Functional Analysis of BARD1 and BRCA1 Variants of Uncertain Significance in
Homology-Directed Repair
Dissertation
Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy
in the Graduate School of The Ohio State University
By
Aleksandra Igorevna Adamovich, B.S.
Biomedical Sciences Graduate Program
The Ohio State University
2019
Dissertation Committee
Jeffrey Parvin, M.D., Ph.D., Advisor
Michael Freitas, Ph.D.
Paul Goodfellow, Ph.D.
Amanda Toland, Ph.D.
Copyrighted by
Aleksandra Igorevna Adamovich
2019
2
Abstract
BRCA1 and BARD1 encode proteins that form a heterodimer critical for mediating tumor suppression. Individuals with deleterious germline mutations in BRCA1 and BARD1 are at an increased risk for breast and ovarian cancers associated with poor disease outcomes.
Although increased cancer screening has helped identify genetic variants, there is not enough information regarding many of them to ascertain clinical risk. These variants are known as variants of uncertain significance (VUS). Functional assays can be used to predict the clinical predisposition of gene variants. The homology-directed repair (HDR) assay is one such assay that has been developed to examine the function of variants in
DNA repair. While BRCA1 has been extensively examined in HDR, with loss of function correlating with pathogenicity, its binding partner BARD1 has not been. In this dissertation, 76 potentially cancer-causing BARD1 truncation and missense variants were identified using whole exome sequencing and examined in the singleton HDR assay. Two were known to be benign variants and found to be functional in HDR, and three were known pathogenic variants found to have loss of function, supporting the notion that the
HDR assay can be used to predict the clinical risk of BARD1 variants. Four variants in the ankyrin repeat domain were found to be non-functional in HDR, indicating that there are DNA repair functions associated with this domain that have not been closely examined. To examine whether BARD1-associated loss of HDR function results in DNA
ii damage sensitivity, cells expressing non-functional BARD1 variants were treated with ionizing radiation or cisplatin. These cells were found to be more sensitive to DNA damage, and variations in the residual HDR function of non-functional variants did not correlate with variations in sensitivity.
While over 130 BRCA1 variants have been examined in the HDR assay, more comprehensive analysis of several BRCA1 functional domains, including the repair- mediating BRCT domain, has not been conducted. In this dissertation, all potential missense and truncation variants in amino acids 1577 to 1863 were generated and subjected to multiplex HDR assays. More than 500 variants between amino acids 1577 and 1768 were characterized from an integrated library of over 2000 variants in the multiplex assay, including thirty variants currently classified as VUS. All of the non- functional missense mutants identified were located in the BRCT domain. Results from the multiplex HDR assay also correlated with results from singleton HDR assays.
Variants classified as functional by the multiplex HDR assay were also functional in assays examining BRCA1 transcriptional activation (TA) and cell proliferation via saturation genome editing (SGE). However, several known pathogenic variants and variants classified as non-functional in the TA and SGE assays were functional in the multiplex HDR assay. Despite discrepancies between functional assays, the majority of result comparisons were consistent. In addition, variants that were non-functional in the multiplex HDR assay were always non-functional in the TA and SGE assays when comparisons were available. These findings help improve understanding of the BARD1
iii and BRCA1 functional domains in DNA repair and identify regions and variants of importance for cancer predisposition.
iv
Dedication
Dedicated to my family.
v
Acknowledgments
I would like to thank my family, my advisor Jeffrey Parvin, my committee, and all past and present members of the Parvin lab for their support and feedback over the years.
vi
Vita
2005-2009…………………………………Dublin Scioto High School
2009-2013…………………………………B.S. Microbiology, The Ohio State University
2013-2014, 2018-2019……………………Distinguished University Fellowship, The
Ohio State University
2014-2015…………………………………HHMI Med into Grad Fellowship
2015 to present……………………………Graduate Research Associate, Department of
Biomedical Informatics, The Ohio State
University
2019………………………………………The Ohio State University College of Medicine
Trainee Research Day Travel Award
Publications
Adamovich, AI and Parvin, JD. (2019). Characterizing BRCA1 Variants Using
Multiplex Functional Assays. Arch. Appl. Med. 1, 1-8.
Adamovich AI, Banerjee T, Wingo M, et al. (2019). Functional Analysis of BARD1
Missense Variants in Homology-Directed Repair and Damage Sensitivity. PLoS Genet.
15, e1008049.
vii
Adamovich AI, Toland AE and Parvin JD. (2019). F-Box Protein-Mediated Resistance to PARP Inhibitor Therapy. Mol Cell. 73, 195-196.
Starita LM, Islam MM, Banerjee T, Adamovich AI et al. (2018). A Multiplex
Homology-Directed DNA Repair Assay Reveals the Impact of More Than 1,000 BRCA1
Missense Substitution Variants on Protein Function. Am J Hum Genet. 103, 498-508.
Fields of Study
Major Field: Biomedical Science
viii
Table of Contents
Abstract ...... ii Dedication ...... v Acknowledgments...... vi Vita ...... vii List of Tables ...... x List of Figures ...... xi Chapter 1. Characterizing BRCA1 and BARD1 Variants Using Functional Assays ...... 1 Chapter 2. Functional Analysis of BARD1 Missense Variants in Homology-Directed Repair and Damage Sensitivity ...... 25 Chapter 3. Multiplex Homology-Directed Repair Analysis of BRCA1 C-terminus Missense Variants ...... 69 Chapter 4. Discussion ...... 102 Bibliography ...... 112 Appendix A. F-box Protein Mediated Resistance to PARP Inhibitor Therapy ...... 136 Appendix B. Variants that Passed Read-Count Thresholding in Multiplex Homology- Directed Repair Analysis of BRCA1 1577-1768...... 142 Appendix C. Primers Used in BRCA1 C-terminus Mutagenesis ...... 156
ix
List of Tables
Table 2-1. GFP expression percentages for BARD1 variant HDR assays...... 32 Table 3-1. Percentages of GFP-positive cells in each HDR assay replicate...... 77 Table 3-2. The number of GFP-positive and GFP-negative cells collected for each HDR sorting replicate...... 78 Table 3-3. The number of variants identified from filtered and unfiltered BRCA1 amino acid 1577-1768 sequence-function data and their depletion scores...... 81 Table 3-4. Functional results for variants tested in the TA, SGE and multiplex HDR assays...... 88
x
List of Figures
Figure 1-1. BRCA1 and BARD1 protein structure...... 3 Figure 1-2. The homology-directed repair (HDR) assay...... 7 Figure 1-3. BRCA1 variants assessed in homology-directed repair (HDR)...... 9 Figure 1-4. In multiplex assays, BRCA1 variant libraries are generated and inserted into reporter cells or organisms...... 19 Figure 2-1. Selection of BARD1 missense variants for functional analysis using sequencing data of cancer patient samples...... 29 Figure 2-2. Functional analysis of 76 BARD1 variants...... 36 Figure 2-3. Functional analysis of 105 BARD1 variants...... 39 Figure 2-4. BARD1 protein sequence differences between mammalian species...... 42 Figure 2-5. Immunoblots of truncated BARD1 variants...... 46 Figure 2-6. HBT-tagged BARD1 variants expressed from a single FRT site function similarly to transiently transfected variants in HDR...... 48 Figure 2-7. Sensitivity of BARD1 variants to ionizing radiation...... 51 Figure 2-8. Sensitivity of BARD1 variants to cisplatin...... 53 Figure 3-1. BRCA1 variant library diversity maps...... 75 Figure 3-2. Sequence-function map of variants in BRCA1 amino acids 1577-1768...... 80 Figure 3-3. Structure of the BRCA1 BRCT domain with labeled potentially intolerant regions...... 83 Figure 3-4. Comparison of multiplex HDR assay results to clinical data and other functional assays...... 85 Figure A-1. Observed mechanisms of PARPi resistance in cancer cells...... 139
xi
Chapter 1. Characterizing BRCA1 and BARD1 Variants Using Functional Assays
A majority of the text from this chapter has been published as a review (Adamovich and Parvin, 2019).
