Introduction

Bile-duct carcinoma (cholangiocarcinoma) which has been generally considered to be less common than hepatocellular carcinoma, has increased in incidence over the last 30 years [1]. In addition, it has very high incidence in some countries where its etiology is linked to parasitic disease of the liver [2]. Cholangiocarcinoma arises from the epithelial cells or cholangiocytes that line bile ducts. The tumor is composed of glandular and tubular structures made of carcinoma cells set in an abundant fibrous stroma. Three main forms of cholangiocarcinoma have been described and categorised based on the position of the tumor in the liver and on the extrahepatic bile duct tree: intrahepatic or peripheral bile-duct carcinoma, hilar adenocarcinoma (Klatskin tumor) and carcinoma of the extrahepatic bile ducts.

Contrary to most cases of hepatocellular carcinoma, the stroma in cholangiocarcinoma is abundant, sclerous, sometimes with calcifications and may be extensive, and submerges the scanty tumoral tubules. The number of modified fibroblasts (myofibroblasts) present is elevated in the intratumoral stroma and is correlated with the degree of tumor fibrosis [3]. Surrounding the tumor, hepatic stellate cells and/or portal fibroblasts also acquire expression of α-smooth muscle actin (αSMA) and seem in direct continuity with intratumoral myofibroblasts [3]. It is well known that these (myo)fibroblastic cells are involved in deposition of extracellular matrix and in matrix remodelling [4]. Stromal cells express matrix metalloproteinase (MMP)-1, MMP-2, MMP-3, MMP-9, or tissue inhibitors of MMPs, TIMP-1 and TIMP-2 [5]. This expression of MMPs and TIMPs is stronger in cholangiocarcinomas that show severe invasion [5]. The stroma surrounding tumoral cells has recently attracted much interest as a factor in tumor growth and metastasis and a possible target for future therapies (for review see [6]). Though several risk factors for developing cholangiocarcinoma have been identified, the pathogenesis of the disease remains poorly understood. Diagnosis of cholangiocarcinoma is often late and patients often have advanced stage disease at the time of diagnosis, leading to a poor clinical outcome. It is therefore important that we understand better the pathogenesis of the disease. In this paper we report the findings of a proteomic study of peripheral cholangiocarcinoma, in comparison with the surrounding non-tumoral tissue, and identify differentially expressed proteins that may play a role in disease development or progression.

In this study we identified several proteins over- or under-expressed in the cholangiocarcinoma tissue compared to normal liver tissue and also looked at the cellular localization of three of these differentially-expressed proteins, 14-3-3 protein, α-SMA and periostin. Two of these (α-SMA and periostin) were expressed in stromal fibroblasts whilst 14-3-3 protein was expressed in cholangiocarcinoma cells. 14-3-3 protein has recently been shown to have a role as an adaptor protein involved in p53 stabilization and thus in regulation of apoptosis [7]. α-SMA is over-expressed in fibrotic tissue and many tumors and has a strong correlation with prognosis in colon cancer for example [8]. Periostin has been detected in several tumor types and has recently been reported to be highly expressed in fibroblasts derived from cholangiocarcinomas and also to correlate with prognosis [9].

With the increased interest in the role of tumor stroma and cancer associated fibroblasts, it is of interest that some of the proteins identified as being differentially expressed were located in the stroma and are markers of myofibroblast differentiation and have been shown to have roles in promoting tumor growth or spread in other cancers.

Materials and Methods

Liver Tissue Samples

Human liver tissue samples used in this study were selected from the files of liver tissue bank of the Department of Pathology (Pellegrin Hospital, Bordeaux, France). All of the cases included came from resected livers with peripheral cholangiocarcinoma from patients at the University Hospital of Bordeaux during the period 2003–2005. Fourteen patient samples were initially screened by histology (see Table 1) and the four most typical samples were chosen for further proteomic study. The samples on which proteomic analysis was performed consisted of three female and one male patient (mean age of 59.5 years). Brief clinical and pathological characteristics of all patients are summarized in Table 1. Cholangiocarcinoma cases were chosen on the basis that they were of typical morphology and arose on a nearly normal background liver, in general without significant fibrosis or inflammation. Furthermore, serological viral markers were negative.

