Results

Overall, 32% of the CRC cell lines showed loss of SMAD4 expression (Table 1, Fig. 1), comprising 14 of 32 (44%) MSI⁻ lines and 0 of 14 MSI⁺ lines (P < 0.004, Fisher's exact test). In no case was a truncated protein band observed (Table 1). Several different types of mutation were found to account for the loss of expression (Table 1). Importantly, sequencing showed that all SMAD4 mutations were present in the homozygous or hemizygous state (Fig. 2), with no underlying wild-type sequence.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Western blot analysis of SMAD4 and β-actin in colorectal cancer cell lines. Shown are the results of Western blot analysis using the B8 antibody against SMAD4, and a monoclonal anti-β-actin for four cell lines (LS180, C99, CX1, and COLO678). Note the complete absence of SMAD4 expression in CX1 and COLO678.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Sequencing results of SMAD4 for HT29/CX1, VACO10, and CACO2. (a)i shows the mutated cell line HT29/CX1 (c.931C>T transition causing Q311X amino acid truncation) compared with (a)ii wild type. (b)i SMAD4 exon 5 of VACO10 showing the c.715C>T transition causing Q239X truncation, compared with (b)ii wild type. (c)i SMAD4 exon 8 of CACO2 showing the c.1051G>C nucleotide change, which results in a D351H missense change, compared with (c)ii wild type. Note the lack of a wild-type sequence underlying a mutated (starred) sequence. The corresponding wild-type base is shown in a box.

For one cell line (COLO678), the whole gene was homozygously deleted, whereas other cell lines showed partial homozygous deletions (exons 1–4 in COLO201/COLO205, and exons 10–11 in SW403). All these mutations were accompanied by allelic loss at SMAD4 (Table 1).

Putative splicing mutations were detected in three cell lines, SW480 and SW620 (derived from a primary tumor and metastasis) and C10. All these lines showed absent SMAD4 protein and allelic loss at SMAD4. Of these three lines, SW620 was tested by using the protein truncation test (PTT) and only normal-length mRNA was detected. Thus, although the mutation in SW620 does not lead to detectable abnormal mRNA splicing, we cannot exclude the possibility that it has pathogenic effects through reduction of normal mRNA levels, which could be assessed by using RNase protection assay.

Nonsense SMAD4 mutations, accompanied by allele loss and absent protein, were detected in two cell lines: HT29/CX1 (Q311X, exon 7) and VACO10MS (Q239X, exon 5). Missense mutations, plus allelic loss, were found in four lines: CACO2 (D351H), C80 (D351H), SW948 (D537Y), and SW1222 (L540R). SMAD4 protein was present in all of these four lines.

Four MSI⁻ cell lines had absent SMAD4 protein and allelic loss, but no detectable SMAD4 mutation (Table 1). None of the 14 MSI⁺ cell lines possessed a pathogenic SMAD4 mutation, a highly significant difference from the MSI⁻ lines (P < 0.005, Fisher's exact test). Known SMAD4 intronic polymorphisms (23) were also detected (data not shown) in MSI⁻ and MSI⁺ lines.

Overall, there was a striking level of allele loss at SMAD4 and DCC in all of the MSI⁻ lines, with 30/32 (94%) showing loss at SMAD4 and 30/32 (94%) with loss at DCC (Table 1). In accordance with the sequencing data, all lines with allelic loss showed complete absence of one microsatellite allele; thus, all copies of 18q21 in each cell were derived from the same chromosome-18 homologue, even where the cancer was polyploid and had more than one copy of 18q. In contrast, 11/12 (92%) MSI⁺ lines showed heterozygosity for at least one marker each at SMAD4 and DCC, indicating that loss of 18q material is not critical for the development and progression of these tumors. The different levels of 18q21 loss in MSI⁻ and MSI⁺ tumors were confirmed in a set of 61 sporadic colorectal tumors, in which mean loss at four markers close to SMAD4 and DCC was 4% in 14 MSI⁺ cancers and 53% in 47 MSI⁻ cancers (P < 0.003, Fisher's exact test; data not shown).

