Identification of a new marker of hepatocellular carcinoma by serum protein profiling of patients with chronic liver diseases


  • Conflict of interest: Nothing to report.


Surface-enhanced laser desorption ionization time-of-flight mass spectrometry (SELDI-TOF MS) is a proteomic technique that enables the profiling of proteins present in any biological material studied. We used this approach to identify new biomarkers of hepatocellular carcinoma (HCC) in the sera of patients with cirrhosis. Sera from 82 patients with cirrhosis, either without (n = 38) or with (n = 44) HCC, were analyzed by SELDI-TOF MS, and the results of the two groups were compared. The most efficient protein peaks leading to discrimination of patients with HCC were selected (receiver operative characteristic curves). The highest-scoring peak combination was established in a first group of serum samples (multinomial regression) and was tested in an independent group. The protein corresponding to the highest discrimination was purified and characterized further. The intensity of 30 protein peaks significantly differed between cirrhotic patients with and without HCC. An algorithm including the six highest-scoring peaks allowed correct classification (presence or absence of HCC) of 92.5% of patients in the test sample set and 90% in the validation sample set. The highest discriminating peak (8,900 Da) was purified further and was characterized as the C-terminal part of the V10 fragment of vitronectin. An in vitro study suggested that the increase of the 8,900-Da fragment in the serum of patients with HCC may proceed from the cleavage of native vitronectin with metalloproteases, a family of enzymes whose activity is enhanced in HCC. In conclusion, global protein profiling is an efficient approach that enabled us to identify a catalytic fragment of vitronectin as a new serum marker of HCC in patients with chronic liver diseases. (HEPATOLOGY 2005;41:40–47.)

Identification of reliable, reproducible, and noninvasive markers of hepatocellular carcinoma (HCC) has a significant impact on public health, especially in light of the high burden of hepatitis C virus chronic infection. Indeed, it is estimated that more than 170 million persons are infected worldwide by hepatitis C virus, and, concomitant with the progression of the disease, HCC is now the major cause of death in patients with cirrhosis in Europe.1, 2 In light of the poor prognosis in advanced HCC, and to improve patient survival significantly, detection of HCC at its earliest stage of development is mandatory. Up to now, HCC diagnosis has relied on several tools combining imaging techniques and measurement of serum alfa-fetoprotein (AFP).3 Although both procedures are relatively efficient for large tumors, the specificity of serum AFP is low, especially against a background of chronic hepatitis.4

Development of new technologies for isolation of serum biomarkers in cancer diseases currently is in progress. The recent development of proteomic array technology, including serum protein profiling, coupling ProteinChip Array with surface-enhanced laser desorption ionization time-of-flight mass spectrometry (SELDI-TOF MS; SELDI-TOF MS ProteinChip technology, Ciphergen Biosystems, Fremont, CA), provides a potentially powerful tool for the discovery of new markers, as recently demonstrated in human immunodeficiency virus infection and in patients with prostate, pancreatic, or ovarian malignancies.5–10 This procedure results in the identification of protein profiles composed of isolated or clustered peaks characterized by molecular weight, enabling the discrimination between different pathological conditions with high sensitivity and significant reproducibility.

The aim of this study was to assess and compare protein expression profiles in sera of patients without or with HCC in the context of chronic liver diseases using SELDI-TOF MS ProteinChip technology. This strategy enabled us to define an optimum discriminatory protein profile in patients with HCC and to identify a new serum marker associated with the presence of HCC.


HCC, hepatocellular carcinoma; AFP, alfa-fetoprotein; SELDI-TOF MS, surface-enhanced laser desorption ionization time-of-flight mass spectrometry; m/z, mass-to-charge ratio; AUC, area under the curve; MMP, metalloprotease.

Patients and Methods

Study Population.