1
Hereditary breast and ovarian cancers (HBOCs) are caused by germline mutations in tumor suppressor genes, and account for 5-10% of breast cancer cases (Larsen et al.,
2014) and 23% of ovarian cancer cases (King et al., 2011). A plurality of HBOCs are caused by mutations in BRCA1 and BRCA2, which account for 25% of hereditary breast cancers (Melchor and Benítez, 2013) and 34-84% of hereditary ovarian cancers (Ramus and Gayther, 2009). Several other genes, including BARD1, are also associated with
HBOC (Couch et al., 2017; Norquist et al., 2016). While a minority of individuals with breast and ovarian cancers have hereditary disease, individuals with pathogenic BRCA1 variants have a 65% lifetime risk for breast cancer and a 39% risk for ovarian cancer
(Eyfjord et al., 2003), as compared to the 12.4% breast cancer risk and 1.3% ovarian cancer risk present in the general population according to the National Cancer Institute.
Individuals with pathogenic BRCA1 and BARD1 variants are also at increased risk for triple negative breast cancers, which have low survival rates (Metzger-Filho et al., 2012;
Shimelis et al., 2018).
BRCA1
The BRCA1 protein has many functions associated with tumor suppression. BRCA1 encodes a 1863 amino acid protein that features an amino-terminal RING domain, a DNA binding domain (DBD), coiled coil (CC) domain, and two tandem carboxy-terminal
BRCT domains (Roy et al., 2012) (Figure 1-1).
2
Figure 1-1. BRCA1 and BARD1 protein structure. The BRCA1 gene encodes an 1863 amino acid protein consisting of an amino-terminal RING domain, DNA binding domain (DBD), coiled coil (CC) domain, and two tandem carboxy-terminal BRCT domains. The BARD1 gene encodes a 777 amino acid protein consisting of an amino-terminal RING domain, a tandem ankyrin (ANK) repeat domain, and two tandem carboxy-terminal BRCT domains. BRCA1 and BARD1 form an obligate heterodimer via their RING and BRCT domains, as indicated by the arrows.
3
The BRCA1 protein forms an obligate heterodimer with another tumor suppressor protein, BARD1, via their RING (Wu et al., 1996) and BRCT (Lee et al., 2015; Simons et al., 2006) domains. This interaction mediates tumor suppressor activity through direct mediation of DNA repair via the recruitment of RAD51 to sites of DNA double-strand breaks (Moynahan et al., 1999; Scully et al., 1996; Snouwaert et al., 1999). In addition, the BRCA1 and BARD1 RING domains function as an E3 ubiquitin ligase (Hashizume et al., 2001; Lorick et al., 1999; Ruffner et al., 2001). The BRCA1 BRCT domains also bind to several phosphoproteins (Rodriguez et al., 2003; Yu et al., 2003). Binding to
Abraxas mediates BRCA1 recruitment to sites of DNA damage (Wang et al., 2007),
BRIP1 is a DNA helicase involved in DNA repair (Cantor et al., 2001; Clapperton et al.,
2004) and CtIP mediates DNA end resection following DNA damage (Yun and Hiom,
2009). BRCA1 has also been shown to activate transcription via its C-terminus
(Chapman and Verma, 1996), mediate checkpoint activation (Xu et al., 2002a), and ubiquitinate γ-tubulin to regulate the centrosome number in the cell (Starita et al., 2004).
BARD1
BARD1 encodes a 777 amino acid protein that consists of an amino-terminal RING domain, an ankyrin repeat domain, and two tandem BRCT domains (Irminger-Finger et al., 2016) (Figure 1-1). In addition to the various tumor suppressor functions mediated by the BRCA1-BARD1 heterodimer, BARD1 regulates several BRCA1-independent processes. BARD1 has been shown to mediate apoptosis through the binding of p53 to the ANK domain (Berardi et al., 2005). The oncoprotein BCL3, which interacts with the 4 transcription factor NF-κB, binds to BARD1 via their ankyrin domains (Dechend et al.,
1999). BARD1 also regulates mRNA processing of 3′ ends by binding and degrading polyadenylation factor CstF-50 (Kleiman et al., 2005). It has also been shown that the
BARD1 BRCT domain interacts with poly(ADP-ribose) (PAR) to recruit the BRCA1-
BARD1 complex to areas of DNA damage (Li and Yu, 2013). BARD1 interaction with
HP1 retains this complex and allows for the accumulation of DNA helicase FANCJ at these sites (Miyoshi et al., 2015; Wu et al., 2016).
Clinical Significance
Thanks to the increased accessibility and affordability of clinical sequencing and cancer screening, a large number of BRCA1 and BARD1 missense variants have been identified.
However, the sequencing data are insufficient to differentiate benign and pathogenic missense variants, and many variants do not occur frequently enough in the population to assign a clinical risk. These are known as variants of unknown clinical significance
(VUS). According to the ClinVar database (Landrum et al., 2018), which records the clinical significance of gene variants, about 50% of known BRCA1 variants, and about
60% of known BARD1 variants, are considered VUS or have conflicting reports of pathogenicity. VUS in BRCA1 and BRCA2 are found in 10-20% of clinical tests (Eccles et al., 2015), indicating that individuals are often informed of their uncertain cancer risk.
Individuals with pathogenic BRCA1 variants are heterozygous for wild-type BRCA1 and are at risk for loss of heterozygosity (LOH), where expression of the wild-type allele is lost and cells become homozygous for the non-functional BRCA1 variant, which may 5 increase cancer risk (Merajver et al., 1995). There exist preventative measures for breast and ovarian cancers such as mastectomy and salpingo-oophorectomy, but these are invasive procedures that should only be done if necessary (Hartmann and Lindor, 2016).
Based on the potential severity of the disease and the exponentially higher risk associated with HBOCs, there is an essential and pressing need to identify which germline variants are associated with disease. Functional assays may help categorize this large number of
BRCA1 and BARD1 VUS by correlating protein function with cancer predisposition.
BRCA1 and BARD1 Functional Assays
While frame-shift or nonsense variants are likely to cause loss of protein function, it is much more difficult to predict the effect of missense variants that change a single amino acid. Numerous assays have been developed to examine BRCA1 function, with one such assay also used to examine BARD1 function, although assay sensitivity and specificity can vary. Many involve DNA repair. The homology-directed repair (HDR) assay examines the ability of variants to mediate homologous recombination (Lu et al., 2015;
Pierce et al., 2001; Ransburgh et al., 2010) (Figure 1-2).
6
Figure 1-2. The homology-directed repair (HDR) assay. (A) The HDR assay utilizes the reporter HeLa-DR cell line containing two non-functional GFP alleles, one of which is interrupted by a cut site for the rare restriction endonuclease I-SceI. Endogenous expression of the gene of interest is depleted using siRNA. An siRNA-resistant gene variant and I-SceI restriction enzyme are co-expressed, the latter inducing a double-strand break. If the variant is functional in HDR, the donor GFP will be used to repair the broken GFP, resulting in green fluorescence that can be quantified using flow cytometry. If the variant is not functional in HDR, other repair methods will be utilized and green fluorescence will not be observed. Variant protein expression is observed via immunoblotting. (B) Photographs of HeLa-DR cells in the HDR assay. Cells treated with control siRNA express GFP, while GFP expression is diminished when BRCA1 expression is depleted, as BRCA1 is necessary for HDR. Figures adapted from (Ransburgh et al., 2010).
7
The HDR assay uses a reporter cell line containing two non-functional GFP coding sequences, one of which is interrupted by a cut site for the rare-cutting restriction endonuclease I-SceI. If variants are functional in HDR, the other GFP allele will be used as a template to repair a double-strand break induced by I-SceI, resulting in functional
GFP expression that can be quantified (Pierce et al., 2001; Ransburgh et al., 2010).
BRCA1 variants have also been tested in a similar assay for single-strand annealing
(SSA), where a double-strand break in a GFP allele is induced in a reporter HeLa-SSA cell line (Towler et al., 2013). DNA resecting reveals homology with an upstream GFP allele, resulting in repair and green fluorescence.
The HDR assay has been used to examine the function of BRCA1 and BARD1 variants, as BRCA1 and BARD1 are found in an obligate heterodimer and both are necessary for mediating DNA repair. Nearly 150 BRCA1 variants have been examined using the HDR assay, and with the exception of splicing variants, the assay has shown 100% specificity and sensitivity (Figure 1-3).
