Novel Changes in Glycosylation of Serum Apo-J in Patients with Hepatocellular Carcinoma

Infection with hepatitis B virus (HBV) or hepatitis C virus (HCV) is the major etiology of hepatocellular carcinoma (HCC; ref. 1–4). Both HBV and HCV cause acute and chronic liver infections, and most chronically infected individuals remain asymptomatic for many years (5). A total of 10% to 40% of all chronic HBV carriers eventually develop liver cancer, and it is estimated that more than one million people worldwide die of HBV/HCV-associated liver cancer (2, 6, 7). Indeed, HBV and HCV infections are associated with more than 80% of all HCC cases worldwide and can be as high as 96% in regions where these viruses are endemic (3).

The progression from liver disease to liver cancer is monitored primarily by serum levels of the oncofetal glycoprotein, α-fetoprotein (AFP), or the core fucosylated glycoform of AFP (AFP-L3). However, AFP can be produced under many circumstances, including in association with other liver diseases (8–10), and it is not present in all patients with HCC (11). Therefore, the use of AFP as a primary screen for HCC has been questioned (12), and more sensitive serum biomarkers for HCC are desired.

Using various proteomic methods to look for biomarkers for HCC, we identified changes in the levels of glycan in total serum and in serum depleted of IgG or of the major acute phase proteins (13–15). In this study, we expanded our search by analyzing glycosylation of an individual protein, Apo-J, as a function of HCC. Apo J, also known as clusterin, is a secreted glycoprotein with a yet unknown function that we (13, 14, 16) and others (17–21) showed to be associated with HCC. Apo J contains 7 potential sites of glycosylation and is generally found in up to 6 isoforms with differing molecular weights and charges (21). Using N-linked glycan analysis, we compared Apo J levels in healthy subjects with those in patients with HCC and found a specific decrease in the amount of (β-1,4) triantennary N-linked glycan in patients with HCC. We analyzed this change in glycosylation in 2 independent cohorts comprising more than 200 patients, and determined the ability of this marker to differentiate HCC from liver disease.

We obtained serum samples from the University of Michigan (n = 60; HCC = 20, cirrhosis = 20, healthy controls = 20), St. Louis University School of Medicine (n = 151; HCC = 76, cirrhosis = 32, HCV infected = 43), and the California Pacific Medical Center (n = 60; HCC = 10, liver disease = 50). We obtained approval for the study protocol from the appropriate institutional review boards, and written informed consent was obtained from each subject. Details on patients from the University of Michigan and St. Louis University School of Medicine are provided elsewhere (22). We obtained demographic and clinical information for each patient from the California Pacific Medical Center.

A blood sample collected from each subject was placed in a serum separator tube and spun for 2 hours; the serum was stored at −80°C until testing. The diagnosis of HCC was made by histopathologic analysis, including all T1 lesions. If histopathologic samples were not available, the diagnosis was made from the results of 2 imaging modalities (ultrasound, MRI, or computed tomography) showing a vascular-enhancing mass greater than 2 cm. Diagnosis of cirrhosis was based on histologic analysis of liver tissue or on clinical, laboratory, and imaging evidence of hepatic decompensation or portal hypertension.

A total of 7 μL of serum was diluted in 330 μL of solubilization buffer containing 7 mol/L urea, 2 mol/L thiourea, 4% CHAPS, 65 mmol/L dithiothreitol (DTT), 5 mmol/L tributylphosphine, and a 0.4% mixture of carrier ampholytes (Servalyt pH 2–4/pH 3–10/pH 9–1, 1:2:1). Samples were vortexed periodically for 1 hour and applied to a 17-cm pH 3–6 nonlinear (NL) immobilized pH gradient strip (Bio-Rad Laboratories). Gel rehydration was carried out for 14 hours at 50 V and focused by using the Protean (Bio-Rad) isoelectric focusing system. After the focusing, gel strips were first reduced then alkylated in 6 mol/L urea, 2% SDS, 30% glycerol, 50 mmol/L Tris (pH 6.8), and either 30 mmol/L DTT or 75 mmol/L iodoacetamide for 10 minutes each. The second dimension was resolved with an 8% to 18% acrylamide–0.8% piperazine diacrylamide gradient gel on a Protean II xi cell (Bio-Rad) with the running conditions set to 20 mA/gel for 20 minutes and 40 mA/gel for 4 hours. Gels were fixed and stained with colloidal Coomassie. Gels were imaged by the Odyssey Infrared Imaging System (Li-Cor Biosystems) and analyzed by NonLinear Dynamics Progenesis Workstation gel imaging software (Nonlinear Dynamics).

