Pathogenic and likely pathogenic variants in PALB2, CHEK2 and other known breast cancer susceptibility genes among 1054 BRCA-negative Hispanics with breast cancer

Abstract

Background:

Breast cancer (BC) is the most common cancer and related cause of mortality among Hispanics, yet susceptibility has been under-studied. BRCA1 and BRCA2 (BRCA) mutations explain less than half of hereditary BC, and the proportion associated with other BC susceptibility genes is unknown.

Methods:

Germline DNA from 1054 BRCA-mutation-negative Hispanic women with hereditary BC (BC diagnosed age < 51 years, bilateral BC, breast and ovarian cancer, or BC 51–70 years with ≥ 2 first- or second-degree relatives with BC diagnosed age < 70 years), 312 local and 887 Multiethnic Cohort controls, was sequenced and analyzed for 12 known and suspected high- and moderate-penetrance cancer susceptibility genes (ATM, BRIP1, CDH1, CHEK2, NBN, NF1, PALB2, PTEN, RAD51C, RAD51D, STK11, TP53).

Results:

We identified 49 (4.6%) pathogenic or likely-pathogenic variants (PV) in 47 of 1054 (4.5%) participants, including 21 truncating frameshift, 20 missense, 5 nonsense and 4 splice variants among: CHEK2 (n=20); PALB2 (n=18); ATM (n=5); TP53 (n=3); BRIP1 (n=2); and CDH1and NF1 (all n=1), and none in NBN, PTEN, STK11, RAD51C, or RAD51D. Nine participants carried the PALB2 c.2167_2168del PV (0.85%) and 14 carried the CHEK2 c.707T>C PV (1.32%).

Conclusions:

Of the 1054 BRCA-negative high-risk Hispanic women, 4.5% carried a PV in a cancer susceptibility gene, increasing our understanding of hereditary BC in this population. Recurrent PVs in PALB2 and CHEK2 represented 47% (23/49) of the total, suggesting founder effect. Accurate classification of variants was enabled by carefully controlling for ancestry, and increased identification of at-risk Hispanics for screening and prevention.

Precis:

Breast cancer (BC) is the most common cancer and related cause of mortality among Hispanics, yet BC susceptibility has been under-studied. We used whole exome sequencing and identified 50 (4.7%) pathogenic or likely pathogenic variants (PV) in 47 of 1054 (4.5%) BRCA-negative Hispanics with familial BC, including recurrent PVs in PALB2 and in CHEK2, where the classification was enabled by carefully controlling for ancestry.

INTRODUCTION

Hispanic Americans are one of the fastest growing racial/ethnic groups in the United States, projected to make up 29% of the total population by 2060.¹ One in twelve Hispanic women will be diagnosed with breast cancer (BC) and it is more likely to be at a later, less curable stage than in non-Hispanic women.^2,3 Yet, there is a dearth of Hispanic-specific healthcare research, particularly in the area of genomic predisposition to BC. Despite clinical studies documenting the efficacy of cancer risk reduction measures in high-risk individuals, low-income, underinsured or ethnic minority individuals have a disproportionate burden of cancer and limited access to genetic cancer risk assessment (GCRA).^4,5

Family history of breast cancer is a consistent risk factor for breast cancer across different racial/ethnic populations. BRCA1 and BRCA2 (BRCA) were identified in the 1990s and mutations in these genes lead to autosomal dominant inheritance of breast and ovarian cancer susceptibility.^6–8 BRCA studies in U.S. Hispanics and Mexican, Colombian and Peruvian populations indicate a relatively high rate of recurrent pathogenic and likely pathogenic variants (PV).^9–15 However, PV alleles in BRCA account for less than 50% of hereditary BC and approximately 70% of identified ovarian cancer susceptibility.^16–18

Since the discovery of BRCA, other genes have been identified to be associated with breast cancer susceptibility.¹⁹ Multi-gene panels are increasingly being used in the clinical setting; these include genes associated with high (e.g. BRCA1) and moderate (e.g. CHEK2) breast cancer risk. Improved understanding of the spectrum of genetic susceptibility and clarification of gene-specific risks for breast cancer may result in enhanced screening, prevention, and therapeutic strategies for these patients and their families.^20,21 Given that racial/ethnic minority patients are underrepresented in many cohorts, it is particularly important to define risks in diverse populations. We analyzed data for 12 known and suspected high- and moderate-penetrance BC susceptibility genes in a large cohort of Hispanics with familial breast cancer.