8
Continued
Figure 1-3. BRCA1 variants assessed in homology-directed repair (HDR). 138 BRCA1 9
Figure 1-3 Continued single missense substitutions were tested for function in the HDR assay (Lu et al., 2015; Ransburgh et al., 2010; Starita et al., 2015; Towler et al., 2013). HeLa-DR cells (Ransburgh et al., 2010) were treated with siRNA specific to the BRCA1 3'-untranslated region (UTR) and empty vector (lane 2) or BRCA1 expression plasmid (lanes 1, 3-140). Cells depleted of endogenous BRCA1 with wild-type BRCA1 rescue (lane 1) were used as a positive control. Cells treated with empty plasmid and BRCA1 3'UTR siRNA were used as a negative control (lane 2). HDR function was characterized by the percentage of GFP-positive cells measured using flow cytometry. Results in each experiment were normalized to the wild-type rescue (lane 1), which was set equal to 1. Missense variants are labeled with colors according to the functional domain they are located in. Variants that are not located in a functional domain are labeled in gray. Truncation variants are labeled as “Trunc.”, and controls as “Cont.” in the legend. Variants that are known to be either benign or pathogenic according to ClinVar are labeled with blue or red dots respectively. Variants with uncertain interpretations are labeled with gray dots.
10
Results were combined from four papers that examined BRCA1 variants in HDR (Lu et al., 2015; Ransburgh et al., 2010; Starita et al., 2015; Towler et al., 2013) and variants were color-coded by functional domain. HDR activity, as quantified by percent GFP- positive cells, was normalized to the activity of cells complemented with wild-type
BRCA1, which was set to one (lane 1). Co-transfection of a plasmid expressing wild-type
BRCA1 fully rescued loss of endogenous BRCA1 expression. Cells depleted of BRCA1
(lane 2) showed a 10-fold drop in DNA repair activity. Known benign variants were labeled with blue dots, known pathogenic variants were labeled with red dots, and VUS were labeled with gray dots. In one BRCA1 study (Lu et al., 2015), germline and tumor samples were evaluated for LOH, and 17 variants in BRCA1 were found to have significantly elevated LOH in the tumor. Of these only six were non-functional in the
HDR assay, suggesting that LOH does not accurately predict loss of function variants.
Non-functional BRCA1 missense variants were located in the RING and BRCT domains, consistent with the locations of known pathogenic variants and suggesting that future analysis could focus on the N- and C-termini. While BARD1 variants have also been examined in the HDR assay, it is not as extensive as the work focused on BRCA1
(Billing et al., 2018; Laufer et al., 2007; Lee et al., 2015). As a result, the roles of the
BARD1 domains in HDR are not as well-characterized. The function of BARD1 variants in HDR will be further discussed in Chapter 2.
Similarly, the rescue of radiation resistance (RRR) assay uses the BRCA1-/- HCC1937 human breast cancer cell line to examine whether BRCA1 variants repair double-strand breaks caused by ionizing radiation (Abbott et al., 1999; Cortez et al., 1999; Ruffner et
11 al., 2001; Scully et al., 1999). A yeast-associated assay also examines BRCA1 missense variants for their ability to affect homologous recombination and cell growth, as deleterious variants have been shown to cause increased homologous recombination in yeast (Caligo et al., 2009).
Other BRCA1 complementation-based functional assays also exist. In the centrosome duplication assay (Kais et al., 2012), cells expressing BRCA1 variants are examined for centrosome numbers, as loss of BRCA1-mediated centrosome regulation results in chromosomal instability (Starita et al., 2004; Xu et al., 1999). Non-functional and known deleterious variants had increased centrosome numbers and unpaired centrioles (Kais et al., 2012). Another complementation assay examines BRCA1 variant function in Brca1- deficient mouse embryonic stem cells via cell proliferation and sensitivity to DNA damage agents such as ionizing radiation and cisplatin (Bouwman et al., 2013; Chang et al., 2009).
Several assays can only be utilized for specific regions of BRCA1 based on the functions examined. BRCA1 RING domain-mediated E3 ubiquitin ligase activity is assayed by examining loss of binding to the E2 ubiquitin enzyme UbcH5a and BARD1 with yeast two-hybrid screenings and ubiquitin ligase activity assays (Morris et al., 2006). Several functional assays have been also designed for the BRCA1 C-terminus. The transcription activation assay fuses the BRCA1 C-terminus to heterologous DNA binding domains in yeast and mammalian two-hybrid assay systems to examine transcription activation function, and non-functional variants are interpreted as likely pathogenic (Chapman and
Verma, 1996; Monteiro et al., 1997; Vallon-Christersson et al., 2001). Results for this
12 assay are identical for most variants in both yeast and mammalian systems (Carvalho et al., 2007; Phelan et al., 2005; Vallon-Christersson et al., 2001). Another yeast assay utilized for the functional analysis of C-terminal variants is the small colony phenotype
(SCP) assay, where functional BRCA1 has been shown to inhibit growth in yeast and non-functional variants do not (Coyne et al., 2004; Humphrey et al., 1997; Millot et al.,
2011). Variants in the N-terminus did not affect colony growth in the SCP assay
(Humphrey et al., 1997). Variants in the C-terminus have also been examined for sensitivity to proteolytic digestion because pathogenic BRCA1 variants have been shown to be unstable and sensitive to digestion (Lee et al., 2010b; Williams and Glover, 2003;
Williams et al., 2001, 2003). Another assay examines whether BRCA1 variants bind phosphopeptides, based on the several phosphoproteins that bind to the C-terminus (Lee et al., 2010b; Williams et al., 2004).
Several functional assays examine BRCA1 localization. In the yeast localization phenotype (YLP) assay, functional BRCA1 accumulates in a single inclusion body in the yeast nucleus, and this localization is disrupted by non-functional and pathogenic variants
(Millot et al., 2011). Another assay examines BRCA1 localization to the nucleus in human cells and nuclear foci formation following DNA damage (Chen et al., 1996;
Scully et al., 1997a, 1997b). Localization of BRCA1 to the cytoplasm has also been associated with breast and ovarian cancer (Chen et al., 1995), and pathogenic variants in the BRCA1 RING and BRCT domains do not form damage-associated foci (Au and
Henderson, 2005; Rodriguez et al., 2004).
13
A large number of functional assays utilize cDNA to examine variant function. However, pathogenic variants may also be caused by splicing defects. Mini-gene splicing assays have been used to identify changes in BRCA1 splicing by inserting BRCA1 exons into a vector, expressing them in cells, and harvesting the RNA to make cDNA and examine transcript changes (Ahlborn et al., 2015; Steffensen et al., 2014).
In addition to biological assays, computational methods have been developed to determine whether variants are non-functional and deleterious. Programs such as Sorting
Tolerant from Intolerant (SIFT), A-GVGD, PolyPhen-2 and Combined Annotation-
Dependent Depletion (CADD) can be used to predict whether certain amino acid substitutions, based on properties such as sequence, structure and evolutionary conservation, affect protein function (Adzhubei et al., 2010; Kircher et al., 2014; Ng and
Henikoff, 2003; Tavtigian et al., 2006). Programs like VarScan, Pindel and GATK can also be used to identify potentially pathogenic variants from sequencing data (McKenna et al., 2010; Shen et al., 2012; Ye et al., 2009). Ideally, the information gained from functional assays of BRCA1 and BARD1 can also be extended to similar proteins.
Functional analysis from BRCA1 BRCT domain variants tested in the transcription activation assay has been used to update the VarCall model and identify potentially pathogenic variants in other proteins containing BRCT domains (Woods et al., 2016).
While in silico and bioinformatics analysis can help evaluate a large number of variants, the conclusions are not always accurate and there is not enough preexisting, validated data in these programs. Such programs are best used in conjunction with functional
14 assays to characterize variants and provide more empirical evidence to improve program output.
Multiplexed Assays
While functional assays are very useful for classifying BRCA1 and BARD1 variants, they are limited by the fact that only a single variant can be expressed in a tested cell population and therefore each needs to be tested one at a time. With each additional variant tested, more time and resources are needed. To accurately examine the function of a protein like BRCA1 or BARD1, all potential variants in a region of interest or even the whole gene need to be examined. That would require testing hundreds or thousands of variants, which is not feasible for the experimental procedures discussed earlier.