The molecular cytogenetic and karyotype data are summarized in Table 1. In the lines with 18q21 loss, the amount of chromosome 18q21 material was always less than the overall ploidy. For example MSI⁻ cell lines with a modal chromosome number of less than 60 (C84, SW837, C99, C125, SW480, and SW620) had only one copy of chromosome 18q21, whereas cell lines with near-triploid karyotypes (SW403, HT29/CX1, HT55, COLO201, COLO205, C32, C106, C80, PC/JW, SW1417, VACO4A, and HCA46) had two copies of chromosome 18q21. Two lines with hypotetraploid (LIM1863) and hypohexaploid (C70) karyotypes had three and four copies of chromosome 18q21, respectively. Typing of extra markers mapping to the p-arm and 18q telomeric region confirmed comparative genomic hybridization (CGH)/spectral karyotyping (SKY) observations (data not shown).

In most cases, the cytogenetic analysis showed the whole 18q arm to be deleted, but there were exceptions. SW1417 was homozygous for microsatellites at SMAD4 and showed loss of SMAD4 protein, but was heterozygous for markers near DCC. The microsatellite data for SW1417 were consistent with an interstitial deletion that targeted SMAD4, but left DCC intact. CGH data for SW1417 also showed a deletion around SMAD4, rather than loss of the whole arm. Conversely, C84 and SKC01 were heterozygous for microsatellites at SMAD4, but homozygous for microsatellites around DCC. Again this was substantiated by CGH data, which showed loss of chromosome 18 distal to 18q21. In these lines, SMAD4 expression was retained and it appeared that DCC or another tumor suppressor was being targeted, leaving SMAD4 intact. The line HT29/CX1 had two apparently identical copies of chromosome 18 with SMAD4 mutations, plus one deleted 18q. This line was heterozygous at D18S474, but homozygous at the rest of the microsatellites and for the SMAD4 mutation, showing that the breakpoint for the deletion almost certainly lay just centromeric to SMAD4. Finally, cell line COLO678 was found to be homozygous for markers mapping to both the SMAD4 and DCC regions, but heterozygous for markers on the p-arm and 18q telomeric region. Cytogenetic data were not available for this line, but presumably it contains an interstitial deletion targeting SMAD4 and DCC without loss of the whole q-arm.

Discussion

This study aimed to clarify the role of SMAD4 in the development of 44 colorectal cancer cell lines. Loss of SMAD4 expression and/or mutation were found in about half of the MSI⁻ lines, but in no MSI⁺ lines. A broad spectrum of SMAD4 mutations was seen; although, for unknown reasons, there were no frameshift changes. No association was made between the presence of a SMAD4 mutation and tumor grade. Mutations appeared to occur more frequently in the C-terminal MH2 domain of SMAD4, although missense mutations in the N-terminal MH1 domain have been shown to cause in vitro protein instability (8, 24). Again, the reasons for this observation are unclear. In five of the fourteen CRC lines with absence of SMAD4 protein, no mutation was detected even after sequencing all exons in forward and reverse orientations, suggesting that cryptic SMAD4 mutations or some other means of inactivating SMAD4 may have occurred. Inactivation of SMAD4 due to hypermethylation has been shown not to occur (25) and, consistent with this finding, we detected SMAD4 mRNA in PC/JW which had absent protein, but no detectable mutation. It is also possible that changes in a protein upstream of SMAD4 can sometimes lead to loss of SMAD4 expression.

In addition to mutations causing loss of protein, four cell lines with retained protein expression had missense SMAD4 mutations in the MH2 region (Table 1). The missense mutations of C80 and CACO2 (D351H) occurred in the three loop–helix of the MH2 and have known pathogenic effects (26). The nearby R361C mutation, found not only in sporadic cancer, but also in the germline of JPS patients, is associated with undetectable levels of protein in JPS polyps, as assessed by using immunohistochemistry (data not shown). Mutations in the three-helix bundle of the MH2, such as R537Y (SW948) and L540R (SW1222), appear not to lead to unstable protein, but, like D351H, are predicted to abrogate protein function (26–28) by impairing the ability of SMAD4 to act as a homo- or heterooligomer and so interfering with signal transduction.

