fax: +81 3 3547 5298
Liver Biology and Pathobiology
Proteomic signature corresponding to alpha fetoprotein expression in liver cancer cells
Article first published online: 30 AUG 2004
Copyright © 2004 American Association for the Study of Liver Diseases
Volume 40, Issue 3, pages 609–617, September 2004
How to Cite
Yokoo, H., Kondo, T., Fujii, K., Yamada, T., Todo, S. and Hirohashi, S. (2004), Proteomic signature corresponding to alpha fetoprotein expression in liver cancer cells. Hepatology, 40: 609–617. doi: 10.1002/hep.20372
- Issue published online: 30 AUG 2004
- Article first published online: 30 AUG 2004
- Manuscript Accepted: 10 MAY 2004
- Manuscript Received: 22 FEB 2004
- Pharmaceuticals and Medical Devices Agency of Japan
Alpha fetoprotein (AFP) has been implicated in the development of hepatocellular carcinoma and is considered to be a diagnostic and prognostic tumor marker. Because elevated expression of AFP is associated with many characteristics of hepatocellular carcinoma tissues, we hypothesized that multiple proteins may function in a coordinated manner with AFP. To identify such proteins, we performed global protein expression analysis, namely a proteomic study. The protein expression profiles of 9 AFP-producing liver cancer cell lines (JHH-5, HuH-1, PLC/PRL/5, Hep3B, HT-17, JHH-7, HuH-7, HepG2, Li-7) and 7 nonproducing liver cancer cell lines (HLE, JHH-6, Sk-Hep-1, JHH-4, HLF, RBE, SSP-25) were generated by fluorescence 2-dimensional difference gel electrophoresis. In fluorescence 2-dimensional difference gel electrophoresis, proteins are labeled with fluorescent dyes before electrophoresis for more accurate quantitative expression analysis. We identified 11 protein spots that distinguished AFP-producing cell lines from nonproducing cell lines by multivariate studies. The spots showed consistent alterations in amount in AFP-producing cell lines (6 up-regulated and 5 down-regulated). An additional 5 liver cancer cell lines (KIM-1, KYN-2, KYN-3, PH5-CH, PH5-T) also were correctly grouped with respect to their AFP production on the basis of the intensity of the 11 protein spots. The proteins corresponding to the 11 selected spots were identified by mass spectrometry and were categorized into 4 groups based on their known role in apoptosis, glucose metabolism, cytoskeletal organization, or translation. In conclusion, we found a novel association of AFP with other proteins. Their interaction should provide insight into the biology of AFP-producing hepatocellular carcinoma cells. (HEPATOLOGY 2004;40:609–617.)
Hepatocellular carcinoma (HCC) represents the vast majority of liver cancer, accounting for more than 90% of all primary liver cancers, and ranks fifth in frequency among all malignancies worldwide.1, 2 Infection with hepatitis B or C virus has been identified as an etiological factor, and the subsequent cellular and histological changes leading to HCC have been extensively studied.3 Genetic alterations, including loss of heterozygosity4, 5 and aberrant DNA methylation,6 also have been implicated in the carcinogenesis of HCC. Nevertheless, the outcome for HCC patients still remains dismal, partly because of the difficulty in establishing an early diagnosis and the frequent recurrence and intrahepatic metastasis after surgery.7 Although systematic chemotherapy for unresectable HCC has been widely used, its efficacy remains low and complications such as significant myelosuppression are observed during the course of treatment.8
Serum alpha fetoprotein (AFP) has been considered to be a hallmark of development of HCC.9 Serum AFP alone has a limited role in the early diagnosis of HCC, because a considerable proportion of HCC patients do not have elevated serum AFP, and serum AFP can increase in patients with diseases other than HCC; it may, however, be a useful prognostic indicator, because the median survival rate of HCC patients with markedly elevated AFP was significantly shorter than that of patients with normal or moderately elevated AFP.10 AFP is a multifunctional glycoprotein belonging to the family of albumin-like proteins.11 AFP has been shown to serve as a dual regulator of growth in a multitude of cell types and cancers involving several mechanisms, including apoptotic regulation.12 AFP also functions as a transporter of estrogens,13 fatty acids,14 and bilirubin15 and modulates immune response in macrophages and T lymphocytes.16 Although the structure and function of AFP have been extensively studied,17 its exact role in the development of HCC has not been defined in detail.
