Identification of Novel Variants in Colorectal Cancer Families by High-Throughput Exome Sequencing

Abstract

Background

Colorectal cancer (CRC) in densely affected families without Lynch Syndrome may be due to mutations in undiscovered genetic loci. Familial linkage analyses have yielded disparate results; the use of exome sequencing in coding regions may identify novel segregating variants.

Methods

We completed exome sequencing on 40 affected cases from 16 multi-case pedigrees to identify novel loci. Variants shared among all sequenced cases within each family were identified and filtered to exclude common variants and single nucleotide variants (SNVs) predicted to be benign.

Results

We identified 32 nonsense or splice-site SNVs, 375 missense SNVs, 1,394 synonymous or non-coding SNVs, and 50 indels in the 16 families. Of particular interest are two validated and replicated missense variants in CENPE and KIF23, which are both located within previously reported CRC linkage regions, on chromosomes 1 and 15, respectively.

Conclusions

Whole-exome sequencing identified DNA variants in multiple genes. Additional sequencing of these genes in additional samples will further elucidate the role of variants in these regions in colorectal cancer susceptibility.

Impact

Exome sequencing of familial CRC cases can identify novel rare variants that may influence disease risk.

INTRODUCTION

Colorectal cancer (CRC) is the third most common cancer and the third leading cause of cancer death in the United States for both men and women (1). Family history is a consistent risk factor (2); without CRC family history, the lifetime risk for an individual is 5% to 6%, but, 10% to 15% if a first-degree relative has CRC (3-5) and 30% to 100% in familial genetic syndromes (6). Lynch Syndrome represents up to 5% of CRCs and results from germline mutations that affect DNA mismatch repair (MMR) genes MLH1, MSH2, MSH6, and PMS2.Tumors from these patients demonstrate a defective MMR (dMMR) phenotype manifested by DNA microsatellite instability (MSI) and absence of MMR protein expression (7, 8).

Beyond the known familial genetic syndromes, linkage studies have implicated several additional regions in CRC susceptibility, including 3q21-24, 4q21, 7q31, 8q13, 8q23, 8q24, 9q22-31, 10p14, 11q23, 12q24, 15q22, and 18q21 (9-21). Genome-wide association studies (GWAS) of CRC have reported evidence of many common risk variants in several genetic regions, including chromosomes 1q41, 3q26, 6p21, 6q25, 8q23, 8q24, 9p24, 10p14, 11q13, 12q13, 12q24, 14q22, 15q13, 16q22, 18q21, 19q13, 20p12, 20q13, and Xp22 (14, 15, 19, 21-30). However, results from linkage studies have not been consistent, and GWAS are not ideal for the identification of rare variants. Hypothesizing that coding regions may harbor rare variants segregating with susceptibility, we sequenced the exomes of 40 affected individuals from 16 CRC families. To our knowledge, this represents the first family-based application of massively-parallel sequencing in this disease (31-36).

MATERIALS AND METHODS

Study Participants

We utilized The Colon Cancer Family Registry (Colon CFR), an NCI–supported consortium established to create an infrastructure for interdisciplinary studies of the genetic and molecular epidemiology of CRC (37-39). Families were enrolled via the Mayo Clinic, Memorial University of Newfoundland, the University of Southern California, or the University of Melbourne (37). Risk-factor data, blood samples, and pathology reports were collected on participants, using standardized core protocols, and germline DNA was isolated from blood. Sixty-six pedigrees were reviewed with ≥2 invasive CRC cases and no evidence of Lynch syndrome, MUTYH mutations (37), or familial adenomatous polyposis. Sixteen families were selected based on the presumption of a genetic predisposition to disease due to 1) large numbers of affected relatives and 2) younger ages at diagnosis (Table 1). Forty affected individuals were chosen for sequencing based on genetic relatedness (preferring distant relatives), including three cases per family where possible (Figure S1). All aspects of this work received institutional review board approval under the policies of the Colon CFR.

Table 1.

Family Characteristics, Sequencing Conditions, and Number of Shared Variants Identified

