Panel sequencing of 264 candidate susceptibility genes and segregation analysis in a cohort of non-BRCA1, non-BRCA2 breast cancer families

Abstract

Purpose

The main aim of this study was to screen epigenetic modifier genes and known breast cancer driver genes for germline mutations in non-BRCA1/2 (BRCAx) breast cancer families in order to identify novel susceptibility genes of moderate-high penetrance.

Methods

We screened 264 candidate susceptibility genes in 656 index cases from non-BRCA1/2 families. Potentially pathogenic candidate mutations were then genotyped in all available family members for the assessment of co-segregation of the variant with disease in the family in order to estimate the breast cancer risks associated with these mutations. For 11 of the candidate susceptibility genes, we screened an additional 800 non-BRCA1/2 breast cancer cases and 787 controls.

Results

Only two genes, CHD8 and USH2A showed any evidence of an increased risk of breast cancer (RR = 2.40 (95% CI 1.0–7.32) and 2.48 (95% CI 1.11–6.67), respectively).

Conclusions

We found no convincing evidence that epigenetic modifier and known breast cancer driver genes carry germline mutations that increase breast cancer risk. USH2A is no longer regarded as a breast cancer driver gene and seems an implausible candidate given its association with Usher syndrome. However, somatic mutations in CHD8 have been recently reported, making it an even more promising candidate, but further analysis of CHD8 in very large cohorts of families or case-control studies would be required to determine if it is a moderate-risk breast cancer susceptibility gene.

Introduction

Breast cancer is the most common cancer and the second leading cause of cancer death in women. Risk increases proportionally with the number of affected relatives with breast cancer and younger age at diagnosis [1]. Highly penetrant mutations in BRCA1 and BRCA2 account <40% of all multiple-case breast cancer families [2], and a small number of families are due to moderate-high penetrance genes in TP53, PALB2, ATM, CHEK2, PTEN, STK11, CDH1 and NF1 [3]. In addition, over 170 breast cancer risk loci have been identified by genome-wide association studies, which explain ~ 18% of the familial relative risk of breast cancer (Michailidou et al., submitted). The “missing” familial relative risk could be explained by extreme poly-genicity, variants not tagged by genome-wide associations such as short tandem repeats or rare mutations in another gene (or genes) conferring a moderately increased risk of breast cancer [4]. In the past decade, intensive efforts have been made to search for additional breast cancer predisposition genes, with limited success [5, 6].

There is increasing evidence that germline and somatic mutations in cancer target the same genes and pathways [7–9]. Whole-exome sequencing studies have identified somatic driver genes involved in DNA methylation and chromatin modification in a number of different cancers, including breast cancer [10–16]. The main aim of this study was to screen epigenetic modifier genes, including human homologs of modifiers of murine metastable epialleles (MOMME) genes which are modifiers of epigenetic silencing [17–20], for germline mutations in non-BRCA1/2 (BRCAx) breast cancer families. Another reason for selecting this class of candidate genes was that we have previously reported that BRCA1-associated breast tumours have a distinct methylome, suggesting that BRCA1 may be involved, directly or indirectly, in de novo methylation [21]. A possible mechanism for the relationship between BRCA1 mutation status and the methylome is suggested by the finding that BRCA1 binds to Oct1-binding sites in the DNMT1 promoter, thereby transcriptionally regulating this maintenance methylation enzyme [22]. Other breast cancer susceptibility genes may share a similar mechanism, thereby altering the methylome. Furthermore, we previously found that BRCAx tumours show two different patterns of methylation, suggesting that mutations in human epigenetic modifying genes may account for some breast cancer families, which share a methylation profile in their tumours. Rare families have been described in which germline epimutations are associated with colorectal cancer [23], but this phenomenon has not been reported in breast cancer families.

We screened 264 candidate susceptibility genes in 656 index cases from non-BRCA1/2 families. Potentially pathogenic candidate mutations were then genotyped in all available family members for the assessment of co-segregation of the variant with disease in the family in order to estimate the breast cancer risks associated with these mutations. For 11 of the candidate susceptibility genes, we screened an additional 800 non-BRCA1/2 breast cancer cases and 787 controls.

