Colorectal cancer (CRC) is the third most commonly diagnosed cancer in males and the second in females with 1.2 million new cases and 608,700 deaths estimated to have occurred worldwide in 2008 1 . In its early stage, CRC represents a curable disease. However, 20–50% of patients with newly diagnosed CRC will develop secondary metastases (mCRC) 2 . A major advance in the treatment of mCRC has been achieved thanks to the development of targeted therapies. Accordingly, two antibodies, cetuximab and panatimumab, which selectively target the extracellular domain of the epidermal growth factor receptor (EGFR), have been approved for the treatment of metastatic diseases. The combination of these targeted molecules with conventional chemotherapy (5-FU, Irinotecan, Oxaliplatin) has led to significant improvement in response rate, progression free survival and overall survival in first line, as well as second or third line treatment of mCRC 3 4 5 6 7 8 . This efficiency constitutes a clinical evidence that activation of EGFR is oncogenic in CRC. However clinical trials have shown a high individual variability of response and outcome in mCRC patients, which has highlighted the need for identification of reliable markers predictive of response to treatment. The only molecular marker predictive of the response of the anti-EGFR mAbs, which has been unambiguously validated in mCRC by numerous studies, is the presence of KRAS activating mutations as a marker of resistance to anti-EGFR 9 10 . However the occurrence of KRAS mutations only accounts for 35–45% of non-responsive patients 11 . Remarkably, the mechanisms of EGFR activation in CRC have not been characterized in most of the patients. This contrasts with the situation observed in lung adenocarcinoma where the key mechanism of EGFR activation, underlying sensitivity to EGFR inhibitors, corresponds to activating mutations within the EGFR tyrosine kinase domain 12 13 . Indeed, in CRC, the amplification of EGFR resulting in overexpression and associated to sensitivity to anti-EGFR is detected in only 10–15% of CRC 14 15 16 17 . Overexpression of the EGFR ligands, amphiregulin and epiregulin, has been reported to be associated to sensitivity to anti-EGFR mAbs 18 19 .

The EGFR gene contains within the 3′ untranslated region (UTR), 281 bp downstream from the stop codon, a polyadenine tract which is polymorphic (A13/A14). Mono or dinucleotide deletions within this polyA tract have been detected in colon cancer cell lines or CRC exhibiting microsatellite instability (MSI) 20 . These deletions have been shown to stabilize EGFR mRNA, to result in EGFR overexpression in vitro and to increase sensitivity to anti-EGFR antibodies in xenografts 20 . This prompted us to investigate, in CRC patients, the oncogenic impact of genetic variations affecting this regulatory region. To this aim, we used two complementary approaches: first, we determined, in a case control study, whether or not the germline EGFR A13/A14 polymorphism constitutes a genetic risk factor for CRC; second we investigated the frequency and impact of somatic mutations of this repeat in CRC.

We genotyped the EGFR polyA repeat in non malignant colorectal tissue or blood from 429 patients with CRC corresponding to different groups of CRC patients selected or not on age of tumor onset and/or familial history and/or MSI. To ensure an acurate genotyping (Figure 1), we used, as calibrators, cloned EGFR polyA repeats the size of which had been determined by sequencing. Allelic frequencies observed in CRC patients and controls are given in Table 1. Allelic frequencies were in Hardy-Weinberg equilibrium in patients and controls. The frequency of the major allele (A13) was estimated in controls and patients to 76.5 and 72.8%, respectively. No statistically significant difference in allelic frequencies of the EGFR polyA repeat was observed between patients and controls and between each group of patients and controls (Table 1).

^a For each allelic frequency, confidence interval is given in brackets.

^b The p value in each patient group corresponds to the comparison with controls (chi-2 test).