The gel spots were excised and denatured with DTT at 100°C for 5 minutes, allowed to cool, and alkylated in the dark for 30 minutes with iodoacetamide. The gel plugs were fixed in a solution of 30% ethanol and 7% acetic acid for 1 hour. The gel plugs were washed by dehydration in acetonitrile (ACN), rehydration in 20 mmol/L ammonium bicarbonate, and dehydration with ACN, then dried in a speed-vac. Glyko N-glycanase-PLUS [peptide:N-glycosidase F (PNGase F); 10 mU] was diluted with 20 mmol/L ammonium bicarbonate (pH 7) and allowed to adsorb into the gel plug. The gel plug was then covered with the same solution and allowed to incubate overnight at 37°C. The glycans were eluted from the gel plug by sonication in Milli-Q water 3 times; the elutant pooled was dried down and labeled with a 2-aminobenzamide dye (Ludger) according to the manufacturer's instructions. Excess dye was removed by using paper chromatography and the glycans were passed through a 0.22-μm syringe filter. Fluorescently labeled glycans were subsequently analyzed by high-performance liquid chromatography (HPLC) by using a TSK-Gel Amide 80 column (Tosoh Bioscience LLC). The mobile phase consisted of solvent A (50 mmol/L ammonium formate, pH 4.4) and solvent B (ACN). The gradient used was as follows: linear gradient from 20% to 58% solvent A at 0.4 mL/min for 152 minutes followed by a linear gradient from 58% to 100% solvent A for the next 3 minutes. The flow rate was increased to 1.0 mL/min, and the column was washed in 100% solvent A for 5 minutes. Following the wash step, the column was equilibrated in 20% solvent A for 22 minutes in preparation for the next sample run. HPLC analysis was carried out by using the Waters Alliance HPLC System, complemented with a Waters fluorescence detector, and quantified by the Millennium Chromatography Manager (Waters Corporation). Glycan structures were identified by calculating the glucose unit value and exoglycosidase digestion, as previously described (23).

Following glycan preparation, the deglycosylated gel plug was dehydrated in ACN and dried. Trypsin-Gold (Promega), diluted to 20 μg/mL according to the manufacturer's recommendations, was absorbed into the dried gel plug on ice. Following rehydration, the gel plugs were covered with 25 mmol/L ammonium bicarbonate buffer and incubated at 37°C overnight. Peptides were extracted twice by using 60% ACN, 0.1% trifluoroacetic acid with sonication, and once with 100% ACN. The extractions were pooled together, dried, and resuspended in 5% ACN, 0.1% trifluoroacetic acid. The peptides were concentrated and desalted by ZipTip C18 (Millipore). Peptides were analyzed by a Dionex UltiMate 3000 Nano liquid chromatography system (Dionex) coupled with a Finnigan LTQ mass spectrometer. Peptides were separated online by using a 75-μm ID × 15-cm C18 analytic column by using 2% ACN/0.1% formic acid (buffer A) and 90% ACN/0.1% formic acid (buffer B). Following the initial wash with 95% buffer A for 10 minutes, peptides were eluted from the column during a 90-minute linear gradient of 5% to 60% of buffer B at a flow rate of 250 nL/min directly into a linear trap quadrupole ion trap mass spectrometer.