Table 1 List and main clinical characteristics of patients involved in the study
Full size table

For all cases, liver tissue specimens were freshly received (within 30–40 min after vascular clamping) and sampled at the Department of Pathology. The specimens of whole tumoral and non-tumoral liver tissue were carefully sampled and representative fragments of tumoral and non-tumoral liver tissue (the latter taken at a distance of at least 2 cm from the tumor) were immediately snap-frozen in liquid nitrogen and stored at −80°C until used for liver protein preparation. Furthermore, the whole tumoral and non-tumoral liver was then fixed in 10% formalin, embedded in paraffin and routinely processed for diagnosis purposes. For immunohistochemistry, samples of tumoral and non-tumoral liver tissue were snap-frozen in isopentane cooled in liquid nitrogen and stored subsequently at −80°C until cryosectioning.

Materials and Reagents for Proteomic Analysis

Isoelectrofocusing (IEF) was carried out on precast 18-cm immobilized pH gradient (IPG) strips (pH 4–7) in an IPGphor II Electrophoresis System (GE Healthcare Bio-Sciences AB, Uppsala, Sweden). Dithiothreitol (DTT) was purchased from Sigma (L’Isle d’Abbeau, France), proteinase inhibitor cocktail tablets were purchased from Roche Diagnostics GmbH (Mannheim, Germany) and other reagents were purchased from GE Healthcare (Orsay, France). All chemical products used for mass spectrometry analysis were purchased from Sigma (St Louis, MO, USA). Sequencing grade modified trypsin used for protein digestion was purchased from Promega (Charbonnières, France).

Preparation of Liver Tissue Protein Samples

Frozen liver tissue samples from either tumors or non-tumoral regions of the liver were placed in liquid nitrogen, and ground thoroughly to a very fine powder with a mortar and pestle. The liver tissue powder (80 to 120 mg) was then homogenized with 5 volumes of lysis buffer (8 M urea, 2 M thiourea, 4% w/v CHAPS, 40 mM Tris, 50 mM DTT, 1 mM EDTA and protease inhibitor cocktail). Liver tissue homogenates were then vortexed, sonicated at 4°C and allowed to sit on ice for 1 h. Crude extracts were centrifuged at 10,000 g for 15 min and the supernatants were purified by precipitation following the manufacturer’s instructions described in the Plus One 2D Clean-Up Kit (GE Healthcare Bio-Sciences AB). Precipitates from each of the non-tumoral and tumoral extracts were subsequently solubilized in the solubilization buffer (8 M urea, 2 M thiourea and 4% w/v CHAPS). Determination of protein concentration was performed as described in the PlusOne 2D Quant Kit (GE Healthcare Bio-Sciences AB). The samples were visualized with a 10% sodium dodecyl sulfate (SDS)—polyacrylamide gel electrophoresis (PAGE) to confirm the quality of the protein extraction and protein concentrations.

Cyanine (Cy) Dye Labeling

To perform two-dimensional fluorescence difference gel electrophoresis (2-D DIGE), the experimental strategy was based on minimal labeling with dye swapping. 50 μg tumoral and non-tumoral protein extracts were labeled with Cy3 or Cy5 and the internal standard which was prepared by mixing an equal amount of all tumoral and non-tumoral samples, was labeled with Cy2. Labeling reactions were carried out according to the manufacturer’s instructions. Each sample was labeled with 400 pmol of CyDye (GE Healthcare Bio-Sciences AB) on ice for 30 min in the dark and reactions were stopped by adding 1 μl of 10 mM lysine. The CyDye-labeled samples (non-tumoral, tumoral and internal standard) were mixed and added to an equal volume of the solubilization buffer containing 130 mM DTT, 1% v/v IPG buffer and a trace of bromophenol blue.