All lines with SMAD4 abnormalities were uniformly hemi- or homozygous at both SMAD4 and DCC microsatellites and for the SMAD4 mutation, showing that all copies of 18q were derived from the same chromosome 18 homologue, even where more than one copy of 18q21 was present in the cell. However, the dosage of 18q21 was always decreased relative to the overall level of ploidy. We found no cancer with more than one independent intragenic mutation in SMAD4. The simplest model to explain these data (Fig. 3) is one in which SMAD4 mutations occur before colorectal tumors acquire chromosomal instability (CIN) and polyploidy/aneuploidy, and are not late changes, as has been suggested (6, 8, 9). We believe that alternative explanations to this model are less plausible, because they would require several independent mutations to inactivate all copies of SMAD4 and/or frequent, independent occurrence of identical SMAD4 mutations in the same cell. However, Fodde et al. (29) have suggested that tetraploidy is a very early event in APC-mutant colorectal tumors, contrary to our favored model. Were the data of Fodde et al. (29) to be confirmed, we would be forced to conclude that some CRCs had lost three copies of 18q, acquired an intragenic SMAD4 mutation, and reduplicated the remaining SMAD4 mutant chromosome.

• Download figure
• Open in new tab
• Download powerpoint

Figure 3

Pathway showing possible sequence of events in tumorigenesis. Following mutations in the APC gene, a subset of tumors have inactivation of mismatch repair genes and so diverge along a pathway that includes mutations of the TGFβRII gene, and is characterized by microsatellite instability (MSI⁺). A subset of the remaining tumors acquire mutations in the SMAD4 gene, accompanied by loss of the wild-type chromosome 18, and this either precedes or causes another subset of tumors to diverge along a chromosomal instability (CIN⁺) pathway with aneuploidy/polyploidy. The remaining tumors are MSI⁻/CIN⁻.

Given that none of 12 MSI⁺ cell lines had SMAD4 mutations or loss of SMAD4 protein, it is most likely that the MSI⁺ pathway had diverged before SMAD4 inactivation occurred (Fig. 3). All eight MSI⁺ lines previously studied have been reported to have TGFBIIR mutations (30, 31), whereas none of the ten MSI⁻ lines studied had TGFBIIR mutations (Fisher's exact test, P < 0.000003). Although inactivation of SMAD4 and TGFBIIR may well not be functionally equivalent, these data certainly suggest that mutations of TGFBIIR occur after the divergence of the MSI^−/+ lineages. Thus, our model shows initial divergence of the MSI⁺ pathway, followed by 18q loss and SMAD4 mutation in the MSI⁻ cancers. Given that limited data suggest a high frequency of 18q loss in MSI⁻CIN⁻ CRCs (32), divergence of the CIN⁻ and CIN⁺ pathways probably occurs after 18q loss and SMAD4 mutation.

Almost all MSI⁻ cell lines showed allelic loss, and even in MSI⁻ lines without defects in SMAD4, the dosage of 18q21 was always decreased relative to the overall level of ploidy: the question therefore remains as to the cause of 18q loss in lines with no evidence of SMAD4 inactivation. Although SMAD2 remains an unlikely target, haplo-insufficiency of SMAD4 is possible and DCC changes, although unlikely, cannot be excluded with certainty. Any other target genes on 18q must, like SMAD4, be altered after the MSI⁺ pathway had diverged [although mutations in these gene(s) might, of course, be common to the MSI⁺ and MSI⁻ pathways]. Given data which suggest that 18q21 loss occurs in colorectal adenomas or close to the adenoma–carcinoma transition (33), but that SMAD4 expression is lost only in CRCs (9), we speculate that 18q21 loss sometimes precedes mutation or silencing of the other copy of one or more target genes in the region.