Antibody to Apo-J/Clusterin (1 μg/well in PBS, pH 7.4) from Sigma-Aldrich, was added to the plate and, following incubation overnight at 4°C, was washed with PBS/0.1% Tween 20 (pH 7.4) and blocked overnight with 1× Carbo-Free blocking buffer (Vector Laboratories). For analysis, 1 μL of serum was diluted in 49 μL PBS with 0.1% Tween-20 and was incubated at room temperature for 2 hours. Subsequently, plates were washed 5 times with 0.1% Tween 20/PBS (pH 7.4), then washed 3 times with lectin incubation buffer (25 mmol/L HEPES, 1 mmol/L MnCl₂, 0.1% Tween 20) before triantennary glycan was detected with a biotin-conjugated Datura stramonium lectin (DSL; Vector Laboratories). Bound lectin was detected by IRDye 800 conjugated streptavidin; signal intensity was measured by the Odyssey Infrared Imaging System (LI-COR Biotechnology). In all cases, signal intensity was compared with signals detected with commercially purchased human serum (Sigma-Aldrich).

Descriptive statistics for patients were compared by scatter plots that included the outliers. All values were reported as mean ± SD unless otherwise stated. Because the data did not follow a typical Gaussian distribution, a nonparametric test (2-tailed, 95% confidence, Mann–Whitney test) was used to determine statistical differences between groups. To determine the optimal cutoff value for each marker, the receiver operating characteristic (ROC) curves were constructed by using all possible cutoffs for each assay. The area under the ROC (AUROC) curves were constructed and compared as described previously. A 2-tailed P value of 0.05 was used to determine statistical significance. All descriptive analyses were carried out by a GraphPad Prism. For combinatorial analysis, AFP values were log transformed to bring the values of all markers into a similar scale, and a centering and scaling approach was taken to normalize data before analysis by a multivariate logistic regression method, using the R package, version 2.8.1.

Samples of pooled sera from healthy patients (n = 20), patients with liver cirrhosis (n = 20), and patients with HCC with a background of cirrhosis (n = 20) were resolved via 2-dimensional gel electrophoresis (2-DE; Table 1). Figure 1A shows the results of representative 2-DE of normal human serum and Figure 1B–D shows the Apo-J α- and β-chain isoforms from the 3 patient groups. The Apo-J protein concentrations in each patient group were comparable (Fig. 1B–D) and fell within normal serum concentrations. In the control sample, the level of Apo-J was 100 μg/mL (±1.4 μg/mL), whereas in the patients with cirrhosis and HCC, the level of Apo-J was 112 μg/mL (±1.4 μg/mL) and 110 μg/mL (±1.4 μg/mL), respectively. Each of the 5 APO-J isoforms was excised and combined, and the N-linked glycans were analyzed by normal phase HPLC following liquid chromatography/tandem mass spectrometry analysis to confirm protein identity and purity (data not shown).

N-linked glycans attached to Apo-J were removed by using PNGase F and labeled with a fluorescent dye before analysis of the N-linked glycans via sequential exoglycosidase digestion (24–26). Figure 2 shows the simplified desialylated glycoprofile of Apo-J for each patient group following treatment with neuraminidase (Arthrobacter ureafaciens). Two peaks that were altered in the HCC sample are indicated with an asterisk in Figure 2A. Sequential exoglycosidase digestion (Supplementary Fig. S1) showed these peaks to be a simple biantennary glycan (A2G2) and a (β-1,4)-triantennary N-linked glycan (A3G3). In the control and cirrhosis groups, the biantennary glycan represented 42% (±4.2%) and 45% (±4.9%) of the total glycan pool, respectively (Fig. 2B). In the HCC group, the biantennary glycan represented 51% (±7.8%) of the total glycan pool. Figure 2B also shows the level of the A3G3 glycan in the 3 patient groups. The level of the A3G3 glycan represented 37% (±4.1%) of the total glycan pool from the control patients and 32% (±4.9%) from the patients with cirrhosis, whereas it represented 21% (3.7%) of the total glycan pool in the HCC group. Two fucosylated glycans, an (α-1,6)-core fucosylated biantennary glycan (FcA2G2) and an (α-1,3)-linked outer arm fucosylated triantennary glycan were also observed but the their levels in the 3 patient groups was not altered.