Multiplexed approaches allow for a greater number of variants to be tested by expressing hundreds to thousands of variants in the same cell population.
Multiplex assays require experimental procedures that allow for the examination of large numbers of variants. Multiplex CRISPR/Cas9 knockout screens have been used to identify essential genes in various cancer cell lines to better target cancer treatments (Hart et al., 2015). A large number of assays utilize deep mutational scanning, where every possible amino acid change at every protein position is generated and functional effects are examined (Fowler and Fields, 2014). Several methods have been developed for generating hundreds or thousands of mutations. Variants can be introduced at random by controlling the PCR mutagenic rate (Wilson and Keefe, 2004). Variants can also be generated through targeted methods such as programmed allelic series (PALS) 15 mutagenesis, where mutagenic primers to the region of interest are synthesized on a microarray and a degradable wild-type template introduces the mutations (Kitzman et al.,
2015). Mutagenic forward and reverse primers for each codon can also be designed for inverse PCR (Jain and Varadarajan, 2014). Mutagenesis by integrate tiles (MITE) uses ligation to introduce a microarray of all potential single nucleotide variants (SNVs) in a region into a full-length construct (Melnikov et al., 2014). SNVs can also be generated using saturation editing by inserting a library of all potential DNA hexamers into a region of interest and inserting the mutagenized region into the genome with CRISPR/Cas9
RNA-guided cleavage (Findlay et al., 2014). It is useful to examine SNVs because they account for the majority of human mutations (Ng et al., 2008), but SNV studies do not provide the same structure-function information available with studies that examine all possible amino acid changes. Functional variants are identified via high-throughput DNA sequencing to examine which variants are depleted by comparing the variant distribution in the input sample versus the experimental sample, and the frequency of a given variant in each pool is used to determine function (Fowler and Fields, 2014; Fowler et al., 2014).
A variety of protein and gene functions have been examined using multiplex assays.
Enhancer activity has been extensively studied via deep mutational sequencing multiplex reporter assays (Inoue and Ahituv, 2015; Muerdter et al., 2015). All possible SNVs of the liver enhancers ALDOB, ECR11 and LTV1 were generated to examine changes in enhancer function (Patwardhan et al., 2012). Thousands of candidate liver enhancers have also been examined for function via lentivirus-based multiplex reporter assays (Inoue et al., 2017). Protein properties have been studied as well - single amino acid variants of
16
PTEN and TPMT have been examined to identify low abundance, potentially pathogenic variants by fusing them to EGFP and sorting cells based on fluorescence (Matreyek et al.,
2018). Protein abundance assays could be applicable to many different proteins, as several tested proteins showed protein abundance differences between variants (Matreyek et al., 2018). Protein stability and folding has been examined with deep mutational scanning via the degradation and loss of cellular export of misfolded TEM-1 and LGK variants (Bacik et al., 2017). Multiplex sequencing has also been used to examine different features of mRNA processing, such as alternative splicing (Julien et al., 2016;
Rosenberg et al., 2015) and 3′ untranslated regions (Shalem et al., 2015; Zhao et al.,
2014).
In addition to examining widely applicable functions, specific proteins have been functionally characterized via multiplex assays. APH(3’)II, which provides resistance to aminoglycoside antibiotics, has been mutated to examine regions necessary for that resistance (Melnikov et al., 2014). Levoglucosan kinase variants have been examined for function in the pyrolysis oil catabolic pathway (Klesmith et al., 2015). Variants in the U- box domain of the ubiquitination factor Ube4b have also been examined to identify residues necessary for E3 ubiquitin ligase activity and ubiquitin transfer (Starita et al.,
2013). Some proteins have also been characterized in multiplex assays to identify pathogenic variants. Variants of the tumor suppressor PTEN have been examined for protein function through variant protein abundance (Matreyek et al., 2018) and examination of variant lipid phosphatase activity (Mighell et al., 2018). PPARγ variants have been examined for inducing CD36 expression in macrophages following stimulation
17 with PPARγ agonists to identify variants that predispose to lipodystrophy (Majithia et al.,
2016). Multiplexing has also been used to examine the function of H-Ras, a signaling protein and known oncogene, by coupling the binding of Ras•GTP and its binding partner
C-Raf to expression of a chloramphenicol resistance factor to evaluate variant fitness
(Bandaru et al., 2017).
Multiplexed Assays for Analyzing the Function of BRCA1 Missense Variants
BRCA1 has been extensively examined via deep functional sequencing using several different functional assays (Figure 1-4).
18
Figure 1-4. In multiplex assays, BRCA1 variant libraries are generated and inserted into reporter cells or organisms. Functional variants are selected for using methods such as ubiquitin activity (Starita et al., 2015), binding to BARD1 (Starita et al., 2015), HDR activity (Starita et al., 2018) or growth fitness (Findlay et al., 2018). After selection, the remaining variants are identified by sequencing the degenerate barcodes attached to them. Depleted variants are identified as depleted based on their presence in reporter cells prior to selection conditions.
19
Variants in the first 300 amino acids of BRCA1, including those in the RING domain, have been examined for their E3 ubiquitin ligase activity and BARD1 binding activity
(Starita et al., 2015). Variants were expressed on the surface of T7 bacteriophage and those that autoubiquinated (Starita et al., 2013) in the presence of ubiquitin were selected.
Binding between the BRCA1 and BARD1 RING domains was examined using multiplexed yeast two-hybrid assays. Examination of E3 ligase activity showed that the zinc ion residues that maintained RING domain structure and residues that interact with
E2 ubiquitin enzyme UbcH5c (Fox et al., 2003) were most frequently non-functional due to mutations. Results were similar with regards to BARD1 binding – variants that affected the zinc ions, and subsequently RING domain structure, were most intolerant to substitution.
HDR has been directly assessed to examine the function of missense variants in the
BRCA1 N-terminus (Starita et al., 2018). Missense variants in the first 192 amino acids of BRCA1 were generated and subjected to the HDR assay - variants were sorted into
GFP-positive and GFP-negative pools, and then the barcodes attached to each variant were sequenced. All of the non-functional variants identified were nonsense variants or were located in the RING domain, which has extensively studied repair functions. Known benign variants were functional in HDR, while all known pathogenic variants, except for the splicing variant R71G (Vega et al., 2001), were non-functional, resulting in 87.5% sensitivity and 100% specificity for the assay. Examining the relationship between HDR function and protein structure also identified regions where glycine substitutions were not tolerated.
20
Multiplex analysis of BRCA1 function has also been measured via the growth of human haploid HAP1 cells, where BRCA1 and other HDR-related genes are necessary for cell survival (Findlay et al., 2018). Instead of generating all potential amino acid variants, the authors examined the function of all potential SNVs. Saturation genome editing (SGE) of the BRCA1 gene in situ was used to generate all possible variants in the 13 exons that make up the RING and BRCT domains (Findlay et al., 2014). SNVs that inhibited
BRCA1 function were selected against by increasing HDR rates in HAP1 cells, and the gDNA and RNA of variants present in the input population was sequenced and compared to the variants present in the population after multiple passages. Using SGE to generate all potential SNVs allowed the authors to examine splicing (Findlay et al., 2014), finding that SNVs that disrupted splice sites were non-functional and had depleted mRNA. Most non-functional missense SNVs also did not have reduced RNA levels, indicating that loss of function was mostly due to protein defects. As previously seen in the multiplex assays of E3 ubiquitin ligase function and HDR, variants that affected RING domain folding, which affects binding to BARD1, and buried hydrophobic residues were non-functional.
When the results of all four multiplexed assays for BRCA1 function were compared, nearly all of the results were consistent (Starita et al., 2018). The multiplexed HDR and
SGE assays had results that were consistent for over 100 variants. The L22S missense variant was found to be functional in the ubiquitin ligase and BARD1 binding assays
(Starita et al., 2015) but was non-functional in HDR, and only five variants had discrepant results when comparing the HDR and SGE assays.
21
The Future of Examining BRCA1 and BARD1 Variant Function
Assays examining BARD1 and BRCA1 functions, such as HDR, transcription activity and nuclear localization, have been used to characterize variants and infer pathogenicity.
