AIM: To evaluate the risk associated with variants of the UNC5C gene recently suspected to predispose to familial colorectal cancer (CRC).

METHODS: We screened patients with familial CRC forms as well as patients with sporadic CRCs. In a first time, we analyzed exon 11 of the UNC5C gene in 120 unrelated patients with suspected hereditary CRC, 58 patients with suspected Lynch-associated cancer or polyposis, and 132 index cases of Lynch syndrome families with a characterized mutation in a DNA mismatch repair (MMR). Next, 1023 patients with sporadic CRC and 1121 healthy individuals were screened for the variants identified in patients with familial cancer.

RESULTS: Of 120 patients with familial CRC of unknown etiology, one carried the previously reported mis-sense mutation p.Arg603Cys (R603C) and another exhibited the unreported variant of unknown significance p.Thr617Ile (T617I). The p.Ala628Lys (A628K) mutation previously described as the main UNC5C risk variant for familial CRC was not detected in any cases of familial CRC of unknown etiology, but was present in a patient with familial gastric cancer and in two Lynch syndrome patients in co-occurrence with MMR mutations. A statistically non-significant increase in cancer risk was identified in familial CRC and/or other Lynch-associated cancers (1/178 patients vs 2/1121 healthy controls, OR = 3.2, 95%CI: 0.29-35.05, P = 0.348) and in sporadic CRCs (4/1023 patients vs 2/1121 healthy controls, OR = 2.2, 95%CI: 0.40-12.02, P = 0.364).

CONCLUSION: We confirm that UNC5C mutations are very rare in familial and sporadic CRCs, but further investigations are needed to justify routine UNC5C testing for diagnostic purposes.

Among the common cancerous diseases, colorectal cancer (CRC) is the second most frequently diagnosed and the second leading cause of death in developed countries, after breast cancer[1]. Approximately 70% of colorectal cancers are sporadic cases caused by additive/multiplicative effects of numerous genetic and environmental low-risk factors. The remaining 30% of cases are inherited forms of the disease caused by highly penetrant germline mutations in single genes[2]. So far, very few clinically well-characterized familial forms of CRC have been associated with highly penetrant germline mutations, accounting for less than 6% of overall CRC cases such as Lynch syndrome with mutations in MMR genes (MLH1, MSH2, MSH6 and PMS2), Familial adenomatous polyposis (FAP) and attenuated FAP associated with mutations in APC, the MUTYH-associated polyposis with biallelic mutations in MUTYH, Peutz-Jegher syndrome with mutations in STK11, and juvenile polyposis syndrome with mutations in SMAD4 or BMPR1A[3,4]. Thus, most inherited CRCs, i.e., about 20%-30% of overall CRCs, are of unknown etiology, suggesting that multiple cancer loci with high penetrance variants may not yet have been discovered.

The finding reported recently by Coissieux et al[5] fits well with this scheme of missing heritability[6]. In their article, the authors highlighted UNC5C as a new gene potentially predisposing to familial forms of CRC[5]. UNC5C is a member of the UNC5H family of transmembrane receptors and functions as a dependence receptor. UNC5C induces apoptosis when unbound to its specific ligand, the laminin-related secreted protein netrin-1, whereas its pro-apoptotic properties are conversely inhibited when it is bound to netrin-1, in which case UNC5C transmits positive signals of proliferation and migration[7]. This activity is the hallmark of all known netrin-1 receptors (UNC5A, UNC5B, UNC5C, UNC5D and DCC) and is thought to be a safeguard mechanism against tumor development[8]. Like the prototypical netrin-1 receptor DCC (deleted in colorectal cancer), the expression of UNC5C is lost or markedly reduced in a large proportion of cancers[9-11], especially in colorectal cancer, because of a loss of heterozygosity or promoter methylation occurring within the UNC5C gene[12-16]. Inactivation of UNC5C would not only contribute to tumor progression, but also to the initiation of CRC formation. This suggests that UNC5C is likely to be a tumor suppressor[13,14], and consequently, that its inactivation could be considered as a selective advantage for tumor cells.