						N Variants (N Private)
Family	N Affected	Mean Age at Diagnosis (Range)	N Sequenced (Relation)	Library Capture	Sequencing Platform	Nonsense, splice site	Indel	Missense	Other	Total
1	8	48.8 (25-72)	2 (first cousins)	36 Mb	GAIIx	1 (0)	4 (4)	15 (14)	39 (32)	59 (50)
2	4	49.0 (31-68)	3 (siblings)	36 Mb	GAIIx	3 (3)	5 (5)	22 (20)	74 (57)	104 (85)
3	6	64.2 (51-69)	3 (siblings)	36 Mb	GAIIx	1 (0)	2 (2)	16 (16)	47(36)	66 (54)
4	6	58.0 (44-89)*	2 (first cousins once removed)	36 Mb	GAIIx	2 (2)	8 (7)	15 (11)	51 (38)	76 (58)
5	6	61.0 (50-79)*	3 (siblings)	36 Mb	GAIIx	1 (1)	4 (2)	28 (28)	61 (54)	94 (85)
6	3	50.8 (39-71)*	3 (siblings)	36 Mb	GAIIx	3 (3)	3 (3)	17 (15)	52 (45)	75 (66)
7	4	66.0 (48-73)	3 (2 siblings, 1 first cousin)	36 Mb	GAIIx	1 (1)	1 (1)	2 (1)	8 (7)	12 (10)
8	6	64.2 (50-79)	3 (avuncular pair, first cousin)	36 Mb	GAIIx†	1 (0)	2 (1)	8 (6)	42 (22)	53 (29)
9	5	51.0 (40-68)*	3 (2 siblings, 1 first cousin)	36 Mb	GAIIx†	3 (1)	3 (1)	7 (4)	51 (27)	64 (33)
10	5	50.2 (28-66)	2 (avuncular pair)	36 Mb	GAIIx†	1 (0)	8 (6)	37 (33)	173 (115)	219 (154)
11	6	66.3 (54-81)*	2 (first cousins)	36 Mb	HiSeq 2000	4 (2)	6 (4)	30 (26)	120 (74)	160 (106)
12	3	49.7 (42-56)	2 (avuncular pair)	36 Mb	HiSeq 2000	2 (2)	9 (8)	59 (57)	169 (112)	239 (179)
13	5	61.4 (53-72)	2 (first cousins)	36 Mb	HiSeq 2000	2 (0)	4 (3)	19 (16)	125 (79)	150 (98)
14	3	49.7 (34-63)	2 (avuncular pair)	36 Mb	HiSeq 2000	9 (7)	7 (5)	28 (25)	161 (111)	205 (148)
15	5	57.4 (45-67)	2 (avuncular pair)	36 Mb	GAIIx, HiSeq 2000†	5 (3)	4 (2)	43 (35)	173 (140)	225 (180)
16	5	59.2 (56-67)	3 (siblings)	50 Mb	HiSeq 2000	5 (5)	8 (8)	53 (51)	291 (240)	357 (304)

^*Age of diagnosis unknown for one or two individual(s); these individuals were excluded from calculation

^†Samples in these families were sequenced twice and the results were pooled for analysis

Library Preparation, Target Capture, and Sequencing

Due to the rapid pace of technological advances during the course of our experiments, library capture and sequencing conditions varied by family (Table 1). Exome capture was completed using Agilent’s 36 Mb (n=37 individuals) or 50 Mb (n=3) All Human Exon chip. Libraries were sequenced once on Illumina’s GAIIx (n=19), twice on a GAIIx (n=8), once on a HiSeq 2000 (n=11) or once each on a GAIIx and HiSeq 2000 (n=2). All samples were run in a single lane of a flow cell; samples run twice were sequenced in separate flow cells and BAM files from separate runs were merged for analysis.

Libraries were prepared following manufacturers’ protocols (Illumina and Agilent). Briefly, 3 μg of genomic DNA was fragmented to 150-200 bp using the Covaris E210 sonicator. The ends were repaired, and an A base was added to the 3′ ends. Paired end DNA adaptors (Illumina) with a single T base overhang at the 3′ end were ligated and the resulting constructs were purified using AMPure SPRI beads from Agencourt. Adapter-modified DNA fragments were enriched by four cycles of PCR using PE 1.0 forward and PE 2.0 reverse primers (Illumina). Concentration and size distributions were determined on an Agilent Bioanalyzer DNA 1000 chip. Whole-exon capture used the protocol for Agilent’s SureSelect Human All Exon kit (36 Mb or 50 Mb). Five hundred ng of the prepared library was incubated with whole-exon biotinylated RNA capture baits supplied in the kit for 24 hours at 65°C. Captured DNA:RNA hybrids were recovered using Dynabeads MyOne Streptavidin T1 from Dynal and DNA was eluted from the beads and purified using Ampure XP (Beckman). Purified capture products were amplified using the SureSelect GA PCR primers (Agilent) for 12 cycles. Libraries were loaded onto paired-end flow cells at concentrations of 6 pM to 8 pM (GAIIx) or 4 pM to 5 pM (HiSeq 2000) to generate cluster densities of 250,000-350,000 per tile (GAIIx) or 300,000-500,000 per mm² (HiSeq 2000) following Illumina’s standard protocol using the Illumina cluster station and paired end cluster kit version 4 (GAIIx) or the Illumina cBot and HiSeq Paired-end cluster kit version 1 (HiSeq 2000).

Illumina GAIIx flow cells were sequenced as 101X2 paired-end indexed reads using SBS sequencing kit version 4 and SCS version 2.5 data collection software; base-calling used Illumina’s Pipeline version 1.5.1. Illumina HiSeq 2000 flow cells were sequenced as 101X2 paired-end reads using TruSeq SBS sequencing kit version 1 and HiSeq 2000 data collection version 1.1.37.0; base-calling used Illumina’s RTA version 1.7.45.0. Results from samples run in duplicate or triplicate were pooled for analysis.

Bioinformatics

Sequences were analyzed using TREAT (Targeted RE-sequencing and Annotation Tool) for sequence alignment, variant calling, functional prediction, and variant annotation (40). Reads were aligned to the human reference genome using BWA and duplicated read levels were evaluated using SAMtools’s rmdup method (41-43). The BWA alignment was improved using the Genome Analysis ToolKit (GATK) (44) local re-alignments. SNVs were called using SNVMix (45), and indels were called by GATK with default parameter settings. A 0.8 SNVMix posterior probability threshold was chosen for filtering based on analysis of a CEU sample sequenced by the 1000 Genomes Project (46). Variants located within the target regions were retained. SIFT (47) and SeattleSeq (http://gvs.gs.washington.edu/SeattleSeqAnnotation131) provided functional annotation. Read depths at each variant position and the average mapping quality score were generated by curating the BAM pile-up files using SAMtools (42). Potential splice variants were defined as those within two bp of exon-intron boundaries; eSplices were those within coding regions. We excluded reads and variants with poor mapping or quality scores (Qphred score <20 and probability score (0.8) and required a minimum number of high-quality reads supporting alternative alleles (10 for SNVs and three for indels). During read alignment, we identified several reads that aligned to off-target coding regions with high mapping scores and expanded the target region to 80 Mb to include high-quality reads in the Agilent capture definition.

