RAD51B Harbors Germline Mutations Associated With Pancreatic Ductal Adenocarcinoma

PURPOSE

Genetic alterations in many components of the homologous recombination, DNA damage response, and repair (HR-DDR) pathway are involved in the hereditary cancer syndromes, including familial pancreatic cancer. HR-DDR genes beyond BRCA1, BRCA2, ATM, and PALB2 may also mutate and confer the HR-DDR deficiency in pancreatic ductal adenocarcinoma (PDAC).

METHODS

We conducted a study to examine the genetic alterations using a companion diagnostic 15-gene HR-DDR panel in PDACs. HR-DDR gene mutations were identified and characterized by whole-exome sequencing and whole-genome sequencing. Different HR-DDR gene mutations are associated with variable homologous recombination deficiency (HRD) scores.

RESULTS

Eight of 50 PDACs with at least one HR-DDR gene mutation were identified. One tumor with BRCA2 mutations is associated with a high HRD score. However, another tumor with a CHEK2 mutation is associated with a zero HRD score. Notably, four of eight PDACs in this study harbor a RAD51B gene mutation. All four RAD51B gene mutations were germline mutations. However, currently, RAD51B is not the gene panel for germline tests.

CONCLUSION

The finding in this study thus supports including RAD51B in the germline test of HR-DDR pathway genes.

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is a malignant disease with a dismal 5-year overall survival rate of 9%.¹ The poor prognosis of PDAC is primarily because of the drug-resistant nature of PDAC and a lack of effective treatment strategies. Therefore, new treatment choices are urgently needed, especially for patients with metastatic disease.

CONTEXT

• Key Objective

• Pancreatic ductal adenocarcinoma (PDAC) has a dismal prognosis, and homologous recombination, DNA damage response, and repair (HR-DDR) genes may mutate and confer the HR-DDR deficiency in PDAC. This study used a 15-gene HR-DDR panel beyond BRCA1, BRCA2, ATM, and PALB2 to examine the genetic alterations in PDAC patients. It identified eight patients with at least one HR-DDR gene mutation with homologous recombination deficiency scores calculated.

• Knowledge Generated

• Eight of 50 PDACs with at least one HR-DDR gene mutation were identified. RAD51B germline mutations were found in four out of the eight patients, while RAD51B currently is not a gene listed on germline test panel.

• Relevance

• Including the RAD51B in germline test panel could allow more PDAC patients with HR-DDR pathway deficiency being identified, and these patients may benefit from HR-DDR pathway related therapies.

Numerous studies have demonstrated that the homologous recombination, DNA damage response, and repair (HR-DDR) pathway is essential for DNA double-strand break repair.^2-4 Accumulated studies have delineated the components in the HR-DDR pathway.^5-7 Genetic alterations in many of the components in this repair pathway are involved in the hereditary cancer syndromes, including familial pancreatic cancer. BRCA1, BRCA2, PALB2, and ATM mutations are among the most studied germline mutations in familial pancreatic cancer.^8-11 PDACs with BRCA1/2 mutations are more sensitive to platinum-based chemotherapy than those without HR-DDR mutations.^12-14 Breast and prostate cancers with BRCA1/2 mutations and ovarian cancers in general demonstrated the response to poly-ADP ribose polymerase (PARP) inhibitor, which inhibits the alternative repair pathway and leads to the synthetic lethality in the tumor cells that have suffered a homologous recombination repair deficiency. In PDAC patients with germline BRCA1/2 mutations, the overall survival of patients with platinum chemotherapy was much longer than those without platinum chemotherapy (22 months v 9 months).¹² In patients with metastatic PDAC who had no progression after a minimum of 4-month first-line platinum-based chemotherapy, the maintenance therapy with a PARP inhibitor, olaparib, compared with a chemotherapy break, prolonged the progression-free survival from 3.8 months to 7.4 months (P = .004).¹⁵

More recently, a 15-gene panel including BRCA1, BRCA2, ATM, BRIP1, BARD1, CDK12, CHEK1, CHEK2, FANCL, PALB2, PPP2R2A, RAD51B, RAD51C, RAD51D, and RAD54L was chosen as a companion diagnostic HR-DDR gene panel for the US Food and Drug Administration–approved olaparib indication in prostate cancer.¹⁶ However, genetic alterations in most of these 15 HR-DDR genes were not characterized in depth in PDACs. Therefore, we conducted a study to examine genetic alterations in these 15 HR-DDR genes in PDACs in our database.