Using a candidate gene strategy, Coissieux et al[5] identified putative high-penetrance risk alleles that almost completely co-segregated with disease phenotype in several families with forms of colorectal cancers of hitherto unknown etiology. Moreover, by screening individuals with a form of CRC distinct from Lynch syndrome, familial polyposis or hamartomatous syndromes, Mehlen’s team identified four putative pathogenic substitutions in UNC5C (NCBI accession number NM_003728), including one in exon 7 c.1057G>A/p.Asp353Asn (D353N) and three in exon 11 [c.1807C>T / p.Arg603Cys (R603C), c.1882_1883GC>AA/p.Ala628Lys (A628K), and c.1888C>G/p.Gln630Glu (Q630E)][5]. No statistically significant differences in frequencies were observed between probands and controls for D353N, R603C, and Q630E, and the effects on CRC risk were even discordant for exon 7 variant D353N between the French and American sub-populations of the study[5]. For this reason, the authors focused the majority of their study on variant A628K, which is located in the region coding for an intracellular ZU-5 domain that is essential for the pro-apoptotic activity of UNC5C[17,18]. They based their demonstration of a genetic predisposition essentially on two points: (1) the co-segregation of the variant with cancerous phenotypes in families with CRC, thus postulating an autosomal dominant mode of inheritance; and (2) the in vitro alteration of the pro-apoptotic function of UNC5C in the absence of netrin-1. In addition, the authors noted a higher frequency of the A628K variant in the patient population than in a control population. From these observations, Coisseux et al[5] inferred a major role of the UNC5C gene in the predisposition to familial forms of CRC. Given the small number of patients concerned by this potentially novel clinical parameter, which resulted in a lack of statistical significance, the authors recommended that other teams replicate their study in samples from independent patient populations.

In order to address the question of the diagnostic potential of UNC5C suggested by Coissieux et al[5], we implemented a two-stage study to evaluate the role of UNC5C in the predisposition to CRC. More specifically, we focused on exon 11 of the UNC5C gene, which codes for the ZU-5 domain and contains A628K, the only variant found to significantly increase CRC risk in the French population studied by Coissieux et al[5]. We started by examining familial forms of cancer; we sought to find UNC5C variants in patients who had undergone prior routine screening for Lynch syndrome predisposition. One third of the patients exclusively presented a familial CRC of unknown etiology, thereby meeting the clinical criteria used by Coissieux et al[5]. The remaining two thirds consisted of two subgroups of patients; one that presented with cancer that fell within the spectrum of Lynch syndrome, albeit with no mutation in MMR genes, and the other that presented with Lynch syndrome with a known germline mutation in an MMR gene. We decided to extend our investigation to these two sub-groups because we were intrigued by the number of different cases of cancer other than CRC displayed by the families reported by Coissieux et al[5], with several patients affected by at least two types of cancer. In the second stage of our study, we tested patients and healthy controls from the ASTERISK population[19,20] to evaluate the possible role of UNC5C in the susceptibility to sporadic CRC. Taken together, the results we report here tend to challenge the hypothesis put forward by Mehlen’s team concerning the pivotal role of UNC5C in the predisposition to familial forms of CRC.

None of the 120 familial CRC patients included in our study who strictly matched Coissieux et al[5]’s clinical criteria were found to carry the main variant A628K, i.e., the variant with the most conclusive anti-apoptotic effect described by the authors (Figure 1). On the other hand, A628K was encountered in one of our patients who presented a gastric adenocarcinoma (Figure 1 and Family 1 in Figure 2). Oddly enough, this patient had three first-degree relatives with a gastric cancer, but no reported family member with colorectal polyps or adenocarcinoma. The only relative who could be tested here was the propositus’s sister, who was also found to be heterozygous for the variant. However, the gastric linitis plastica she had developed was a histologically different type of cancer compared to that of her brother, in whom CDH1 had been ruled out by molecular analysis.

Open in New Tab   Full Size Figure   Download Figure

Figure 1 Flowchart of search strategy and results. The numbers of exon 11 UNC5C variants found are reported for each sub-group of individuals included in the study. CRC: Colorectal cancer; MMR: Mismatch repair; VUS: Variant of unknown significance.

