Evaluating ACMG and COSMIC set of genes

We evaluated the clinical implications of the high-impact, very rare variants by comparing the existence of these variants in two gene sets: the Catalogue of Somatic Mutations in Cancer (COSMIC; http://cancer.sanger.ac.uk/cosmic) and the American College of Medical Genetics and Genomics (ACMG; https:// www.acmg.net/). COSMIC's Cancer Genes Census catalogues genes that exhibit mutations that are causally implicated in cancer pathogenesis (see Supplementary Material for the complete list). Of all COSMIC genes harbouring germline cancer mutations (n = 87) associated with cancer predisposition, six high-impact variants in five cancer predisposition genes were noted (Supplementary Table S2). Five of the variants were singletons, and one was a doubleton: rs34295337 in ERCC3, a gene associated with xeroderma pigmentosum type B (Ma et al., Reference Ma, Siemssen, Noteborn and van der Eb1994), which is a rare autosomal recessive disease that is associated with skin cancer (Paszkowska-Szczur et al., Reference Paszkowska-Szczur, Scott, Serrano-Fernandez, Mirecka, Gapska, Górski, Cybulski, Maleszka, Sulikowski, Nagay, Lubinski and Dębniak2013). One variant, rs11571833, in the BRCA2 gene, was described previously as being associated with an increased risk of developing a variety of cancer types including lung, breast, prostate, gastric and aerodigestive tract cancer (Wang et al., Reference Wang, McKay, Rafnar, Wang, Timofeeva, Broderick, Zong, Laplana, Wei, Han, Lloyd, Delahaye-Sourdeix, Chubb, Gaborieau, Wheeler, Chatterjee, Thorleifsson, Sulem, Liu, Kaaks, Henrion, Kinnersley, Vallée, LeCalvez-Kelm, Stevens, Gapstur, Chen, Zaridze, Szeszenia-Dabrowska, Lissowska, Rudnai, Fabianova, Mates, Bencko, Foretova, Janout, Krokan, Gabrielsen, Skorpen, Vatten, Njølstad, Chen, Goodman, Benhamou, Vooder, Välk, Nelis, Metspalu, Lener, Lubiński, Johansson, Vineis, Agudo, Clavel-Chapelon, Bueno-de-Mesquita, Trichopoulos, Khaw, Johansson, Weiderpass, Tjønneland, Riboli, Lathrop, Scelo, Albanes, Caporaso, Ye, Gu, Wu, Spitz, Dienemann, Rosenberger, Su, Matakidou, Eisen, Stefansson, Risch, Chanock, Christiani, Hung, Brennan, Landi, Houlston and Amos2014; Delahaye-Sourdeix et al., Reference Delahaye-Sourdeix, Anantharaman, Timofeeva, Gaborieau, Chabrier, Vallée, Lagiou, Holcátová, Richiardi, Kjaerheim, Agudo, Castellsagué, Macfarlane, Barzan, Canova, Thakker, Conway, Znaor, Healy, Ahrens, Zaridze, Szeszenia-Dabrowska, Lissowska, Fabianova, Mates, Bencko, Foretova, Janout, Curado, Koifman, Menezes, Wünsch-Filho, Eluf-Neto, Boffetta, Fernández Garrote, Polesel, Lener, Jaworowska, Lubiński, Boccia, Rajkumar, Samant, Mahimkar, Matsuo, Franceschi, Byrnes, Brennan and McKay2015; Thompson et al., Reference Thompson, Gorringe, Rowley, Li, McInerny, Wong-Brown, Devereux, Li, Trainer, Mitchell, Scott, James and Campbell2015; Meeks et al., Reference