In recent years, new developments in multiplexed approaches have been applied to study
BRCA1 variants as well to characterize BRCA1 functional domains more broadly and comprehensively. Multiplexed assays examining ubiquitin ligase activity, BARD1 binding activity, HDR, and cell fitness examine the BRCA1 N- and C-termini, and have had a general consensus of functional results.
With the existence of multiple functional assays for BRCA1 and multiplexing making it more feasible to characterize thousands of variants at once, this begs the question of what is the “best” assay to use for identifying pathogenic variants. BRCA1 has been shown to be involved in functions such as DNA repair, centrosome regulation and transcription, so which of these functions is the best for predicting clinical risk? It appears that no one assay is perfect, and they all serve different purposes because function is context-specific.
Testing the same variants in the HDR and SSA DNA repair processes has shown that all known pathogenic variants were defective in both processes (Towler et al., 2013).
However, the SSA assay also identified several benign variants as defective, indicating that different regions of the BRCA1 protein are involved in both processes. There were also differences in what was considered a non-functional variant between the HDR and centrosome duplication assays (Kais et al., 2012). Similarly, more variants were non- functional in the multiplexed E3 ubiquitin ligase activity assay than in the multiplexed
BRCA1-BARD1 RING binding assay, and evidence suggests that ubiquitin ligase 22 activity is not needed for tumor suppressor function (Laufer et al., 2007; Shakya et al.,
2011; Starita et al., 2015). In addition, multiple BRCA1 BRCT missense variants have been tested via the same assays – sensitivity to proteolysis, binding to BRIP1 phosphopeptides, and transcription activation via fusion to a heterologous binding domain – and complete concordance occurred in assays about 55% of the time (Lee et al.,
2010b). Another factor that many functional assays do not address is splicing variants, as many of the assays discussed utilize cDNA and do not account for them. The variety of responses regarding BRCA1 function that are provided by these assays show that there is no one true functional assay, and that multiple assays are needed to fully characterize a gene like BRCA1.
BARD1 function has been examined using the HDR assay, and loss of variant function consistently correlates with pathogenicity. BARD1 has not been as extensively researched as BRCA1, and the multiplexed HDR assay would be helpful for characterizing BARD1 functional domains. As several BARD1 functions are dependent on interaction with BRCA1, it is likely that functional assays developed for BRCA1 can be used to examine BARD1 function as well. Multiplex assays examining E3 ubiquitin ligase function and BRCA1/BARD1 binding could be conducted using BARD1 variants.
In addition, BARD1 variants could be tested using the multiplex SGE assay, as it uses a cell line where HDR-associated genes are necessary for cell survival. Non-multiplex assays, such as nuclear localization or stem cell complementation, may also be utilized to characterize BARD1 more effectively.
23
Ultimately, functional assays should augment genetic information and enable a clinician to advise individuals about their risk for breast and ovarian cancer. As more functional assays are conducted, more confidence can be placed on their predictive value. All four published multiplexed BRCA1 assays had results that were overall consistent with each other. We suggest that the confidence in the predictive value is higher for the variants that have common results, as indicated by the more than 100 variants whose function was consistent between the HDR and SGE assays (Starita et al., 2018). When results differ, it could be due to technical error, or possibly due to real differences in the amino acid requirements of the two assays. The multiplexing of additional BRCA1 functional assays, as well as the application of already-validated multiplex assays to BARD1 and other domains of BRCA1, would allow for better understanding of both BRCA1 and BARD1 and provide improved classification of variants.
24
Chapter 2. Functional Analysis of BARD1 Missense Variants in Homology-Directed Repair and Damage Sensitivity
Aleksandra I. Adamovich1, Tapahsama Banerjee1, Margaret Wingo1, Kathryn Duncan1,
Jie Ning2, Fernanda Martins Rodrigues2, Kuan-lin Huang2, Cindy Lee1, Feng Chen2, Li
Ding2, and Jeffrey D. Parvin1*
1 Department of Biomedical Informatics, Ohio State University Comprehensive Cancer
Center, Ohio State University, Columbus, OH
2 McDonnell Genome Institute, Washington University School of Medicine, Washington
University in St. Louis, St. Louis, MI
Address correspondence to [email protected]
Author contributions: A.A. performed experiments, analysis and wrote the manuscript,
T.B., M.W., K.D. and C.L. helped perform clonogenic and/or functional assays, J.N. cloned mutants, K.H., F.C. and F.M.R. performed data analysis, L.D. and J.P. supervised experiments and edited the manuscript.
25
Introduction
Variants in BRCA1 and BRCA2 account for a plurality of hereditary breast and ovarian cancer (HBOC) cases, and are associated with risks of 50–85% for breast cancer and 15–
40% for ovarian cancer (Antoniou and Easton, 2003; Godwin et al., 2010; Levy-Lahad and Friedman, 2007; Parmigiani et al., 2007). BARD1 forms an obligate heterodimer with BRCA1, which functions as both an E3 ubiquitin ligase (Hashizume et al., 2001;
Wu et al., 1996) and as a direct mediator of homologous recombination for the recruitment of RAD51 to the sites of DNA double-strand breaks (Moynahan et al., 1999;
Scully et al., 1996; Snouwaert et al., 1999). Truncated BARD1 variants have been identified in breast and ovarian cancers (De Brakeleer et al., 2010; Ghimenti et al., 2002;
Lee et al., 2010a) and germline variants in the BARD1 gene are associated with increased cancer risk (Hart et al., 2017). Still, for both BRCA1 and BARD1, the functional and clinical consequences are often unknown for sequence changes that replace the encoded amino acid residue.
Both BRCA1 and BARD1 are tested on clinical gene panels for breast and ovarian cancer susceptibility. Many BRCA1 variants, as well as a few BARD1 variants, have been determined to be clinically pathogenic. However, many more variants, which are generally missense substitutions, do not occur frequently enough in the population to assign a cancer risk and are classified as variants of uncertain significance (VUS). The
ClinVar database (Malheiro et al., 2017) gathers information on pathogenic and benign variants, but most variants in its database are VUS. A gene panel testing 25 breast cancer- associated genes found 42% of all tests have findings of a VUS in one or more genes, 26 indicating many people have such variants and there is a growing need for their classification (Tung et al., 2015). Datasets such as the Cancer Genome Atlas (TCGA) gather information on missense variants, but are unable to be used for the accurate prediction of the cancer predisposition of a specific VUS. Assays examining homology- directed repair (HDR) function have demonstrated that known pathogenic BRCA1 variants are non-functional in HDR, while benign variants are functional (Lu et al., 2015;
Ransburgh et al., 2010; Starita et al., 2015; Towler et al., 2013). BARD1 consists of an amino-terminal RING domain, three ankyrin repeat domains, and two carboxy-terminal
BRCT domains (Bowcock et al., 2002; Wu et al., 1996). Previous work in our lab has examined the HDR function of 29 BARD1 variants, focusing on the RING and BRCT domains (Lee et al., 2015).
In this study, we identified 76 BARD1 missense and truncation variants that were potentially cancer-associated from a large dataset containing exome-sequencing data on matched germline and tumor samples (Huang et al., 2018; Lu et al., 2015), and tested them for HDR function. Several HDR-deficient variants were identified in both the ankyrin repeat and BRCT domains. To examine the effects caused by loss of HDR function, cells expressing HDR-deficient BARD1 variants were treated with DNA damaging cisplatin or ionizing radiation. Cells expressing HDR-deficient variants were more sensitive to DNA damage and formed significantly fewer colonies than cells expressing wild-type BARD1. Although cells expressing HDR-deficient variants were more sensitive to damage than wild-type cells, quantitative variations in HDR deficiency did not correlate with differences in sensitivity to DNA damage agents. The results of this
27 study reveal functional domains of BARD1 and suggest that the functional analysis of
BARD1 HDR activity is predictive of breast and ovarian cancer risk.
Identification of BARD1 variants as potential cancer-associated loss of function variants.