METHODS

Study Population

We included women with breast cancer who previously had tested negative for carrying a BRCA1 or BRCA2 (BRCA) mutation through clinical gene testing or through a HISPANEL of 114 recurrent Hispanic pathogenic mutations ¹⁰. For the BRCA-negative women, further inclusion criteria were: age less than 51 years at BC diagnosis, bilateral breast cancer, breast and ovarian cancer, OR age at diagnosis between ages 51 and 70 with a family history of BC in 2 or more first or second-degree relatives diagnosed at less than 70 years of age. These participants were selected from self-identified Hispanic participants from 3 high-risk registry studies including the City of Hope (COH) Clinical Cancer Genomics Community Research Network (CCGCRN)^12,22, the University of California San Francisco (UCSF) Clinical Genetics and Prevention Program and the University of Southern California (USC) Norris Comprehensive Cancer Center clinical genetics program. All individuals were consented and enrolled into the study through center-specific Institutional Review Board approved protocols. We performed whole exome sequencing (WES) on DNA from 1078 Hispanics with familial BC who met inclusion criteria and 312 Hispanic controls from Southern California.

We also included WES data from 887 participants from the Multiethnic Cohort (MEC) without breast cancer (approximately half had diabetes) who were self-described Hispanics and had undergone WES at the Broad Sequencing Center. These controls are a subset of ExAC Hispanic samples.

After sequencing, we excluded 9 breast cancer cases and 5 controls with less than 20-fold average coverage. We also searched for duplicates and relative pairs using PLINK and removed 2 cases and 2 controls identified as duplicates and 5 cases and 3 controls identified as first-degree relatives. Lastly, we excluded 8 cases with previously undetected BRCA PVs (negative on HISPANEL). In total, we had 1054 cases and 1189 controls

Library Construction, Hybridization, and Massively Parallel Sequencing

We used KAPA Hyper Preparation Kits (Kapa Biosystems, Inc., Wilmington, MA) to generate libraries from 500ng DNA. Each library was assigned a 6-digit DNA barcode sequence and linked to a unique subject identifier. Eight libraries were pooled and hybridized to the SureSelect Clinical Research Exome (Agilent, Santa Clara, CA) kit to capture the exons of all known human transcripts. One-hundred base-pair paired-end sequencing on the HiSEQ 2500 Genetic Analyzer (Illumina Inc., San Diego, CA) was performed in the COH Integrative Genomics Core (IGC) to an average fold coverage of 65X. Paired-end reads from each sample were aligned to human reference genome (hg37) using the Burrows-Wheeler Alignment Tool (BWA, v0.7.5a-r405) under default settings, and the aligned binary format sequence (BAM) files were sorted and indexed using SAMtools.^23,24 The same FASTA reference file had been used for aligning the MEC control samples. The sorted and indexed BAMs were processed by Picard MarkDuplicates (http://broadinstitute.github.io/picard/) to remove duplicate sequencing reads. Variant calling from the BAM files from the IGC and the Broad were processed together at UCSF. Following local realignment of reads around in-frame insertions and deletions (indels) and base quality score recalibration by The Genome Analysis Toolkit (GATK), GATK HaplotypeCaller was utilized to call variants (https://software.broadinstitute.org/gatk). Variants were considered high confidence if coverage at the site was >10 reads and, for heterozygous calls, if the alternate allele was seen ≥4 times. DNA from eight MEC participants were sequenced at both COH and Broad with >99.8% concordance for variant calling.

Selection of genes for analysis

We selected 12 known and suspected high- and moderate-penetrance breast and ovarian cancer susceptibility genes (ATM, BRIP1, CDH1, CHEK2, NBN, NF1, PALB2, PTEN, RAD51C, RAD51D, STK11, and TP53) for this study, based in part on inclusion on clinical multi-gene panels and as actionable genes in the NCCN guidelines.²⁵

Quality Filtering

Samples with average coverage less than 20-fold across the target region were filtered out (N=9). Variants with a call quality less than 20, read depth less than 10, less frequent allele depth of less than 4 or allele fraction ratio less than 30% were filtered out for low quality.