We then screened 179 patients with CRC for somatic mutations of the EGFR polyA repeats, by comparing, for each patient, the PCR profile obtained from tumor to that from paired non malignant tissue (Figure 2). As shown in Figure 2B, somatic EGFR polyA mutations could easily be detected by a clear shift of the EGFR fluorescent peak observed in tumors. In the 99 MSS CRC, we observed no somatic EGFR polyA mutation. In contrast, we detected an EGFR polyA mutation in 47/80 (58.8%) MSI CRC. The detected mutations always corresponded to adenine deletion and no gain was observed. The number of deletions ranged from 1 to 4 adenines and the total number of deletions observed on both alleles was: 1 (25.5%), 2 (27.7%), 3 (17%), 4 (12.8%), 5 (10.6%), 6 (2.1%), 7 (2.1%) and 8 (2.1%). There was no significant difference (chi-2 test, p = 0.70) in somatic mutation frequency (Table 2) in patients with A13/A13, A13/A14 and A14/A14 genotypes.

To address the specificity of somatic mutations affecting the EGFR 3′UTR polyA tract in MSI CRC, we evaluated in 10 MSI with EGFR mutations and 10 MSS CRC samples the frequency of mutations within two other 3′UTR polyA tracts sharing structure similar to that of the EGFR: a polyA(15) in RAB31 (member RAS oncogene family) and a polyA(14) in ATP6V1G1 (ATPase V1 subunit G1). In all MSI CRC samples with EGFR polyA tract mutations, we also found mutations of RAB31 and ATP6V1G1 polyA tracts, but no mutation was observed in MSS tumors.

We analyzed the potential impact of the EGFR 3′UTR polyA tract mutations on mRNA secondary structure through bioinformatics prediction. Successive deletions of adenine was not predicted to result in any significant alteration of the mRNA structure and, in particular, there was no modification of predicted binding sites for miRNAs (hsa-mir-146a/b, hsa-mir-133b, hsa-mir-7-1/2) or regulating proteins (HuR: AU-rich elements).

We then determined the impact on EGFR expression of the somatic EGFR polyA tract mutations detected in MSI CRC, using real-time PCR quantitation of mRNA and RT-QMPSF (Figure 3). These two methods applied to 11 CRC with EGFR polyA mutation and 10 CRC without mutation yielded identical results (r = 0.75, see Additional file 2: Figure S1A). In 10/11 mutated and 10/10 non mutated samples, we observed, as illustrated in Figure 3, that the level of EGFR mRNA was lower in malignant tissue, as compared to paired normal tissue, although the difference was not significant. In the remaining mutated sample, we observed a slight increase (×1.1) of EGFR expression in tumor by comparison to normal tissue. There was no influence of the total number of adenine deletions on EGFR mRNA levels, even in a sample exhibiting up to 7 adenine deletions (see Additional file 2: Figure S1B). In 8 tumor samples harboring two EGFR alleles of different size and in 10 non malignant tissues from patients with a heterozygous genotype, we could compare the EGFR allelic expression by calculating the mRNA ratios corresponding to the short / long allele. In both cases, we did not observe an obvious allelic expression imbalance, but only a slight increase of expression of the short allele, as compared to the long one (mean 1.11 and 1.15, respectively).

Finally, we evaluated whether the germline EGFR polyA repeat polymorphism or mutational status in tumor influence the risk of tumor recurrence in 64 patients with a localized form of CRC (stage I, II and III) followed for at least two years. There was no difference in the percentages of recurrence according to the germline polyA polymorphism (p = 0.72), nor according to the existence or not of a somatic mutation (p = 0.72). In 18 patients with metastatic disease (stage IV) treated by anti-EGFR (cetuximab or panitumumab), the disease control rate was not influenced by the polyA tract polymorphism (p = 0.78).