Meeks, Song, Michailidou, Bolla, Dennis, Wang, Barrowdale, Frost, McGuffog, Ellis, Feng, Buys, Hopper, Southey, Tesoriero, James, Bruinsma, Campbell, Broeks, Schmidt, Hogervorst, Beckman, Fasching, Fletcher, Johnson, Sawyer, Riboli, Banerjee, Menon, Tomlinson, Burwinkel, Hamann, Marme, Rudolph, Janavicius, Tihomirova, Tung, Garber, Cramer, Terry, Poole, Tworoger, Dorfling, van Rensburg, Godwin, Guénel, Truong, Stoppa-Lyonnet, Damiola, Mazoyer, Sinilnikova, Isaacs, Maugard, Bojesen, Flyger, Gerdes, Hansen, Jensen, Kjaer, Hogdall, Hogdall, Pedersen, Thomassen, Benitez, González-Neira, Osorio, Hoya Mde, Segura, Diez, Lazaro, Brunet, Anton-Culver, Eunjung, John, Neuhausen, Ding, Castillo, Weitzel, Ganz, Nussbaum, Chan, Karlan, Lester, Wu, Gayther, Ramus, Sieh, Whittermore, Monteiro, Phelan, Terry, Piedmonte, Offit, Robson, Levine, Moysich, Cannioto, Olson, Daly, Nathanson, Domchek, Lu, Liang, Hildebrant, Ness, Modugno, Pearce, Goodman, Thompson, Brenner, Butterbach, Meindl, Hahnen, Wappenschmidt, Brauch, Brüning, Blomqvist, Khan, Nevanlinna, Pelttari, Aittomäki, Butzow, Bogdanova, Dörk, Lindblom, Margolin, Rantala, Kosma, Mannermaa, Lambrechts, Neven, Claes, Maerken, Chang-Claude, Flesch-Janys, Heitz, Varon-Mateeva, Peterlongo, Radice, Viel, Barile, Peissel, Manoukian, Montagna, Oliani, Peixoto, Teixeira, Collavoli, Hallberg, Olson, Goode, Hart, Shimelis, Cunningham, Giles, Milne, Healey, Tucker, Haiman, Henderson, Goldberg, Tischkowitz, Simard, Soucy, Eccles, Le, Borresen-Dale, Kristensen, Salvesen, Bjorge, Bandera, Risch, Zheng, Beeghly-Fadiel, Cai, Pylkäs, Tollenaar, Ouweland, Andrulis, Knight, Narod, Devilee, Winqvist, Figueroa, Greene, Mai, Loud, García-Closas, Schoemaker, Czene, Darabi, McNeish, Siddiquil, Glasspool, Kwong, Park, Teo, Yoon, Matsuo, Hosono, Woo, Gao, Foretova, Singer, Rappaport-Feurhauser, Friedman, Laitman, Rennert, Imyanitov, Hulick, Olopade, Senter, Olah, Doherty, Schildkraut, Koppert, Kiemeney, Massuger, Cook, Pejovic, Li, Borg, Öfverholm, Rossing, Wentzensen, Henriksson, Cox, Cross, Pasini, Shah, Kabisch, Torres, Jakubowska, Lubinski, Gronwald, Agnarsson, Kupryjanczyk, Moes-Sosnowska, Fostira, Konstantopoulou, Slager, Jones, Antoniou, Berchuck, Swerdlow, Chenevix-Trench, Dunning, Pharoah, Hall, Easton, Couch, Spurdle and Goldgar2016; Vijai et al., Reference Vijai, Topka, Villano, Ravichandran, Maxwell, Maria, Thomas, Gaddam, Lincoln, Kazzaz, Wenz, Carmi, Schrader, Hart, Lipkin, Neuhausen, Walsh, Zhang, Lejbkowicz, Rennert, Stadler, Robson, Weitzel, Domchek, Daly, Couch, Nathanson, Norton, Rennert and Offit2016). Two variants, one in the DICER1 gene and one in the NF1 gene, were novel. The NF1 gene harboured one additional high-impact variant. Notably, NF1 germline mutations underlie the neurofibromatosis type 1 phenotype, a disease that is reportedly diagnosed at higher rates in the AJP than in the European population (Garty et al., Reference Garty, Laor and Danon1994).