Two-Dimensional Electrophoresis (2-DE) and Image Analysis

Analytical 2-DE was performed as follows: the CyDye-labeled protein samples were electrophoresed in the first dimension on IPG pH 4–7 gels. After rehydratation at room temperature for 12 h in a reswelling tray (GE Healthcare Bio-Sciences AB), IEF was carried out at 300 V for 750Vh, at 1,000 V for 3,000Vh and at 10,000 V for 65,000Vh in the dark. The gel strips were then equilibrated in 15 ml of the equilibration buffer A (50 mM Tris-HCl pH 8.8, 6 M urea, 30% v/v glycerol, 2% w/v SDS, 50 mM DTT) for 15 min with gentle shaking, and then in 15 ml of the equilibration buffer B (50 mM Tris-HCl pH 8.8, 6 M urea, 30% v/v glycerol, 2% w/v SDS, 2.5%w/v iodoacetamide, with a trace of bromophenol blue). The equilibrated strips were loaded on to the top of 10% SDS-PAGE (24x20 cm) and sealed with 1% w/v agarose. Separation in the second dimension was carried out in Tris-glycine buffer [10]. After 2-DE, gels were scanned with a Typhoon TRIO scanner (GE Healthcare Bio-Sciences AB) using filters appropriate for each dye’s excitation and emission wavelength. To obtain an adequate amount of the proteins from the individual spots for identification, 400 μg of the tumoral samples and 400 μg of the non-tumoral samples were separately run on 2D electrophoresis and were stained using a colloidal CBB G-250 procedure [11]. Preparative gels were scanned with an Image Scanner II (GE Healthcare Bio-Sciences AB).

Statistical Analysis

After the images were scanned, GE Healthcare DeCyder software 5.01 was used for differential gel analysis. The 2-D image of tumoral liver tissue was compared with that of the non-tumoral liver tissue via the internal standard sample. Specifically the protein expression analysis was performed for each of the four gels in parallel using the differential in-gel analysis (DIA) module of DeCyder using a value of 2,500 as the initial estimate of protein spots present. DIA analysis allows for the direct comparison of intensities of specific protein spots between different samples within the same gel. In this case, protein intensities were compared between the tumoral proteome and non-tumoral proteome. These DIA analyses were collated into a single analysis using the biological variation analysis module of DeCyder, and final values for the expression ratio of specific protein spots between tumoral liver tissue and non-tumoral liver tissue were determined for >2 or <−2 fold differences. The statistical significance of each expression level was calculated using Student’s t test on the logged ratio. Results obtained by DeCyder software were compared to those obtained with Progenesis SameSpot software (version 3.2, Nonlinear Dynamics). Significantly differentially expressed spots between tumoral and non-tumoral tissue were selected using a type one error risk of 1%, at least a power of 80% and that they were found on all the images.

Digestion of Proteins

Protein spots were excised from 2-DE gels that had been stained with Coomassie blue G-250. The excised spots were first destained by washing in milliQ distilled water then dehydrated in 50 μl of acetonitrile (ACN) followed by rehydration in 50 μl of 100 mM ammonium bicarbonate for 15 min at 37°C. An equivalent volume of ACN was then added to the mix and further incubated for 15 min at 37°C. Samples were then dried by vacuum dessicator. Sequencing grade modified trypsin was prepared from a stock solution at 0.1 μg/μl and diluted in 25 mM ammonium bicarbonate to a final concentration of 10 ng/μl. Dehydrated spots were incubated in 25 μl of trypsin at 10 ng/μl (a total of 250 ng per spot) overnight at 37°C. The supernatant was then collected in a 0.5 ml microfuge tube and the digested peptides extracted sequentially in 50 μl of 40% ACN—1% formic acid (FA), then 10 μl of 25% ACN—1% FA and finally 25 μl of 60% ACN. All samples were then dried by evaporation using a vacuum dessicator.

Mass Spectrometry (MS)

After trypsin digestion and evaporation, the peptides were resolubilised in 6 μl of Switchos solvent (2% ACN, 0.05% TFA) for analysis by nano-LC MS/MS using a nano-chromatography liquid LC Packings system (Dionex, Amsterdam, Holland) coupled to a 2000-QTRAP mass spectrometer (Applied Biosystems, Courtaboeuf, France). For each sample, 5 µl was injected into a pre-column (C18 PepmapTM 300 μm ID x 5 mm) using the Switchos unit. After desalting for 3 min with Switchos solvent, the pre-column was switched online with the analytical column (C18 PepmapTM 75 μm ID x 150 mm) pre-equilibrated with 100% solvent A (2% ACN—0.1% FA). Peptides were eluted from the pre-column into the analytical column and then into the mass spectrometer by a linear gradient of 0% to 50% of solvent B (90% ACN—0.1% FA) over 65 min at a flow rate of 300 nl/min.

Data acquisition was carried out using the IDA (Information-Dependant Acquisition) software of Analyst 1.4.2 (Applied Biosystems). The data from MS and MS/MS were continuously recorded with a cycle duration of 3 sec. For each MS scan, two precursors were selected for fragmentation on the basis of their intensity (greater than 20,000 cps), their charge state (2+, 3+) and if the precursor had already been selected for fragmentation (dynamic exclusion). The collision energies were automatically adjusted according to the charge state and ionic mass of selected precursors.

