Early Colorectal Cancers Provide New Evidence for a Lynch Syndrome-to-CMMRD Phenotypic Continuum

Abstract

Lynch syndrome (LS) is the most common hereditary colorectal cancer (CRC) syndrome, caused by heterozygous mutations in the mismatch repair (MMR) genes. Biallelic mutations in these genes lead however, to constitutive mismatch repair deficiency (CMMRD). In this study, we follow the diagnostic journey of a 12-year old patient with CRC, with a clinical phenotype overlapping CMMRD. We perform molecular and functional assays to discard a CMMRD diagnosis then identify by exome sequencing and validation in a cohort of 134 LS patients, a candidate variant in the MLH1 UTR region in homozygosis. We propose that this variant, together with other candidates, could be responsible for age-of-onset modulation. Our data support the idea that low-risk modifier alleles may influence early development of cancer in LS leading to a LS-to-CMMRD phenotypic continuum. Therefore, it is essential that larger efforts are directed to the identification and study of these genetic modifiers, in order to provide optimal cancer prevention strategies to these patients.

1. Introduction

Lynch syndrome (LS; OMIM 120435) is the most frequent hereditary colorectal cancer (CRC) syndrome that accounts for around 2.8% of cases. It is caused by germline mutations in the mismatch repair (MMR) genes or 3′ deletions in EPCAM. LS patients have a 60–80% lifetime risk of developing CRC and other tumours that typically develop around the fifth decade of life and are characterized by microsatellite instability (MSI) and the loss of expression of the corresponding MMR protein [1].

Biallelic MMR germline mutations lead however to constitutive mismatch repair deficiency syndrome (CMMRD; OMIM 276300), characterized by haematologic, brain and colorectal tumours in childhood/adolescence. Diagnostic criteria for the clinical suspicion of CMMRD were defined by the European Care for CMMRD Consortium (C4CMMRD) [2].

However, there is still a current gap in the molecular understanding of MMR-associated phenotypes, and this is reflected in the difficulties in genetic and clinical diagnosis. In this work, we present a male patient from a LS family who developed CRC at 12 years of age. Although he carried a potentially spliceogenic MSH2 variant in addition to the familial mutation, functional analyses clearly reject CMMRD. Therefore, we evaluated other possible scenarios that could explain the exceptionally early age of CRC onset. Our analyses support the idea of a phenotypic continuum between the classical Lynch and CMMRD syndromes that could be modified by multiple genetic factors.

2. Results

MSH2 predictive testing confirmed that the patient had inherited the pathogenic c.1076+1G>A variant from his mother (Figure S1). Molecular analyses revealed MSI and loss of MSH2/MSH6 expression in the tumour but not in the normal adjacent tissue. A second somatic hit in MSH2, c.1035G>A, p. (W345Ter), was found in 42% of tumour reads, together with the original germline variant.

Considering the early presentation, we contemplated the idea that the patient indeed suffered from CMMRD. Germline trio analyses identified only a previously unreported intronic variants of unknown significance (VUS) in MSH2 (c.1077-11A>G). Although in silico tools predicted a splicing effect, minigene experiments showed that this VUS generates only full-length transcripts (Figure S2), and hence, it is very unlikely the second MSH2 hit. Furthermore, we could observe neither microsatellite instability (MSI) in DNA from germline (gMSI)/LCL (evMSI), nor the resistance pattern typical of CMMRD-deficient cells (Figures S3 and S4), and hence a molecular CMMRD phenotype was rejected.

Because phenotypes similar to CMMRD can arise from high-penetrance germline mutations in other genes [3], we searched for paternally-inherited or de novo variants in other hereditary CRC genes, but could only find the MLH1 c.-93G 5′UTR variant in homozygosis (Table S1).

For whole-exome sequencing (WES) analysis, we used the somatic mutational signatures as a proxy to identify pathways relevant to tumorigenesis in this patient. We also looked for tumour mutations in the hereditary cancer genes to check whether the early-onset phenotype could be a result of somatic events, but found no high-impact loss-of-function variants that could account for the phenotype. Interestingly, the tumour mutational pattern was driven by signatures 1, 6, 20, 12, 26 and 3 (Figure S5a). The normal tissue profile was however dominated by signature 12 (Figure S5b). This supports the idea that signature 1 in the cancer (which may seem contradictory with the patient’s young age) is due to the hypermutated nature of the tumour derived from the MSI phenotype. It would also be consistent with the observation of other signatures related to MMR-deficiency (signatures 6, 20 and 26). Additionally, we observed that the tumour somatic mutation burden (TMB, calculated as the total number of mutations/Mb) was 81.8, which is in accordance with a hypermutated tumour expected for LS. The adjacent normal tissue presented a TMB of 18.6.