Variant Filtering and Analysis

Shared variants (shared among all sequenced cases within each family) with MAF <0.05 (dbSNP Build 130) were identified and categorized as either a nonsense or splice SNV, missense SNV, other SNV (synonymous variants and non-coding), or a frame-shifting indel variant. We took two analysis approaches based on the following two questions: what novel genes harbor variants that may cause predisposition to CRC and what variants can be found in genes and regions previously associated with CRC? And what variants can be found in genes and regions previously associated with CRC? First, as an agnostic approach, we excluded variants not likely to be disease-causing, based on the following criteria: 1) shared in ≥4 pedigrees (likely representing artifacts or reference sequence annotation errors or newly identified common variants); 2) MAF ≥0.01 in CEU populations (HapMap, 1000Genomes, and Beijing Genome Institute); 3) annotation errors of nonsense and splice-site variants (variants incorrectly identified as nonsense or splice-site were correctly categorized then subjected to the standard exclusion criteria); 4) prediction of pathogenicity (missense and indel variants predicted to be benign by PolyPhen or tolerated by SIFT); 5) indel variants in splice sites that did not alter the splice site. Second, we examined variants in a priori candidate genomic regions using less stringent filtering criteria. All variants present that were shared among family members, even those with a low probability of causing disease, were included. These included: 1) 27 known CRC susceptibility genes (AKT1, APC, AXIN1, AXIN2, BLM, BMPR1A, BRCA1, BRCA2, CHEK2, GALNT12, MCC, MLH1, MLH3, MSH2, MSH3, MSH6, MUTYH, MYH11, PMS1, PMS2, PTEN, SMAD4, SMAD7, STK11, TGFB1, TGFBR2, and TP53); 2) previously identified linkage regions (3q21-24, 4q21, 7q31, 8q13, 8q23, 8q24, 9q22-31, 10p14, 11q23, 12q24, 15q22, and 18q21), and 3) GWAS regions (1q41, 3q26.2, 8q24, 9p24, 10p14, 11q23, 12q13, 14q22, 16q22, 18q21, 19q13, and 20p12) (9-27). In 15 families, we examined regions with family-specific dominant or recessive LOD scores > 1.3 found in multipoint linkage analysis using MERLIN version 1.1.2 (48) and genotype data from Affymetrix Linkage 2.0 or Illumina Linkage Panel 12 arrays, as described previously (12). Expected sharing was calculated as in described in Feng et al. and compared to actual sharing for each family.(49)

Technical Validation and Replication of Select Variants

Variants prioritized from the whole exome sequencing were validated using Sanger sequencing. Primers were designed using GRCh37/hg19 reference assembly for selected nonsense, splice-site, and missense variants identified in families 1, 2, 3, 8, 9, 12, and 16. Sequencing was performed on 16 individuals within the families that had been WES to validate the variants and on 31 available relatives with DNA to look at segregation of the variant with additional CRC- and polyp-affected and unaffected relatives. Briefly, 25 ng of leukocyte DNA was amplified in a 15 μl PCR containing 7.5 μl of GoTaq master mix (Promega, Madison, WI, USA) and 5 pmoles of each primer (available on request). Reactions were cycled on a Biorad iCycler (Biorad, Hercules, CA, USA) using the following profile: 94°C for 2 minutes, followed by 45 cycles of 94°C for 15 seconds, 60°C for 15 seconds and 72°C for 15 seconds, cycling was finalized at 72°C for 5 minutes. PCR reactions were subsequently cleaned up using Montage PCR96 Cleanup plates (Millipore, Bedford, MA, USA) according to the manufacturer’s guidelines. PCR product (0.5μl) was then used in an 8μl sequencing reaction comprising 0.4μl BigDye Terminator v3.1, 1.4μl 5x reaction buffer and 1.5pmoles of either primer. Reactions were cycled for 96°C for 1 minute, followed by 25 cycles of 96°C for 10 seconds, 50°C for 5 seconds and 60°C for 90 seconds. Prior to running on an ABI3100 genetic analyzer (Applied Biosystems, Foster City, CA, USA), sequencing reactions were cleaned up using Xterminator reagent (Applied Biosystems, Foster City, CA, USA) according to the manufacturer’s instructions. Resultant sequences were analyzed using SeqMan software (DNASTAR, Madison, WI, USA). Nonparametric linkage analyses used MERLIN version 1.1.2 and non-parametric Kong & Cox LOD (NPL) scores were computed for validated SNVs. (48)