Methods

Patient cohort

We selected 656 non-BRCA1/2 families from the Kathleen Cuningham Foundation Consortium for Research into Familial Breast cancer (kConFab) [24] using the following criteria: (i) no known pathogenic mutation in BRCA1 or BRCA2 in any member of the family, and (ii) no known protein-truncating mutation in PALB2 or the ATM V2424G mutation. In addition, we prioritized family selection (i) on the basis of age of diagnosis of the individual to be sequenced, (ii) including any family with a case of pancreatic cancer (n = 94), (iii) families who had previously had the most sensitive and complete testing of BRCA1 and BRCA2, (iv) the availability of at least two germline DNA samples from related family members, and (v) ranking based on the probability calculated by BOADICEA [25] of carrying a BRCA1/2 mutation. Families with multiple cases of breast or ovarian cancer are eligible for recruitment into kConFab under several different criteria [24]. These 656 families had an average of 4.9 breast cancer cases per family (with a range of 0–19 breast cancer cases, and 1–5 ovarian cancer cases). From these families, we selected the youngest breast cancer case for whom germline DNA was available as the index case for sequencing. We also selected 800 early-onset breast cancer cases (mean age of diagnosis 45.5 years; range 20–62) and 787 controls with germline DNA from the Breast Cancer Family Registry [26](700 cases and 589 controls) and the University of Utah (100 cases and 198 controls). Cases with previously identified known pathogenic variants in BRCA1 or BRCA2 were excluded.

Exome sequencing

We performed exome sequencing of germline DNA from 12 members of five kConFab families. The libraries were prepared using NimbleGen SeqCap EZ exome capture kit v.2 and sequenced on the Illumina HiSeq 2000 machine. Paired-end reads were aligned to hg19 genome using BWA.[27] Alignments were de-duplicated with Picard and realigned and recalibrated using GATK tools [28]. The variants were identified using UnifiedGenotyper followed by variant quality score recalibration and variant filtering according to GATK protocol. Variants were annotated with ANNOVAR and 1000 Genomes data (May 2011). We identified loss of function (LoF; nonsense, at canonical splice sites, and frameshift indels) variants, and all variants shared between fourth- or fifth-degree relatives within each family. We filtered out variants present in the November 2010 release of the 1000 Genomes Project and dbSNP v.132, in UTRs and ncRNA, overlapping known indels, and variants in genes with known LoF variants in the human population [29]. A LoF mutation in BRCA2 gene was identified in one sample and so this family was excluded from further analysis. For the capture experiment, we selected 26 genes with variants shared between distant relatives in four families, and 79 genes with LoF variants identified in 11 samples.

Candidate gene selection

We selected a list of 103 genes involved in epigenetic regulation (59 selected from the literature—Group 1; 44 using the GO terms ‘chromatin organization’ or ‘chromatin reorganization’—Group 2), and supplemented it with 105 candidate genes from whole-exome sequencing of four non-BRCA1/2 families (Group 3; 79 loss of function and 26 variants shared between distant family members) and 56 additional breast cancer driver genes (Groups 4 and 5 [30, 31]) known at the time (Table 1).

Table 1.