To further examine whether the changes in glycosylation observed in the pooled samples could be observed on individual samples and in an independent cohort, we used a lectin fluorochrome-linked immunoassay (FLISA). Because no lectins that solely bind biantennary glycans are available, we used DSL, which has been shown to bind to (β-1,4)-linked triantennary glycan (27–29). In this assay, Apo-J was captured by using a monoclonal antibody, and the level of A3G3 glycan was determined by using DSL. We noted that the lectin FLISA detects the amount of triantennary glycan present on an equal quantity of captured molecules from each patient sample and is performed in a manner that is independent of the total amount of protein in a sample from any given patient. Thus, even if protein levels are different in different individuals, this assay measures only the relative proportion of altered glycoprotein. This method was first tested on the pooled samples used in Figures 1 and 2; the results are shown in Figure 2C. As these results show, we observed a decrease in the level of A3G3 glycan, similar to that observed when using glycan sequencing of Apo-J. This result was further analyzed in a patient cohort consisting of 151 patients using the lectin FLISA (Table 1). In this cohort, 43 patients were infected with HCV; 32 patients were infected with HCV and had liver cirrhosis; and 76 patients had liver cancer (Table 1). Figure 3A shows a scatter plot of the level of DSL-reactive Apo-J in the 3 patient groups. Values are given as fold increase over the level in commercially purchased “normal” sera. We found a clear statistical difference between the HCC group and the cirrhotic group (P < 0.0001) but not between the HCV-infected and the cirrhosis group (P = 0.6184). The mean level of DSL-reactive Apo-J was 0.91-fold (±0.30) in the HCC group, 1.2-fold (±0.16) in the cirrhosis group, and 1.2-fold in the HCV infected group (±0.20). Surprisingly, we did not see any difference in the mean level of DSL-reactive Apo-J in HCC patients at different stages of HCC. Patients with stage 1 HCC, defined as a single lesion of less than 2 cm (11), had a 0.855-fold (±0.14) level of DSL-reactive Apo-J whereas patients with stage 4 HCC (those with multiple lesions >6 cm in diameter) had a mean of 0.779 (±0.13), which was not different from that observed in stage 1 patients (P = 0.4204). In this cohort, DSL-reactive Apo-J distinguished between HCC and cirrhosis with an AUROC of 0.852. This result was greater than the results obtained with AFP in this cohort (0.7646). In this cohort, AFP had a mean value of 60.1 ng/mL in the HCC patients and 10.4 ng/mL in the cirrhotic patients.

Figure 4A shows a perspective plot of the relative levels of DSL-reactive Apo-J, AFP, and another marker previously examined in this cohort, fucosylated kininogen (22). Most HCC patients (represented by solid red dots) were found in the lower back corner of the plot, indicating that they had high levels of AFP and fucosylated kininogen but low levels of DSL-reactive Apo-J. In contrast, the cirrhotic samples (represented by open circles) cluster in the bottom left hand corner of the plot, indicative of low levels of the individual markers.

When DSL-reactive Apo-J was combined with AFP by using a multivariate logistic regression analysis, the performance was improved to 0.9107. Similarly, as Figure 4B shows, when combined with another potential marker, fucosylated kininogen, the AUROC was 0.9139. When all 3 markers were included (Fig. 4B) in a multivariate logistic regression model, the AUROC increased to 0.944.

The performance of DSL-reactive Apo-J alone and in combination with fucosylated kininogen was tested in an independent cohort of 60 patients, 10 of whom had HCC and 50 had liver disease (Table 1). In this cohort, using the algorithm determined in the first sample set, the AUROC of DSL-reactive Apo-J and fucosylated kininogen that differentiated HCC from liver disease was 0.9129, with an optimal sensitivity of 87% and a specificity of 96%, similar to what was obtained in the initial training set.