Recently, global gene-expression studies have been shown to have great value in identifying the genes associated with various clinical states of liver cancer, such as early intrahepatic recurrence,18 and have been used to develop diagnostic and prognostic biomarkers. Kawai et al19 performed a genome-wide gene expression study and found that AFP-producing HCC cell lines shared a distinct expression profile of genes, including those related to apoptosis, cell cycle, cell–cell interaction and oncogenesis. Similarly, classification of HCC cell lines based on unbiased gene expression profiles also resulted in 2 major subgroups corresponding to AFP expression level, suggesting that the elevated AFP may be a part of the molecular signature that determines the major subtypes of HCC.20 These studies provide clues to the particular genetic pathways in which AFP is involved and contribute to further understanding of the biological and clinical significance of AFP in HCC.
In this report, we performed global protein expression analysis, a proteomic study, to identify the proteins that function in a common network with AFP. Because higher expression of AFP often is associated with greater tumor size, histological undifferentiation, and portal vein thrombosis,10 and it is unlikely that any single protein is directly responsible for such a variety of characteristics, we hypothesized that multiple proteins may function in a coordinate manner with AFP. Although discordance of messenger RNA and protein expression has been reported,21–24 the DNA microarray data for AFP-producing cells were not examined at the protein level, and a proteomic study has not yet been performed. In this study, we generated protein expression profiles using fluorescence 2-dimensional difference gel electrophoresis, in which proteins are labeled with fluorescent dyes before electrophoresis, and used a multivariate method to capture a proteomic signature corresponding to AFP production.
Materials and Methods
The cell lines JHH-4, JHH-5, JHH-6, JHH-7, HuH-1, PLC/PRF/5, HuH-7, HLE, and HLF were obtained from the Human Science Research Resources Bank (Sennai, Osaka, Japan). The cell line SK-Hep-1 was obtained from the American Type Culture Collection (Rockville, MD). Two cholangiocellular carcinoma cell lines, SSP-25 and RBE, and the HepG2 cell line were obtained from the Riken Gene Bank (Ibaragi, Tsukuba, Japan). The HT17 and Hep3B cell lines were from the Cell Resource Center for Biomedical Research, Tohoku University (Sendai, Miyagi, Japan). The KYN-2, KYN-3, and KIM-1 cell lines were provided by Dr. Masamichi Kojiro of Kurume University. Li-7 and 2 immortalized hepatocyte cell lines, PH5-T and PH5-CH, were previously established in the National Cancer Center Research Institute.25, 26 All cell lines were HCC ones, unless otherwise specified. To minimize the effects of culture condition on protein expression, the cells were maintained with the recommended culture media, and proteins were extracted when the cells reached approximately 80% confluence.
Preparation of Fluorescence-Labeled Protein Samples.
The cells were washed with phosphate-buffered saline and treated with 10% trichloroacetic acid for 30 minutes on ice. Then, the cells were collected into a tube, were pelleted by a brief centrifugation, and were washed with ice-cold phosphate-buffered saline. The cells were then incubated with urea lysis buffer, consisting of 7 M urea, 2 M thiourea, 3% CHAPS, and 1% Triton X-100 for 30 minutes on ice. After centrifugation at 15,000 rpm for 30 minutes, the protein concentration of recovered supernatant was measured with a Protein Assay Kit (Bio-Rad Laboratories, Hercules, CA) and adjusted to 1 mg/mL with urea lysis buffer. We then made an internal control sample by mixing the protein samples of all cell lines used. After the pH of the protein samples had been adjusted to 8.0 with 30 mM Tris-HCl, the individual samples and the internal control sample were labeled for 30 minutes with 200 nmol of 1-(5-carboxypentyl)-1′-propylindocarbocyamine halide (Cy3) and 1-(5-carboxypentyl)-1′-methylindocarbocyamine halide (Cy5; Amersham Biosciences, Little Chalfont, Buckinghamshire, UK), respectively. The labeling reaction was terminated by incubation with 0.2 mM lysine for 10 minutes on ice. The labeled samples were treated with an equal volume of urea lysis buffer containing 130 mM DTT and 2% ampholine for 15 minutes on ice. The labeled internal sample and the samples of the individual cell lines then were mixed together. The final volume was adjusted to 430 μL with urea lysis buffer containing 65 mM DTT and 1.0% ampholine. The procedure of sample labeling and mixing is illustrated in Fig. 2.