BARD1 missense variants with potential cancer predisposition were identified in a set of
10,389 TCGA cancer samples from 33 cancer types using whole exome sequencing
(Figure 2-1A) (Huang et al., 2018; Lu et al., 2015). Sixty-two rare germline variants and
14 somatic variants were found with variant calling. The variant allele frequency (VAF) and loss of heterozygosity (LOH) of germline variants were also examined to identify variants that could be functionally important. Six variants—S339T, T343I, V523A,
N450H, G451fs and L239Q—were identified as having significantly higher LOH, indicating they had an increased likelihood of being pathogenic (Figure 2-1B). Variants are listed by amino acid change instead of nucleotide change. At the beginning of this study, variants were selected from a cohort of 4,034 samples (Kandoth et al., 2013; Lu et al., 2015) that later became part of a larger set of 10,389 samples (Huang et al., 2018;
Kumar et al., 2018). Because of changes to selection criteria and data analysis, most, but not all, of the variants analyzed in this study were present in the larger data set, which was used to update the variant calling (Figure 2-1C). While several variants were not present in our newer data set, they are still likely present in the exome sequencing data.
Analyzed variants were present in 24 of the 33 cancer types examined, not just breast or ovarian cancer, as might be predicted for a BRCA1 binding partner (Figure 2-1D). 28
Continued
Figure 2-1. Selection of BARD1 missense variants for functional analysis using sequencing data of cancer patient samples. (A) BARD1 missense variants of interest were identified in a cohort of 4,034 samples from 12 cancer types and a larger set of 10,389 TCGA samples from 33 cancer types with whole exome sequencing (Huang et al., 2018; Lu et al., 2015). (B) Identification of LOH in BARD1 through comparison of VAF in 29
Figure 2-1 Continued tumor and normal samples. Each dot depicts one variant. The diagonal line denotes neutral selection of the germline variant where the normal and tumor variant allele frequencies (VAFs) are identical. LOH was considered significant at False Discovery Rate (FDR) ≤ 0.05. (C) Number of samples containing each of the 76 BARD1 variants in the 10,389 cohort (Huang et al., 2018). (D) Number of samples affected by each BARD1 variant for each of the 33 cancer types.
Analysis of BARD1 variants in Homology-Directed Repair (HDR).
Seventy-six BARD1 missense variants, a majority of which were located in the ankyrin repeat and BRCT domains or between these domains, were tested for function in the homology-directed repair (HDR) assay (Figure 2-2A). For the HDR assay, a cell line that has two non-functional GFP coding sequences integrated into its DNA is used to examine DNA repair function. One of these GFP-encoding genes contains a recognition site for the rare-cutting restriction endonuclease I-SceI. When the I-SceI expression plasmid is transiently transfected into these cells, a double-strand break is made in one of the GFP sequences. If homology-directed repair uses the second GFP coding sequence as a template to repair across the double-strand break, then the encoded GFP is rendered functional (Pierce et al., 2001; Ransburgh et al., 2010). We used a HeLa-derived cell clone called HeLa-DR, which has the GFP-encoding recombination substrate integrated at a single site. After transfection of the I-SceI expression plasmid, 10–20% of the cells were converted to GFP-positive (Ransburgh et al., 2010). Endogenous BARD1 expression was depleted in HeLa-DR cells by two rounds of transfection of a siRNA that targets the 3’-UTR of the BARD1 mRNA. Simultaneously with the silencing of 30 endogenous BARD1, BARD1 variants were expressed from transiently transfected plasmids that were resistant to the siRNA. Two days following the first transfection, the siRNA and plasmid were transfected again into the cells along with the plasmid that expresses the I-SceI endonuclease. Three days after the second transfection, the number of GFP-positive cells was determined using flow cytometry (Table 2-1). Full HDR activity was observed under conditions of mock depletion of BARD1 by transfection with a control siRNA (Figure 2-2A, bar 1) and by depletion of BARD1 using the 3’-UTR targeted siRNA with rescue by transfection of a plasmid that expressed wild-type
BARD1 (Figure 2-2A, bar 3). Cells depleted of BARD1 and transfected with an empty vector had a 25-fold decrease in HDR activity measured as the percentage of GFP- positive cells (Figure 2-2A, bar 2). We set the level of GFP expression following a double-strand break to a value of 1 relative to wild-type rescue (Figure 2-2A, bar 3) to facilitate comparison between experiments.
31
Experiment 1 Experiment 2 Experiment 3 Sample GFP Sample GFP Sample GFP % % % Control 5.8 Control 8 Control 11.71 Empty 0.1 Empty 0.3 Empty 0.75 WT 6 WT 7.5 WT 9.55 E223G 6.5 S241C 8.5 E67K 10.78 E223G 6.9 S241C 9 E67K 10.6 E223G 6.7 S241C 7.9 E67K 10.59 D230E 6.8 I249V 7.1 M104I 11.37 D230E 6.2 I249V 7 M104I 11.09 D230E 7.1 I249V 8.1 M104I 10.67 L239Q 6.8 I258T 8.8 S109R 10.61 L239Q 7.3 I258T 9.4 S109R 11.63 L239Q 6.8 I258T 9.5 S109R 11.83 R322H 9.9 N118S 12.06 R322H 9.8 N118S 10.96 R322H 10.3 N118S 11.61 N326D 8.2 K140N 10.58 N326D 10.2 K140N 10.97 N326D 8.4 K140N 10.78 S151N 11.22 S151N 12.47 S151N 11.16 V154fs 1.15 V154fs 1.25 V154fs 1.33 D190N 10.96 D190N 11.66 D190N 11.3 D190Y 11.64 D190Y 10.59 D190Y 10.73 R194K 11.07 R194K 11.36 R194K 12.16
Continued
Table 2-1. GFP expression percentages for BARD1 variant HDR assays.
32
Table 2-1 Continued
Experiment 4 Experiment 5 Experiment 6 Sample GFP Sample GFP Sample GFP % % % Control 15.73 Control 12.62 Control 14.56 Empty 0.39 Empty 0.57 Empty 0.63 WT 13.35 WT 9.71 WT 13.78 N326S 13.61 I434F 8.76 N488S 10.77 N326S 13.24 I434F 8.41 N488S 11.04 N326S 13.25 I434F 8.39 N488S 10.96 S339N 11.64 A435V 10.09 H506R 11.15 S339N 12.76 A435V 10.27 H506R 12.65 S339N 11.5 A435V 10.38 H506R 11.77 S339T 12.16 L447C 10.66 V507A 2.86 S339T 12.23 L447C 10.17 V507A 2.14 S339T 13.24 L447C 10.15 V507A 2.38 S342N 11.32 N450H 9.4 V510A 12.12 S342N 11.88 N450H 9.64 V510A 10.95 S342N 11.23 N450H 10.76 V510A 12.18 T343I 11.39 G451fs 1.68 A518V 8.26 T343I 11.8 G451fs 1.86 A518V 9 T343I 12.41 G451fs 1.73 A518V 8.89 T351M 10.64 A460T 4.32 V523I 12.68 T351M 10.5 A460T 4.24 V523I 11 T351M 11 A460T 4.87 V523I 11.73 L359fs 0.83 L465F 5.11 V523A 11.14 L359fs 0.81 L465F 4.71 V523A 9.81 L359fs 0.59 L465F 5.2 V523A 10.11 E361D 13.56 L480S 2.08 P530L 3.19 E361D 12.4 L480S 2.22 P530L 2.93 E361D 13.29 L480S 2.25 P530L 2.98 S389C 12.9 H483R 9.05 Y533F 12.05 S389C 12.49 H483R 8.94 Y533F 11.52 S389C 12.69 H483R 8.91 Y533F 11.07 H606D 0.77 H606D 0.68 H606D 0.59
Continued
33
Table 2-1 Continued
Experiment 7 Experiment 8 Experiment 9 Sample GFP Sample GFP Sample GFP % % % Control 15.23 Control 13.22 Control 10.07 Empty 0.75 Empty 0.54 Control 11.25 WT 13.38 WT 14.3 Control 11.41 K540N 11.92 H606D 0.76 Empty 0.46 K540N 10.57 H606D 0.85 Empty 0.55 K540N 11.01 H606D 0.84 Empty 0.5 S551* 1.97 D612V 14.39 WT 12.25 S551* 1.85 D612V 13.57 WT 12.14 S551* 2.29 D612V 13.58 WT 11.23 H556D 11.3 L625I 10.57 C62S 13.24 H556D 11.57 L625I 9.37 C62S 11.75 H556D 11.15 L625I 9.98 C62S 11.42 S558P 11.91 R641Q 13.55 I69M 13.36 S558P 11.15 R641Q 13.04 I69M 13.12 S558P 11.27 R641Q 13.48 I69M 13.34 Q564* 2.1 R642G 13.68 V85L 11.69 Q564* 1.84 R642G 13.01 V85L 13.29 Q564* 1.82 R642G 13.39 V85L 11.01 S660R 0.7 R664T 11.9 N98S 11.03 S660R 0.51 R664T 12 N98S 11.82 S660R 0.49 R664T 11.98 N98S 12.22 R565C 7.47 G698D 0.57 S103N 12.28 R565C 7.59 G698D 0.57 S103N 11.38 R565C 7.34 G698D 0.67 S103N 11.58 R565H 10.13 P707S 5 S575N 11.69 R565H 9.96 P707S 4.57 S575N 10.83 R565H 10.07 P707S 5.06 S575N 11.24 G574D 9.69 G753D 2.77 S616N 11.81 G574D 9.71 G753D 2.76 S616N 12.54 G574D 9.99 G753D 3.09 S616N 11.19 T598I 8.59 V767fs 0.56 G656R 8.87 T598I 7.63 V767fs 0.76 G656R 8.93 T598I 7.28 V767fs 0.73 G656R 7.48