Peptide Identification

For protein identification, the results of analysis by nano-LC MS/MS were used to search the SwissProt database (Swiss-Prot Release 51.6, 6 February 2007, 257 964 sequences) using the Mascot software (version 2.2, Matrix Science, UK) with the following criteria: species Homo sapiens, 0.5 Da tolerance for peptide and peptide fragment mass, a single missed cleavage site allowed during trypsin digestion, and carbamidomethylation of cysteine residues (due to alkylation of -SH groups by iodoacetamide) and oxidation of methionine as variable modifications. Protein identification was validated if at least 2 peptides had a score of greater than 25, or one peptide had a score of greater than 50 at a confidence level of at least 95%.

Western Blot

6 μg from each protein sample extracted for 2-DE analysis was run on a 10% SDS-polyacrylamide gel and transferred to a nitrocellulose membrane (Hybond ECL, GE Healthcare, Orsay, France). The membrane was blocked in TBST buffer (20 mM Tris-HCl pH 7.6, 137 mM NaCl, 0.1% Tween 20) containing 1% blocking buffer (Roche Diagnostics, Meylan, France) before adding the periostin [osteoblast specific factor-2] antibody (rabbit polyclonal, BioVendor, Heidelberg, Germany) at a 1:10,000 dilution in TBST buffer with 0.5% blocking buffer. Membranes were incubated overnight at 4°C. After washing, the membranes were incubated with a secondary peroxidase-conjugated swine anti-rabbit antibody (1:1,000 in TBST buffer with 0.5% blocking buffer). Signal was detected with a BM chemiluminescence blotting kit (Roche).

Immunohistochemistry

All samples had tissue embedded for frozen sectioning. This tissue was taken from the periphery of the tumor sample so that it contained both tumor tissue and non-tumoral liver which was histologically normal. Liver samples snap-frozen in isopentane/liquid nitrogen were sectioned on a cryostat at 5 μm, fixed for 5 min in 4% paraformaldehyde and then stained by immunohistochemistry using primary antibodies directed against α-smooth muscle actin (mouse mAb, Dako, Carpinteria, USA), periostin [osteoblast specific factor-2] (rabbit polyclonal, BioVendor, Heidelberg, Germany) and 14-3-3 B/a protein (rabbit mAb, Novus Biologicals, Littleton, USA). Detection of labeling was carried out using Alexa-labeled secondary antibodies; Alexa 594 goat-anti mouse IgG (Invitrogen SARL, Cergy Pontoise, France) and Alexa 488 goat anti-rabbit IgG (Invitrogen). Slides were mounted in aqueous mounting medium (Dako, Carpinteria, USA) for viewing and photography. For staining by immunoperoxidase, incubation in the primary antibody was followed by incubation in secondary antibodies linked to HRP (Envision™ + system, Dako) and the positive staining was detected using diaminobenzidine. Slides were counterstained with hematoxylin for viewing and photomicroscopy. Specificity of staining was confirmed by incubation in non-immune serum and in the absence of the primary antibody.

Results

Identification of Differentially Expressed Proteins

Analysis of 2-D gels showed that there were approximately 2,400 spots detected on protein samples from 8 biopsies from four patients (Fig. 1). Of these, analysis using DIGE showed that 172 protein spots exhibited significant variations (at least 2-fold changes) at a p < 0.01 confidence level and at least a power of 80%. 49 protein spots were up-regulated in the tumoral liver tissue compared to the non-tumoral liver tissue. Nano-LC MS/MS analysis of these spots led to the identification of 70 different proteins. Among 123 down-regulated protein spots, 51 spots corresponding to 68 different proteins were identified. The list of proteins that were found to be under-expressed or over-expressed in the tumoral liver tissue compared with the non-tumoral liver tissue, are shown respectively in Table 2 and Table 3. The expression ratios of these proteins and their Swiss-Prot identification numbers are also specified.