Variant Filtering

Variants with a frequency greater than 2% in the 1000 Genomes Project, National Heart, Lung, and Blood Institute (NHLBI), Exome Sequencing Project (ESP) exome, or ExAC databases, were removed. Variant call format files were evaluated using Ingenuity Variant Analysis (IVA) version 4 (Qiagen Inc, Alameda, CA). IVA used the following content versions: Ingenuity Knowledge Base (Hogwarts 160211.000), 1000 Genome Frequency (v5b)²⁶, Exome Variant Server (ESP6500SI-V2)²⁷, Exome Aggregation Consortium²⁸ data set (ExAc, release 0.3), PhyloP^29,30, Sorting Intolerant from Tolerant (SIFT)³¹, the Human Gene Mutation Database (HGMD, 2015.4), COSMIC (v75)³², and Clinvar.³³ American College of Medical Genetics and Genomics (ACMGG) guidelines were applied to the variants using the IVA ACMGG calling algorithm.³⁴ IVA categorizes variants based on standard ACMGG variant calling recommendations in addition to running in silico models as described above. All ACMGG-called pathogenic or likely pathogenic variants, as well as the remaining frameshift variants, stop codon changes, or variants that disrupt a splice site up to two bases into the intron, were individually evaluated by the research team using the available literature and ClinVar to make a final call.³³

Ancestry Estimation

The Clinical Research Exome included a custom panel of 180 ancestry informative single nucleotide polymorphisms (SNPs). Based on a previous publication³⁵, these markers were selected to be informative for ancestry in a mixed European, Native American and African population. In addition, we selected 7691 variants common to our WES data and a dataset of Axiom arrays including African (N=90), European (N=90) and Native American (N=51) populations. We selected unlinked markers by linkage disequilibrium pruning in PLINK, identifying a subset of 4544 variants for ancestry estimation. We estimated genetic ancestry using ADMIXTURE, and performed analyses with both supervised (specifying the ancestral populations) and unsupervised (including the data from ancestral populations, but not specifying the identity of ancestral populations).

To determine genetic ancestry in the ExAC data, we used the same ancestral reference samples and selected a subset of independent variants (12,758) that overlapped between the Axiom arrays and the ExAC dataset. We then entered the allele frequency of these markers into a likelihood function assuming that the ancestry is a 3-component model (African, European and Native American ancestry). Assuming the markers are independent, the allele frequency for all of the variants in ExAC is a multinomial likelihood function that can be maximized for African, European and Native-American components.

Association Analysis

We performed two types of association tests. First, we performed association analyses using ExAC controls, excluding TCGA samples, and selected the subset of cases that matched the global ancestry of the ExAC Hispanics. The ExAC samples had been selected to eliminate those with lowest Native American ancestry; therefore, we removed cases with the highest European ancestry until the mean European and Native American ancestry were <1% different than ExAC. The average genetic ancestry in the 547 remaining cases was 57.7% Native American, 7.0% African and 35.3% European. The average genetic ancestry in the in ExAC was 58.7% Native American, 5.4% African and 35.9% European. We then used Fisher’s exact test to compare the allele frequency in cases to ExAC controls. We only performed the comparisons for PVs observed three or more times in cases and at least once in the ExAC dataset. We calculated odds ratios, under the assumption that each of the carriers of the variants we tested in the ExAC dataset is heterozygous, a reasonable assumption for variants with frequency <0.5% in the population.

To validate our results, we also performed an analysis using individual data on the 1054 cases and 1189 controls for whom we had individual sequence data. This analysis used the jointly called genotype results so that the informatics pipeline was identical. To adjust these analyses for genetic ancestry, we added genetic ancestry as a covariate into logistic regression models rather than matching. To calculate 95% CI’s and p-values for these low frequency variants, we used bootstrap sampling, sampling 10,000 times. All analyses were conducted in R.