Two-Dimensional Polyacrylamide Gel Electrophoresis.
For first-dimension separation, immobilized pH gradient gels (24 cm length; pI range, 3–10; Amersham Biosciences) were rehydrated with labeled protein samples for 12 hours at 20 °C. Isoelectric focusing was performed with IPGphor (Amersham Biosciences) for a total of 80 kVh at 20°C. The immobilized pH gradient gels were equilibrated for 15 minutes in equilibration buffer containing 6 M urea, 50 mM Tris (pH 8.8), 30% glycerol, 1.0% sodium dodecyl sulfate, and 16 mM DTT, and then for another 15 minutes in the same buffer containing 122 mM iodoacetamide instead of DTT. Equilibrated immobilized pH gradient gels were transferred onto 9% to 15% gradient polyacrylamide gels and embedded with agarose. The second-dimension electrophoresis was performed at 17 W for 15 hours at 20°C. We divided the Cy3-Cy5 mixture in 3 gels and calculated the average spot intensity.
After electrophoresis, the gels were scanned at the appropriate wavelengths for Cy3 and Cy5 with a 2920 2D Master Imager (Amersham Biosciences). The exposure time was determined to ensure that the maximum spot intensity was not saturated. For every spot, the Cy5 intensity was normalized to the Cy3 intensity in the same gel. The BVA mode of DeCyder software (Amersham Biosciences) was used to normalize the intensity of all spots and calculate the average spot intensity among 3 gels.
Machine-Learning Method and Multivariate Analysis.
The 1334 protein spots that appeared in at least 80% of Cy3 images were selected and subjected to statistical analysis. The spot intensity was standardized by subtraction of the average value and by division by the root mean square, and then transformed logarithmically. Hierarchical clustering analysis and principal component analysis were performed using GeneMaths software (Applied Maths, Sint-Martens-Latem, Belgium). Cross-validation and spot ranking were performed with Impressionist software (Gene Data, Basel, Switzerland). The prediction model for AFP production was created with the Visual Data Mining system (Suuri-ken, Tokyo, Japan).
Western Blotting for AFP.
To explore the expression of intracellular AFP, 30 μg of protein lysate was labeled with Cy3 as described in Preparation of Fluorescence-Labeled Protein Samples and separated by sodium dodecyl sulfate -polyacrylamide gel electrophoresis with 12% polyacrylamide or 2D-polyacrylamide gel electrophoresis. The gels were scanned at appropriate excitation and emission wavelengths for Cy3, and the amount of protein loaded onto the gels was normalized by the sum of band intensities of the entire lane. The proteins then were transferred to polyvinylidene difluoride membranes and incubated with anti-AFP antibody (1:200 dilution; ZYMED, San Francisco, CA). The blots were reacted with anti-mouse immunoglobulin G antibody conjugated with peroxidase (1:1,000; Amersham Biosciences), and the AFP signal was monitored by an enhanced chemiluminescence system (Amersham Biosciences). The intensity of the AFP signal was quantified with an LAS-1000 image analyzer (Fujifilm, Tokyo, Japan).
Mass Spectrometric Protein Identification.
For preparative purposes, 500 μg of unlabeled protein sample was separated by 2D-polyacrylamide gel electrophoresis and stained with SYPRO Ruby dye (Molecular Probes, Eugene, OR). The image of the preparative gel was matched to that of the analytical gels and the spots of interest were excised. The gel plugs were washed twice for 5 min with water, then with 100% vol/vol acetonitrile. After complete drying, the protein in the gel was digested at 37°C overnight with trypsin (Promega, Southampton, UK) in 50 mM ammonium bicarbonate, with gentle agitation. Peptides then were extracted from the gel plugs with 50 μL of 50% acetonitrile in 0.1% trifluoroacetic acid and were concentrated to approximately 10 μL. A 1-μL sample of extracted peptides was mixed with saturated dihydroxybenzoic acid in 50% acetonitrile/0.1% trifluoroacetic acid and spotted onto a target plate. Peptide mass fingerprinting analysis was performed by matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MS) with a Q-Star Pulser-i equipped with the orthogonal injection/matrix-assisted laser desorption/ionization ion source (Applied Biosystems, Framingham, CA). Mass spectra were processed with the Analyst QS program (Applied Biosystems), and a search of the Swiss-Prot database was performed for peptide mass fingerprinting with a mass tolerance of less than 50 ppm.