Fig. 1
figure 1

2-D DIGE gel images. The image A shows the internal standard master. Indicated numbers correspond to identified spots (refer to Tables 2 and 3). The three proteins selected for further analyses are marked with * (periostin), ** (smooth muscle actin) and *** (14-3-3 isoforms). The image B associates Cy3-labeled non-tumoral sample image (red) with Cy5-labeled tumoral sample image (green). 2-DE was performed using a pH range of 4-7 in the first dimension and SDS-PAGE (10%) in the second

Full size image
Table 2 Under-expressed proteins
Full size table
Table 3 Over-expressed proteins
Full size table

Proteins Under-Expressed in the Tumoral Liver Tissue Compared with the Non-Tumoral Liver Tissue

The proteins that were found to be under-expressed in the tumoral liver tissue compared with the non-tumoral liver tissue are shown in Table 2. Many of these proteins were involved in cellular metabolism while a number were cytoskeletal proteins. Several of these proteins have also been reported to be differentially expressed in either hepatocellular carcinoma or other tumor types, for example alcohol dehydrogenase and aldehyde dehydrogenase [12], carbamyl phosphate synthase [13], selenium binding protein 1 and DNA helicase RuvB-like protein 2 ([14–16]).

Proteins Over-Expressed in the Tumoral Liver Tissue Compared with the Non-Tumoral Liver Tissue

As shown in Table 3, proteins that were found to be over-expressed in the tumoral liver tissue compared with the non-tumoral liver tissue could be categorised by functions into several groups including cytoskeletal proteins, apoptosis/survival and cellular metabolism. Some of the proteins identified have been reported to be differentially expressed in hepatocellular carcinoma, for example elongation factor 1 delta, ERO-1 like protein and ezrin ([17–19]). Proteomic screening confirmed the over-expression of several proteins that have previously been shown by other methods to be highly expressed in tumor or tumoral stroma. An example of one of these proteins was α-smooth muscle actin, a marker of stromal cell activation that has been linked to poor prognosis in colon cancer [8] and is present in stromal myofibroblasts in several tumor types [3]. The proteomic screen also showed over-expression of novel proteins such as the 14-3-3 proteins which are phospho-serine/phosphothreonine-binding proteins that may promote cell survival and have been suggested to be a possible target for therapies aimed at enhancing killing of cancer cells [20]. Other proteins that were over-expressed and could be postulated to have a link to tumorigenesis or cancer growth or spread included pigment epithelium derived factor precursor, a potent anti-angiogenic protein that has the potential to limit tumor growth [21]. Interestingly, we also found increased expression of periostin in the tumoral liver tissue compared to non-tumoral liver tissue and this protein has been described in other tumor types for example breast cancers, melanoma and colon cancers ([22–24]). The elevation of these proteins in tumoral liver tissue versus non-tumoral liver tissue was also shown to be the case in 4 out of 4 patient samples as shown in Fig. 2. In the case of α-smooth muscle actin, 14-3-3 proteins and periostin, immunohistochemistry was carried out on multiple samples to confirm both the elevated expression in the tumoral liver tissue and the localization of the expression to particular cell populations (see below).

Fig. 2
figure 2

Quantitative analyses illustrating 14-3-3 isoforms a, aortic smooth muscle and cytoplasmic 1 actin isoforms b and periostin c overexpression in tumoral liver (T) compared with non-tumoral liver (NT) (dotted lines and full line show the data for the 4 patients and the mean respectively). On the right, enlarged regions of 2-D gel images showing pixel volume distributions between non-tumoral and tumoral samples. 2-D DIGE analysis was performed using DeCyder software. The Cy3 to Cy5 average ratios (R) were calculated using the biological variation analysis module of DeCyder. An independent Student’s test was applied to all samples, yielding a p-value within the 99th percentile confidence level for these three proteins

Full size image

Western blot analysis of periostin expression in the same samples as those used for 2-D DIGE, confirmed that perisotin was highly expressed in four out of four tumor samples while the normal non-tumoral liver from the same patients showed undetectable periostin levels (Fig. 3). Periostin appeared as a single band of approximately 75 kDa.