Continued
34
Table 2-1 Continued
Experiment 7 Experiment 8 Experiment 9 Sample GFP Sample GFP Sample GFP % % % F677L 8.19 F677L 8.33 F677L 9.19 V713M 7.94 V713M 8.72 V713M 8.44 T719A 6.81 T719A 8.04 T719A 7.87 R731H 11.11 R731H 12.27 R731H 10.42 R731C 10.36 R731C 10.65 R731C 9.66
35
Continued
Figure 2-2. Functional analysis of 76 BARD1 variants. (A) 76 BARD1 single missense 36
Figure 2-2 Continued substitutions were tested for function in the HDR assay. HeLa-DR cells (Ransburgh et al., 2010) were treated with control siRNA (lane 1) or siRNA specific to the BARD1 3'- untranslated region (UTR) (lanes 2-80) and empty vector (lanes 1,2) or BARD1 expression plasmid (lanes 3-80). Two positive controls were used: cells treated with empty vector and control siRNA (lane 1), and cells depleted of endogenous BARD1 with wild-type BARD1 rescue (lane 3). Cells treated with empty plasmid and BARD1 3'UTR siRNA were used as a negative control (lane 2). HDR function was characterized by the percentage of GFP-positive cells measured using flow cytometry. Results in each experiment (±S.E.M.) were normalized to the WT rescue (lane 3), which was set equal to 1. Results represent three independent transfections per BARD1 plasmid. Variants that are benign and pathogenic according to ClinVar are labeled blue and red respectively. Variants with conflicting interpretations are labeled gray. HDR-deficient variants are marked by an asterisk and classified by having HDR function less than 0.6 and p < 0.01 when compared with endogenous BARD1 (control siRNA) using the Student’s t-test. (B) BARD1 variants tested in the HDR assay were examined for their expression relative to endogenous BARD1. Replicates were pooled together to examine BARD1 expression. The BARD1 protein is indicated with an arrow, as the BARD1-specific band migrated more slowly than a cross-contaminating band. The endogenously expressed BARD1 in control transfections (lanes 1, 14, 23, 29, 33, 41, 53, 65, 77, 90, 100) and the depleted BARD1 without rescue (lanes 2, 15, 24, 30, 34, 42, 54, 66, 78, 91, 101) can be compared with the expression of variant BARD1 proteins as indicated. A tagged BARD1 V507A plasmid was used to confirm BARD1 expression (lanes 31 and 32, upper arrow) and migrated more slowly than endogenous BARD1 (lane 29, lower arrow). All missense variants had expression higher than the endogenously expressed BARD1.
37
The 76 variants tested were from across the full coding sequence of BARD1. Variants whose HDR activity was significantly different from endogenous BARD1 and whose expression was greater than or equal to endogenous BARD1 were considered to be repair-deficient (Figure 2-2A, B). The eight variants located in the RING domain, as well the 22 in the region between the RING and ankyrin repeat domains, all had HDR activity similar to wild-type. Previous work in our lab (Lee et al., 2015) examined the HDR activity of 29 BARD1 missense variants, including additional variants in the RING domain. We combined the current HDR results with the previously published results into a single table containing 105 variants (Figure 2-3). In this previous work, the variants
L44R, C53W, and C71Y in the RING domain were found to be defective in HDR due to defective binding to BRCA1.
38
Continued
Figure 2-3. Functional analysis of 105 BARD1 variants. 105 BARD1 single missense 39
Figure 2-3 Continued substitutions were tested for function in the HDR assay. Results from Lee at al. 2015 are included with variants tested in this chapter. HeLa-DR cells (Ransburgh et al., 2010) were treated with control siRNA (lane 1) or siRNA specific to the BARD1 3'-untranslated region (UTR) (lanes 2-109) and empty vector (lanes 1, 2) or BARD1 expression plasmid (lanes 3-109). Two positive controls were used: cells treated with empty vector and control siRNA (lane 1), and cells depleted of endogenous BARD1 with wild-type BARD1 rescue (lane 3). Cells treated with empty plasmid and BARD1 3'UTR siRNA were used as a negative control (lane 2). HDR function was characterized by the percentage of GFP-positive cells measured using flow cytometry. Results in each experiment (±S.E.M.) were normalized to the WT rescue (lane 3), which was set equal to 1. Results represent three independent transfections per BARD1 expression plasmid. Variants that are benign and pathogenic according to ClinVar are labeled blue and red respectively. Variants with conflicting interpretations are labeled gray. HDR-deficient variants are marked by an asterisk and classified by having HDR function less than 0.6 and p < 0.01 when compared to endogenous BARD1 (control siRNA) using the Student’s t-test.
40
Surprisingly, in the current study, four of the 17 variants in the ankyrin domain, which has no known DNA repair function, were found to express full-length BARD1 and be defective in HDR. Variants A460T, L465F, L480S, and P530L had HDR activity lower than 0.6, which was significantly lower than cells expressing endogenous BARD1. The five variants located between the ankyrin repeat and BRCT domains were proficient in
DNA repair with the exception of R565C, whose HDR activity was just below the cutoff of 0.6. Of the 19 missense variants tested in the BRCT domain, which is known to be involved in recruiting and retaining the BRCA1-BARD1 heterodimer to areas of DNA damage, five were found to be defective in HDR (Li and Yu, 2013; Miyoshi et al., 2015).
The variants S660R and G698D had HDR function comparable to cells transfected with empty vector. The variants T598I, P707S, and G753D had activity higher than empty vector but still significantly lower than endogenous BARD1. A larger fraction of residues conserved across several mammalian species were mutated in repair-deficient variants
(9/10) than in functional ones (38/55) (Figure 2-4). Five truncation variants were also tested, and all were about as equally defective as the empty vector in the HDR assay.
Previous work has suggested that filtering using high LOH could be used to identify
BRCA1 variants defective in HDR (Lu et al., 2015). However, BARD1 variants that were found to have high LOH (Figure 2-1B) were all functional, with the exception of truncation variant G451fs.
41
Continued
Figure 2-4. BARD1 protein sequence differences between mammalian species. 42
Figure 2-4 Continued
Human (Homo_sapiens), domestic cat (Felis_catus), domestic dog (Canis_lupis_familiaris), mouse (Mus_musculus), gray short-tailed opossum (Monodelphis_domestica) and domestic sheep (Ovis_aries) BARD1 protein sequences were aligned using Clustal Omega to examine conserved residues. Asterisks (*) indicate positions with a single, fully conserved residue. Colons (:) indicate residue conservation between groups of strongly similar properties. Periods (.) indicate conservation between groups of weakly similar properties. Residues that were mutated in HDR-deficient, non- truncating variants are highlighted in yellow. Residues mutated in HDR-functional variants are highlighted in green.