With the patients' consent, serum samples were collected retrospectively at one institution (Department of Hematology and Department of Hepatology, Beaujon Hospital) from two sets of patients: (1) patients with cirrhosis without a suspected or malignant nodule (n = 38), and (2) patients with HCC that developed during the course of chronic liver diseases (n = 44). Diagnosis of HCC relied on the presence of a malignant liver nodule, as established on imaging techniques (n = 4) or by histological analysis (n = 40) of liver biopsies, surgical resections, or explanted livers. In this group, sera were obtained at the time of HCC diagnosis and were frozen immediately and stored at −80°C before use. For each patient, clinical data, including age, sex, cause of disease, staging of the chronic liver disease, and AFP measurement, were collected. For patients with HCC, size of tumors and histoprognostic features (grade of differentiation, presence of perilesional capsule, vascular invasion, and satellite nodules) were available for 32 and 25 patients, respectively.

Ciphergen ProteinChip SELDI-TOF MS Analysis.

To determine the best conditions for identifying the most discriminating serum protein profiles between patients with and without HCC, different experimental conditions (weak cationic exchange ProteinChip Array and binding buffer containing 50 mM sodium acetate, pH 5; weak anionic exchange ProteinChip Array, binding buffer containing sodium acetate 50 mM, pH 6, and immobilized metal ion affinity capture ProteinChip Array loaded with zinc; and binding buffer containing NaCl 0.5 M 1× phosphate-buffered saline) were tested. For this purpose, 14 serum samples (10 obtained from patients with overt HCC and 4 obtained from patients with hepatitis C virus–induced cirrhosis without HCC) were tested. These preliminary assays resulted in the selection of immobilized metal ion affinity capture ProteinChip Array loaded with zinc as the most effective chip array.

Next, one dilution (1:10) of each of the 82 serum samples was processed on immobilized metal ion affinity capture arrays according to the manufacturer's protocols. Briefly, the array spots were preactivated with 100 mM ZnCl2 for 15 minutes at room temperature, followed by one wash with H2O. Each serum sample was first diluted 1:10 with 7 M urea, 2 M thiourea, 4% Chaps, and 1% DTT. Five microliters of each diluted serum were spotted onto preactivated immobilized metal ion affinity capture arrays chips and incubated with 95 μL of binding buffer (phosphate-buffered saline 1×, 0.5 M sodium chloride, and 0.1% Triton X-100) for 30 minutes. After two washes with binding buffer (5 minutes each) and one wash with binding buffer without Triton X-100 and HEPES 1 mM, the air-dried arrays were saturated with sinapinic acid in 0.5% trifluoroacetic acid and 50% acetonitrile before being read on the instrument (Ciphergen ProteinChip Reader; Ciphergen Biosystems). All samples were tested during the same experiment.

The arrays were analyzed with the Ciphergen ProteinChip Reader (model PBS II). The mass spectra of proteins were generated by using an average of 195 laser shots at a laser intensity of 225 to 250 arbitrary units. For data acquisition of low–molecular weight proteins, the detection size range was between 3 and 30 kd. Detector sensitivity was set at 10 and laser intensity was set at 225. For the high–molecular weight proteins, the detection size range was between 30 and 150 kd. The detector sensitivity was set at 10 and the laser intensity was set at 250. The mass-to-charge ratio (m/z) of each of the proteins captured on the array surface was determined according to externally calibrated standards (Ciphergen Biosystems). According to the manufacturer, the mass accuracy of the spectrometer is 0.1%.

Intra-ProteinChip Array reproducibility was checked by spotting eight different aliquots of one sample on the same array, and inter-ProteinChip Array reproducibility was checked by including one given sample on every different array. The intra- and inter-ProteinChip Array coefficients of variation were assessed for all protein peaks of more than background according to the setting of detection. The means of intra- and inter-ProteinChip Array coefficient of variations were 10% and 25%, respectively.

Data Mining.