METHODS

Patients and Specimens

Genomic variants of 15 genes in the FoundationOne companion diagnostic HR-DDR gene panel,¹⁶ including BRCA1, BRCA2, ATM, BARD1, BRIP1, CDK12, CHEK1, CHEK2, FANCL, PALB2, PPP2R2A, RAD51B, RAD51C, RAD51D, and RAD54L, were used in our study. Eight deidentified PDAC patients with one or more HR-DDR gene variants in the tumors were identified from 50 consecutive patients from the Pancreatic Cancer Precision Medicine Center of Excellence registry database. Patient consent was waived by John Hopkins Medicine Institutional Review Board. These patients were treated at the Johns Hopkins Hospital between March 2016 and February 2018. Tumor samples were obtained from the primary tumor through surgery or endoscopic ultrasound-guided fine-needle aspiration biopsy. Normal samples for controlling were obtained from adjacent normal tissues of the primary tumor.

DNA Extraction, Library Preparation, and Sequencing

Genomic DNA was extracted by using the AllPrep DNA/RNA Kit (Qiagen, Germantown, MD) according to the manufacturer's instructions. To construct whole-exome capture libraries, genomic DNA was randomly fragmented into 200 approximately 250  bp fragments, and the fragments were purified and ligated by specific adaptors according to instructions of the MGIEasy Universal DNA Library Prep Set, and then captured with the MGIEasy Exome Capture V4 Probe Set (approximately 59 Mb; MGI, Shenzhen, China). The whole-genome sequencing (WGS) libraries were constructed according to instructions of the MGIEasy Universal DNA Library Prep Set. All constructed libraries were sequenced on a DIPSEQ platform (BGI). We performed both whole-exome sequencing (WES) and WGS if there was sufficient DNA. WES achieved an average coverage of 268× (90-420) in normal samples and 665× (83-1,251) in tumor samples. For WGS data sets, the normal samples achieved an average coverage of 46× (34-80), and the tumor samples achieved an average coverage of 49× (36-72).

Sequencing Data Analysis

The raw sequencing data were processed by SOAPnuke v 2.0.7¹⁷ to filter low-quality reads and adaptor contaminations. The clean reads were processed on the basis of the UCSC human reference genome (hg19) using Sentieon pipeline¹⁸ with the Sentieon driver. The Sentieon DNA Software package is a speed-up software program that rebuilt the phasing algorithms following Genome Analysis Toolkit (GATK) best practices: read alignment, mark duplication reads, indel realignment, base quality score recalibration, and variant calling.

Germline Variant Calling

For patients with family history of PDAC, the germline variants including single-nucleotide polymorphism (SNP) and short insertion and deletions (indels) were identified using Sentieon's Haplotyper algorithm on WES data. The raw SNPs were then filtered by GATK Variant Quality Score Recalibration. Raw indels were filtered using the GATK VariantFiltration module with the parameter QD < 2.0 || FS > 200.0 || ReadPosRankSum < –20.0 || SOR > 10.0 || InbreedingCoeff < –0.8.

Somatic Variant Calling

First, single-nucleotide variations (SNVs) and indels were detected using WES and WGS data. There were six widely used somatic mutation callers: Lancet v1.0.7,¹⁹ Strelka2 v2.9.2,²⁰ Muse v1.0,²¹ Mutect2 (GATK4.0.6),²² Somticsniper v1.0.5.0,²³ and Svaba v0.2.1.²⁴ The variant calls were refined by intersecting the callings from Mutect2, Strelka2, Lancet, Muse, and Somticsniper for SNVs, and Strelka2, Lancet, Mutect2, and Svaba v0.2.1 for indels. The high confident variants were identified by at least two variant callers.