We evaluated the biological impact, in patients with CRC, of germline or somatic genetic variations occurring within the EGFR 3′UTR polyA tract. First, we observed that the EGFR polyA allelic frequency in 429 CRC patients was similar to that observed in a control sample. Considering the genetic heterogeneity of CRC, we constructed the patient sample with 4 different groups selected or not on the basis of age of tumor onset or familial history or MSI status. The first group, composed of 179 CRC patients unselected on age of tumor onset or familial history, has been, on purpose, enriched in patients with MSI tumors, which had been shown in the original study of Yuan et al. 20 to exhibit a high rate of somatic EGFR mutations. The second group, constituted of 62 patients without detectable mutations within MMR or adenomatous polyposis genes but whose personal or familial history was suggestive of an increased genetic risk, was analyzed to determine whether or not the EGFR polyA polymorphism could act as a genetic risk factor for CRC. We also analyzed a series of 100 patients with Lynch syndrome to evaluate if the EGFR polyA polymorphism could act as a modifier risk factor in patients harboring a MMR gene mutation. Finally, the last group corresponded to 88 unselected sporadic CRC. In none of these groups, could a significant difference in EGFR allelic frequencies with controls be detected, suggesting that the EGFR 3′UTR polyA polymorphism does not modify the genetic risk for CRC. It could be argued that the size of the patient sample or that of the different groups was insufficient to detect a significant difference, but the allelic frequency between patients and controls were remarkably similar (Table 1). We also screened for somatic mutations of the EGFR polyA tract in the group of 179 CRC patients, whose genotypes had been characterized and found that somatic mutations, corresponding to deletions, were detected in 59% of the 80 MSI tumors but in none of the 99 MSS tumors. This confirms, on a larger sample, the results observed by Baranovskaya et al. 28 , Yuan et al. 20 and Deqin et al. 29 who had reported, from a series of 40, 16 and 36 MSI CRC a mutation detection rate of 92.5%, 69% and 81%, respectively. Nevertheless, we obtained two results which argue against an oncogenic effect of these somatic mutations: first, the adenine deletions occurring in the 3′UTR polyA tract did not show any specificity with respect to EGFR since they could also be observed in 2 others genes not involved in CRC: RAB31 and ATP6V1G1; therefore the high frequency of somatic EGFR polyA mutations reported in MSI tumors by other studies and this work probably reflects a particular sensitivity of mononucleotide tracts to defective DNA mismatch repair system, as recently reported for the polyT(20) tract of the MT1X gene 30 ; second, we found that these mutations did not result into a significant increase of EGFR expression. In a study focused on the CA repeat located within the EGFR first intron, Baranovskaya et al. 28 have also observed, in agreement with our results, that EGFR expression was decreased in MSI CRC. In a sample composed of 16 MSI endometrial adenocarcinomas, Deqin et al. 29 have reported that tumors with EGFR polyA deletions exhibit a slight (1.6) but nevertheless not significant increase of EGFR expression, as compared to that without mutations. Our observation contrasts with results obtained by Yuan et al. 20 . Indeed, these authors had reported, in colon MSI cancer cell lines, that a deletion within the EGFR polyA tract increases in vitro the EGFR mRNA stability. In CRC patients, we observed that, in the majority of the tumor samples with somatic EGFR mutations (91%), the total level of EGFR mRNA was not increased but, in contrast, decreased and this result was obtained using two independent methods. The discrepancy observed between both studies highlights the need to confirm in clinical samples results previously obtained with cell lines which may not be representative of the complexity of gene regulation in clinical samples, because of the genetic drift occurring during in vitro culture.

This study has raised several arguments showing that genetic variations affecting the EGFR polyA repeat are not involved in CRC development: (i) The EGFR polyA polymorphism does not constitute a genetic risk factor for CRC; (ii) somatic mutations of this repeat are commonly observed in MSI CRC, but their frequency reflects a sensitivity of this type of repeat to MSI and not a specific selective advantage; (iii) somatic EGFR polyA mutations do not result into an EGFR mRNA increase in colorectal tissue. Therefore, genotyping of the EGFR polyA tract has no clinical utility to identify patients with a high risk for CRC or patients who could benefit from anti-EGFR antibodies.

The authors have no conflict of interest to declare.

Conception and design: NSV, TF, RS. Development of methodology: NSV. Acquisition of data: NSV, DS, DT, FLP, SB. Technical support: AL, FB. Analysis and interpretation of data: NSV, DS, FDF, PM, JBL, TF, RS. Study supervision: TF. Writing, review and/or revision of the manuscript: NSV, DS, TF, RS. All authors read and approved the final manuscript.