RESULTS

Comparison of Exome Capture and Sequencing Platforms

We completed germline exome sequencing of forty cases from 16 familial CRC families (Table 1). Cases were selected to be distant relatives to decrease the number of shared, non-susceptibility variants. Sequencing technologies advanced rapidly during our work; both capture and sequencing technologies were updated, providing an opportunity for comparison across platforms. As expected, more variants were identified in samples captured with the 50 Mb chip than the 36 Mb chip. Most variant types were increased modestly, with three notable exceptions: intergenic indels increased by 21.6-fold and indels near the 3′ and 5′ ends of a gene were increased by 15-fold and 4.3-fold, respectively, when using the 50 Mb compared to the 36 Mb chip (Table S1). These increases likely represent the expanded target of the 50 Mb chip; similar increases are expected for future versions targeting more UTR regions. Samples run twice on a GAIIx, showed an approximately two-fold increase in the number of reads and variants identified compared to those run once (Figure S2a, Table 1). Samples sequenced twice on a GAIIx also increased the coverage similar to that of samples run once on a HiSeq 2000 (Figure S2b).

Agnostic Search for Novel Loci Identifies Candidate Genes

As described in the Methods, variant filtering was applied to the full whole-exome sequence dataset. We found that, on average, affected cases within families shared 33 variants (Table 1). There was great disparity in the number of shared variants by family and platform, with as few as four shared variants in Family 7 and up to 70 in Family 12. The majority of shared variants (75.8%) were synonymous or non-coding (intronic or intergenic). Missense variants represented 18.5% of all variants, while indels and nonsense or splice-site variants represented 3.6% and 2.1% of variants, respectively. On average, related cases shared approximately three nonsense or splice-site variants within a family, of which two were private.

Thirty-five nonsense or splice-site variants shared among affected family members were identified, each in a unique gene (Table 2). Although most of these variants were found in only one family, multiple families shared the one variant observed in each of SHROOM3, CDC27, ARSD, H2BFM, and TMC2.

Table 2.

Shared Nonsense and Splice Site SNVs

Chr	Position	rsID	Gene	DNA Change	AA Change	Families
1	85,546,961	-	WDR63	G/T	GLU>stop	6
	148,932,885	rs1048214	LOC645166	C/T	GLN>stop	5
	152,277,622	-	FLG	G/T	SER>stop	9
2	166,904,221	-	SCN1A	G/T	TYR>stop	16
3	40,231,748	-	MYRIP	A/T	LYS>stop	15
	75,787,516	-	ZNF717	C/G	Splice Site	4
4	77,660,829	rs73826426	SHROOM3	C/A	TYR>stop	8, 10
6	49,425,475	-	MUT	G/A	ARG>stop	6
	121,560,230	-	C6orf170	G/A	ARG>stop	2
	168,226,602	-	C6orf124	C/T	TRP>stop	7
7	76,751,068	rs71555938	CCDC146	G/C	TYR>stop	16
10	123,846,924	-	TACC2	C/T	GLN>stop	4
	135,490,903	rs36130162	DUX1	C/T	ARG>stop	16
11	1,857,515	-	SYT8	G/A	Splice Site	14
12	4,870,307	rs61758971	GALNT8	C/T	GLN>stop	14
	109,690,964	-	ACACB	G/A	Splice Site	14
13	24,243,246	-	TNFRSF19	C/T	ARG>stop	12
14	21,779,981	-	RPGRIP1	A/G	Splice Site	16
	58,832,019	rs62621193	ARID4A	G/A	Splice Site	14
15	75,562,499	-	GOLGA6C	C/T	GLN>stop	14
	78,807,407	-	AGPHD1	T/A	TYR>stop	14
16	24,873,990	-	SLC5A11	G/A	TRP>stop	2
	84,495,318	rs4782970	ATP2C2	A/C	Splice Site	15
17	45,234,277	-	CDC27	A/C	Splice Site	13, 14
18	14,513,786	rs8095431	POTEC	T/C	Splice Site	4
19	11,943,225	-	ZNF440	C/T	ARG>stop	12
	40,195,184	-	LGALS14	G/A	Splice Site	16
	43,699,204	-	PSG4	C/A	GLU>stop	4
	56,459,551	-	NLRP8	C/T	ARG>stop	6
20	2,597,716	-	TMC2	A/T	Splice Site	1, 9
	30,226,904	-	COX4I2	T/G	Splice Site	14
22	44,287,073	-	PNPLA5	G/A	GLN>stop	2
X	2,832,668	-	ARSD	A/G	Stop>GLN	9, 15
	55,185,663	-	FAM104B	T/G	Splice Site	15
	103,294,760	rs2301384	H2BFM	C/T	GLN>stop	11, 13, 14

There were 375 missense variants and 70 indels shared among affected family members after filtering. Three hundred fifty-eight of the missense SNVs (95%) and 62 of the indels (89%) were private, 10 missense SNVs and 8 indels were present in two families, and seven missense SNVs were found in three families (Table 3). Two genes had two variants each (CTBP2 and MUC6); the variants were shared between the same families. In both genes, the variants were <50 bp apart and likely due to an inherited haplotype in the families. Seventeen genes had >1 missense SNV, including six variants in CDC27 (Table S2). Private missense and indel variants are shown in Tables S3 and S4, respectively.

Table 3.

