Serial analysis of gene expression identifies putative metastasis-associated transcripts in colon tumour cellines

We have used serial analysis of gene expression (SAGE) to identify gene expression differences between a primary colon tumour cell line (SW480) and an isogenic lymph-node metastasis cell line (SW620). Differential expression was confirmed for the following genes: keratin K5, cystatin S, serum amyloid A, the human homologue of yeast ribosomal S28 and the p32 subunit of human pre-mRNA splicing factor SF2. Expression of confirmed differences were also analysed in other metastatic cell lines. © 2000 Cancer Research Campaign

ribosomal S28, the p32 subunit of human pre-mRNA splicing factor SF2, keratin type II K5, cystatin S and serum amyloid A (SAA) all showed differential expression in the metastatic cell line (SW620) relative to the primary tumour cell line (SW480) ( Figure  1A, lanes 1 and 2) ( Table 2). The cDNA products generated from the RT-PCR analysis were purified and used for Northern blot analysis ( Figure 2). All five cDNA probes hybridized to transcripts of the expected size and differential expression was also further confirmed.

Analysis of expression in other cell lines
We examined the level of expression of the five confirmed differentially expressed genes in other colon carcinoma cell lines by semi-quantitative RT-PCR. These included a pair of cell lines: SW1116, a human colon adenocarcinoma line and 498LI (Lieberman et al, 1991), a metastatic variant of SW1116 selected in vitro ( Figure 1B, lanes 1 and 2) ( Table 2). Three metastatic cell lines, LoVo, Colo201 and T84, all originating from primary colon tumours, were also examined ( Figure 1A, lanes 3-5). The expression data is summarized in Table 2.

DISCUSSION
Our SAGE analysis has allowed a comparison between highly expressed genes and, thus far, we have further examined the expression of ten genes showing differential expression. We have found that a SAGE analysis based on approximately 5000 tags per cell type results in a number of false-positives, despite the 'housekeeping' genes GAPDH and β-actin (Cleveland et al, 1980;Fort et al, 1985) showing a similar tag number ( Table 1). All of the false-positives that we identified were represented by <10 tags in  abundance. Therefore, increasing the number of tags sequenced is likely to reduce or eliminate the number of false-positives. We have confirmed differential expression of five genes between SW480 and SW620 (Table 2). Cystatin S and keratin type II K5 were not expressed at a significant level in the four other metastatic cell lines that we examined (LoVo, Colo201, T84 and 498LI) or in the primary tumour cell line SW1116. Cystatin S is a member of the cysteine proteinase inhibitor superfamily (Bobek et al, 1991) and other members, such as cystatin M, have been previously associated with metastatic spread (Sotiropoulou et al, 1997). Keratin K5 is a type II cytoskeletal intermediate filament protein (Fuchs and Weber, 1994) and the cytoskeletal structures are known to play a major role in cell motility and invasion, proliferation, differentiation and in the transduction of extracellular signals (Sherbet and Lakshmi, 1997).
Serum amyloid A (SAA) showed varying degrees of loss of expression in all of the metastatic cell lines that we examined. SAA is an acute-phase reactant and is found in the circulation as an apolipoprotein bound to HDL (high density lipoprotein) (apoSAA) (Strachan et al, 1989). The potential role that SAA may have in metastasis is currently unclear but the observed increase in expression of SAA in response to p53-induced apoptosis (Polyak et al, 1997) may be significant.
We also observed increased expression of the p32 subunit precursor of pre-mRNA splicing factor SF2, and the human homologue of yeast ribosomal S28 in the metastatic cell line(s) relative to their primary tumour cell line counterpart(s) ( Table 2). The interaction of p32 with the pre-mRNA splicing factor SF2 (Krainer et al, 1991;Honore et al, 1993) and its ability to bind hyaluronic acid (HA) (Deb and Datta, 1996) suggests a potential role in metastasis. Increased expression of the human homologue of yeast ribosomal S28 has also been found in colon tumours as compared to normal tissue . This indicates that increased expression of S28 may be correlated with colon cancer progression.
In conclusion, we have shown here that analysis of a moderate number of SAGE tags (approximately 10 000 in total) has allowed us to identify a number of differentially expressed genes that may be involved in colon tumour metastasis. However, our analysis has also shown that, with a moderate level of SAGE tags, comparison of the gene expression profiles must be interpreted with caution as tag abundances <10 do not consistently represent true differences in expression. This agrees with another recent SAGE analysis of endothelial cells which consisted of over 12 000 tags (de Waard et al, 1999). The five differentially expressed genes that we have confirmed have not been implicated in colon tumour progression  T/M is the ratio of tags from the primary tumour cell line SW480 (T) compared to the metastatic cell line SW620 (M). Genes written in bold were confirmed by independent methods (see text). *The 'housekeeping' genes GAPDH and βactin showed a similar tag abundance as expected Cell lines written in bold are metastatic and those written in roman are the primary colon tumour cell lines. Fold differences were calculated based on βactin ratios and relative to expression in primary tumour cell lines. The results are based on repeated experiments with an overall average standard error of the mean of 4.1%. Cell lines SW620, LoVo, Colo201 and T84 are described relative to SW480, and 498LI is described relative to SW1116; nd = not detected; a Fold differences are described as decreased expression in the metastatic cell lines relative to the primary tumour cell lines; b Fold differences are described as increased expression in the metastatic cell lines relative to the primary tumour cell lines previously and may contribute to our understanding of metastasis. However, it is clear that our study requires further analysis in order to show a direct association between the identified differentially expressed genes and the metastatic phenotype.