Serum levels of Apo-J have been reported to be associated with both the presence of HCC and the metastatic potential of the cancer (17, 19, 21). In addition, we previously identified Apo-J as being part of the serum fucosylated proteome in patients with HCC, which suggested a difference in glycosylation of Apo-J. To further examine the glycosylation of Apo-J, we carried out N-linked glycan analysis of Apo-J following separation by 2-DE. Serum Apo-J from patients with HCC had less triantennary N-linked glycan compared with Apo-J from healthy controls or patients with liver cirrhosis. The change in glycosylation was analyzed by a lectin ELISA in which a saturating amount of Apo-J was captured and the relative level of triantennary N-linked glycan was determined by a lectin (DSL) specific for this glycan structure. Like the N-linked glycan data, the lectin ELISA data confirmed the decrease in triantennary glycan in a larger sample set. The ability of DSL-reactive Apo-J to differentiate cirrhosis from HCC was greater than that of AFP. When used in combination with AFP through multivariate analysis, DSL-reactive Apo-J could distinguish cirrhosis from cancer with an AUROC of 0.9107, with an optimal sensitivity of 86% and a specificity of 95%. Analysis of the multivariate model in a second, independent patient cohort was used to confirm the multivariate analysis, and a similar AUROC was observed (0.9139). Efforts are underway to develop a chip-based platform that will allow for the simultaneous analysis of all 3 proteins from a single sample.

Although alterations in fucosylation have been observed on serum proteins as a function of HCC, this report describes a decrease in triantennary branching with the development of liver cancer. It is interesting to note that the decrease in triantennary glycan associated with Apo-J is accompanied by an increase in the simple biantennary glycan associated with Apo-J from HCC patients. The increase in the biantennary does not appear as strong as the triantennary decrease; however, this does not rule out the possibility of the 2 events being related. One common denominator that could link these 2 events would be a decrease in the level of the N-acetylglucosaminyltransferase IV (GnT-IV) enzyme, which can extend the biantennary glycan chain to the triantennary form (30). However, it is unclear whether this reduction in branching is a global event and seen on many proteins, as would be expected if there were a reduction in the level of GnT-IV. This is currently under investigation.

Apo-J has been implicated as a biomarker of HCC. Previous reports have indicated that serum levels of Apo-J decreased in patients with cirrhosis (compared with healthy controls) but increased in those with HCC. In our limited analysis of patients with cirrhosis and HCC, we did not see any statistical difference in the level of Apo-J in the HCC group compared with the non-HCC groups (data not shown). It should be noted that our plate-based approach for the detection of Apo-J with triantennary glycan was conducted under saturating conditions and examined equal amounts of captured Apo-J.

In summary, we have discovered a novel glycan modification on Apo-J and developed a lectin FLISA to measure this change. DSL-reactive Apo-J could distinguish HCC from cirrhosis to the same degree as AFP. However, when used in combination with AFP or AFP and another potential marker, fucosylated kininogen, DSL-reactive Apo-J performed better than AFP alone. We postulate that these markers could be used to supplement AFP as a general screen in those patients at high risk for HCC development, either alone or in combination with ultrasound, which is the current practice. These markers may also be useful in monitoring HCC patients following treatment as well as identifying the patients with recurrence.

These data must be confirmed in larger cohorts of patients to determine whether these markers are true indicators of early HCC and to compare the accuracy of the data with data of AFP in patients of diverse gender, ethnicity, and etiologies of liver disease and to determine its role in HCC surveillance.

No potential conflicts of interest were disclosed.

This work was supported by grants R01 CA120206 and R01 CA136607 from the National Cancer Institute (NCI), and Grant UO1 CA084951 from the NCI's Early Research Detection Network (EDRN), the Hepatitis B Foundation, and an appropriation from The Commonwealth of Pennsylvania.

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