The ACMG recommendation for reporting incidental findings in clinical sequencing includes 56 genes (22 genes intersect with COSMIC genes; see Supplementary Material for the complete list). High-impact variants were noted in two ACMG genes. The first variant (rs11571833) in the BRCA2 gene was already described and discussed above. The second variant, rs200563280, results in a premature stop codon in the RYR1 gene, a gene that is associated with malignant hyperthermia (Robinson et al., Reference Robinson, Carpenter, Shaw, Halsall and Hopkins2006). Thus, the rate of actionable incidental findings in the AJP is 1·56%, similar to the estimate for Europeans at approximately 2% (Amendola et al., Reference Amendola, Dorschner, Robertson, Salama, Hart, Shirts, Murray, Tokita, Gallego, Kim, Bennett, Crosslin, Ranchalis, Jones, Rosenthal, Jarvik, Itsara, Turner, Herman, Schleit, Burt, Jamal, Abrudan, Johnson, Conlin, Dulik, Santani, Metterville, Kelly, Foreman, Lee, Taylor, Guo, Crooks, Kiedrowski, Raffel, Gordon, Machini, Desnick, Biesecker, Lubitz, Mulchandani, Cooper, Joffe, Richards, Yang, Rotter, Rich, O'Donnell, Berg, Spinner, Evans, Fullerton, Leppig, Bennett, Bird, Sybert, Grady, Tabor, Kim, Bamshad, Wilfond, Motulsky, Scott, Pritchard, Walsh, Burke, Raskind, Byers, Hisama, Rehm, Nickerson and Jarvik2015). None of the above variants were mentioned in a recent study, based on the same dataset that expanded the recommendations for an AJP screening panel (Baskovich et al., Reference Baskovich, Hiraki, Upadhyay, Meyer, Carmi, Barzilai, Darvasi, Ozelius, Peter, Cho, Atzmon, Clark, Yu, Lencz, Pe'er, Ostrer and Oddoux2016).

AJP-specific variants

We next examined AJP-specific variants. We defined variants as AJP specific if they were unique (i.e. novel) or very rare (minor allele frequency (MAF) <1%) in the NJPs, but more prevalent in the AJP (MAF >1%). Of the total AJP variants, 17,977 (8%) were AJP specific. To confirm that our dataset is enriched with variants that are unique to the AJP, we performed the same analysis on 128 verified Europeans from the Personal Genome Project (PGP) (Church, Reference Church2005; see Supplementary methods). Only 8748 variants (3·6% of the PGP dataset) were more than 1% in the PGP dataset but not in NJPs (both European and non-European populations).

We then looked at genes that are enriched for moderate- to high-impact variant groups that are AJP specific. This analysis yielded 5142 variants. Most genes harboured up to one such variant, 840 genes exhibited two variants and 196 genes displayed three or more moderate- to high-impact variants (Supplementary Fig. S2). After QC (see Supplementary Methods), three outlier genes were filtered out (Supplementary Fig. S2). In this analysis, virtually no correlation between the number of variants and the genomic length of the gene was observed (Pearson's correlation = 0·1). Next, we examined the residual variation intolerance score (RVIS) (Petrovski et al., Reference Petrovski, Wang, Heinzen, Allen and Goldstein2013) in order to identify genes under purifying selection that harbour unique or prevalent mutations in the AJP. Briefly, RVIS measures the tolerance of a gene to contain damaging variation. Genes with a low RVIS are predicted to be less tolerant to variation, and hence are more likely to exhibit a phenotype due to non-synonymous variants. The APC gene harboured a high number of AJP-specific variants (n = 7) and is in the lowest 0·2 percentile of RVIS (Fig. 2(a)). Mutations in the APC gene are associated with a specific form of inherited predisposition to colorectal cancer. Overall, colorectal cancer is more prevalent in the AJP than in NJPs (Feldman, Reference Feldman2001). Notably, the p.I1307K missense mutation in APC (rs1801155), which has been previously shown to moderately increase colorectal cancer risk in the AJP (Woodage et al., Reference Woodage, King, Wacholder, Hartge, Struewing, McAdams, Laken, Tucker and Brody1998), was among the identified variants (MAF = 0·047), and was recommended for inclusion in AJP screening (Baskovich et al., Reference Baskovich, Hiraki, Upadhyay, Meyer, Carmi, Barzilai, Darvasi, Ozelius, Peter, Cho, Atzmon, Clark, Yu, Lencz, Pe'er, Ostrer and Oddoux2016). However, additional susceptibility variants were detected in the APC gene, suggesting that other variants may contribute to the increased prevalence of colorectal cancer in the AJP. Other genes with low RVIS and harbouring four AJP-specific damaging variants are ABCA12, TULP4, DNMT1, DMXL1 and HECW1. To the best of our knowledge, the prevalence of the phenotypes associated with these genes (Supplementary Table S3) is not significantly higher in the AJP compared with other NJPs. Hence, the clinical implications and significance of this seemingly high rate of damaging variants in these genes warrant further investigation in additional extended Ashkenazi Jewish studies.