Open in New Tab   Full Size Figure   Download Figure

Figure 2 Pedigrees of the French families the proband of which was found to carry exon 11 UNC5C variant A628K (families 1-3), R603C (family 4) or T617I (family 5). A black filled-in circle or square shows that the individual developed a colorectal cancer, whereas a grey filled-in symbol indicates that the individual presented a cancer encompassed by the Lynch syndrome tumor spectrum. The type of cancer is precised below the affected individuals; the age at diagnosis is noted by figures next to the cancer type. Other figures below symbols indicate either the current age or the age of death of the individual. The variants carried by the individuals are noted below their respective symbols. The absence of any genotype means that no DNA sample was available for testing. CRC: Colorectal cancer; WT: Wild-type genotype.

The two cases of familial CRC from our study that presented the A628K variant were Lynch syndrome with characterized variants in MMR genes (Figure 1; Families 2 and 3 in Figure 2). In patient II.3 from family 2, UNC5C A628K co-occurred with a deleterious non-sense mutation of MSH2 c.1861C>T (p.Arg621*), which has been reported several times in the national UMD-MSH2 mutations database (http://www.umd.be/MSH2/) and in the International Society for Gastrointestinal Hereditary Tumours database (http://www.insight-group.org/mutations/). In patient II.2 from family 3, UNC5C A628K was found in co-occurrence with the variant of unknown significance (VUS) MLH1 c.1754T>G (p.Leu585Arg) that has also been reported in specific MMR databases. Despite its almost perfect co-segregation with the cancer phenotype in this particular family, this MLH1 variant could not be classified as likely pathogenic or even definitely pathogenic because the endometrial tumor in patient II.2’s sister was found to be microsatellite stable (MSS).

While the segregation study was not very informative in family 2, an absence of a co-segregation of the UNC5C A628K variant with cancer phenotype was observed in family 3 (Figure 2). On one hand, the propositus’s three siblings II.12, II.11 and II.6, who were 57, 59 and 68 years of age, were heterozygous for variant A628K and were all asymptomatic. On the other hand, the two siblings who were non-carriers of the UNC5C variant, II.13 and II.3, developed respectively two colorectal cancers at 41 and 55 years of age, and an endometrial cancer at 41 years of age.

Since no patients with a hitherto molecularly unexplained CRC were found to carry the A628K variant alone, risk calculation for CRC only was not possible. Nonetheless, by comparing A628K distribution between ASTERISK healthy controls and patients with familial colorectal or other Lynch-associated forms of cancer, we observed a non-significant increase in risk for cancer of the Lynch syndrome spectrum (OR = 3.2, 95%CI: 0.29-35.05, P = 0.348; Table 2). Even when considering molecularly unexplained cancer cases in combination with familial CRC cases with a known MMR mutation, the association of A628K with cancer risk did not achieve statistical significance (OR = 5.5, 95%CI: 0.91-32.87, P = 0.063). Further investigations in sporadic CRC cases from ASTERISK indicated a non-significant increase in risk of A628K for CRC (OR = 2.2, 95%CI: 0.40-12.02, P = 0.364; Table 2), which is very similar to the risk calculated in the US cohort included in Coissieux et al[5]’s study, with the difference that patients from this cohort had a familial form of CRC.

Two UNC5C variants were found in patients with CRC of hitherto unknown etiology: D353N in patient III.3 of family 4 and c.1850C>T/p.Thr617Ile (T617I) in patient II.1 of family 5 (Figure 2). No segregation study could be done in their respective families, in the absence of available samples and/or compliant relatives. To date, variant T617I has never be reported in any of the known mutations, neither in the dbSNP (http://www.ncbi.nlm.nih.gov/projects/SNP/) nor in the Exome Variant Server (http://evs.gs.washington.edu/EVS/). According to the bioinformatic programs embedded in Alamut^® software (Interactive Biosoftware, Rouen, France), including Align GVGD, Polyphen-2, SIFT and MutationTaster, amino acid 617 of UNC5C is conserved across mammalian species and a change in the protein would be benign.

Because a study limited to the A628K variant might be too restrictive, we estimated the risk for cancer associated with heterozygosity for any of the four variants R603C, T617I, A628K, and Q630E, i.e., the UNC5C exon 11 variants described by Coissieux et al[5] plus the novel T617I variant identified in our study. Specifically, we compared distribution of these variants in the familial cases of our study with that of the individuals from the NHLBI Exome Sequencing Project (ESP), assuming that Caucasian ESP individuals are genetically similar to our ASTERISK healthy control individuals. Even though we observed an increase in the risk of familial CRC without MMR mutations for subjects with any of the four variants (OR = 2.0, 95%CI: 0.47-8.20, P = 0.361; Table 2), the association was not statistically significant.

