levels was previously identified as a method to assist in the differentiation of BA from other types of NC (27). LP-X is a unilamellar liposome with high free cholesterol and phospholipid levels that accumulates in the serum during biliary tract obstruction (28,29). LP-X is also found in individuals with deficiency in lecithin cholesterol acyl transferase (LCAT), an enzyme that produces cholesterol ester from free cholesterol and lecithin. Interestingly, Apo C-III inhibits lecithin cholesterol acyl transferase (30), which is consistent with the view that LP-X may increase in infants with BA because of an increase in Apo C-III.
It is tempting to speculate what role Apo C-II and Apo C-III may play in BA progression, given what we know about the function of these 2 proteins; however, our data indicate that Apo C-II and Apo C-III are good candidate biomarkers for the diagnosis of BA in infants. How the expression of these biomarkers changed during disease progression was not addressed in our study, but future studies focusing on this topic may be clinically useful.
Two studies similar to our study were performed. Neither of these studies identified Apo C-III as a candidate BA biomarker; this is unique to our study. One study only compared BA samples to healthy samples (18). Therefore, the present study did not determine whether Apo C-II expression could distinguish between BA and non-BA NC samples. Also, they only used 2 control samples for the present study. The other study compared BA and non-BA NC samples; however, they did not include HC (17). Also, our study included a larger number of samples. Therefore, we believe our study is the most comprehensive BA biomarker screen performed to date. Furthermore, our study used a different procedure that allowed for screening of low-abundance and low-molecular-weight proteins. This different procedure is also easier to perform because prefractionation and albumin depletion are not required. Also, use of serum without pretreatment affords a high-throughput and cost-effective approach.
In summary, we have identified a set of protein markers that distinguish infants with BA from non-BA NC and healthy infants, Apo C-II and Apo C-III. This panel of markers is likely to distinguish infants with BA from non-BA NC and healthy infants. Further studies need to be performed with larger sample sizes to verify the specific protein markers and with additional populations or using prediagnostic serum to confirm the importance of these findings as diagnostic markers of infants with BA.
1. Hartley J, Harnden A, Kelly D. Biliary atresia. BMJ 2010; 340:c2383.
2. Santos JL, Carvalho E, Bezerra JA. Advances in biliary atresia: from patient care to research. Braz J Med Biol Res 2010; 43:522–527.
3. Chitsaz E, Schreiber RA, Collet JP, et al. Biliary atresia: the timing needs a changin’. Can J Public Health 2009; 100:475–477.
4. Wadhwani SI, Turmelle YP, Nagy R, et al. Prolonged neonatal jaundice and the diagnosis of biliary atresia: a single-center analysis of trends in age at diagnosis and outcomes. Pediatrics 2008; 121:e1438–e1440.
5. Khalil BA, Perera MT, Mirza DF. Clinical practice: management of biliary atresia. Eur J Pediatr 2010; 169:395–402.
6. Hay SA, Soliman HE, Sherif HM, et al. Neonatal jaundice: the role of laparoscopy. J Pediatr Surg 2000; 35:1706–1709.
7. Tang ST, Li SW, Ying Y, et al. The evaluation of laparoscopy-assisted cholangiography in the diagnosis of prolonged jaundice in infants. J Laparoendosc Adv Surg Tech A 2009; 19:827–830.
8. Luo J, Qian JH, Yu JK, et al. Discovery of altered protein profiles in epithelial ovarian carcinogenesis by SELDI mass spectrometry. Eur J Gynaecol Oncol 2008; 29:233–238.
9. Tomosugi N. Discovery of disease biomarkers by ProteinChip system; clinical proteomics as noninvasive diagnostic tool. Rinsho Byori 2004; 52:973–979.
10. Chen L, Ho DW, Lee NP, et al. Enhanced detection of early hepatocellular carcinoma by serum SELDI-TOF proteomic signature combined with alpha-fetoprotein marker. Ann Surg Oncol 2010; 17:2518–2525.
