Cell culture and tissue samples

Human HCC cell line HepG2 was cultured in Dulbecco's modified Eagle's medium (Invitrogen, Carlsbad, CA) with 10% fetal bovine serum under 5% CO₂ in a humidified incubator at 37°C. Cells were harvested at 80–90% confluence. Normal liver tissue was obtained from a patient who underwent hepatectomy because of hepatic hemangioma. Clinical HCC samples and corresponding non-tumorous liver tissues were derived from HCC patients and kept frozen in liquid nitrogen immediately after separation. All samples were collected with informed consent in accordance with the standards of Institutional Human Subjects Protection Review Board.

SAGE and LongSAGE

Total RNAs were extracted from HepG2 and liver tissues using TRIzol reagent (Invitrogen, Carlsbad, CA) and then treated with DNaseI (Rhoch Diagnostics, Almere, The Netherlands) according to the manufacture's protocol. Polyadenylated mRNAs were then isolated using Oligotex Direct mRNA Midi Kit (QIAGEN). Three SAGE libraries were constructed starting with 200 ng polyA⁺ mRNA from HepG2, normal liver tissue and HCC tissue respectively according to SAGE protocol [20, 22]. LongSAGE was performed with the same amount of polyA⁺ mRNA from the same HCC tissue used in SAGE following the procedures described previously [21]. Sequencing of SAGE and LongSAGE tags were carried out using BigDye Terminator Cycle Sequencing Kits and ABI 3700 DNA Sequencers (Applied Biosystems, Foster City, CA). The sequence and frequency of 14-bp tags from SAGE and 21-bp tags from LongSAGE were extracted from the raw sequence files of concatenated di-tags using SAGE2000 analysis software version 4.5 (kindly provided by Dr. K. Kinzler, Johns Hopkins University School of Medicine).

Computational Analysis

To exclude sequencing errors, only tags detected at least twice in the four SAGE libraries (three SAGE libraries and one LongSAGE library) were included for further analysis. SAGEmap reliable mapping (http://www.ncbi.nlm.nih.gov/SAGE, UniGene Build #182) [23] was used to establish tag-to-gene assignments. Comparison analysis between SAGE libraries was performed using the IDEG6 software http://telethon.bio.unipd.it/bioinfo/IDEG6/[24]. Differentially expressed tags were selected according to the cut-off threshold (p value < 0.05, fold change > 3) calculated according to pairwise comparison algorithm [25]. The Cluster and Treeview programs were used to generate the average linkage hierarchical clustering and visualize changes of gene expression [26]. Functional classification of genes was performed using DAVID program http://david.abcc.ncifcrf.gov/ obtained from NIAID (National Institute of Allergy and Infectious Disease).

Real-time quantitative RT-PCR

Total RNAs were extracted from HCC samples and corresponding nontumorous liver tissues using TRIzol reagent (Invitrogen) followed by treating with DNaseI according to the manufacture's protocol. 2 μg of total RNA was reverse transcribed in a total volume of 20 μl containing 200 units of SuperScript II RNase H^- Reverse Transcriptase (Invitrogen), 50 mM Tris-HCl (pH8.3), 75 mM KCl, 3 mM MgCl₂, 10 mM dithiothreitol, 500 μM dNTPs each, 500 ng oligo(dT)₂₃ primer and 40 units of RNaseOUT at 42°C for 50 min, followed by inactivating at 70°C for 15 min.

For real-time quantitative RT-PCR, 1 μl of the first strand cDNA and 5 μl of 2× SYBR Green PCR mix (Applied Biosystems, Foster City, CA) was added to a final volume of 10 μl. The thermal cycles were performed on LightCycler (Roche) following conditions at 95°C for 30 s, 40 cycles at 95°C for 5 s, 60°C for 5 s and 72°C for 30 s. Each reaction was performed in triplicate. The threshold cycle (Ct) value was used to calculate the expression ratio of target genes to internal control gene (GAPDH) with the formula 2^{Ct(GAPDH)-Ct(target gene)}. Two-side student's t test was applied in the statistical analysis of quantitative RT-PCR data.

Summary of SAGE data

Correlation analysis of SAGE data

As one set of SAGE data of normal liver is available in the public Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/, accession number GSM785), we compared our data of normal liver with it to evaluate the correlation between these two independent libraries. The result indicated that the data from public database correlated well with our data, with Pearson's correlation coefficient 0.74 (Figure 1A).

We also compared the SAGE and LongSAGE data which were obtained from the same HCC tissue sample of the same patient. We extracted the 14-bp SAGE tags from the 21-bp LongSAGE tags. The converted SAGE library could be considered as a technical replicate. Totally 6,815 overlapping tags were detected between these two libraries. Thus among the 20,313 unique tags in the original SAGE library, only 33.5% of them are detected by the second library derived from LongSAGE tags. Most of the missed tags are low abundance tags for which detection by random sampling is a stochastic process. Another reason might be possible sequencing errors that resulted in many singleton tags. However, analysis of these overlapping tags showed remarkable consistency between these two libraries with Pearson's correlation coefficient 0.97 (Figure 1B). This result reassured us the reproducibility of SAGE method on measuring gene expression.