Results

The demographic and clinical features of the study population are summarized in Table 1. Among 1054 high-risk BRCA-negative Hispanics, 49 (4.6%) PVs were identified in 7 of the 12 known or suspected breast cancer susceptibility genes (n) including: CHEK2 (20), PALB2 (18), ATM (5), BRIP1 (2), TP53 (3), and CDH1, and NF1 (all=1) (Table 2). No PVs were observed in NBN, PTEN, RAD51C, RAD51D or STK11. In Supplemental Table 1, the distribution of the 25 unique PVs identified in the cases by type of mutation are shown. In Supplemental Table 2, the individual PVs found in both cases and controls are shown with the ACMGG criteria ³⁶ for the call as well as information on age at diagnosis, family history (yes/no), and ER/PR/HER2 status if available. In Supplemental Table 3, all frameshift, splice, nonsense, and non-synonymous variants observed in cases and controls and the ANNOVAR output are shown. Of 405 variants, 143 (35%) were not noted previously in ClinVar.

Table 1.

Characteristics of the 1054 Hispanic breast cancer cases

Personal History of Breast Cancer (BC)	Median age in years	Range
Age at DX 1st BC	42	18–70
Age at DX 2nd BC	49	28–74
	No.	%
No. w/2nd BC	98	9.3
No. w/ovarian cancer (OC)	12	1.1
No. w/other cancer	54	5.1
Family History of BC/OC	No.	% of 933
None	574	61.5
1 FD/SD w/BC	237	25.4
2 FD/SD w/BC	92	9.9
3+ FD/SD w/BC	30	3.2
OC	94	10.1
Unknown family history	121
Tumor ER status	No.	% of 840
Positive	597	71.1
Negative	235	28.0
Indeterminate/borderline	8	1.0
Unknown	214
Tumor PR status	No.	% of 819
Positive	504	61.5
Negative	305	37.2
Indeterminate	10	1.2
Unknown	235
Tumor HER2 status	No.	% of 584
Positive	147	25.2
Negative	428	73.3
Indeterminate/inconclusive	9	1.5
Unknown	470

FD/SD = first degree/second degree relative; PR = progesterone receptor; ER = estrogen receptor; HER2 = Her2/neu amplication

Table 2.

Number of women with pathogenic and likely pathogenic variants in BC susceptibility genes in 1054 Hispanic breast cancer cases

Gene	frameshift	nonsense	missense	splicing	Total variants
ATM	3*	1		1^	5
BRIP1	2*				2
CDH1				1	1
CHEK2	1	1	16	1**	20
NF1	1				1
PALB2	14^	3		1	18
TP53			3		3
	21	5	20	4	49 (4.6%)

^*One individual had frameshift variants in both ATM and BRIP1

^{^}One individual had a PALB2 frameshift variant and an ATM splicing variant

PVs were detected in 47 (4.4%) breast cancer cases; 3 of 47 carried 2 PVs (Table 2). The ages at diagnosis and tumor characteristics were not significantly different between the 47 participants with PVs compared to the 1007 without any PVs; median age at diagnosis was 42 years (range from 18 to 70 years) overall and 43 years (age range 26–68) for women identified to carry a PV. A family history of breast cancer was reported for 51.1% of the PV carriers compared to 37.8% of non-carriers (p = 0.074).