Missense and Indel Variants Shared in Multiple Families

Chr	Position	rsID	Gene	DNA Change	AA Change	Families
1	117,142,736	-	IGSF3	A/G	ILE>THR	1, 4, 12
	145,293,515	rs12565078	NBPF10	A/G	ASN>SER	11, 15
	148,023,040	-	NBPF14	G/C	SER>CYS	9, 7, 4
	154,171,908	-	C1orf189	TC/-	FS	9, 15
6	31,324,603	rs66519358	HLA-B	T/-	FS	9, 10
7	76,619,625	rs2302541	PMS2P11	C/T	ARG>CYS	12, 14, 15
	99,434,077	rs61469810	CYP3A43	A/-	FS	8, 11
10	118,215,310	-	PNLIPRP3	-/A	FS	4, 14
	126,673,560	-	ZRANB1	-/A	FS	5, 15
	126,678,112	-	CTBP2	T/C	ASN>SER	2, 15
	126,678,148	-	CTBP2	G/C	ALA>GLY	2, 15
11	1,017,337	-	MUC6	T/C	THR>ALA	4, 16
	1,017,338	-	MUC6	C/A	GLN>HIS	4, 16
	5,172,795	-	OR52A1	-/C	FS	5, 11
	71,529,890	-	FAM86C & DEFB108B *	A/T	ILE>LYS	6, 15
12	9,581,791	rs4763566	DDX12	T/C	LYS>GLU	11, 13, 15
14	19,378,312	rs61969158	OR11H12	T/G	VAL>GLY	11, 13, 14
16	85,132,883	-	FAM92B	A/C	PHE>VAL	6, 10
19	41,622,107	rs11399890	CYP2F1	-/C	FS	10, 14
	44,778,796	-	ZNF233	T/-	FS	12, 13
	58,385,748	-	ZNF814	G/A	ALA>VAL	9, 10
22	18,846,088	rs9605845	GGT3P & DGCR6 *	A/G	MET>THR	8, 13
X	2,832,715	rs73632953	ARSD	T/C	LYS>ARG	9, 15
	55,185,656	rs5003001	FAM104B	C/A	ARG>ILE	8, 10, 15
	68,725,640	rs1171942	FAM155B	T/C	LEU>PRO	10, 11, 14

^*Intergenic variants, identified are the two closest genes

Synonymous, intronic, and intergenic variants were the most abundant, with 1,394 shared among affected family members after filtering. Most of these were private (86%), while 191 were detected in ≥2 families. Over half of the variants were in genes without any other variant present (n=837); the remaining 557 variants were found in 152 genes (range 2-31 variants per gene). Summarizing across variant types, 46 genes had at least 4 variants (Table 4).

Table 4.

Genes with Multiple Variants Shared in Affected Family Members: Number of Variants

Gene(s)	Nonsense and Splice	Missense	Indel	Other	Total
ZNF717	1	-	-	30	31
ANKRD30BL	-	-	-	25	25
FRG1B, NCAPG2	-	-	-	18	18
CDC27	1	6	-	8	15
CTBP2	-	4	-	9	13
KIR2DS4	-	-	-	10	10
ROCK1P1, TTTY23/GYG2P1 *	-	-	-	9	9
HLA-DRB1	-	1	1	6	8
MST1P2, MUC12	-	-	-	8	8
ARSD	1	1	-	5	7
ACHE, BAGE/BAGE4, KCNJ12, MST1P9	-	-	-	7	7
FAM104B	1	2	-	4	7
BCL8, KIR3DL3, LOC642846, MUC3A, NBPF10,  RACGAP1P	-	-	-	6	6
AQP7P1, SIGLEC16, CROCCP2, FANK1,  KIR2DL,1 KRT16P2/TNFRSF13B*, KRTAP5-4	-	-	-	5	5
C6orf10	-	2	-	3	5
NBPF12	-	3	-	2	5
CFTR	-	3	-	1	4
ADAM6, HLA-DRB5, HLA-DRB6, HSP90AB2P,  MED12, NBPF1, NBPF9, PCDHB17, POLA1,  RPGR, TBC1D3P2, WASH2P	-	-	-	4	4

^*Intergenic variants, identified are the two closest genes

Pedigree structures of five families suggested recessive inheritance; these families were separately investigated to identify genes with homozygous variant alleles or compound heterozygosity (Table S5). In Family 2, five genes were identified with multiple variants. Three had only non-coding variants, while one (CTBP2) harbored two missense variants and a 5′ UTR variant and another gene (PDE4DIP) harbored two indels. In Family 3, three genes had multiple variants. One gene contained only non-coding variants; the remaining genes had a missense and a non-coding variant (PYROXD1) or a missense variant and an indel (PTPN9). In Family 6, three genes harbored multiple variants; however, all variants were non-coding. In Family 11, 19 genes had multiple variants. Fourteen of these genes had only non-coding variants, one had two indels (HLA-DQA1), two had missense variants (DDX12, MUC2), and the remaining two had a combination of variants (NBPF10, ZNF717). In Family 13, 11 genes harbored multiple variants; variants in ten of the genes were non-coding, while GGT3P had one missense and one non-coding variant.

Technical Validation of Select Variants

Thirty-one variants identified in seven families were selected for technical validation and segregation studies by Sanger sequencing. Additional variants in the families were not tested due to the presence of homologous sequences, because the variant had been identified as a common sequencing error, or because the gene was hypervariable. Of the 31 variants tested, 27 were validated in the previously exome sequenced individuals and four were found to be a false positives, including two nonsense variants (SCN1A and SHROOM3) and two indels (B3GNT6 and RBMX) (Table 5).