The data were analyzed with ProteinChip Software version 3.0.2 (Ciphergen Biosystems). For each comparison, the raw intensity data were normalized by using the total ion current of all profiles. The peak intensities were normalized to the total ion current of m/z between 1,500 and 30,000 Da for the low–molecular weight range and between 1,500 and 150,000 for the high–molecular weight range.

To characterize protein peaks of potential interest, serum profiling of patients with HCC (n = 44) and patients without HCC (n = 38) were compared. Mean peak intensity of each protein was calculated and compared (nonparametric test) in each group of serum samples, and their diagnostic value for the diagnosis of HCC was assessed by determination of the area under the curve (AUC) with receiver operator characteristic curves.

An index that combined the most discriminatory independent peaks was generated in a first set of 40 randomly selected serum samples (18 patients without and 22 patients with HCC). The best index for discrimination was the logistic regression function that combined the most discriminatory independent factors. Peaks of potential interest included in the model were those selected in the previous analysis. This index then was validated in a testing set of 42 serum samples obtained from the remaining 20 patients without and 22 patients with HCC. Finally, the combining index was tested further in the serum samples taken together.

Isolation and Identification of the Protein Marker.

To characterize fully the most discriminating peak, selected sera were fractionated first using BioSepra IMAC-Zn HyperCel spin columns (BioSepra, Cergy, France). Each of the fractions eluted from the spin columns was analyzed further on a hydrophilic NP20 ProteinChip Array (1 μL per spot) to monitor the elution and recovery of the protein of interest. The eluted fractions in which the marker was the most abundant were concentrated down to 10 μL using a Speed Vac and were loaded onto 16% tricine one-dimensional sodium dodecyl sulfate–polyacrylamide gel electrophoresis (Invitrogen, Carlsbad, CA). The samples were run in tricine sodium dodecyl sulfate buffer according to the manufacturer's instructions and were stained using Invitrogen SilverQuest protocols. Control serum samples of patients with cirrhosis without HCC were prepared in a similar way and were run in parallel to the liver cancer serum samples for comparison. The band corresponding to the potential marker—present only in the liver cancer serum samples—was excised from the gel, cut into pieces, and introduced into a siliconized 1.5-mL tube. Gel pieces were destained (4 times, 15 minutes each) with 50 μL of 100 mM ammonium bicarbonate, acetonitrile 50:50 (vol/vol). The gel pieces were dehydrated with acetonitrile (10 minutes) and then dried in Speed Vac (20 minutes). Sequencing grade modified porcine trypsin (Promega France, Charbonnières-les-bains, France) at 10 ng/μL in 25 mM ammonium bicarbonate was added to the gel pieces and was incubated overnight at 37°C.

The resulting peptide digest was added to two spots of an NP20 ProteinChip Array (2 μL digest with 1 μL of saturated CHCA [alpha-cyano-4-hydroxy-cinnamic acid] in ACN-H2O 50:50 [vol/vol] containing 0.1% trifluoroacetic acid). The trypsin digest of a blank gel piece was analyzed in parallel for comparison. Tandem MS data were acquired on a Micromass QTOF II (Manchester, United Kingdom) tandem quadruple TOF mass spectrometer equipped with a SELDI-TOF MS ProteinChip Interface PCI 1000. Ions were created using a pulsed nitrogen laser operating at 30 pulses per second. Nitrogen gas was used for collisional cooling of formed ions, and argon gas was used for collision-induced dissociation experiments. The system was calibrated externally in the MS/MS mode using the parent ion and four selected fragments of adrenocorticotropic hormone (ACTH) human fragment 18-39 (m/z = 2465-1983). The MS/MS fragment data were exported as Sequest files and were used for database searches with Mascot ( using the National Center for Biotechnology Information and SwissProt databases.

Real-Time Reverse-Transcriptase Polymerase Chain Reaction of Vitronectin in HCC.