Fig. 2. (a) Genes with a low residual variation intolerance score (RVIS) are less tolerant to rare functional variants. Only six genes had a very low RVIS and four or more high to moderate Ashkenazi Jewish population (AJP)-specific variants, including the APC gene, which had the lowest RVIS and highest number of variants at seven. (b) Histogram of the number of AJP-specific deleterious variants in a gene. While most of the genes had two or fewer of these variants, eight genes had three to five variants.

To assess the effect of the AJP-specific variants on protein function, we used the MetaLR (Dong et al., Reference Dong, Wei, Jian, Gibbs, Boerwinkle, Wang and Liu2015) ensemble tool, which integrates different prediction tools using logistic regression to predict whether a variant is deleterious (see Supplementary Methods). Overall, we obtained 649 AJP-specific deleterious variants in 580 different genes. Only eight genes had at least three AJP-specific deleterious variants (Fig. 2(b) and Supplementary Table S3): APC, ABCA12, LRP2, EPPK1, HGFAC, ACAD11, HLCS and NOX1. APC and ABCA12 were discussed; the HGFAC (three variants) gene is a member of the peptidase S1 protein family and is associated with pancreatic cancer (Kitajima et al., Reference Kitajima, Ide, Ohtsuka and Miyazaki2008), a cancer type that is known to be more frequent among the AJP (Feldman, Reference Feldman2001). The EPPK1 gene (four variants) encodes a protein that belongs to the plakin family and is related to ‘vacterl association’ disorder (Hilger et al., Reference Hilger, Schramm, Pennimpede, Wittler, Dworschak, Bartels, Engels, Zink, Degenhardt, Müller, Schmiedeke, Grasshoff-Derr, Märzheuser, Hosie, Holland-Cunz, Wijers, Marcelis, van Rooij, Hildebrandt, Herrmann, Nöthen, Ludwig, Reutter and Draaken2013). The phenotype of this disorder encompasses Fanconi anaemia, a phenotype that is diagnosed at a higher frequency in the AJP compared with NJPs (Kutler & Auerbach, Reference Kutler and Auerbach2004), and hence, these variants may contribute to these higher occurrence rates. The other genes are associated with different types of rare diseases, but to the best of our knowledge, these conditions are not diagnosed at an increased rate in the AJP (Supplementary Table S3).