Fig. 3
figure 3

Western blot for periostin in non-tumoral (NT) and tumoral (T) liver tissue from the same 4 patients used for 2-D DIGE analysis. Numbers refer to individual patients. All of the tumor samples (T1, T2, T3, T4) show very strong periostin expression at approximately 75 kDa while non-tumoral liver samples show undetectable periostin levels (NT1, NT2, NT3, NT4)

Full size image

Immunohistochemistry

Routine histology performed on samples of tumoral liver tissue showed typical features of peripheral cholangiocarcinoma. The surrounding non-tumoral liver tissue was in general relatively normal in appearance with only slight signs of nonspecific inflammation in a normal appearing liver parenchyma. Immunohistochemistry performed to detect three proteins that were found to be differentially expressed in the tumoral liver tissue confirmed high levels of expression of α-smooth muscle actin, 14-3-3 proteins and periostin in the cholangiocarcinoma tissue but not in the adjacent non-tumoral liver tissue in all of the samples that were examined. 14-3-3 proteins were found to be differentially expressed in the tumoral biliary cells themselves and not expressed in the tumoral stroma or in hepatocytes of the surrounding non-tumoral liver (Fig. 4a and b). Some expression was detected in normal intrahepatic biliary epithelial cells. α-Smooth muscle actin was found to be strongly expressed in myofibroblasts in fibrous stroma within the tumor and in myofibroblasts that infiltrate between the carcinoma cells (Fig. 4c and d). In non-tumoral liver tissue only low levels of α-smooth muscle actin expression were seen; however, expression was normally found in structures where it would be expected to be present, such as vascular smooth muscle cells surrounding arterioles in the portal areas. There was no evidence of fibrotic changes with associated high levels of α-smooth muscle actin-positivity in surrounding non-tumoral liver parenchyma. The extracellular matrix protein periostin was found to be highly expressed in the tumor stroma and in and around myofibroblasts present within the tumoral liver tissue (Fig. 4e and f). Periostin staining was not seen in non-tumoral liver sections from the same subjects. Taking into account the similar location of periostin and of α-smooth muscle actin which suggested that periostin was produced by activated myofibroblasts within the tumoral liver tissue, we carried out double immunostaining to colocalise these two proteins. Double immunostaining showed that stromal myofibroblasts were positive for both α-smooth muscle actin and periostin but that the periostin had a wider distribution, we believe, because it is secreted into the extracellular matrix in addition to being localised within the myofibroblasts themselves (Fig. 5a, b and c).

Fig. 4
figure 4

Immunostaining of differentially expressed proteins in cholangiocarcinoma (a, c, e) compared to adjacent non-tumoral liver tissue (b, d, f). 14-3-3 protein staining is seen in the tumoral cells within the cholangiocarcinoma (CC) but not in stromal cells and non-tumoral liver tissue (NTL) (a). It can be seen that non-tumoral liver tissue shows no visible expression of 14-3-3 protein (b). Strong staining for α-smooth muscle actin is observed in stromal myofibroblasts within the tumoral tissue (c). α-Smooth muscle actin staining is only seen in few smooth muscle cells associated with blood vessels in the non-tumoral liver parenchyma (d). Periostin staining shows a similar distribution to α-smooth muscle actin staining with periostin being detected in stromal connective tissue and myofibroblasts (e), while in the non-tumoral liver tissue (f), there is no apparent staining for periostin. Magnification x200

Full size image
Fig. 5
figure 5

Immunofluorescence comparing the distribution of α-smooth muscle actin to that of periostin. In the cholangiocarcinoma tissue, there is strong staining for α-smooth muscle actin in the stromal cells within the tumor (a). A similar distribution is seen with periostin staining (b). Merge image for the two proteins (c) shows that α-smooth muscle actin and periostin are colocalized in stromal cells, though periostin (green) has a slightly wider distribution than α-smooth muscle actin (red) as it is both intracellular in myofibroblasts and deposited in the extracellular matrix. Magnification x200

Full size image

Discussion

In this study we have compared the proteome of peripheral cholangiocarcinoma to that of the surrounding non-tumoral liver tissue. This represents a novel study of differentially expressed proteins in cholangiocarcinoma which to our knowledge has not been previously reported. We have used a strategy of examining a small number of samples with tumor and control tissue from the same subject and then further analysis by immunohistochemistry and Western blotting of some of the proteins found to be differentially expressed. Cholangiocarcinoma is a relatively rare cancer and thus sample sizes tend to be small, however, we assume that by choosing the most histologically typical tumors that were on a relatively normal liver background, we have identified proteins that are consistently and significantly differently expressed between cancerous tissue and the normal liver. The analysis of 2-D gels showed a large number of differentially-expressed proteins in the tumoral liver tissue when compared to the surrounding non-tumoral and morphologically normal liver that was taken from the same patients. A number of these proteins could be predicted to play roles in tumorigenesis or tumor growth. Some of the identified proteins have been shown to be differentially expressed in experimental models or other tumor types such as colon or liver cancer ([15, 16, 25, 26]). Overall the proteins found to be over-expressed or under-expressed in tumoral tissue fell into the categories shown in Fig. 6. Over-expressed proteins largely consisted of cytoskeleton, metabolism and cell cycle, apoptosis or survival proteins, while the majority of proteins found to be under-expressed in tumor tissue were proteins involved in metabolic pathways.

