DISCUSSION
We have made an effort to classify variants in the low-moderate penetrance CHEK2 gene. For that, we analyzed the whole coding region of CHEK2 in a large HC cohort, performed in-depth literature review and have defined specific cut-offs for ACMG criteria to allow classification of variants with low effect. Furthermore, we applied different combinatorial rules that enabled us to compare classification rates. Concluding that the Bayesian model is the most optimal framework to classify variants to a greater extent.
From our experience in variant classification and after a comprehensive literature review, we propose two adaptations of the ACMG criteria. Regarding PS4 we propose to score PS4_moderate for low-moderate penetrant genes if an OR is given between 1.5 and 5, with a p value of <0.01, when the phenotype is in accordance with the previously described. In relation to PM2 evidence, in our laboratory we use an extremely conservative approach and assign PM2 only if the variant is absent or present in less than 1 out of 100,000 alleles in gnomAD (0.001% of maximum frequency) for high penetrant genes. However, we propose to assign PM2_supporting when the variant is ≤1 out of 20,000 alleles.
Variants meeting PVS1 criterion tend to be easier to classify as LP/P. For instance, the founder mutation c.1100delC is the most studiedCHEK2 mutation and it has a prevalence of 0.26% in NFE population. CHEK2 c.1100delC has a moderate penetrance (Meijers-Heijboer et al., 2002; Oldenburg et al., 2003), conferring an increased BC risk for overall population (OR= 2.89, 95% CI, 2.63–3.16) (Liang et al., 2018) and for carriers with familial BC (OR= 3.21, 95% CI, 2.41-4.29) (Liang et al., 2018). It has been reported absent in Spanish population (Bellosillo et al., 2005), or with frequencies of 0.93% in Basque population, 0.36% in Galician population and 0.3% in a study of BRCA -negative HBC Basque and Catalan families (Fachal, Santamariña, Blanco, Carracedo, & Vega, 2013; Gutiérrez-Enríquez, Balmaña, Baiget, & Díez, 2008; Martínez-Bouzas et al., 2007). In our larger cohort, only one case was identified (0.08%, 1 out of 1,251 BC affected cases), confirming its low prevalence in our population. Moreover, in a recent study analyzing 15 truncating CHEK2variants in 213 patients and 29 control carriers, the BC risk OR was 3.11 (95% CI, 2.15-4.69) (Decker et al., 2017). Here we identified 10 proband carriers of truncating variants, 8 of which developed the first tumor before the age of 50, consistent with previous findings of early cancer development in carriers of truncated variants (Decker et al., 2017; Han et al., 2013). Nonetheless, the median age at first cancer diagnosis in our study was not very different amongst carriers of truncating and missense LP/P variants, being 42 (range 25-65) and 40 (range 22-51) years, respectively. Bilateral BC has been mainly reported in c.1100delC carriers (M Kriege & J M Collée, 2014), and truncating variants in this gene have been associated to other non-breast second primary tumor diagnosis in a study using multigene panel testing (Fostira et al., 2020). In our cohort, 4 cases with two or multiple cancers were carriers of truncating variants and only one was carrier of a LP missense, confirming a higher aggressiveness of truncating variants over missense variants.
Conflicting results are common for missense hypomorphic variants and represent one of the biggest challenges we faced for CHEK2variant classification due to the lack of more sensitive functional assays and the use of different controls, complicating replication and therefore bypassing PS3 application. The c.470T>C founder mutation conveys a moderate susceptibility for overall cancer (OR= 1.39; p<0.00001) and for BC only (OR= 1.58; p<0.00001) in a large meta-analysis (Han et al., 2013). Its pathogenicity has been established for ovary cystadenomas in young Polish carriers (OR = 2.6; p=0.006) (Szymanska-Pasternak et al., 2006) and is associated to a 2-fold risk of non-Hodgkin lymphoma, colon, kidney, thyroid and prostate cancers (Cybulski et al., 2004). We found it in a male patient diagnosed of testicular cancer at 25 years. Interestingly, in a recent study of 448 Croatian testicular cancer patients it was found in 5.1% of them, resulting in an OR of 3.93 (95% CI, 1.53-9.95) even when its population frequency is of 1-2% (AlDubayan et al., 2019). Of note, when applying ACMG-AMP guidelines, c.470T>C remains as VUS even applying PS4_moderate. To our knowledge, c.470T>C is the most studied CHEK2 missense variant, but as shown in Table S3, it has conflictive interpretations of pathogenicity at almost all functional studies, therefore PS3 was ruled out, remaining as VUS in the ACMG context. However, we were able to classify it as ERA according to the risk allele-based classification (Senol-Cosar et al., 2019). Of note, this variant is classified as LP by GeneDx, and as P by Ambry, Color and Invitae diagnostic laboratories (Table S3), which could convey errors in clinical interpretation. PS3 was also not possible to apply for 2 other missense variants: c.190G>A and c.1427C>T.CHEK2 c.190G>A is a fairly frequent variant found in 0.03% of NFE by gnomAD, with partial reduction of Thr68 phosphorylation, auto-phosphorylation and Cdc25C phosphorylation, but DNA repair assays in yeast are discordant (Table S3). Variant c.1427C>T is another relatively frequent variant present in 0.05% of NFE (gnomAD). It has been reported to affect DNA damage response in yeast at intermediate-high level. In addition, it shows reduced SOX phosphorylation almost equally to c.1100delC. However,in vivo and in vitro studies of KAP1 phosphorylation from the same group showed discordant results of pathogenicity (Table S3). As noted in Table S2, lack of robust association studies and meta-analysis of these variants hampered the possibility of applying risk allele-based classification. Both remained as VUS in any classification framework, although are classified as LP by at least 2 different reputable sources (Table S3).
To summarize, we describe here a comprehensive CHEK2 mutational analysis in a large Spanish cohort of HC patients, providing full data of the actual prevalence of CHEK2 pathogenic variants in our population. The frequency of LP/P variants in the HBC suspected cases in the whole gene analysis was 1.3% (9 out of 689), similar to the reported by Couch et al (Couch et al., 2017) in a study of 58,798 BC patients, in which they found 1.41% of truncating variants and 2.22% of LP/P CHEK2 missense variants. Interestingly, 3 young CRC cases carried an LP/P CHEK2 variant and none of them had any additional pathogenic variant in our NGS panel analysis. By this means,CHEK2 represents the most frequently mutated gene after MMR genes in our hereditary non-polyposis CRC (HNPCC) cohort. CHEK2c.1100delC was reported in 6 out of 234 HNPCC families from Poland (Meijers-Heijboer et al., 2003). In their study, 3 of them also carried germline MMR P variants. In addition, c.470T>C has been found in familial CRC. To our knowledge this is the largest Spanish dataset presenting the sequencing of the whole CHEK2 coding region together with the first attempt to apply ACMG-AMP guidelines for this gene. We detailed different strategies that can be helpful to classify VUS using different frameworks with the aim of being of help not only for the curation of CHEK2 variants but also for other genes. We hope our work serves as a starting point to better tune ACMG criteria in the case of low-penetrance and low effect size variants associated with disease risk.