Expression of AFP in Liver Cancer Cell Lines.
We examined the expression of AFP in 19 HCC cell lines and 2 cholangiocellular carcinoma cell lines. Western blotting revealed that 11 HCC cell lines expressed AFP, whereas the other cell lines did not (Fig. 1A). The expression level of AFP as a function of relative expression ratio between the AFP signal generated by enhanced chemiluminescence and the total Cy3 intensity of bands is presented as a histogram of the 21 cell lines (Fig. 1B). The degree of AFP expression varied even among the AFP-producing cell lines.
Proteomic Pattern of Fluorescence-Labeled Proteins.
The spot intensities were analyzed by fluorescence 2-dimensional difference gel electrophoresis in a quantitative way. Protein samples of cell lines were mixed together to create the internal control sample and labeled with Cy3. Protein samples of individual cell lines were labeled with Cy5. Cy3-labeled and Cy5-labeled protein samples were mixed and coseparated by 2D-polyacrylamide gel electrophoresis in the same gel. Two-dimensional profiles of the internal control sample and individual cell line samples were obtained by scanning the same gel with wavelengths unique to Cy3 or Cy5. The fluorescent dyes are engineered so that the migration of a given protein labeled with Cy3 or Cy5 is practically identical in 2-dimensional gels. Because the Cy3 image is produced from the common internal control sample, normalization of Cy5 image spot intensity to the intensity of the corresponding Cy3 image spot in the same gel avoided electrophoretic artifacts and enabled quantitative analysis. The 2-color image of Cy3-labeled or Cy5-labeled proteins showed that the location of spots was identical between the 2 images and that the intensity of each spot was distinct, reflecting the differential expression level in the internal control sample and individual cell line (Fig. 2A).
Figure 2B shows a representative protein expression profile with approximately 2,000 protein spots. For multivariate analysis, we selected 1,334 spots that were present in the Cy3 images of at least 80% of gels. The number of selected spots in each cell line is summarized in Fig. 2D. We did not filter the spots with respect to expression level. The 11 spots circled in Fig. 2B were selected later by a machine-learning method as essential spots for classification of liver cancer cell lines according to their AFP production.
We examined the expression of AFP on 2-dimensional gel. Western blotting identified AFP as 6 protein spots on the membrane. However, the spots corresponding to AFP were not observed on the Cy3 image (Fig. 2C), because the expression level of AFP was below the detection limit of fluorescence 2-dimensional difference gel electrophoresis, and 1 of the spots was behind the spot for heat-shock protein 70 (HSP 70). Therefore, the AFP spots were not included in our study.
To observe the degree to which the protein expression profiles were characteristic to each cell line, we observed the overall correlation between the protein expression profiles of the cell lines (Fig. 2D). Correlation matrix revealed that the protein expression profiles of AFP-negative cell lines were more homogeneous than those of AFP-positive cell lines. This different degree of correlation may reflect the differentiation status of the HCC cell lines; the AFP expression level was higher in poorly differentiated HCC,27 and the protein expression profile of well-differentiated cells may be more uniform than that of poorly differentiated cells.
Supervised Classification and Spot Identification.
We used a machine-learning method to find the spots that best discriminated AFP-producing cells from nonproducing cells. In the leave-one-out crossvalidation, 9 AFP-producing cell lines and 7 nonproducing cell lines were used as test samples. Fifteen cell lines were randomly selected and were used to generate the classifier, and the classification performance was validated using 1 cell line that was not used to generate the classifier. The probability of obtaining crossvalidation error by chance was obtained by repeating this crossvalidation procedure using 1,000 random permutation trials. We first examined 4 classifier algorithms, including a linear Support Vector Machine (SVM), Sparse Linear Discriminant Analysis (SLDA), Fisher Linear Discriminant Analysis, and K-Nearest Neighbors. Figure 3A illustrates the results of crossvalidation based on the SVM algorithm, indicating that the classifier categorized the cell lines into the groups in the training set with low crossvalidation error rate. SLDA also resulted in correct classification (data not shown). The other algorithms, Fisher Linear Discriminant Analysis and K-Nearest Neighbors, misclassified the cell lines (data not shown).