Table 5.

Validation of Identified Variants

GRCh37 Chr:Position	Gene Name	dbSNP130	Variant Type	Family	Technical validation of exome sequenced CRC- affected carriers
20:2,597,716 4:100,130,075 4:104,030,143 19:48,735,017 4:57,204,689	TMC2 ADH6 CENPE CARD8 AASDH	- rs149932401 - rs146319637 -	Splice-site Missense Missense Frameshift Frameshift	1	2/2 2/2 2/2 2/2 2/2
6:121,560,230 16:24,873,990 22:44,287,073 3:187,003,786 3:186,331,094 12:70,088,219 11:64,543,927	C6orf170 SLC5A11 PNPLA5 MASP1 AHSG BEST3 SF1	- - - - - - rs34514973	Nonsense Nonsense Nonsense Missense Missense Frameshift Frameshift	2	3/3 3/3 3/3 3/3 3/3 3/3 3/3
19:55,327,891 15:75,798,025	KIR3DL1 PTPN9	rs71367103 -	Upstream Frameshift	3	3/3 3/3
4:77,660,829	SHROOM3	rs73826426	Nonsense	8, 9	0/3
13:24,243,246 19:11,943,225 19:44,778,796 16:336,700 2:196,661,361 7:128,587,351 7:15,601,409 20:34,215,234	TNFRSF19 ZNF440 ZNF233 PDIA2 DNAH7 IRF5 AGMO CPNE1	- - - rs201624048 - rs60344245 - rs76294482	Nonsense Nonsense Frameshift Frameshift Frameshift Deletion Frameshift Frameshift	12	2/2 2/2 2/2 2/2 2/2 2/2 2/2 2/2
19:40,195,184 2:166,904,221 14:21,779,981 15:69,732,770 3:178,960,766 9:43,844,264 11:76751,603 X:135,960,146	LGALS14 SCN1A * RPGRIP1 KIF23 KCNMB3 CNTNAP3B B3GNT6 * RBMX *	- - - - rs143962239 - - -	Nonsense Nonsense Splice-site Missense Frameshift Frameshift Frameshift Frameshift	16	3/3 0/3 3/3 3/3 3/3 3/3 0/3 0/3

^*Variants in bold were not validated and considered false positives.

Segregation Analysis of Validated Variants

For all variants validated, additional affected and non-affected family members, and family members with polyps were Sanger sequenced to determine co-segregation (Table 6). Only one variant (PTPN9) was not replicated in any of the new samples; others were replicated in one to six additional family members. Several of the variants appeared to segregate with affection status, such as TMC2, ADH6, CENPE, AASDH, C6orf170, AHSG, SF1, RPGRIP1, and KIF23. Particularly interesting are the variants in CENPE and KIF2. Both are very rare; the KIF23 variant is seen only once in the ESP database of European Americans, while the variant in CENPE is not present in any public database. Non-parametric LOD scores were calculated for validated SNVs. The maximum possible NPL score was <2.5 for all families (Table S6). No variants had an observed NPL LOD score >1, possibly due to the few individuals and families with available data for analysis.

Table 6.

Replication and Segregation of Validated Variants

					Segregation
GRCh37 Chr:Position	Gene Name	dbSNP130	Variant Type	Family	Additional CRC-affected carriers	unaffected carriers	Polyp- affected carriers
20:2,597,716	TMC2	-	Splice-site		2/3*	0/1	3/4
4:100,130,075	ADH6	rs149932401	Missense		3/3*	0/1	3/4
4:104,030,143	CENPE	-	Missense	1	3/3*	0/1	3/4†
19:48,735,017	CARD8	rs146319637	Frameshift		2/3*	1/1¥	2/4
4:57,204,689	AASDH	-	Frameshift		2/3*	0/1	4/4
6:121,560,230	C6orf170	-	Nonsense		0/0	0/2	4/7
16:24,873,990	SLC5A11	-	Nonsense		1/1*	1/2	5/7
22:44,287,073	PNPLA5	-	Nonsense		1/1*	1/2	3/7
3:187,003,786	MASP1	-	Missense	2	0/0	1/2	2/7
3:186,331,094	AHSG	-	Missense		1/1*	1/2	3/7
12:70,088,219	BEST3	-	Frameshift		1/1*	0/2	5/7
11:64,543,927	SF1	rs34514973	Frameshift		1/1*	1/2	4/7
19:55,327,891	KIR3DL1	rs71367103	Upstream	3	2/2	1/1	n/a
15:75,798,025	PTPN9	-	Frameshift		0/2	0/1	n/a
13:24,243,246	TNFRSF19	-	Nonsense		1/1	1/4	1/1
19:11,943,225	ZNF440	-	Nonsense		1/1	0/4	0/1
19:44,778,796	ZNF233	-	Frameshift		1/1†	3/4	1/1
16:336,700	PDIA2	rs201624048	Frameshift	12	1/1	1/4	0/1
2:196,661,361	DNAH7	-	Frameshift		1/1	1/4	1/1
7:128,587,351	IRF5	rs60344245	Deletion		1/1	4/4	1/1
7:15,601,409	AGMO	-	Frameshift		1/1	1/4	0/1
20:34,215,234	CPNE1	rs76294482	Frameshift		0/1	3/4	0/1
19:40,195,184	LGALS14	-	Nonsense		0/1	1/3	n/a
14:21,779,981	RPGRIP1	-	Splice-site		1/1	0/3	n/a
15:69,732,770	KIF23	-	Missense	16	1/1	0/3	n/a
3:178,960,766	KCNMB3	rs143962239	Frameshift		0/1	2/3	n/a
9:43,844,264	CNTNAP3B	-	Frameshift		1/1	3/3	n/a

^*Includes one obligate carrier

^†at least 1 individual is homozygous for the variant

^¥has stomach cancer

All individuals with available DNA in each family (excluding the original WES samples) were tested for each variant (Family 1, n=8; Family 2, n=10; Family 3, n=3; Family 12, n=6; Family 16, n=5). Only results of successful sequencing are reported in the table.