The present study was conducted to replicate recent publication findings of a major role for the UNC5C gene in the predisposition to familial forms of colorectal cancer of hitherto unknown etiology[5]. In the first stage of our study, we attempted to determine if the high CRC risk conferred by UNC5C variants could be confirmed in familial forms of CRC in our study. We focused on exon 11 of the gene, which codes for the intracellular ZU-5 domain, because this exon contains three of the four variants shown by Coissieux et al[5] to significantly alter the pro-apoptotic activity of UNC5C.

At least four of our observations diverge from those made by Coissieux et al[5], which weakens the hypothesis of a familial CRC predisposition due to UNC5C: (1) A628K was found in a family with gastric cancer (family 1); (2) A628K co-occurred with mutations in MMR genes (families 2 and 3); (3) A628K did not co-segregate with cancer phenotypes (family 3); and (4) no exon 11 variant was significantly associated with CRC risk.

Of interest, we found A628K mutation in a family with history of gastric carcinoma. This possible association between the new tissue location and A628K concurs with Coissieux et al[5]’s observation of several forms of cancer in the index cases or their relatives in their study, which can be explained by the extended inclusion criteria used in our study. We indeed followed the Amsterdam I, II or Bethesda guidelines[21-23] for Lynch syndrome screening, which include tumor sites encompassed by the clinical spectrum of Lynch syndrome (stomach, endometrium, etc.). Moreover, a down-regulation of UNC5C expression has already been reported in multiple forms of cancer, e.g., breast, ovary, uterus, stomach, lung and kidney cancers[12] and UNC5C methylation has been reported in gastric carcinomas[24].

Surprisingly, A628K was identified only in two Lynch syndrome cases with mutations previously identified in MMR genes. This fortuitous finding in two different families further complicates the interpretation of a possible association between UNC5C variants and risk of CRC, i.e., the co-segregation study is informative in only one family and no genotype-phenotype correlation was observed. Of course, the lack of co-segregation between A628K and cancer phenotype is not sufficient to definitively invalidate the hypothesis of a genetic predisposition according to a dominant mode of inheritance, considering likely phenocopies and low penetrance. It is also possible that carriers of the variant may have not yet developed any cancer. In fact, a heterozygous mutation in a predisposition gene greatly increases the probability of developing a cancer at an early age, but it does not mean that carriers of this mutation will necessarily develop a malignancy before 50-60 years of age. In contrast, as regard to the already known high-penetrance CRC predisposition genes, the likelihood of finding three asymptomatic carriers as well as two non-carriers affected with early onset cancer in the same family is relatively small. For instance, in Lynch syndrome, the cumulative risk of developing a CRC is 30%-80% in MMR mutations carriers and the mean age at CRC onset is 45 years[25]. It is also worth noting that co-segregation between A628K and CRC was not perfect in the families studied by Coissieux et al[5]. For example, at least one non-carrier, albeit with CRC, was found in each family, whereas a 64-year old carrier was reported as asymptomatic, underlining the aforementioned problem of phenocopies and suggesting that the penetrance of the variant may not be as high as originally thought.

In our study, we were not able to replicate the marginally significant association for an increase in CRC risk with UNC5C exon 11 variants reported by Coissieux et al[5] (OR = 8.8, 95%CI: 1.47-52.93, P = 0.0315). The divergence in the results observed between the two studies is in fact slighter than it may appear, as the groups of A628K carriers are very small in both cases. The power of association studies is indeed relatively limited for investigating rare variants, and achieving strong statistical significance would either require a very large sample size or necessitate more powerful methodologies such as burden tests. Thus, Genome-Wide Associations studies (GWAS) have failed to highlight loci already known in familial forms of cancer and other complex traits, due to their inability to detect highly-penetrant, rare variants[4,6]. On the contrary, by adopting a candidate gene strategy, Coissieux et al[5] were able to detect associations for some of these rare variants. However, the associations between the identified variants with CRC risk are somewhat fragile. This illustrates the difficulties to identify new high penetrance genetic variants involved in predisposition to CRC because of the high locus heterogeneity. The estimated 20%-30% of missing heritability for CRC can therefore be explained by a possible mosaic of rare familial variants, with an accumulation of private mutations. UNC5C could be one of the genes involved, but the identification of additional families is needed to confirm this. This could be facilitated by a more precise clinical definition of UNC5C-associated CRC cases.