To determine the set of spots that best distinguished the 2 groups, we applied a spot ranking method in which the crossvalidation was performed with all combinations of spots with multiple independent classifier algorithms, and the misclassification error rate was plotted as a function of the number of best spots. We tested linear SVM and SLDA as classifiers, and the algorithms of support vector machine weight, recursive feature elimination, sparse linear ranking, supervised gene shaving, ANOVA, and Kruskal Wallis test to rank the spots according to the degree of contribution to the classification. Figure 3B shows the results of spot ranking using Support Vector Machine Weight as a ranking method. The crossvalidation error rate reached 0% when particular sets of 11 and 67 spots were used with SVM and SLDA, respectively. The other ranking algorithms did not identify spot sets that had a classification error rate of 0%. Because the 67 spots selected by SLDA included the 11 spots selected by SVM, we concluded that the combination of SVM as a classifier and Support Vector Machine Weight as a ranking method could best select the essential features of the expression profile representing AFP production.
We confirmed the performance of the 11 spots as a classifier using unsupervised classification methods. Figure 3C shows the dendrogram of cell lines created by hierarchical clustering analysis using the 11 spots. The cells were clustered in 2 major trees according to their AFP productivity. The 11 protein spots also were divided into 2 major trees: the intensity of 6 spots was up-regulated and the intensity of 5 spots was down-regulated in AFP-producing cell lines (Fig. 3C). Principal component analysis showed that the 16 cell lines were also divided into 2 major groups corresponding to their AFP production on the basis of the 11-spot profile (Fig. 3D).
Use of the 11-Spot Set to Predict AFP Production by HCC Cell Lines.
We further examined the predictive performance of the 11 spots using the cell lines that were not used to create the classifier. Those included 2 AFP-producing cell lines (KIM-1 and KYN-2) and 3 nonproducing cell lines (KYN-3, PH5-CH, and PH5-T). Using the selected 11-spot set, hierarchical clustering (Fig. 4A) and principal component analysis (Fig. 4B) grouped these additional cell lines correctly into either the AFP-producing or the nonproducing cell line group. We then used these 11 spots to construct a scoring model to predict AFP production and applied this model to an independent test set of 5 HCC cell lines. All training samples and the additional 5 cell lines were classified according to their AFP production (Fig. 4C).
These results indicate that the profiles of the 11 spots may represent the cellular phenotypes differently shared between the AFP-producing cell lines and the nonproducing cell lines.
Mass Spectrometric Identification of the 11 Spots and Their Expression in the Cell Lines.
A database search with an MS and tandem mass spectrometry (MS/MS) study of tryptic peptides of spot 2,226 (Fig. 5A, B) resulted in identification of galectin-I (Fig. 5C–E). MS analysis was performed on the other protein spots, and the Mascot scores obtained allowed positive protein identification (Table 1).