We then restricted the germline search to genes related to four categories linked to CRC, and found four rare germline candidate variants, including a likely pathogenic non-synonymous change in IGF1 p. (A118T), and an intronic variant in FANCC (c.1073-4G>A) (Tables S2 and S3). We also performed a hypothesis-free rare variant analysis prioritization using eDiVa. Our top hit is a missense variant in the PTPN4 gene p. (Y126C), yet the IGF1 p. (A118T) variant still appears at lower rank (Table S4). Because variants in IGF1 have been reported as LS modifiers, we looked into other previously-described LS modifier genes. Additional candidate modifier variants in CCND1 p. (P241P) and MTHFR p. (R519C) were found (Tables S5 and S6).

A summary of the paternally-inherited candidate variants in this patient can be found on

Table 1. Germline candidate variants. Germline changes found by exome analysis that could potentially have a modifying effect in the LS environment and could therefore affect age of cancer onset.

Chromosome: Position	Reference Allele	Alternative Allele	Genotype	Gene	Variant	Location	HGMD	Automated InterVar	dbSNP ID	gnomAD NFE	CADD Phred	DANN Score	GERP++ Score	Interpro Domain
1:11852412	G	A	het	MTHFR	NM_005957:c.1555C>T,	exonic	-	Uncertain significance	rs45496998	0	34	0.999	4.19	-
					p. (R519C)
2:120639370	A	G	het	PTPN4	NM_002830:c.A377G,	exonic	-	Uncertain significance	NA	0	6.13	0.998	5.41	-
					p. (Y126C)
3:37034946	G	A	hom	MLH1	NM_000249:c.-93G>A *	UTR5	DFP	-	rs1800734	0.222				-
9:97876996	C	T	het	FANCC	NM_000136:c.1073-4G>A	splicing/intronic	-	-	rs147695697	0	-	-	-	-
11:69462910	G	A	het	CCND1	NM_053056:c.723G>A, p.(P241P)	exonic	DFP	Benign	rs9344	0.465	-	-	-	-
12:102813337	C	T	het	IGF1	NM_000618:c.352G>A,	exonic	-	Likely pathogenic	rs151098426	0.001	24.1	0.998	5.85	Insulin, conserved site
					p. (A118T)

Hom: homozygote; het: heterozygote; HGMD: Human Gene Mutation Database class; DFP: Disease-associated polymorphism with additional supporting functional evidence; frequency: gnomAD NFE variant frequency in gnomAD all in non-Finish Europeans; CADD; DANN, GERP++: in silico predictors of pathogenicity. All variants are paternally-inherited with the exception of *, where mother and father are heterozygous. No de-novo variants were found that fulfilled our prioritization criteria.

. These six candidates were selected due to previous evidence supporting their role in cancer risk predisposition, and were genotyped in our LS cohort of 134 LS patients. We only observed variant alleles for MLH1 and CCND1. Albeit non-significant, there was a trend towards a younger age-of-onset in patients with one or more MLH1 risk alleles (μ_GG = 47.66, range (26–85), n = 74; μ_GA/AA = 44.89, range (25–67, n = 46; p = 0.112) (Figure S6). CCND1 analyses did not show evidence for a younger debut.

In parallel, because low-penetrance variants have also been proposed as risk modifiers, we also calculated the polygenic risk score (PRS) to account for the genetic risk explained by low penetrance alleles (Table S7). The patient and his mother exhibited a normalised PRS of 0.408 and 0.472, respectively, which fall into the population-expected range (percentiles 47 and 49), so although some of these variants could be individual risk modifiers, we cannot link the early onset observed in this patient with an outstanding contribution of lower penetrance alleles.

3. Discussion

In this work, we present a LS patient who exhibits an extremely early age-of-onset of CRC at only 12 years. We initially hypothesized that the patient could have CMMRD, and identified a likely spliceogenic intronic VUS in MSH2. Although we cannot fully exclude other explanations (a low-frequency mosaic mutation in the healthy colonic mucosa or a hypomorphic leaky splice effect of the VUS not revealed by the minigene experiments), the molecular analyses performed do not support a CMMRD molecular phenotype and diagnosis.

We then hypothesised that the presence of other high/moderately penetrant mutations may be responsible for the early onset of cancer [4]. Nevertheless, we could only find a variant in the 5′UTR of the MLH1 gene in homozygosis, which could have important consequences on the risk of developing CRC (odds ratio 1.3 and 2.6 for heterozygous and homozygous, respectively, Thomas R et al. under review). This variant has been extensively reported in the literature as a low-penetrance allele in MSI cancers [5], as it is related to the epigenetic regulation of the MLH1 CpG island and shore [6]. We find that there is a non-significant trend towards a younger age of onset in patients carrying the risk variant in our cohort of LS patients. Notably, there was another 24-year-old patient in our LS cohort that presented a CMMRD-overlapping phenotype, who presented the MLH1 variant in homozygosis.