Quantitative real-time reverse-transcriptase polymerase chain reaction was performed as described previously.11 Briefly, cDNA was made from total RNA extracted from cases of HCC (n = 8), cirrhosis (n = 8), normal liver (n = 8), and HepG2 cells, respectively, using the acid-phenol guanidium method. We quantified transcripts of the TBP (TATA box-binding protein) gene as the endogenous RNA control. Each sample was normalized on the basis of TBP content. Results, expressed as N-fold differences in target gene expression relative to the TBP gene (termed Ntarget), were determined by the following formula: Ntarget = 2δCtsample, where the δCt value of the sample was determined by subtracting the average Ct value of the target gene from the average Ct value of the TBP gene. The Ntarget values of the samples subsequently were normalized such that the mean ratio of the normal liver would equal a value of 1. The nucleotide sequences of the primers used for polymerase chain reaction amplification were as follows: (1) VTN-U (5′-GCACCCCTGAGACCCCTTC-3′) and VTN-L (5′-CCCTTGCATGACTCTTGGTCAG-3′), with a VTN-specific product size of 74 bp; and (2) TBP-U (5′-TGCACAGGAGCCAAGAGTGAA-3′) and TBP-L (5′-CACATCACAGCTCCCCACCA-3′), with a TBP-specific product size of 132 bp. Polymerase chain reaction was performed using the SYBR Green PCR Core Reagents kit (Perkin-Elmer Applied Biosystems, Courtaboeuf, France). Experiments were performed with duplicates for each data point.

In Vitro Degradation of Vitronectin by Metalloprotease (MMP)-2.

Reagents were purchased from the following commercial sources: human vitronectin (VWR International, Strasbourg, France), Pro MMP-2 (Euromedex, Souffelweyersheim, France). Before incubation with vitronectin, pro MMP-2 (1.8 μg) was activated by incubation with 4-aminophenylmercuric acetate 1 mM (Sigma, St. Louis, MO) overnight at 37°C, as previously described.12 Digestion of vitronectin was carried out by incubation of the substrate at 37°C for up to 24 hours (0, 4, 8, and 24 hours) with MMP-2 in an enzyme-to-substrate ratio of 1:20 in 50 mM Tris-HCl, pH 7.5, containing 0.15 M NaCl, CaCl2 10 mM, 0.05% Brij 35, and 0.02% NaN3. At each time point, 2 μL of the mix reaction were discarded and the products were analyzed by SELDI-TOF MS as follows: each sample was diluted (vol/vol) with saturated sinapinic acid in 0.5% trifluoroacetic acid and 50% acetonitrile. One microliter then was processed on NP20 (normal phase) arrays according to the manufacturer's protocols (Ciphergen Biosystems). The detector sensitivity was set between 8 and 10 and the laser intensity was set between 195 and 220.


Study Population.

The group of patients with HCC included 44 patients (40 men and 4 women; mean age, 62 years; range, 46-76 years). All patients displayed chronic liver disease related to hepatitis C virus infection (n = 15), hepatitis B virus infection or coinfection hepatitis C virus–hepatitis B virus (n = 6 and 2, respectively), alcohol consumption (n = 8), nonalcoholic steatohepatitis (n = 3), hemochromatosis in 4 patients, and unknown cause in 6 patients. Staging of chronic liver diseases was assessed in 40 patients according to the METAVIR system: no fibrosis (F0) was observed in 1 patient with nonalcoholic steatohepatitis, portal fibrosis without septa (F1) in 2 patients, portal fibrosis with few septa (F2) in 5 patients, portal fibrosis with numerous septa (F3) in 6 patients, and cirrhosis (F4) in 26 patients.13 The size of HCC was available for 32 patients (mean size, 6.1 cm; range, 1.3-22 cm). Multifocal HCC was present in 4 patients. Histological grade of the tumor was well, moderately, or poorly differentiated in 6, 18, and 1 patient, respectively. The mean value of the AFP measurement was available in 33 patients with HCC (mean value, 14,415; range, 2-317,500).