|Spot Ranking*||Spot No.||Fold Difference† (P Value)||Identification (Protein Name, Accession No.)‡||Score§||Function∥||Known Connection to HCC/Other Cancer||Reference¶|
|1||874||5.12 (1.2e-008)||Glucose-6-phosphate 1-dehydrogenase (P11413)||1131||G||Yes/yes||28, 29|
|2||2118||9.19 (0.00032)||Ubiquitin-like protein SUMO-1 conjugating enzyme (P50550)||202||A||No/no||—|
|3||2226||0.27 (7.4e-008)||Galectin 1 (P09382)||240||A||No/yes||—, 30|
|4||2081||0.29 (5.3e-009)||Mitochondrial dicarboxylate carrier (Q9UBX3)||1141||G||No/no||—|
|5||1955||1.97 (7.4c-006)||BH3 interacting domain death agonist (P55957)||427||A||Yes/yes||31, 32|
|6||865||2.64 (0.00014)||Aldehyde dehydrogenase 1A1 (P00352)||251||G||Yes/no||33, —|
|7||1169||3.15 (5.4e-010)||Isocitrate dehydrogenase (O75874)||228||G||Yes/yes||34, 35|
|8||1395||0.38 (0.00038)||Annexin A1 (P04083)||364||A||Yes/yes||36, 37|
|9||905||1.59 (0.00013)||Keratin, type II cytoskeletal 7 (P08729)||440||C||Yes/yes||38, 34|
|10||1451||0.45 (0.00014)||60S acidic ribosomal protein P0 (P05388)||242||T||Yes/yes||40, 41|
|11||142||0.43 (8.5e-006)||Vinculin (P18206)||597||C||No/yes||—, 42|
We demonstrated the expression levels of the identified proteins between the cell lines, and the 2-dimensional images for AFP-producing and nonproducing cell lines are shown in Fig. 6. The Cy5 intensity standardized by Cy3 intensity also was shown for all the cell lines used. It is important to note that global messenger RNA expression studies on liver cancer cell lines did not identify these proteins.21, 22
Elevated expression of serum AFP has been shown to be associated with various malignant characteristics of HCC (greater tumor size, bilobar involvement, undifferentiated tissue types, massive or diffuse types, and portal vein thrombosis) and poor prognosis.10 These pathological and clinical features could be explained by a common protein network with AFP expression. Taking annexin A1 as an example, this protein is a major substrate for tyrosine kinases such as epidermal growth factor receptor and serine/threonine kinases such as protein kinase C43, 44 and has been implicated in cellular signal transduction.45 Enhanced expression of annexin A1 was observed in poorly differentiated HCC tissues compared with their well-differentiated counterparts, and the nontumorous region contained significantly lower amounts of annexin A1, suggesting that annexin A1 is related to the histological grade of HCC and is involved in the malignant transformation process.36 The histological undifferentiation observed in AFP-positive HCC may be explained by the enhanced expression of annexin A1.
In summary, we identified the differential expression of 4 apoptosis-related proteins in AFP-positive cells. Two of them are known to be associated with HCC (BH3-interacting domain death agonist and annexin A1) and the other 2 have not been implicated in the mechanisms of HCC (ubiquitin-like protein SUMO-1 conjugating enzyme and galectin I). It has been reported that AFP has growth-suppressive effects on cells by inducing apoptosis.46 The expression of apoptosis-related proteins in AFP-producing cells may explain the apoptotic effects of AFP on the cells. Because the aberrant regulation of apoptosis is known to contribute to the development of HCC,47 clarification of the association of apoptosis-related proteins with AFP will be useful for understanding the mechanisms of HCC development.
Aberrant glucose metabolism also has been reported in HCC,28 and we identified 4 proteins involved in glucose metabolism. The higher expression of glucose-6-phosphate 1-dehydrogenase, aldehyde dehydrogenase 1A1, and isocitrate dehydrogenase has been reported in liver cancer and hepatoma cells.33, 34 However, mitochondrial dicarboxylate carrier in liver has not been previously considered.29
The other proteomic approach, such as MS-based separation techniques, which may be complement to fluorescence 2-dimensional difference gel electrophoresis, will identify the other proteins associating with AFP. Those proteins will also provide novel insights into the biology of HCC.
- 4Accumulation of genetic changes during development and progression of hepatocellular carcinoma: loss of heterozygosity on chromosome arm 1p occurs at an early stage of hepatocarcinogenesis. Genes Chromosomes Cancer 1995; 13: 163–167., , , , , , et al.
- 6Genetic instability and aberrant DNA methylation in chronic hepatitis and cirrhosis—a comprehensive study of loss of heterozygosity and microsatellite instability at 39 loci and DNA hypermethylation on 8 CpG islands in microdissected specimens from patients with hepatocellular carcinoma. HEPATOLOGY 2000; 32: 970–979., , , , , .
- 27Comparison of clinicopathological features of patients with hepatocellular carcinoma seropositive for alpha-fetoprotein alone and those seropositive for des-gamma-carboxy prothrombin alone. J Gastoenterol Hepatol 2001; 16: 1290–1296., , , , , , et al.