Of the 49 PVs identified in 47 cancer cases (Table 2), 25 were distinct (Supplemental Tables 1 and 2). Among the 25 distinct PVs, 11 were frameshift leading to truncation, 6 were missense, 4 were nonsense and 4 were splice site. Among controls, there were a total of 18 PVs of which 8 (0.44) were in ATM and 4 were the recurrent mismatch variant (see Supplemental Table 2). Three of the 25 PVs occurred in 3 or more individuals (Table 3) and included 2 recurrent frameshift variants in PALB2 and 1 recurrent missense variant in CHEK2. These three variants were significantly more commonly observed in the breast cancer cases than in the ExAC Hispanic controls (Table 3). The recurrent PALB2 c.2167_2168del: p.M723fs was detected in 9 individuals. In ancestry matched comparison to ExAC controls the variant was associated with an odds ratio (OR) of 12.9 (95% CI: 3.5 – 51.2) (Table 3). In analyses of individual data from cases and controls (Table 3), we also observed a highly significant association with this variant (p<0.0001). Carriers of this variant were significantly more likely to have been diagnosed with ovarian cancer compared to cases not carrying this variant (Supplemental Table 4; p=0.004). PALB2 c.2411_2412del: p. S804fs, detected in three individuals, was associated with an OR of 27.5 (95% CI: 2.1 – 1431.2) (Table 3). If both recurrent PALB2 mutations are combined, the OR for breast cancer was 13.9 (95% CI: 4.4 – 47.7). The CHEK2 PV (c.707T>C: p.L236P) was detected in 14 cases. In an ancestry-matched comparison to the ExAC database values for Hispanics, the OR for BC risk was 3.2 (95% CI: 1.5–6.5; p=0.002) (Table 3). In gnomAD, the allele frequency of L236P is consistent ~0.002 in Latinos and 0 in all other populations consistent with a founder variant, likely of Native American ancestry.

Table 3.

Frequency of pathogenic and likely pathogenic variants observed in three or more individuals compared to ExAC and to our own sequenced controls.

Analysis using ExAC Controls
Gene	Variant	Cases with variant/ total ancestry matched cases (%)#	ExAC controls with variant/ total controls in ExAC (%)&$	OR (95% CI)	P Value
CHEK2	c.707T>C: pL236P	12/ 612 (1.96)	35 / 5603 (0.63)	3.2 (1.5–6.5)	0.0016
PALB2	c.2167_2168del: p.M723fs	9 / 612 (1.14)	5 / 5608(0.09)	12.9 (3.5–51.2)	0.00005
PALB2	c.2411_2412del: p. S804fs	3 / 612 (0.49)	1 / 5601 (0.02)	27.5 (2.1–1431.2)	0.0035
Analysis using individually sequenced controls from City of Hope and the Multiethnic Cohort
Gene	Variant	Cases with variant / total (%)*	Controls with variant / total (%)	OR (95% CI)	P Value
CHEK2	c.707T>C: pL236P	14 / 1045 (1.34)	4 / 1189(0.34)	4.1 (1.5 – 22.0)	0.039
PALB2	c.2167_2168del: p.M723fs	9 / 1045 (0.86)	0 / 1189 (0)		<0.0001
PALB2	c.2411_2412del: p. S804fs	3 /1045 (0.29)	1 / 1189 (0.08)	3.7 (0.0 - >100.0)	1.0

^#Number of cases has been reduced in order to match the ancestry of cases with ExAC controls.

^&The carrier frequency was calculated based on the assumption that all of the alleles for these variants in ExAC are found in heterozygous carriers.

^$This variant is at genome position (22:29107982A/G); Build37 and corresponds to rs587782471. In ExAC and gnomAD it is annotated as Leu279Pro.

^*Only 1045 of the 1054 cases were included in the joint calling.

Analysis of individual data in cases and controls, adjusting for ancestry found a similar and significant estimate for association with breast cancer (OR 4.1; 95% CI: 1.5 – 22.0; p=0.039) (Table 3). All CHEK2 c.707T>C carriers had ER+ BC (Supplemental Table 4) which was significantly higher compared to women not found to carry this variant (p=0.024).

Discussion

This is the first study of Hispanic women with breast cancer that reports on the spectrum and frequency of PVs in breast cancer susceptibility genes beyond BRCA. Similar to our previously reported observation of recurrent BRCA variants among U.S. Hispanics^12,15,37 and Mexican breast and ovarian cancer patients^9,10, we found that three recurrent PVs represented 52% of the total. One recurrent PALB2 PV was responsible for 50% (9/18) of all PALB2 PVs. We previously demonstrated that Hispanics with this PALB2 variant (c.2167_2168del: p.M723fs) shared a common genotype; Italian breast cancer families with the same variant had a different founder genotype.³⁸