Search of the Known Susceptibility Genes and Regions also Identifies CENPE and KIF23

In addition to the agnostic search for novel loci, we investigated 27 known or suspected high-risk and familial CRC genes and several candidate regions. We required that all affected family members shared the variants as in our previous analyses. However, we did not exclude any variants beyond that, as the genes and regions we were targeting are well documented risk regions and we didn’t want to overlook any potential candidate variants. Our selection of non-MMR families was effective - no shared variants were observed in MLH1, MLH3, MSH2, MSH3, MSH6, or PMS1. Two SNPs in PMS2 were identified in Families 5, 6, and 9 (Table S7). However, both SNPs were common and expected to be tolerated. BRCA2 had missense (n=3) and synonymous or intronic (n=3) variants with a MAF between 1 - 5%; five of the SNVs were found only in Family 12, increasing the likelihood that the region containing the variants was inherited as a haplotype block. MCC harbored two SNVs, a missense variant resulting in a glycine-to-arginine substitution in Family 13, and a non-coding variant found in five families. BRCA1 harbored a GLN to ARG substitution (rs1799950) predicted to be damaging; the same rare allele has been associated with a decreased risk of developing breast cancer, however it has not been described in colon cancer previously (50). APC, AXIN2, GALNT12, MYH11, and TP53 each had one non-coding variant. No SNVs or indels were shared among affected family members in the remaining 12 HCC genes (Table S7).

Previously, we reported four regions linked to CRC with HLOD scores greater than 3.0 in 356 families, including 15 of the currently studied families (12). Two regions, 4q21.1 and 15q22.31, harbored variants (Table S8). The 4q21.1 region contained five shared variants, including the SHROOM3 nonsense variant, which found to be a false positive, and four non-coding variants. The 15q22.31 region contained two missense variants (CGNL1 and KIF23) and five non-coding variants. No variants were found in the other linkage regions examined. Family-specific linkage analysis yielded LOD scores >1.3 in 10 regions in five of the families sequenced (Table S9). None of the regions contained a gene with a shared nonsense or splice variants and six of the regions harbored only non-coding variants. The linkage peak on chromosome 4 harbored 2 missense SNVs in Family 1, one each in ADH6 and CENPE. In Family 2, two linkage peaks on chromosomes 1 and 3 harbored missense variants in WDR47, AHSG, and MASP1.The linkage peaks in Family 5 contained two missense variants (CFTR and ZC3HC1) and two non-coding variants. In addition, although it is expected that variants responsible for disease in densely affected families differ from modest penetrance variants, we investigated SNVs within the +/-500 Kb regions surrounding SNPs shown to be associated in GWAS with CRC risk (14, 15, 19, 21-27). One indel variant was found in TPD52L3 within the 9p24 region in Family 5; however, this family does not carry the identified risk allele at rs719725. Twenty-one missense and 45 synonymous or non-coding variants were also identified in the GWAS regions, however, many were common (MAF>5%) and not likely to contribute to CRC genesis (Table S10). Thus, we were able to identify two variants of interest (CENPE and KIF23) in the regions previously implicated in CRC risk that were also identified by our earlier agnostic search. These two variants were validated and replicated in the affected families, strengthening the results.

DISCUSSION

This analysis of whole-exome sequence data in 16 high-risk CRC families demonstrates the utility of massively-parallel exome sequencing to identify novel candidate genes for complex diseases. We have enumerated potential novel variants as well as those in prior candidate genes and regions. After excluding variants not shared among affected family members, common variants (MAF≥0.01), and those expected to be benign or tolerated, several remained, including protein-truncating mutations in genes involved in: cell shape and motility (ZRANB1); mitosis (CDC27, CENPE, DDX12, HAUS6/FAM29A, HIST1H2BE, KIF23, TACC2, and ZC3HC1); transcription regulation (CTBP2, IRF5, MED12, RNF111, SF1, TLE1, TLE4, TRIP4); and the immune response (BTNL2, BAGE, CARD8, FANK1, KIR2DL1, KIR2DS4, KIR3DL3, MASP1, and NLRP8) (51-77), as well as numerous missense and indel variants.

It is likely that some of the identified genes are causal. We divided the variants into three categories, based on the likelihood of causing a loss of protein or protein function: the most likely to be causal (nonsense and splice site), those with an elevated risk of being deleterious (missense) and those with the lowest likelihood of being damaging (synonymous and non-coding). Given the lengthy list of candidate genes, the possibility of false positive results, and the paucity of functional information, additional targeted sequencing studies in a large set of independent cases and controls is warranted. Targeted sequencing of novel candidate genes (e.g., those with nonsense or splice site variants) in upwards of 1,000 familial CRC cases would be an informative next step.