The group of cirrhotic patients without HCC comprised 38 patients (29 men and 9 women; mean age, 56 years; range, 34-76 years). Patients displayed liver cirrhosis related to hepatitis C virus infection (n = 25), hepatitis B virus infection or coinfection with hepatitis C virus–hepatitis B virus (n = 6 and 2, respectively), and alcohol consumption (n = 5). Measurement of AFP was available for 11 patients in this group (mean, 4.4; range, 2-14).

Serum Protein Profiling Associated With Cirrhosis Without or With HCC.

Using the immobilized metal ion affinity capture ProteinChip Array, more than 250 protein peaks ranging from 3 to 40 kd were generated for each serum, with 62 expressed in common in more than 20% of all serum samples (Fig. 1). Comparing peak intensities of serum from patients with HCC with those without HCC, mean intensity differed significantly for 30 peaks; 13 of them were significantly higher in the group of patients with HCC, whereas the other 17 peaks were higher in the group of patients without HCC. Diagnostic values of each of these 30 differentially expressed peaks were high, with an AUC of the receiver operator characteristic curve ranging from 0.63 to 0.85 according to the different protein peaks. The mean value of amplitude of these peaks with their AUC values for the two groups of patients is given in Table 1. The most performant peak for the diagnosis of HCC corresponded to an overexpressed protein in HCC with an m/z of 8,900 Da. The AUC of this protein was 0.85, its sensitivity was 85%, and its specificity was 75% for a peak intensity cutoff value of 6. By comparison, in the same group of patients with an AFP measurement available, the AFP value had an AUC of 0.72. Sensitivity and specificity were 60% and 73%, respectively, for a cutoff value of 20.

Figure 1.

Surface-enhanced laser desorption ionization time-of-flight mass spectrometry protein profile (2,500-12,500 mass-to-charge ratio) obtained from serum from a patient with liver cirrhosis and without hepatocellular carcinoma.

Table 1. Discriminatory Peaks and Mean Value Between Groups
m/zP ValueHepatocellular Carcinoma (n = 38)Cirrhosis (n = 44)AUC
8,900<.000141.91 ± 25.9219.26 ± 18.280.85
9,054<.000117.41 ± 12.754.26 ± 4.750.78
8,869<.000138.90 ± 31.089.19 ± 8.160.82
9,099<.000118.96 ± 15.606.83 ± 6.270.81
27,967<.00010.35 ± 0.350.80 ± 0.610.78
33,143<.00014.98 ± 3.248.70 ± 3.250.80
99,768<.00010.18 ± 0.110.32 ± 0.140.79
66,392<.000127.92 ± 16.244.58 ± 130.78
59,313<.00013.04 ± 2.395.73 ± 1.330.82
8,078.000229.79 ± 34.353.13 ± 2.010.73
7,901.00038.12 ± 6.714.24 ± 2.190.73
4,080.00045.24 ± 7.090.46 ± 0.980.72
33,233.00056.77 ± 3.139.35 ± 3.220.72
145,867.00060.005 ± 0.0050.08 ± 0.0050.72
8,267.000710.53 ± 10.053.50 ± 4.600.73
10,805.00082.03 ± 3.882.53 ± 2.130.71
82,380.0010.19 ± 0.270.41 ± 0.380.70
8,132.00117.90 ± 16.545.96 ± 3.820.71
4,486.00310.46 ± 8.275.18 ± 3.560.69
65,827.00314.92 ± 7.1119.36 ± 6.020.69
101,592.0070.17 ± 0.120.22 ± 0.0080.67
44,528.0081.27 ± 0.551.56 ± 0.510.66
78,941.0090.64 ± 0.480.86 ± 0.240.66
89,684.020.74 ± 0.541.04 ± 0.410.70
8,538.0212.73 ± 11.578.95 ± 9.440.65
14,581.031.34 ± 0.881.70 ± 0.870.65
15,997.0416.34 ± 15.9725.51 ± 20.130.64
9,508.045.76 ± 4.273.85 ± 2.860.64
11,654.046.71 ± 3.858.09 ± 3.240.63
9,353.059.92 ± 9.277.46 ± 8.360.63