Fig. 6
figure 6

Diagrammatic representation of the functions of proteins that were identified as either over-expressed (a) or under-expressed (b) in the tumoral compared to non-tumoral liver tissue. Assignments were made on the basis of information provided in the SWISS-PROT website (www.expasy.ch)

Full size image

Of the over-expressed proteins, those that may have a role in tumorigenesis or that have been linked to other cancers include: 60 S acidic ribosomal protein P0 over-expression which has been linked to proliferation in breast and liver cancer [27], alpha-actinin 4 which is also differentially expressed in hepatocellular carcinoma [28], annexin V which has been shown to be elevated in chemically induced hepatocarcinoma [29]; as have elongation factor 1 delta [17] and ezrin [19]. Other proteins that have also been shown to be elevated in hepatocellular carcinoma include fibrinogen gamma chain precursor [30], glutathione S transferase [31], glycerol-3-phosphate dehydrogenase [32], histone-binding protein RBBP4 [33] and nucleophosmin [34]. Lamin-A/C was elevated in sera of patients with a hepatocellular carcinoma [35]. Other proteins have been shown to be associated with different tumor types such as L-lactate dehydrogenase B chain which is elevated in lung cancer [36], niban which is over-expressed in renal tumors [37], and non-syndromic hearing impairment protein 5 which shows altered expression in colorectal cancer [38].

Of particular interest were proteins that have not yet been shown in liver cancers such as 14-3-3 proteins and periostin. 14-3-3 proteins have been shown to have roles in suppressing apoptosis or regulating cell survival ([20, 39–41]) and these may represent a novel indicator of cancer transformation in hepatocytes or biliary epithelium. In addition to its role in regulation of apoptosis, 14-3-3 sigma protein, also known as stratifin, is now known to be secreted via exosomes and induces expression of the matrix metalloproteinases, MMP-1 and MMP-3. This may also link its over-expression to increased cell migration or capacity for invasion [42].

Periostin is a secreted protein that has integrin binding sites and has been shown to be expressed by tumor cells and tumor stroma in melanoma and colon cancer ([23, 24]). We have shown elevated expression in cholangiocarcinoma and this expression was then localised to the tumor stroma (myofibroblasts) by immunohistochemistry. Recent examination of fibroblasts isolated from cholangiocarcinoma has shown that these cells are capable of producing periostin and that high levels of periostin production correlate with poor prognosis [9]. Further studies using in situ hybridization for periostin mRNA expression may clarify whether periostin is expressed in cholangiocarcinoma uniquely in the tumor-associated myofibroblasts. It has been suggested recently that the stroma may play an important role in tumor growth and progression and periostin may be a novel candidate protein for regulation of cancer cell behaviour. Periostin has been shown to induce cell migration and also epithelial-mesenchymal transition in pancreatic cancer cells [43], thus it may represent a novel mediator of tumor growth in cholangiocarcinoma. In addition, periostin has been reported to be elevated in serum of some patients with cancers including thymoma and non-small cell lung carcinoma [44]. The detection of periostin or other over-expressed proteins such as 14-3-3 preoteins in serum may therefore prove to be useful markers for detection of malignancy.

In view of the fact that cholangiocarcinoma has shown a recent increase in incidence, along with the difficulty in early diagnosis, we believe that studies such as this may be useful in identifying proteins that may have a role in cancer development or progression. The use of differentially expressed proteins such as those identified in this study may provide the possibility of finding markers for detection of cholangiocarcinoma, although this is complicated by the fact that some of these proteins are differentially expressed in other malignancies. Identifying a unique pattern of expression or the possibility that these proteins may represent serum or bile markers should also be considered. Further studies are required to examine possible links between these proteins and disease outcome or prognosis as well as comparisons between the proteome of hepatocellular carcinoma, non-malignant disease of the liver (cirrhosis) and stroma of metastatic tumors.