The following criteria eliminated possible germline-induced artifacts: (1) For variants deposited in the dbSNP database, < 19 supporting reads in the paired-normal sample or not presented in the COSMIC²⁵ database. For variants that did not deposit in dbSNP, less than eight supporting reads in the normal sample; (2) more than 0.1% minor allele frequency in the 1000 Genomes Project; and (3) more than 0.1% minor allele frequency in the EXAC and the annotation of the ClinVar is not pathogenic.

FACETS v0.5.11,²⁶ an allele-specific copy-number algorithm, was used to identify the copy-number variation for the WES and WGS data with default parameter setting. The structural variants (SVs) were called for WGS data by Manta v1.5.0²⁶ with default parameters. Moreover, a simple reciprocal inversion format was transformed to single inverted sequence junctions by a supplement script provided with Manta.

Determining the Clinical Significance of Genomic Variants

All SNVs and indels, germline or somatic, of HR-DDR genes were categorized into four groups: (1) clinically significant; (2) potentially clinically significant_a; (3) potentially clinically significant_b; and (4) not clinically significant. First, clinically significant variants were defined as SNV/indel variants with annotations of pathogenic or likely pathogenic in ClinVar²⁷ (GRCh37, database date December 10, 2020). Second, potentially clinically significant_a variants were defined as truncated variants (nonsense, frameshift, or splice intervening sequence ± 1 or 2) with annotations of the variant of unknown significance (VUS) or without an entry in ClinVar. Third, the pathogenicity of missense variants with annotations of VUS or with no entry in ClinVar was predicted with MetaSVM,²⁸ MetalLR,²⁸ and FATHMM-MKL.²⁹ The missense variants were categorized as potentially clinically significant_b if any of these three algorithms had an outcome of deleterious. Pathogenicity of indels with annotations of VUS or had no entry in ClinVar were predicted with SIFT.³⁰ The indel variants were categorized as potentially clinically significant_b if the outcomes of SIFT were deleterious. Fourth, not clinically significant variants were defined as SNV/indel variants annotated as benign or likely benign in ClinVar.

Homologous Recombination Deficiency Score

The scar-based homologous recombination deficiency (scarHRD) R package,³¹ a genomic scar-based algorithm, was performed to calculate the homologous recombination deficiency (HRD) score. The score was the sum of the loss of heterozygosity (LOH), telomeric allelic imbalance, and large-scale state transitions. In our study, a reliable estimation of ploidy and purity was estimated and manually confirmed on the basis of three tools (FACETS, PyLOH v1.1,³² and ABSOLUTE v1.2³³). A combination of ploidy, purity, and allele-specific copy-number profile was used as the scarHRD input to measure the HRD score. scarHRD scores were estimated separately using WES and WGS data.

For the whole-genome sequenced tumors, we also estimated the HRD scores by classifier of homologous recombination deficiency (CHORD),³⁴ which was independent of genomic scar information. Briefly, CHORD used a random forest-based classifier and involved the counts of three types of mutation contexts as input features for the model: (1) SNVs subdivided by base substitution type; (2) indels stratified by the presence of sequence homology, tandem repeats, or the absence of either; and (3) SV, stratified by type and length. A sample was considered HRD if the HRD probability (sum of the probability of belonging to the BRCA1 and BRCA2 classes) was ≥ 0.5.

Calculation of Tumor Mutation Burden and Neoantigen Prediction

The tumor mutation burden (TMB) was calculated by the total number of nonsynonymous SNV and INDEL per megabase in the coding region. HLA-allele typing was predicted by OptiType v1.3.1³⁵ using the WES results. Nonsynonymous SNVs and INDEL were translated into 9 approximately 11 amino acids with a sliding window method as the candidate mutated peptides. NetMHC v4.0³⁶ and NetMHCpan v4.1³⁷ were applied to predict the HLA-binding peptides. The final neoantigens were retained with binding affinity < 500 nM predicted by at least one method. Neoantigen burden was estimated as the number of peptides that bind to the HLA-A allele.