Candidate genes screened in kConFab index cases

RefSeq_gene	Group	References
ABCA13	3	–
AHSP	3	–
AKT1	4	[30]
AKT2	4	[30]
ANKRD17	3	–
ANKRD30A	3	–
APC	4	[30]
ARHGAP31	3	–
ARHGAP5	5	[31]
ARID1A	1	[48]
ARID1B	1	[16, 49]
ARID2	1	[50]
ASB4	3	–
ASH1L	3	–
ASIC3	3	–
ASXL1	2	–
ATR	5	[31]
ATRX	1	[51, 5]
AURKC	1	[52, 53]
BABAM1	2	–
BAP1	1	[54]
BAZ1B	1	[18]
BBS10	3	–
BCOR	2	–
BMPER	3	–
BRCC3	1	[55]
BRD7	1	[56]
C11orf30	5	[31]
C14orf21	3	–
C15orf42	3	–
C17orf50	3	–
C17orf78	3	–
CABIN1	2	–
CACNA1B	3	–
CACNA2D2	3	–
CADPS2	3	–
CALHM3	3	–
CAND2	3	–
CASC5	2	–
CASP8	4	[30]
CCND1	4	[30]
CD163L1	3	–
CDC73	2	–
CDHR2	3	–
CDKNlB	4	[30]
CELF2	3	–
CELSR1	5	[31]
CEP192	5	[31]
CHD1	1	[57]
CHD2	1	[57]
CHD3	1	[57]
CHD4	1	[57]
CHD5	2	–
CHD6	2	–
CHD7	1	[57]
CHD8	1	[57]
CHD9	1	[57]
CLIC6	3	–
CREBBP	1	[58, 59]
DAXX	1	[51, 60]
DAZAP2	3	–
DGKH	3	–
DLEC1	3	–
DLG5	5	[31]
DNAH5	5	[31]
DNMT1	1	[61]
DNMT3A	1	[62, 63]
DNMT3B	1	[64]
DPYD	3	–
DSCR6	3	–
EP300	1	[65]
EP400	2	–
EPX	3	–
ERBB2	4	[30]
EYA1	2	–
EZH2	1	[66–68]
FAM104B	3	–
FAM208A	1	[69]
FGFR1	4	[30]
FIGNL1	3	–
FLJ43860	3	–
FMN2	5	[31]
FOXO3	1	[70]
FREM1	3	–
FSHR	2	–
GATA2	2	–
GATA3	4	[30]
GGT1	3	–
GH1	5	[31]
GPR112	5	[31]
GPR125	3	–
H3F3A	1	[60]
HDAC1	1	[18]
HDAC4	2	–
HDAC9	2	–
HECW1	5	[31]
HMGA1	1	[71]
HMGA2	1	[72]
HNF1A	2	–
HRNR	5	[31]
HSPB1	3	–
HUWE1	2	–
IDH1	1	[73]
IDH2	1	[73]
IER2	3	–
IGSF3	3	–
IKZF1	2	–
IL17B	3	–
ITPR2	3	–
JAK2	2	–
JMJD1C	2	–
JMJD1C	2	–
KCNK15	3	–
KDM2A	2	–
KDM3A	2	–
KDM3B	2	–
KDM5A	2	–
KDM5B	2	–
KDM5C	1	[74]
KDM6A	1	[74]
KIAA0556	5	[31]
KIAA2018	5	[31]
KRAS	4	[30]
LAMA2	5	[31]
LILRA1	3	–
LIX1	3	–
LPHN2	5	[31]
LRP2	5	[31]
LYRM5	3	–
MAML1	3	–
MAP2K4	4	[30]
MAP3K1	4	[30]
MAP3K13	4	[30]
MDM2	4	[30]
MDN1	5	[31]
MED12	1	[75]
MEN1	1	[51]
MESP1	3	–
MLL	2	–
MLL2	1	[76, 77]
MLL3	1	[78, 77]
MLL4	2	–
MLL5	2	–
MORC3	1	[79]
MRPL37	3	–
MTOR	1	[51]
MYB	1	[80]
MYC	4	[30]
NCOR1	1	[81]
NEB	3	–
NELL1	5	[31]
NET1	3	–
NF1	4	[30]
NIPBL	2	–
NLRP2	3	–
NOS1	2	–
NOTCH1	3	–
NPM1	1	[82]
NSD1	2	–
NUMBL	3	–
NUP210	3	–
OPRM1	3	–
PAM	3	–
PAX5	2	–
PBRM1	1	[74]
PCDHB11	3	–
PDE4DIP	5	[31]
PDGFRL	3	–
PGLYRP4	3	–
PIK3C2G	3	–
PIK3CA	5	[31]
PKD1L1	5	[31]
PLEKHB1	3	–
PLIN4	3	–
POC5	3	–
PPARGC1A	2	–
PRDM6	3	–
PRKAA2	2	–
PRKCA	2	–
PRKCB	2	–
PRKCQ	5	[31]
PUM2	3	–
RALGAPA1	5	[31]
RANBP17	3	–
RASA3	3	–
RB1	1	[71, 80]
RBL1	2	–
RBM44	3	–
REST	2	–
RIF1	1	[83]
RLF	1	[83]
RNF180	3	–
RP1L1	3	–
RPTN	3	–
SALL1	2	–
SAMD14	3	–
SATL1	5	[31]
SCAMP2	3	–
SETD1A	2	–
SETD2	1	[74]
SETX	3	–
SLC22A5	3	–
SLC25A2	3	–
SLC37A4	3	–
SMAD4	4	[30]
SMARCA1	1	[71]
SMARCA2	1	[84]
SMARCA4	1	[85]
SMARCA5	1	[61]
SMARCB1	1	[7]
SMARCC2	1	[86]
SMARCD1	1	[87]
SMCHD1	1	[17]
SMG5	5	[31]
SPEN	1	[88]
SPHKAP	5	[31]
SRA1	3	–
SRCAP	2	–
SSH2	3	–
ST8SIA5	3	–
SUPT3H	3	–
SYNE1	5	[31]
SYNE2	5	[31]
TAF1	2	–
TAF1L	2	–
TBX3	4	[30]
TCTN1	3	–
TDO2	5	[31]
TEK	3	–
TET1	1	[89]
TET2	1	[90]
TMEM106C	3	–
TMEM131	3	–
TMEM144	3	–
TMEM37	3	–
TPR	3	–
TPTE2	3	–
TRIM28	1	[19]
TRRAP	2	–
TSHZ3	5	[31]
TXNDC2	3	–
TYRO3	3	–
UBAP2L	5	[31]
UBR5	5	[31]
UGT1A8	5	[31]
USH2A	5	[31]
UTP14C	3	–
UTS2R	3	–
WDR36	3	–
XRN1	5	[31]
ZBTB34	3	–
ZNF142	5	[31]
ZNF217	3	–
ZNF227	3	–
ZNF256	3	–
ZNF418	3	–
ZNF438	3	–
ZNF514	3	–
ZNF518B	3	–
ZNF558	3	–
ZNF570	3	–
ZNF703	4	[30]
ZNF764	3	–
ZNF850	3	–