Furthermore, to examine whether the genes harbouring AJP-specific deleterious variants were previously implicated as AJP-prevalent phenotypes, we queried VarElect (http://varelect.genecards.org/) using the term ‘Ashkenazi’. VarElect can prioritise genotype–phenotype associations based on various databases. Of the 580 queried genes, 14 genes harbouring 17 variants (Table 1) were found to be directly related to the ‘Ashkenazi’ term, denoting conditions that are common to the AJP. Five of the 17 variants are considered to be pathogenic by the Clinvar database, four of the variants were also included in the recent recommendation for the AJP screening panel (Baskovich et al., Reference Baskovich, Hiraki, Upadhyay, Meyer, Carmi, Barzilai, Darvasi, Ozelius, Peter, Cho, Atzmon, Clark, Yu, Lencz, Pe'er, Ostrer and Oddoux2016) and four of the genes are included in the AJP screening panel, but for different variants. To verify our results, we did the same for the 128 European individuals looking at European-specific variants, meaning genes with variants that were very rare in the non-European population but not in the European population (423 genes), and tried to find genes that were related to the ‘Ashkenazi’ phenotype. Although 20 genes were found to be related, none of the variants in them was found to be pathogenic by Clinvar, which further supports our results. Taken together, these results suggest that additional variants, among these 17 variants, are plausibly causal and hence should be further investigated.

Table 1. Ashkenazi Jewish population-specific deleterious predicted variants in genes that relate to Ashkenazi Jews according to VarElect.

^a A score given by VarElect to show how much the gene was found to be related to the Ashkenazi phenotype.

^b The diseases that were related to this gene in the context of the Ashkenazi phenotype according to VarElect.

^c Diseases that were related to the variant according to Clinvar.

^d Is the gene or the variant also found in a recent recommended AJP screening panel?

AJP = Ashkenazi Jewish population; RVIS = residual variation intolerance score.

Using the Ashkenazi Jewish database in an analysis of Ashkenazi Jewish early BC patients

The major objective of clinical sequencing is to identify the causative mutation from amongst numerous detected variants. To that end, non-synonymous variants with rare allele frequencies are considered initially as plausible causative mutations. Since the AJP is not included in any of the public databases of international sequencing efforts, the MAFs of closely related populations such as Europeans (Haas et al., Reference Haas, Winter, Lim, Kirby, Blumenstiel, DeFelice, Gabriel, Jalas, Branski, Grueter, Toporovski, Walther, Daly and Farese2012; Lee et al., Reference Lee, Durr, Majczenko, Huang, Liu, Lien, Tsai, Ichikawa, Goto, Monin, Li, Chung, Mundwiller, Shakkottai, Liu, Tesson, Lu, Brice, Tsuji, Burmeister, Stevanin and Soong2012; Rees et al., Reference Rees, Ng, Ruppert, Turner, Beer, Swift, Morken, Below, Blech, Mullikin, McCarthy, Biesecker, Gloyn and Collins2012) are often utilised as surrogates. We evaluated the advantages of using AJP-specific MAFs when screening the WES data of Ashkenazi Jewish samples. Of the 55,416 high- and moderate-impact mutations, 57·7% were classified as very rare based on the general European MAF versus 50·6% based on the AJP MAF, leading to out-filtration of approximately 3900 variants (Fig. 3(a)). Likewise, based on the maximum MAF (MMAF) of all NJPs, 50·1% of the variants were classified as very rare, compared to 40·8% when including the AJP. These results are in line with Carmi et al. (Reference Carmi, Hui, Kochav, Liu, Xue, Grady, Guha, Upadhyay, Ben-Avraham, Mukherjee, Bowen, Thomas, Vijai, Cruts, Froyen, Lambrechts, Plaisance, Van Broeckhoven, Van Damme, Van Marck, Barzilai, Darvasi, Offit, Bressman, Ozelius, Peter, Cho, Ostrer, Atzmon, Clark, Lencz and Pe'er2014). For rare variants (MAF <5%), the advantage of using AJP-specific MAFs is somewhat less significant (1·2% difference), in line with the notion that population-specific variants are predominantly very rare (1000 Genomes Project Consortium et al., Reference Auton, Brooks, Durbin, Garrison, Kang, Korbel, Marchini, McCarthy, McVean and Abecasis2015).