This study represents only one point in the journey to identifying genetic predisposition to CRC. CRC is highly heterogeneous and polygenic; unlike the very distinctive Mendelian diseases for which whole-exome sequencing has been successful. Studies directed at identifying candidate susceptibility genes for familial CRC are not readily yielding causal variants (78-81). We debated family selection strategies. Without defined criteria about the optimal selection methods, identification of families and individuals best suited for exome sequencing proved more challenging than expected. We based selection on what we believed would maximize the chance of including families with a high risk genetic predisposition based upon widely held tenets: multiple, closely related affected relatives and younger ages at diagnosis (49, 82).To investigate this further, we compared the observed proportion of shared variants to the expected proportion of shared variants. Significant increases in nonsense, missense, and indel variants (Table S11) strengthened our belief that the methods for choosing families was suitable.

For our families, we chose to sequence two individuals when distantly related cases were available and three individuals when only closely related individuals (siblings) were available, following the recommendations of Feng and colleagues for studying complex diseases by sequencing (49). Model-based approaches, such as estimation of expected lod scores, could also have been used (83). Missing information on earlier generations meant that sequenced samples may be connected through unaffected relatives (in one avuncular pair, additional data became available confirming this to be the case), demonstrating the challenges of incomplete penetrance and phenocopies in studying cancer and complex traits. Sequencing of non-affected family members to help distinguish between causal and benign variants was also discussed; however, the penetrance for CRC in families that met Amsterdam criteria (84), but do not have MMR defects (Type X), is lower than for Lynch Syndrome (85). This makes sequencing unaffected relatives less useful, compared to disorders with complete or very high penetrance.

Our study has weaknesses. First, it involves only a small number of families, chosen to include those with no evidence of Lynch syndrome, MUTYH mutations, or Familial Adenomatous Polyposis. Several other candidate families were considered, however, funding was only available for a small number of families. As sequencing costs decline, combining with other collections will be more feasible and needed to identify additional genes of interest. Second, for each family, only a limited number of affected individuals were available for sequencing. The relationships of those selected were preferentially chosen to be cousins or avuncular. However, for several families, the only individuals with DNA available were siblings, reducing power to detect causal variants. Third, we had a false discovery rate of ~13%, which is higher than expected based on previous studies. This may in be due to the fact that rare variants, such as the ones we choose to validate, have a lower rate of validation than more common variants (86), highlighting the critical need for Sanger validation. It is interesting to note that the three false positive variants identified were all in the same family (Family 12), which was the only one utilizing the 50 Mb capture system. Multiple factors may have contributed to the false positives identified in this family, including degraded DNA for the individuals tested, increased target size, resulting in localized areas of decreased coverage, or misalignment due to poor probe design. Fourth, samples were not all subject to the same capture or sequencing conditions, resulting in increased coverage for samples sequenced toward the study’s end. It is possible that some variants detected in later families were present in the earlier families, but went undetected, skewing our perceptions of the allele frequencies. Differences in capture and sequencing technologies also likely affect public databases; numerous variants identified had little available frequency information. Lastly, the most appropriate method to filter for causal variants in complex diseases is unknown. We first narrowed the number of variants by filtering those not shared among the affected family members, which may have excluded causal variants that don’t perfectly co-segregate with disease. We used several strategies, including examining candidate genes and regions, looking for genes with multiple variants, and agnostic searching for novel loci. We restricted our search to rare variants, hypothesizing that genes important for the development of CRC will harbor several private variants

In summary, we have completed exome sequencing of 40 familial CRC cases from 16 families and identified and technically validated several candidate CRC variants. Follow-up studies to determine the frequency of variants in many of the identified genes are currently underway. Further sequencing and functional studies will be needed to confirm the identified genes and determine their role in the genesis of CRC.

Supplementary Material

1

Figure S1. Pedigrees of Exome Sequenced Families. Numbers under selected symbols reflect the age of diagnosis for cases, age when polyps were discovered for individuals with polyps, and the last known age of contact for unaffected individuals. Full shaded symbols indicated affected individuals; half shaded symbols represent individuals with polyps. Gray circles indicate that the individual was included in validation and replication studies. Cases with whole-exome sequencing data are denoted with ‘WES’. ‘MSS’ indicates individuals with microsatellite stable tumors. Loss of MMR protein is indicated by the name of the gene. Gender was made anonymous to protect patient confidentiality.

2

Figure S2. Mapping and Coverage of the Target Region. A) Mean number of reads, reads mapped, and reads mapped on-target, in millions. Black bars, GAIIx, sequenced once (n=21). Hatched bars, GAIIx, sequenced twice (n=8). White bars, HiSeq 2000, 36 Mb capture kit (n=8). Gray bars, HiSeq 2000, 50 Mb capture kit (n=3). For additional capture and sequencing details, please refer to the methods. B) Percent of probes in the capture region with 10x, 20x, 30x, and 40x coverage. Diamonds, GAIIx, sequenced once (n=21). Squares, GAIIx, sequenced twice (n=8). Triangles, HiSeq 2000, 36 Mb capture kit (n=8). Circles, HiSeq 2000, 50 Mb capture kit (n=3). For additional capture and sequencing details, please refer to the methods.