Targeted sequencing and selection of putative mutations

We performed custom-designed targeted sequencing covering the coding exons of all 264 candidate genes (Table 1). Targets were captured using the Agilent Target Enrichment kit by Axeq Technologies Inc. according to the manufacturer’s instructions. For the 11 candidate genes (ATR, AURKC, BAZ1B, CASP8, CHD5, DNMT3A, FSHR, KDM5B, MAP3KI, RBL1, TET2) screened in the additional cases and controls, we captured the targets using the Roche SeqCap EZ Choice Library (cat# 06266339001) and Seq-Cap EZ Reagent Kit Plus v2 (NimbleGen #06-953-247-001) according to the manufacturer’s instructions. All sequencing were done by 100 bp paired-end reads on the Illumina HiSeq 2000 platform. High-quality sequencing data with an average depth of 359.4-fold (ranging from 31.97 to 1215.27 per exon) were obtained. Sequences were aligned to human reference genome (GRCh37) using BWA [27]. Reads marked as PCR or optical duplicates were removed from consideration by Picard (http://picard.sourceforge.net). Variants were called from the targeted exons, and also from the off-target reads up to 25 nucleotides into the introns to capture potential splice site mutations. Single nucleotide variants (SNVs) and small indels were called by GATK [28]. After calling, to eliminate possible falsepositive variants, we applied strict quality control filtering of excluding variants with <30 reads or variants with reference: alternative reads ratio <0.2 or >5.0. We used ANNOVAR [32] to annotate the functional consequences of variants, as well as to retrieve allele frequencies from the public databases (dbSNP137, 1000 Genomes Project April 2012 release, complete genomes, and NHLBI-ESP 6500 exomes). Variants were removed from our analysis if they have been reported in those public databases. The potential functional importance of non-synonymous variants was predicted by the following tools: SIFT [33], Polyphen2 [34], LRT [35], MutationTaster [36] and CADD [37]. These tools are not totally independent as most include the analysis of evolutionary conservation. Any non-synonymous variant that was described by at least three of SIFT [33], Polyphen2 [34], LRT [35], or MutationTaster [36] as damaging or probably damaging, or had a CADD score >15, was regarded as putatively deleterious, and selected for segregation and casey–control analysis. Protein-truncating mutations within the last exon of each gene were removed from our candidate list since they are predicted not to activate nonsense-mediated mRNA decay (NMD). The impact on splicing was predicted for intronic variants within 25 bp from the exon/intron boundary using stand alone perl scripts of MaxEnt [38] (http://genes.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq.html). Intronic variants were classified as putatively pathogenic if they fulfilled the following criteria: (a) the predicted value of either 5′ or 3′ splicing motif was >3, and (b) the variation between MaxEnt values of variant and wildtype sequences was ≥30%.