Fig. 3. Frequencies of high- to moderate-impact variants (a) and deleterious variants (b) by populations' minor allele frequency (MAF) (orange – very rare; yellow – rare; green – common). By joining the Ashkenazi Jewish population (AJP) MAF to the non-Jewish population (NJP) MAF and using the maximum MAF, the percentages of very rare variants were reduced by 10% and 13%, respectively. (c) Filtration of very rare variants of 49 Ashkenazi Jewish (AJ) breast cancer patients. Adding the AJ MAF filtered an additional 57 (36%) of the variants, demonstrating the utility of using the same population database.

AFR = African; EAS = East Asian; EUR = European; SAS = South Asian.

Similarly, potentially deleterious variants are prioritised in clinical NGS applications. Based on the AJP MAF, 79·0% of deleterious variants, based on MetaLR, were considered very rare, whereas 89·6% were considered very rare based on the European MAF (Fig. 3(b)). Furthermore, combining the AJP MAF with the NJP MMAF substantially improved filtering from 85·9% of the variants classified as very rare to just 72·9%. Since the MAFs of numerous populations, but not the AJP, are included in the MetaLR model, adding the AJP MAF can significantly improve the filtering of deleterious variants. Taken together, these significant population-specific differences in rare variants indicate that by utilising AJP-specific MAFs, finer filtration and lower false-positive rates can be achieved in Ashkenazi Jewish sequencing studies.

Importantly, we evaluated the utility of the AJP-specific screening approach using the independent WES data of 49 Ashkenazi Jewish samples derived from high-risk BC cases who do not harbour mutations in the predominant underlying genes – BRCA1 and BRCA2. Of the 2638 predicted deleterious variants, 81·3% were very rare according to the European MAF, compared to 77·5% using the AJP MAF. Similarly, combining the AJP with the NJP MMAF improved filtering by approximately 10% from 75·9% to 64·5% (Supplementary Fig. S3).

In our actual disease gene analysis of the Ashkenazi Jewish BC sample, we screened for very rare variants that are potentially deleterious by MetaLR and are present in at least three BC cases, resulting in 450 potentially deleterious variants. Filtering by using the European MAF resulted in 189 variants in 148 genes, while using the MMAF of the Ashkenazi Jewish and Europeans filtered an additional 69 variants, resulting in 120 potential variants (36%). In comparison, using the MMAF of Europeans and 128 individuals from African, EAS or SAS populations resulted in minor additional filtering of only seven, two and 13 variants, respectively (Supplementary Fig. S4). Using all populations' MMAFs (AJP + NJP) versus only the NJP MMAF resulted in 100 variants in 72 genes compared to 157 variants in 126 genes (36%) (Fig. 3(c)). We then used VarElect to search for genes related to the keyword ‘breast’. The MSH6 gene scored highest using VarElect (Supplementary Table S4) and by the MetaLR deleterious score (0·88). The protein coded by this gene is a member of the DNA mismatch repair MutS family, and rare variants in this gene are associated with familial BC (Wasielewski et al., Reference Wasielewski, Riaz, Vermeulen, van den Ouweland, Labrijn-Marks, Olmer, van der Spaa, Klijn, Meijers-Heijboer, Dooijes and Schutte2010). Mutations in MSH6 are traditionally associated with Lynch syndrome (Baglietto et al., Reference Baglietto, Lindor, Dowty, White, Wagner, Garcia, Vriends, Cartwright, Barnetson, Farrington, Tenesa, Hampel, Buchanan, Arnold, Young, Walsh, Jass, Macrae, Antill, Winship, Giles, Goldblatt, Parry, Suthers, Leggett, Butz, Aronson, Poynter, Baron, Le Marchand, Haile, Gallinger, Hopper, Potter, de la Chapelle, Vasen, Dunlop, Thibodeau and Jenkins2010), a syndrome that seems to encompass BC susceptibility according to recent publications (Win et al., Reference Win, Lindor and Jenkins2013). This finding requires further examination of a larger cohort in order to draw better conclusions about the role of these variants in BC predisposition.