11. Fan Y, Shi L, Liu Q, et al. Discovery and identification of potential biomarkers of papillary thyroid carcinoma. Mol Cancer 2009; 8:79.
12. Liu D, Cao L, Yu J, et al. Diagnosis of pancreatic adenocarcinoma using protein chip technology. Pancreatology 2009; 9:127–135.
13. Wang JX, Qin P, Liu QL, et al. Detection and significance of serum protein marker of Hirschsprung disease. Pediatrics 2007; 120:e56–e60.
14. Zhang Q, Wang J, Dong R, et al. Identification of novel serum biomarkers in child nephroblastoma using proteomics technology. Mol Biol Rep 2011; 38:631–638.
15. Sandoval JA, Dobrolecki LE, Huang J, et al. Neuroblastoma detection using serum proteomic profiling: a novel mining technique for cancer? J Pediatr Surg 2006; 41:639–646.
16. Li Y, Altarescu G, Renbaum P, et al. Non-invasive prenatal diagnosis using cell-free fetal DNA in maternal plasma from PGD pregnancies. Reprod Biomed Online 2009; 19:714–720.
17. Wang H, Malone JP, Gilmore PE, et al. Serum markers may distinguish biliary atresia from other forms of neonatal cholestasis. J Pediatr Gastroenterol Nutr 2010; 50:411–416.
18. Lee CW, Lin MY, Lee WC, et al. Characterization of plasma proteome in biliary atresia. Clin Chim Acta 2007; 375:104–109.
19. Monteoliva L, Albar JP. Differential proteomics: an overview of gel and non-gel based approaches. Brief Funct Genomic Proteomic 2004; 3:220–239.
20. Minter RM, Bi X, Ben-Josef G, et al. LPS-binding protein mediates LPS-induced liver injury and mortality in the setting of biliary obstruction. Am J Physiol Gastrointest Liver Physiol 2009; 296:G45–G54.
21. Alpini G, Lenzi R, Zhai WR, et al. Bile secretory function of intrahepatic biliary epithelium in the rat. Am J Physiol 1989; 257:G124–G133.
22. Whitington PF, Malladi P, Melin-Aldana H, et al. Expression of osteopontin correlates with portal biliary proliferation and fibrosis in biliary atresia. Pediatr Res 2005; 57:837–844.
23. Onori P, Franchitto A, Alvaro D, et al. Immunohistochemical features of bile duct epithelial cells in normal and experimental liver conditions. Ital J Anat Embryol 2001; 106:371–378.
24. Maurya P, Meleady P, Dowling P, et al. Proteomic approaches for serum biomarker discovery in cancer. Anticancer Res 2007; 27:1247–1255.
25. Koga S, Miyata Y, Ibayashi H. Plasma lipoproteins and apoproteins in primary biliary cirrhosis. Hepatology 1985; 5:286–292.
26. Andersson Y, Majd Z, Lefebvre AM, et al. Developmental and pharmacological regulation of apolipoprotein C-II gene expression. Comparison with apo C-I and apo C-III gene regulation. Arterioscler Thromb Vasc Biol 1999; 19:115–121.
27. Tazawa Y, Yamada M, Nakagawa M, et al. Significance of serum lipoprotein-X and gammaglutamyltranspeptidase in the diagnosis of biliary atresia. A preliminary study in 27 cholestatic young infants. Eur J Pediatr 1986; 145:54–57.
28. Miller JP. Dyslipoproteinaemia of liver disease. Baillieres Clin Endocrinol Metab 1990; 4:807–832.
29. Felker TE, Hamilton RL, Havel RJ. Secretion of lipoprotein-X by perfused livers of rats with cholestasis. Proc Natl Acad Sci U S A 1978; 75:3459–3463.
30. Albers JJ, Lin J, Roberts GP. Effect of human plasma apolipoproteins on the activity of purified lecithin: cholesterol acyltransferase. Artery 1979; 5:61–75.