Family sequencing and segregation analysis

Segregation analysis for all putative pathogenic mutations was carried out by either Sanger sequencing, using the standard sequencing procedure of BigDye Terminator v3.1 Cycle Sequencing Kit in the PE Applied Biosystem (PE Applied Biosystem), or by Fluidigm 96.96 Dynamic Array integrated fluidic circuits (IFCs) on the BioMark™ HD System (Fluidigm Corporation) according to the manufacturer’s instructions, or by iPLEX genotyping [39] on the MassARRAY system (Agena Bioscience). The primers can be provided on request.

Statistical analyses

We calculated the penetrance for breast cancer in carriers of the susceptibility genes assuming a constant relative risk with age. Models were fitted under maximum likelihood theory using a modified version of the LINKAGE genetic analysis package [40]. Non-carriers were assumed to be at population risks specific to Australia with incidence rates taken from cancer registry data obtained from Cancer Incidence in Five Continents, VIII (IARC, Lyon). We estimated risk ratios (RR, the age-specific breast cancer incidence rate in carriers divided by the relevant population rate). Reported breast cancers with unknown age at diagnosis were excluded from all analyses. Individuals with cancers other than breast (including ovarian cancer) were treated as unaffected at the age of their cancer diagnosis. To test for an association of pooled variants in a gene with breast cancer risk in the case–control study, we estimated odds ratios (OR) and corresponding 95% CIs based on the Fisher’s exact test.

Results

We identified 420 putative pathogenic mutations in 313 index cases using the following criteria: (a) not reported in any of these public databases (dbSNP137, 1000 Genomes Project April 2012 release, Complete Genomes, and NHLBI-ESP 6500 exomes); (b) not present in more than five index cases, since these were not likely to be high risk variants; (c) protein-truncating mutations (PTC, including frameshift insertion/deletion and stopgain) not in the last exon; (d) missense variants predicted as “deleterious” by three out of four tools, or with a CADD score >15, and were present in genes in which we also identified PTCs; (e) intronic variants called from ±25 bp region of the exon–intron boundary had to be predicted by MaxEntScan to possibly create a new splice site and in genes in which we also identified PTCs or predicted deleterious missense mutations. From these 313 families, there were 2178 germline DNAs from family members available for segregation analysis. Of the 420 putative pathogenic mutations identified, we could not design assays for 36, and failed to validate 9 variants in the index case. However, the remaining 375 variants were successfully validated by Fluidigm or iPLEX and 259 families (some of which had more than one putatively pathogenic variant), were informative for segregation analysis (Supplementary Table 1).