The most performant peak combination for diagnosis of HCC was determined in a first set of sera from cirrhotic patients with or without HCC using regression analysis. The resulting algorithm, including six peaks, allowed correct classification of 92.5% of patients (37 of 40) according to the presence of HCC (AUC = 0.98). The six peaks included in the algorithm are labeled in Table 1. The performance of this combination was tested in the remaining independent set of 42 serum samples. It correctly classified 92% of samples with an AUC value of 0.93. Finally, the algorithm was validated in all samples taken together, where 90% of sera were ranked correctly according to the presence or absence of HCC (AUC value, 0.92). For a cutoff value of 2, sensitivity was 85% and specificity was 91%.

Correlations With Clinicopathological Data.

In patients with HCC, there was no relationship between 8,900-Da peak intensity and patient age, sex, AFP level, or cause of disease. Although no association was observed with histoprognostic features of HCC, a significant positive correlation was found by linear regression between 8,900-Da peak intensity and size of tumors in the 32 patients for whom data were available (coefficient of correlation, 0.67; P < .01; Fig. 2).

Figure 2.

Correlation between the 8.9-kd peak intensity and tumor size in 32 hepatocellular carcinomas.

Characterization of the 8,900-Da Marker of HCC.

To characterize the protein corresponding to the peak of 8,900 Da, selected sera were enriched in an 8,900-Da peak using BioSepra IMAC-Zn HyperCel spin columns. Figure 3 shows SELDI-TOF MS analysis of collected fractions and one-dimensional gel polyacrylamide gel electrophoresis of the eluted fraction from sera of patients with HCC in comparison with serum of patients without HCC. After overnight trypsin digestion of the 8,900-Da gel band, digested peptides were analyzed using matrix-assisted laser desorption time-of-flight (MALDI-TOF). Figure 4 shows the protein profile of the digest with a major peak (molecular weight = 1314.64). MS/MS fragment data for this peptide were exported as Sequest files and were used for database searches with Mascot ( using the NCBI and SwissProt databases. We found that the sequence of this peptide matched with the C-terminal fragment of the vitronectin precursor (serum spreading factor, S-protein, primary accession number, P04004), with an individual ion score of 46, indicating extensive homology (P < .05). Similar analysis performed with the other two polypeptides (m/z = 1571.76 and 1613.78) confirmed this result, with overall sequence coverage attaining 34% of the vitronectin V10 subunit (Fig. 5).

Figure 3.

(A) Isolation of the 8.9-kd peak through column affinity. Surface-enhanced laser desorption ionization time-of-flight mass spectrometry analysis of collected fractions after purification on IMAC-Zn HyperCel spin column (1) flow through, (2, 3, 4) successive washes, and (5, 6, 7, 8) successive eluates. Red bar indicates the 8,900-Da potential hepatocellular carcinoma (HCC) marker. (B) One-dimensional polyacrylamide gel electrophoresis after silver staining of eluated fraction no. 8 of serum of patients with cirrhosis and HCC (lanes 25) and serum of patients without HCC (lanes 710). Lanes 1, 6, and 10 are molecular weight markers. Arrow indicates the 8,900-Da band.

Figure 4.

Peptide mass fingerprinting of the 8,900 Da after trypsin digestion overnight and matrix-assisted laser desorption ionization time-of-flight mass spectrometry (MALDI TOF MS) analysis. Arrows indicate three major peptides (1314.64, 1571.76, 1613.78) used subsequently for MS/MS analysis. m/z, mass-to-charge ratio.

Figure 5.

(A) Sequencing of the three major peptides (1571.76, 1314.64, 1613,78 Da) through MS/MS analysis, and (B) sequence coverage of the V10 fragment of vitronectin.