RESULTS

Whole-Exome Sequencing of the HR-DDR Genes

Twelve HR-DDR gene variants according to WES were found in eight PDACs (Table 1). The average age of these eight patients was 67 years (range, 61-74 years). All eight patients had a family history of PDAC. One PDAC had three variants of the HR-DDR genes; two PDACs had two HR-DDR gene variants, and all other five PDACs had only one HR-DDR pathway variant. Within these 12 gene variants, the majority were missense (n = 9, 75%), followed by two frameshifts (17%) and one missense (13%). Both of the two frameshift variants were predicted to be prematurely truncated variants, which are anticipated to result in a loss of function in BRCA2 and RAD51B, respectively, as previously described.³⁸ One PDAC carried a somatic mutation in BRCA2 with a variant allele frequency (VAF) of 0.06. The highest variant frequency was found in RAD51B (n = 4, 50%), followed by BRIP1 (n = 2, 25%) and RAD54L (n = 2, 25%). LOH was only found in BRIP1 21:g.30699137A>G and ATM 11:g.108216545C>T in PDAC from patient 1. The clinical significance of the HR-DDR gene variants was determined by using the algorithm as described above. Three variants were clinically significant, one was potentially clinically significant_a, and the remaining eight were potentially clinically significant_b (Table 1).

TABLE 1.

Homologous Recombination, DNA Damage Response, and Repair Gene Variants in Pancreatic Ductal Adenocarcinoma

Structure Variation Analysis Via WGS

Structure variation was analyzed in five PDACs whose WGS results are available. The median number of total structure variations in the five PDACs was 165. The tumor (patient 1) with three HR-DDR gene variants had the highest SVs (No. = 286), followed by the tumor (patient 2) with two HR-DDR pathway variants (No. = 206). The SVs and subtype results were summarized in Table 2. One tumor (patient 1) has a breakpoint in intron 11 of the RAD51B gene, resulting in the translocation of the RAD51B gene (Fig 1).

TABLE 2.

SV in Pancreatic Ductal Adenocarcinoma Tumors With Homologous Recombination, DNA Damage Response, and Repair Gene Variants

FIG 1.

The structural variant of the RAD51B translocation identified in one patient. PDAC according the WGS result. The schema depicts the location, the fusion partner, the orientation, and the breakpoint junction fragment. One breakpoint on chr16:64133026 is located at a distance of 844631 bp from the CDH11 gene; and a second breakpoint on chr14:68943896 is located in intron 10 of the RAD51B gene and anticipated to truncate RAD51B at the 346th amino acid (alanine). The breakpoint junction fragments on chromosome 14 and chromosome 16 are labeled in blue and pink, respectively. PDAC, pancreatic ductal adenocarcinoma; WGS, whole genome sequencing.

HRD Score of PDACs With HR-DDR Gene Variants

HRD scores of all eight patients were assessed by using the scarHRD analysis of the WES results (Table 3). The median scarHRD score of the eight PDACs was 21.3. The highest score was detected in the PDAC (patient 7) with two clinically significant BRCA2 variants, whereas the lowest score was detected in the PDAC (patient 5) with one clinically significant CHEK2 variant. HRD scores were also calculated by using both scarHRD and CHORD analyses of the WGS results that are available with five PDACs. The median scarHRD (WGS) and CHORD score of PDAC with HR-DDR pathway variants were 22 and 0, respectively. The PDAC tumor (patient 3) with the highest scarHRD score (WGS) had a RAD51B, a potentially clinically significant_a variant. The highest CHORD score was found in the PDAC tumor (patient 4) with a BRIP1 potentially clinically significant_b variant (Table 3).

TABLE 3.