In the single gene analysis, assuming the constant relative risk with age, only two genes, CHD8 and USH2A showed any evidence of an increased risk of breast cancer (RR = 2.40 (95% CI 1.0–7.32) and 2.48 (95% CI 1.11–6.67), respectively) (Table 2). We identified seven families with putative pathogenic mutations in CHD8 (six missense variants and one in-frame deletion) and 10 families with putative pathogenic mutations in USH2A (two protein truncating, seven missense and one in-frame deletion). We also performed group analyses for Groups 1 and 2, Group 3, and Groups 4 and 5, but this did not identify any groups in which variants were associated with breast cancer risk (Table 3). In the case–control screening, we identified a small number of putatively pathogenic mutations, but found no evidence of an association with breast cancer risk.

Table 2.

Results of segregation analysis in kConFab families

Gene	Group	# Families	Log likelihood
ABCA13	3	15	0.85 (0.32, 1.95)
ANKRD30A	3	1	0.01a (0.01a, 2.17)
APC	4	8	2.96 (0.86, 26.0a)
ARID1A	1	7	0.58 (0.05, 1.83)
ASH1L	3	5	1.24 (0.43, 2.86)
ATR	5	4	1.00 (0.17, 5.65)
BAZ1B	1	3	>26.0
C15orf42	3	5	0.74 (0.06, 2.55)
CASC5	2	2	1.49 (0.10, 8.82)
CASP8	4	1	>26.0
CDHR2	3	1	<0.01
CELSR1	5	5	0.50 (0.04, 1.73)
CEP192	5	9	1.22 (0.35, 3.39)
CHD5	2	3	0.96 (0.16, 3.28)
CHD7	1	5	1.17 (0.42, 2.87)
CHD8	1	7	2.40 (1.0, 7.32)
CHD9	1	9	0.62 (0.22, 1.34)
DGKH	3	4	1.75 (0.80, 4.57)
DLEC1	3	5	0.01a (0.01a, 3.86)
DNAH5	5	16	0.86 (0.36, 1.79)
DNMT3A	1	1	>26.0
EP400	2	5	1.65 (0.66, 4.21)
HECW1	5	3	1.41 (0.42, 4.13)
HUWE1	2	9	2.9 (0.8, 15.0)
IL17B	3	2	0.85 (0.07, 2.52)
ITPR2	3	9	1.59 (0.58, 4.61)
KDM5B	2	6	7.50 (0.24, 26.0a)
LAMA2	5	19	1.32 (0.75, 2.29)
LPHN2	5	5	1.60 (0.76, 3.33)
LRP2	5	7	0.73 (0.17, 2.06)
MAP3K1	4	6	1.16 (0.09, 16.17)
MDN1	5	11	0.01a (0.01a, 0.03)
NCOR1	1	5	1.59 (0.06, 26.0a)
NEB	3	25	0.87 (0.40, 1.65)
NELL1	5	1	1.0 (0.04, 22.26)
NET1	3	3	1.32 (0.45, 3.30)
NIPBL	2	5	1.22 (0.36, 2.86)
NOS1	2	6	0.97 (0.14, 4.39)
NOTCH1	3	7	1.82 (0.67, 5.22)
NUP210	3	6	0.58 (0.04, 2.14)
PAX5	2	3	1.52 (0.68, 3.21)
PDE4DIP	5	1	>26.0
PIK3C2G	3	1	2.78 (0.64, 25.47)
PKD1L1	5	3	0.94 (0.10, 2.61)
PLIN4	3	1	0.78 (0.05, 7.54)
PUM2	3	3	1.77 (0.38, 8.95)
RALGAPA1	5	5	1.22 (0.39, 3.29)
RBM44	3	3	0.30 (0.05, 1.02)
RIF1	1	6	0.39 (0.03, 1.52)
RNF180	3	2	0.01a (0.01a, 1.85)
SETX	3	7	1.37 (0.48, 3.80)
SLC22A5	3	3	6.66 (0.03, 26.0a)
SMG5	5	2	0.01a (0.01a, 6.44)
SPHKAP	5	2	0.67 (0.09, 4.44)
SUPT3H	3	1	0.01a (0.01a–2.3)
SYNE1	5	22	0.88 (0.45, 1.50)
SYNE2	5	9	2.14 (0.94, 5.93)
TDO2	5	1	0.01a (0.01a, 1.73)
TET2	1	3	2.21 (0.07, 26.0a)
TMEM131	3	9	1.61 (0.77, 3.78)
TPR	3	4	1.06 (0.31, 3.05)
UBAP2L	5	4	0.70 (0.15, 2.73)
UBR5	5	1	0.01a (0.01a–6.6)
USH2A	5	10	2.48 (1.11, 6.67)
XRN1	5	2	<0.01
ZNF558	3	2	1.28 (0.33, 3.22)
TOTAL		366