Similarly, there was a recurrent CHEK2 PV, c.707T>C; pL236P, that was observed 14 times (representing 70% of CHEK2 PV carriers in our study), with an OR of approximately 3.2 (the more conservative point estimate we found.) This variant is listed five times as a variant of uncertain significance (VUS) and one time as likely benign in ClinVar (as deposited by six commercial laboratories), although Myriad Genetics distributed revised clinical test reports reclassifying it as likely pathogenic in October of 2017 (personal communication; JNW). Only through a large study focused on Hispanics are we able to report its definitive association with breast cancer, as this variant is rarely seen in individuals of European ancestry. In fact, this variant is only found in Latin American populations in ExAC and not in African or African Americans, Europeans, East Asians or South Asians; this allele frequency is strongly suggestive that the variant originates from Indigenous American populations. The estimated carrier frequency of this variant in ExAC Hispanics (excluding TCGA) was ~0.6%. Therefore, this variant seems comparable in its prevalence and effect size to the CHEK2 1100delC mutation in European ancestry populations.^39–42 This association demonstrates the importance of investigating under-studied populations, as well as the value of multi-center collaborations.

The incidence of clinically actionable PVs in this cohort of high-risk Hispanic BC was 4.7%, and was largely driven by the PALB2 and CHEK2 founder PVs. The overall prevalence of PV’s is similar to other studies which have reported on the use of panels in high risk women of mostly European ancestry.^43–45 However, the relative contribution of individual genes in our study is different from the distribution reported in women of European ancestry. When considering the number of unique PVs (Supplemental Table 3), PALB2 had the highest number. There was a relative paucity of ATM PVs (10% of total) compared to published reports in European ancestry populations.^39,40,45

The relatively large proportion of founder mutations in PALB2 and CHEK2 in our study and in BRCA1 and BRCA2 found in our previous studies^9–13,15 are indicative of substantial founder effects that contributed to the modern-day Latin-American population. These observations are consistent with more systematic population genetic studies of Native American populations^46,47 which demonstrate a profound bottleneck that occurred with the settling of the Americas.

Our study has several strengths. First we have sequenced a large number of Hispanic breast cancer cases, adding significantly to what is known about breast cancer susceptibility in this population. Second, we performed analyses carefully controlling for ancestry. Although ExAC is commonly used for rare variant association studies, there are limitations to its use. In particular, ancestry differences between cases and the ExAC controls could bias any associations. This is particularly true in Hispanics where ancestry is quite heterogeneous among individuals ^48,49. To overcome this bias, we subsampled our cases to match the ancestry of the ExAC controls. Furthermore, we performed a second analysis of our cases and a smaller dataset of controls adjusting for genetic ancestry as a covariate. To reduce any potential differences from sequencing coverage and informatics, we jointly analyzed the individual case and control data using the same informatics pipeline. The two analyses provided similar results suggesting that we were able to remove potential ethnicity differences from our associations. Since 937 of the controls from the second analysis of individual sequence results were included in the ExAC dataset, the second analysis is not an independent statistical validation. Rather it is experimental validation that one can match cases on aggregate ExAC ancestry data for case-control analyses of ORs.

There are limitations to this study. First, we did not investigate large genomic rearrangements; the frequency of their occurrence is not reported for many of these genes, thus we may be underestimating the heritability accounted for by known breast cancer susceptibility genes. We previously reported that large genomic rearrangements accounted for 12% (22/189) of all BRCA mutations identified in a study of 746 high risk Hispanic breast cancer patients in the U.S.,¹² a higher proportion than seen in non-Hispanics.⁵⁰ Second, the histopathological data are limited so that we do not have sufficient statistical power to investigate associations with mutation status.

In conclusion, as there are gaps in incorporation of GCRA for women with BC in the US,^51,52 this study contributes to our understanding of the spectrum of genetic susceptibility and risks for breast cancer in Hispanics; this important information can enable enhanced screening, prevention, and therapeutic strategies for patients and their families. For example, detection of a PALB2 or CHEK2 PV would prompt a recommendation for intensified breast screening, including annual contrast enhanced breast MRI, starting at 30–35 years of age⁵³ and increased colon cancer screening in the setting of a CHEK2 PV. Insights about recurrent variants continue to illuminate historical population genetics for Hispanics, while raising the possibility that the presence of frequent founder variants may enable economical genetic screening⁹ in Hispanic populations.