Real-Time Reverse-Transcriptase Polymerase Chain Reaction of Vitronectin in HCC.

Taking into consideration the increase in the 8,900-Da fragment of vitronectin in sera of patients with HCC, we examined the vitronectin gene expression level by real-time reverse-transcriptase polymerase chain reaction in HCC tissue and HepG2 cells by comparison with liver with cirrhosis and normal liver. After normalization to the normal liver value, the vitronectin gene expression level was 0.96 in cirrhosis, 0.37 in HCC, and 0.0 in HepG2 cells. These results suggest that the increase in the serum concentration of the catalytic fragment of vitronectin observed in patients with HCC is not related to vitronectin gene expression upregulation in tumoral cells.

Identification of the 8,900-Da Marker After Vitronectin Digestion by MMP-2.

Because vitronectin gene expression is not upregulated in HCC tissue, we wondered whether the increased level of the catalytic fragment of vitronectin in sera of patients with HCC may be related to degradation of vitronectin by enzymes overproduced in HCC. Because MMP activity, and especially that of MMP-2, is upregulated in HCC and vitronectin is a substrate of MMP-2, we sought to determine whether in vitro digestion of vitronectin by MMP-2 gave rise to an 8,900-Da fragment.12, 14, 15 For this purpose, proteolytic fragments of native vitronectin digestion by MMP-2 were analyzed by SELDI-TOF MS in a time-course experiment. Before MMP-2 digestion, the SELDI-TOF MS spectrum showed only the 75-kd and 65-kd peaks corresponding to the native MMP-2. As expected, after MMP-2 digestion, an 8,900-Da peptide was detected successfully in reaction mixtures obtained after incubation of vitronectin with MMP-2 for 8 and 24 hours. Intensity of the 8,900-Da peak was maximal at 8 hours (Fig. 6).

Figure 6.

Surface-enhanced laser desorption ionization time-of-flight mass spectrometry ProteinChip spectra before (upper panel) and after (lower panel) in vitro metalloprotease-2 digestion of native vitronectin. After in vitro digestion, an 8,900-Da peak is detected (gray bar).


The aim of the present study was to identify serum protein biomarkers of HCC that developed in the context of chronic liver diseases, using a proteomic approach. For this purpose, we screened serum protein profiles obtained from patients with or without HCC using the SELDI- TOF MS ProteinChip system, a recently developed technology in proteomic profiling. This technology allows researchers to draw a proteomic profile rapidly from little quantity of starting material. Using this approach, we could identify approximately 250 peaks in each serum sample. The number of peaks that can be identified by this approach does not cover the whole serum proteome. This is related to several potential technical limits. A more complete proteome should be obtained by depleting sera of the most abundant proteins, preliminary fractionation of sera before analysis by SELDI-TOF MS, or by testing several different ProteinChip Arrays. However, these technical limitations are counterbalanced by the high efficiency and ease of use of the system that makes the SELDI-TOF MS ProteinChip system a useful tool for clinical proteomics. Using this approach, and comparing all samples together, we found 30 peaks, the levels of which were significantly different according to the presence or absence of HCC. More importantly, a combination of six of these peaks showed a high level of performance for diagnosis of HCC, leading to the accurate classification of more than 90% of patients in both a training set and in an independent testing set of samples. It is of note that three of these peaks have a close m/z ratio (8,900 Da, 9,054 Da, 9,099 Da) that according to the mass accuracy of the TOF mass spectrometer (0.1%) may correspond to either the same protein, to different post-translational modifications of the same protein, or to different molecules.