Group 1: genes involved in epigenetic regulation selected from the literature

Group 2: using the GO terms ‘chromatin organization’ or ‘chromatin reorganization’

Group 3: candidate genes from whole-exome sequencing of five non-BRCA1/2 families

Groups 4 and 5: breast cancer driver genes [30, 31]

^aEstimate at boundary

Table 3.

Segregation analysis: candidate genes stratified by group

Groups	Description	Included genes	# Families	Log likelihood (95% CI)
1 + 2	Genes involved in epigenetic regulation selected from the literature and using GO terms	ARID1A, BAZ1B, CHD7, CHD8, CHD9, DNMT3A, NCOR1, RIF1, TET2, CASC5, CHD5, EP400, HUWE1, KDM5B, NIPBL, NOS1, PAX5	85	1.17 (0.86, 1.56)
3	Candidate genes from whole-exome sequencing	ABCA13, ANK4D30A, C15orf42, CDHR2, DGKH, DLEC1, IL17B, ITPR2, NEB, NET1, NOTCH1, PIK3C2G, PLIN4, PUM2, RBM44, RNF180, SLC22A5, SUPT3H	124	1.17 (0.94, 1.46)
4 + 5	Driver genes [30, 31]	APC, CASP8, MAP3K1, ATR, CELSR1, CEP192, DNAH5, HECW1, LAMA2, LPHN2, LRP2, MDN1, NELL1, PDE4DIP, PKD1L1, RALGAPA1, SMG5, SPHKAP, SYNE1, SYNE2, TDO2, UBAP2L, UBR5, USH2A, XRN1	157	1.10 (0.89, 1.35)
1 + 2 + 4 + 5			242	1.11 (0.94, 1.32)

Discussion

We carried out a large screen of candidate BRCAx genes, mostly epigenetic modifier genes and known breast cancer driver genes, by analysing 264 genes in 656 multiple-case families, followed by segregation analysis of 375 putative, pathogenic missense, splice site and truncating variants in 259 informative families. One of the limitations of this study is that since we began the project, the list of breast cancer driver genes has been extended [16, 41–43], so there are at least 90 known driver genes that we did not screen. We found some evidence that mutations in CHD8 (seven families) and USH2A (10 families) are associated with breast cancer risk. These genes were selected as candidates as an epigenetic modifier [44] and a putative cancer driver [45], respectively. Mutations in USH2A cause Usher syndrome type IIa and retinitis pigmentosa. USH2A is no longer considered a breast cancer driver gene [16, 41–43], but has been shown to have a high false discovery rate in mutation calling [46]. CHD8 encodes a member of the chromodomain-helicase-DNA binding protein family and plays a role in transcriptional regulation, epigenetic remodelling, promotion of cell proliferation, and regulation of RNA synthesis [47]. Mutations in CHD8 have been recently reported in over 5% of breast tumours [46], making it a very plausible candidate for breast cancer susceptibility. However, the p value for testing the association between variants in CHD8 and breast cancer risk is only nominally significant (p = 0.05), and given that we tested 66 genes, clearly would not be significant after correction for multiple testing. Eleven of the genes, ATR, AURKC, BAZ1B, CASP8, CHD5, DNMT3A, FSHR, KDM5B, MAP3KI, RBL1 and TET2, were also screened in approximately 800 cases and controls, but none were found to be associated with breast cancer risk. In conclusion, we have some evidence that mutations in CHD8 might be associated with an elevated risk of breast cancer, warranting further investigation.