It is quite tempting to draw parallels between UNC5C and DCC. In fact, a possible tumor suppressor role was postulated for both netrin receptors for several years[7,13,18,26]. However, no conclusive proof was produced until recently. With the development of murine models, it has been shown that, beside their role in tumor progression, UNC5C and DCC could also participate in the early events of colorectal carcinogenesis[1,2,13,27]. The hypothesis of tumor suppressor gene function is also supported by the fact that the UNC5C and DCC genes are located in frequently rearranged chromosomal regions and also the frequent hypermethylation of their promoters in colorectal tumors[15]. Nevertheless, for DCC, this putative role is considered as very controversial, notably because of its much rarer somatic mutations in colorectal cancers reported in literature or in specific databases (e.g., COSMIC), as compared to the nearby SMAD4 gene for example[28-34]. In addition, DCC has never been related to a familial form of colorectal malignancy, in contrast to SMAD4 - the mutations of which can cause Juvenile Polyposis Syndrome[28-34]. A query in the COSMIC database[28] showed that somatic UNC5C mutations are as infrequent in colorectal adenocarcinomas as DCC. Moreover, no clearly deleterious germline mutations, such as nonsense or frameshift mutations, were reported in UNC5C in patients with CRC, whereas these types of mutations are quite frequent in familial forms of cancers caused by tumor suppressor genes.

This raises the issue of whether UNC5C is the right candidate gene to explain colorectal cancer in the five families described by Coissieux et al[5]. In fact, the good co-segregation of the A628K variant with cancer phenotype is puzzling. In Coissieux’s study, all carriers of this variant shared a 108 kilo-base-pair minimum critical region within 4q22.3 determined by 34 SNPs. Genotyping a subset of these SNPs (rs65322545, rs6846193, rs17023311, rs1434534, rs3775048, rs11737434 and rs2289043) in our study patients was sufficient to discriminate carriers of variant A628K from carriers of other UNC5C exon 11 variants (data not shown). However, the power displayed in linkage analysis by 34 SNPs is quite low, because of a limited heterozygosity. Consequently, the minimum critical region defined by Coissieux et al[5] could not be as specific and significant as expected, leading us to postulate that UNC5C may not be the right susceptibility gene. The rare variants identified by Mehlen’s team may instead be markers that are in linkage disequilibrium with a real deleterious variant located, for example, within any other gene in the neighborhood of UNC5C. When considering the DCC/SMAD4 example, that other gene could be the nearby BMPR1B gene, which is centromeric to UNC5C. The risk for CRC conferred by variants of BMP genes has already been highlighted by GWAS[32,33], and a recent study pointed to a significant increase in CRC risk associated with genetic variation in BMPR1A, BMPR1B, BMPR2, BMP2 and BMP4[34].

In conclusion, our study shows that UNC5C is not a major CRC-predisposing gene that could compare for instance with MMR genes or APC. UNC5C exon 11 allelic variants were rare, with an incomplete penetrance in the families of our study. However, we were able to confirm a tendency towards an increase in cancer risk, not only for familial but also for sporadic CRC, which means that UNC5C alleles can be classified at least as low-moderate penetrance alleles. Previous studies have highlighted UNC5C as a potential tumor suppressor gene, but the simple presence of UNC5C variants in the families included in our study does not seem sufficient to confer a high risk for CRC. One may hypothesize that, in the families included in Coissieux et al[5]’s study, either A628K is in complete linkage disequilibrium with a causative mutation located outside UNC5C, or UNC5C participates in an oligogenic mechanism of carcinogenesis. In any case, it is probably too preliminary to apply Coissieux et al[5]’s finding to genetic counseling and routine genetic CRC-predisposition testing. The presence of variant UNC5C A628K in patients with gastric cancer in family 2 (Figure 2) also suggests that further investigations should be performed to address the impact of UNC5C on digestive cancers.

Sébastien Küry and Stéphane Bezieau are very grateful and indebted to Professor Jean-Paul MOISAN for having given them the opportunity to work in the field of molecular genetics. He will miss them.