Subsequently, we considered other genes described as genetic modifiers of LS, and found interesting variants in IGF1, CCND1, MTHFR, FANCC and PTPN4. The IGF1 gene codes for a growth factor determinant in cell cycle control. Variants in this gene have been linked to an early onset of CRC in LS [7].

Cyclin D1 also plays a relevant role in cancer, and the variant found in this patient codes for a synonymous change that affects splicing [8] and has been related to abnormal cell proliferation [9]. This SNP has been extensively studied in the context of LS, although with ambiguous results [10,11]. The MTHFR gene codifies for the rate-limiting enzyme that regulates folate availability and two of its most common SNPs are low penetrance alleles for CRC [12,13]. Lastly, PTPN4 belongs to the superfamily of protein tyrosine-kinases and phosphatases, frequently mutated in cancers. Although not much is known about PTPN4, it has been suggested that mutations in another family member, PTPN12, could cause susceptibility to CRC [14]. Finally, FANCC belongs to the Fanconi anaemia DNA repair pathway, which has been proposed to play a role in inherited predisposition to CRC [15].

Altogether, we propose that the combination of several low-risk modifier alleles may be responsible for the CMMRD phenotypic overlap in this patient. The presence of these or other genetic modifiers could potentially explain the higher prevalence of childhood cancers in LS patients [16]. Nevertheless, we cannot exclude that other epigenetic or environmental factors may also play a role in early CRC development.

4. Materials and Methods

We initially studied a patient from a LS family who developed CRC at 12 years of age. Upon colonoscopy and histological analyses, the presence of a stage II adenocarcinoma in the caecum (T2N0M0) was confirmed. Additional clinical findings compatible with a CMMRD diagnosis included a neurofibroma in the back (although not histologically confirmed). No café-au-lait macules (CALMs) were observed. The patient and his family received informed consent, according to the tenets of the Declaration of Helsinki, and then MSH2 predictive testing and somatic sequencing of the complete coding region of MSH2 were performed to confirm the LS diagnosis. Multi-gene targeted sequencing and multiplex ligation-dependent probe amplification (MLPA) were used to identify other potential germline (epi)mutations in other cancer susceptibility genes: MSH6, MLH1, PMS2, APC, MUTYH, POLE, POLD1 and NF1. In silico splicing predictor analyses and minigene assays were performed to evaluate the splicing effect of the MSH2 variant of unknown significance (VUS) c.1077-11 A>G. Germline and ex vivo MSI, and toxicity tolerance to MNNG were additionally assayed to evaluate MMR deficiency (see online information for detailed description of these assays).

Whole-exome sequencing (WES) was performed from blood germline DNA for the patient and both parents, to identify the variants responsible for the early onset. Additionally, the tumour and matched normal colonic tissue were also analysed to characterize somatic mutation features. Median coverage was 55× and 150× for the germline and somatic tissues. An additional resequencing panel with Ion PGM was also performed on the MSH2 gene. This panel yielded an average sequencing depth in the normal somatic tissue of 3225×, which would be adequate for somatic mosaicism analyses if we consider a variant allele frequency (VAF) of 10% as a threshold for somatic mosaicism calling. We selected either de novo or paternally-inherited alleles, with loss-of-function, high functional impact or known modifier effects as candidates. All candidate changes were confirmed by Sanger sequencing. The EDiVa bioinformatics tool (https://ediva.crg.eu/) was used to obtain a variant list ranked by potential pathogenicity.

We validated the effect of these candidate variants on age of CRC onset by genotyping in a cohort of 134 LS patients (26 with pathogenic variants in MLH1, 88 in MSH2 and 20 in MSH6; age range 25–85).

Because it has been described that low-penetrance alleles may also act as genetic modifiers of CRC risk, we also genotyped the 37 GWAS-described non-exonic risk variants that had attained genome-wide significance (p value ≤ 5 × 10⁻⁸) to generate the polygenic risk score (PRS) (see online material). The mutational signature in the tumour was obtained with MuSiCa [17]. Because LS tumours are expected to be hypermutated, we assessed the mutational burden as the number of somatic mutations/megabase. A detailed description of the methods is found in Appendix A.

5. Conclusions

We recommend that patients with a clinical CMMRD-overlapping phenotype be subject to molecular testing to discard CMMRD. Then, further efforts should be made, given the current genomics era, into the identification of modifier genes and variants. In this regard, although there have been studies trying to identify LS modifiers for the past 20 years [18,19], these have yielded inconsistent results. We must make an effort in order to design robust studies with appropriate sample sizes that can assess the effects of these genetic modifiers on age of onset. All of these could prove effective in finally bringing the search for modifier alleles in the Lynch-to-CMMRD phenotypic continuum forward.