We further compared the data of HCC with that of HepG2 to find out how much the gene expression profile has changed in cell line comparing with the HCC tissue. The difference between HCC and HepG2 was very notable, with Pearson's correlation coefficient 0.55, even greater than difference between HCC and normal liver (Pearson's correlation coefficient 0.70) (Figure 1C and 1D). This result suggested that the gene expression profile of cell line has changed too much to be regarded as the reference of its corresponding tissue.

Identification and functional analysis of genes differentially expressed between normal liver and HCC

In order to identify genes differentially expressed between normal liver and HCC, we used the IDEG6 software to compare the normal liver library with both the HCC library and the second HCC library derived from LongSAGE. Only genes consistently up-regulated or down-regulated in both comparisons were considered to be differentially expressed. Totally 224 genes were identified to be differentially expressed, with the significance p < 0.05 and fold change > 3 (Additional file 1). Figure 2 shows the expression patterns of these altered genes in normal liver, HCC and HepG2. The top 20 up-regulated and down-regulated genes ranked by fold change were listed in Table 4 and Table 5 respectively. Consistent with previous results [18, 28], our data also indicated that GPC3 gene was significantly up-regulated in HCC with fold change of 69. Among the 224 differentially expressed genes, the expression level of 104 genes was significantly higher in HCC than in normal liver (p < 0.05), while 120 genes was significantly lower in HCC (p < 0.05). In addition to these 224 annotated tags, our list of differentially expressed tags also includes 99 tags that could be matched to multiple genes, and 109 tags that were differentially expressed but could not be mapped to any known gene (data not shown). Some of these novel tags are extremely highly expressed in HCC. For example, the tag TAAGTTTGGG is detected 24 and 17 times in two HCC libraries but is not detected at all in the other two libraries. It will be of interest to further study these tags. In the current study, however, we will focus on the annotated tags.

Functional analysis of the 224 differently expressed genes was performed using DAVID program. The 104 up-regulated genes were grouped into eight functional clusters and one unclassified cluster (Additional file 2). These genes activated in HCC were involved in biological processes of biosynthesis, cell proliferation, signal transduction, transport, response to external stimulus, and cellular metabolism et al. Genes related to biosynthesis include 10 ribosomal proteins. Consistent with our observation, increased expression pattern of ribosomal proteins in HCC were demonstrated by previous studies [29, 30]. Genes participate in cell proliferation and signal transduction processes, such as MCM7 and IGFBP1 were also reported to be up-regulated in HCC by other groups [31, 32].

The 120 down-regulated genes were assembled into seven functional Gene Ontology (level 3) categories including blood coagulation, cell death, cellular metabolism, transport, signal transduction et al., and one unclassified group (Additional file 3). We also tested the over-representation of GO terms of all levels in this list. The most significant cluster is 8 genes belongs to a subcategory related to acute inflammatory response with p < 0.0005 after Benjamini correction of multiple testing. This cluster includes complement factor I (CFI) and several genes (C1S, C1QA, and C8B) encoding subcomponents of complement component 1 and 8. Another significantly enhanced cluster includes 6 genes related to blood coagulation (p < 0.03). These results indicated the loss of liver function in HCC.

Taken together, functional classification of genes differentially expressed in HCC revealed dysregulated pathways which may contribute to the disease. Similar to our results, microarray studies carried out by other groups also indicated genes involved in protein biosynthesis and cell signaling were up-regulated in HCC, while genes expressed at lower level in HCC included complement proteins and clotting factors [33, 34]. Thus, data obtained from SAGE and microarray methods could cross-validate each other and reveal some consistent events in HCC despite its heterogeneity.

Validation of differentially expressed genes

By real-time quantitative RT-PCR analysis, increased expression was confirmed in 8 of these 10 genes. As shown in Figure 3, CCL20, S100P, PEG10, SGCE, XAGE-1, COL4A1, ZNF83, and TM4SF1 was significantly up-regulated in HCC compared to the corresponding nontumorous liver (p < 0.05). Expression level of PEG10 and SGCE was further examined in more samples (totally 32 pairs of HCC and corresponding nontumorous liver). Similar expression patterns of these two genes were observed in HCC, with Pearson's correlation coefficient 0.71 (Figure 4). No significant differences in expression level of TNPO2 and ZFYVE1 was detected between HCC and adjacent nontumorous liver, reflecting the heterogeneous feature of HCC. These results further validated the quantitative data of SAGE profiles and provided novel candidate genes which may help to illustrate the molecular mechanism of hepatocarcinogenesis.