Homologous Recombination, DNA Damage Response, and Repair Gene Variants and Corresponding HRD Scores

Neoantigen and TMB Analysis Via WES

Neoantigen load and TMB were assessed in all eight PDACs with HR-DDR gene variants. The average tumor neoantigen load predicted by netMHC according to neoepitopes in length from 9 to 11 amino acids was 53. The average TMB was 1.34. Both the highest tumor neoantigen level and TMB were associated with the PDAC from patient 3 that carried a RAD51B frameshift variant. The tumor (patient 5) that carried a CHEK2 missense variant had no predicted neoantigen and also demonstrated the lowest TMB (0.03/Mb; Table 4). Four tumors with the RAD51B variants had an average neoantigen load of 62.3 and an average TMB of 1.75. Therefore, HR-DDR–deficient PDACs also have a low TMB and neoantigen load.

TABLE 4.

Neoantigen Load and Tumor Mutation Burden of Pancreatic Ductal Adenocarcinoma Tumors With Homologous Recombination, DNA Damage Response, and Repair Gene Variants

DISCUSSION

This comprehensive analysis of HR-DDR–deficient PDAC showed the BRCA2 mutation was associated with a high HRD score, as anticipated. However, CHEK2 mutation was associated with a zero HRD score. The ranges of TMB and neoantigen load in HR-DDR–deficient PDACs were lower than those of the high-TMB or microsatellite instability-high tumors, suggesting that HR-DDR deficiency does not increase the rate of missense mutations. Thus, HR-DDR–deficient PDACs are unlikely to respond to the single-agent immune checkpoint inhibitors. In addition, as different HR-DDR gene mutations were associated with variable HRD scores, their sensitivity to the PARP inhibitors would be anticipated to be different.

Interestingly, RAD51B alterations were present in four of eight PDACs in this study. RAD51B is a member of the RAD51 protein family. RAD51 family members are evolutionarily conserved proteins essential for DNA repair by homologous recombination. The RAD51B protein has been shown to form a stable heterodimer with the family member RAD51C. The RAD51B gene mutations have been reported to be associated with leiomyoma^39,40 and hereditary breast ovarian cancer syndrome.^41-43 Two patients with somatic RAD51B variants in the tumors were previously identified in a combined data set of 3,584 patients.³⁴ In this study, all four RAD51B gene mutations appear to be germline mutations. However, currently, RAD51B is not the gene panel for germline tests. The finding in this study thus supports including RAD51B in the germline test of HR-DDR pathway genes.

Nevertheless, this study is limited by its small sample size and descriptive nature. A larger study is warranted in the future.

DATA SHARING STATEMENT

The raw sequencing data were deposited at NCBI database with the project accession number PRJNA737570.

AUTHOR CONTRIBUTIONS

Conception and design: Mouen Khashab, Richard A. Burkhart, William R. Burns, Amol Narang, Atif Zaheer, Elliot K. Fishman, Robert Anders, Jin He, Lei Zheng, Dongbing Liu, Kui Wu, Daniel A. Laheru

Financial support: Jin He

Administrative support: Dea Cunningham, Richard A. Burkhart, Jin He, Kui Wu

Provision of study materials or patients: Dea Cunningham, Dung T. Le, Vikesh Singh, Richard A. Burkhart, Jin He

Collection and assembly of data: Fanfan Xie, Ding Ding, Cong Lin, Dea Cunningham, Michael Wright, Valerie Lee, Ross Donehower, Dung T. Le, Michael Pishvaian, Eun Ji Shin, Vikesh Singh, Alison P. Klein, Amy Hacker-Prietz, Thomas McPhaul, Richard A. Burkhart, Atif Zaheer, Elizabeth D. Thompson, Robert Anders, Jun Yu, Jin He, Lei Zheng, Daniel A. Laheru

Data analysis and interpretation: Fanfan Xie, Ding Ding, Cong Lin, Michael Wright, Ammar A. Javed, Nilo Azad, Ana De Jesus-Acosta, Michael Pishvaian, Eun Ji Shin, Anne Marie Lennon, Alison P. Klein, Nicholas J. Roberts, Amol Narang, Atif Zaheer, Elliot K. Fishman, Robert Anders, Jin He, Christopher L. Wolfgang, Lei Zheng, Dongbing Liu, Daniel A. Laheru

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/po/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).