Previous studies that focused on HCC proteomics used two-dimensional PAGE, a time-consuming method requiring high technical expertise and significant amounts of proteins, thereby limiting the number of samples to be investigated.16, 17 Comparative serum proteomic profiling using the ProteinChip system is a fast, easy-to-use monitoring platform leading to rapid characterization of the serum protein profile in a several minutes with high throughput capacity. The system uses functionalized arrays to capture individual proteins from complex mixtures according to specific properties, yielding a significant number of protein peaks identified by their m/z. Using this technology, our study led to the identification of 30 peaks, the intensity of which significantly differed in sera of patients with and without HCC. Interestingly, most of the samples gave a better performance than AFP, the best marker of HCC to date. In addition, an algorithm combining six peaks offered high performance in separating patients according to the presence of HCC both in a training set and in an independent testing set of samples. Our results confirmed a previous report that was able to identify, using the same technology, serologic markers that distinguish patients with HCC from patients with chronic liver diseases, regardless of AFP concentrations.18 It is of note that in this study, one peak had an m/z ratio very close to 8,900 Da. All these data support the reliability of global proteomic profiling in the discovery of new biomarkers useful for the clinical management of patients with HCC.

Although SELDI TOF MS ProteinChip technology specifies the m/z of each peak, characterization of the protein may be achieved using peptide mass fingerprinting and analysis of digestion fragments by mass tandem spectrometry. We applied such a combined approach after enrichment, isolation, and digestion from selected sera of a putative candidate HCC marker (8,900 Da) giving the best performance, as assessed by its AUC (0.85). Sequencing data unambiguously led to the characterization of this peak as part of the C-terminal fragment of the vitronectin precursor. Vitronectin, a member of the pexin family, is a multifunctional glycoprotein present in both tissue and serum that is involved in cell adhesion, migration of tumoral cells via its ability to bind mainly αV integrin, and matrix remodeling.19, 20 It has been shown that hepatocytes are the main cellular source of vitronectin in the normal liver, resulting from transcription from a single gene leading to a 75-kd protein. Within liver cells, the 75-kd protein is cleaved partially by furin into two fragments of 65 and 10 kd, respectively. Vitronectin is secreted and detected in the serum as either a single chain of 75-kd form or a clipped form composed of two chains (65 and 10 kd, respectively) held together by a disulfide bond. Immunohistochemical study of HCC has shown that vitronectin is detected mainly in the stroma, where it can interact with several extracellular matrix components.21 Because we demonstrated that vitronectin gene expression was downregulated in HCC tissue, we postulated that the occurrence of the 8,900-Da carboxy terminal fragment in sera of patients with HCC may derive from an increase in vitronectin catabolism associated with liver carcinoma. Several studies already reported a significant increase in MMP-2 gene expression and activity in HCC at the invasive border.14, 15 Interestingly, it has been shown that vitronectin is a potential substrate of the MMP family members, especially MMP-2, -3, -7, and -9, suggesting that MMPs may participate in the turnover of vitronectin in tissues.12 Using in vitro experiments, we demonstrated that human vitronectin digestion by MMP-2 resulted in the production of an 8,900-Da fragment, as detected by SELDI TOF MS. Our results, showing an increase in a carboxy terminal fragment of the circulating vitronectin with a size matching the resulting digested peptide of vitronectin in the presence of MMP-2, suggest that, in HCC, vitronectin, normally produced by peritumoral cells, could be partially hydrolyzed by MMP-2 at the margin front of HCC, a site at which MMP-2 activity is upregulated. According to this, the 8,900-Da biomarker may reflect an invasive process. As proof of this concept, we found a positive correlation between 8,900-Da peak intensity and tumor size.

In conclusion, our study enabled the detection of several protein peaks associated with HCC in sera of patients with chronic liver diseases. Among them, an 8,900-Da peak corresponding to part of the carboxyterminal fragment of vitronectin seemed to be the most performant. These results are promising in terms of identification of new biomarkers of HCC using proteomic profiling. Further development of a specific antibody against the 8,900-Da protein should, in the near future, enable development of an enzyme-linked immunosorbent assay and routine testing of the performance of this marker in comparison with other ancillary markers.