43
Testing the HDR function of BRCA1 variants has shown that, with the exception of variants that impact mRNA splicing, known pathogenic variants of BRCA1 are HDR- defective, while known benign variants are not (Lu et al., 2015; Ransburgh et al., 2010;
Towler et al., 2013). Similarly, the BARD1 variants S241C and E361D, which have been found in patients with breast cancer and are benign according to ClinVar, are functional in HDR (Figure 2-2A, blue dots). Truncation variants V154fs, S551*, and Q564*, where the asterisk indicates a stop codon, are listed as pathogenic according to ClinVar and were non-functional in the HDR assay (Figure 2-2A, red dots). Several other variants that were tested have been identified in breast cancer patients and have conflicting interpretations of pathogenicity (Figure 2-2A, gray dots).
We tested whether the level of expression of any BARD1 variants could have affected their HDR activity. Expression of BARD1 variants was examined via immunoblot
(Figure 2-2B). The relative expression of the endogenous BARD1 and siRNA-depleted
BARD1 were shown (Figure 2-2B lanes 1, 2, 14, 15, 23, 24, 29, 30, 33, 34, 41, 42, 53,
54, 65, 66, 77, 78, 90, 91, 100,101). Though the expression levels of missense variants differed, they all expressed at higher levels than the endogenous BARD1. As an example,
BARD1 L480S (lane 63) had lower expression than the plasmid encoded wild-type (lane
55), but both had more intensely labeled bands than the endogenously expressed BARD1
(lane 53). The variant BARD1 H606D is present in the immunoblots in Figure 2-2B (lane
103), but is not listed in Figure 2-2A because full length protein was not detected, and it was found to contain a nonsense mutation at codon 125. Similarly, BARD1 L359fs is present in Figure 2-2B (lane 50) and Figure 2-5 (lanes 5, 14) but is not listed in Figure 2-
44
2A because it was a miss-call during variant selection. Frameshift and nonsense codon variants (Figure 2-2B lanes 10, 60, 81, 84, 112) lacked full length BARD1. Truncation variants G451fs, S551*, Q564* and V767fs expressed truncated BARD1, while variant protein V154fs was not detected (Figure 2-5). We infer that repair defects observed in truncation variants with poor protein expression were due to the absence of protein instead of expressed, non-functional variant protein. We conclude that for the missense variants, a low level of HDR activity was not due to low expression of the BARD1 protein.
45
Figure 2-5. Immunoblots of truncated BARD1 variants. Truncated BARD1 variants tested in the HDR assay were examined for their expression relative to endogenous BARD1. Replicates were pooled together to examine BARD1 expression. The BARD1 protein is indicated with an arrow. The endogenously expressed BARD1 in control transfections (lanes 1, 10) and BARD1 WT rescue (lanes 3, 12) were compared with the expression of truncated BARD1 variant proteins. Variants were run on 4-12% Bis-Tris (top) and 6% acrylamide (bottom) gels due to a contaminating band that co-migrated with BARD1 WT on 4-12% Bis-Tris gels. Variants G451fs (lanes 6, 15), S551* (lanes 7, 16), Q564* (lanes 8, 17) and V767fs (9, 18) expressed truncated protein. Variant V154fs (lanes 4, 13) does not contain the residue recognized by the BARD1 antibody used, and may express truncated protein. 46
Comparison of BARD1 HDR activity with sensitivity to DNA damage by ionizing radiation and cisplatin.
BARD1 variants A460T, P707S, G753D, and V767fs were selected for further analysis, as they covered a range of HDR activities below 0.6 when transiently expressed. The selected variants and wild-type BARD1 were tagged with the His-Biotin-Tobacco Etch
Virus (HBT) tag (Tagwerker et al., 2006) and integrated into the FRT site of a HeLa-DR derivative cell line called HeLa-DR-FRT (Starita et al., 2018). The advantage of these
FRT site-containing cells was that the BARD1 gene was stably expressed from a single site and should have consistent levels of expression. We tested the stably expressed variants in the HDR assay to confirm repair proficiency was the same as the transiently expressed variants. The HDR activities of these variants from Figure 2-2 are shown in isolation (Figure 2-6A). The same repair trends were observed in both transiently expressed and stably expressed BARD1 variants (Figure 2-6A, B). BARD1 A460T- integrated cells had the most residual repair activity among these variants, and cells integrated with BARD1 P707S, G753D and V767fs had decreasing levels of repair proficiency respectively. Expression of the integrated BARD1 variants was also greater than or equal to that of endogenous BARD1 (Figure 2-6C). BRCA1 expression in variant-integrated cell lines was similar in cells expressing endogenous BARD1 and the various defective BARD1 variants (Figure 2-6C). Thus, none of the changes in HDR observed with these BARD1 variants were attributable to changes in BRCA1 expression.
47
Continued Figure 2-6. HBT-tagged BARD1 variants expressed from a single FRT site function similarly to transiently transfected variants in HDR. (A) Of the 76 BARD1 variants examined for HDR functionality (Figure 2-2A), BARD1 A460T, P707S, G753D, and V767fs were selected for further study based on their range of HDR activity. (B) BARD1 variants integrated into a single FRT site in HeLa-DR-FRT cells (Starita et al., 2018) functioned similarly to the same variants expressed in HeLa-DR cells by transient 48
Figure 2-6 Continued transfection. HBT-tagged BARD1 WT and variants were integrated into HeLa-DR-FRT cells via the FRT site for consistent single-site expression. Cells were treated with control or BARD1 3'UTR siRNA and examined via the HDR assay. Unintegrated HeLa-DR-FRT cells were used as a control. For comparison, results from transiently transfected cells were selected from Figure 2-2 (Figure 2-6A), and the results from expressing the integrated BARD1 variants are shown. While HDR activity was lower in cells containing integrated BARD1 as compared with transiently expressed BARD1, the integrated variants exhibited the same trend as the transiently expressed variants—BARD1 A460T was the most proficient, followed by BARD1 P707S, G753D, and V767fs. (C) Endogenous BARD1 was knocked down in treated stable cell lines, and BRCA1 and BARD1 expression were examined. HBT-tagged BARD1 variants migrated more slowly on electrophoresis gels than endogenous BARD1, as indicated by the upper (HBT- tagged) and lower (endogenous) arrows. HBT BARD1 was unaffected by the 3'UTR- targeted siRNA, while endogenous BARD1 was depleted. All of the BARD1 variants expressed at higher levels than the endogenously expressed BARD1 (middle). BRCA1 protein levels were not affected in variant-integrated cells (top).
49
We examined whether the quantitative loss of HDR proficiency correlated with the sensitivity of cells to extrinsic DNA damage. Clonogenic cell sensitivity assays were performed on HeLa-DR-FRT cells expressing integrated BARD1 wild-type and variants, as well as endogenous-only unintegrated cells. Cells were depleted of endogenous
BARD1 or BRCA1 and subjected to ionizing radiation (IR) (Figure 2-7) or cisplatin
(Figure 2-8). Depletion of BARD1 or BRCA1 from the endogenous-only cells
(E/siBARD and E/siBRCA) was used to determine the effect of a non-rescued HDR defect on sensitivity to IR and provided a baseline for DNA damage sensitivity (Figure 2-
7A, bottom). E/siBARD and E/siBRCA cells formed significantly fewer colonies after IR than control cells (E/siCON). BARD1 variant-integrated cells depleted of endogenous
BARD1 (Variant/siBARD) all formed significantly fewer colonies following IR than the same cells treated with control siRNA (Variant/siCON) (Figure 2-7A, top). For ease of comparison, we included results from Variant/siBARD, E/siBARD, E/siBRCA and
WT/siBARD cells on one graph (Figure 2-7B). Immunoblots were done to confirm knockdown of endogenous BARD1 (Figure 2-7C). While expression of each BARD1 variant differed, it was still greater than or equal to that of endogenous BARD1.
50
Continued
Figure 2-7. Sensitivity of BARD1 variants to ionizing radiation. (A) Endogenous-only cells and cells expressing BARD1 WT, A460T, P707S, G753D, and V767fs were treated with control, BARD1 3'UTR or BRCA1 3’UTR siRNA and X-ray irradiated at doses of 1, 2, 4 and 6 Gy. Colonies were counted after 12 days of growth followed by staining with crystal violet. Results in each experiment (±S.E.M.) were carried out in triplicate and 51
Figure 2-7 Continued converted to logarithmic scale. In each plot, the dashed line indicates control siRNA (siCON), and the solid line indicates depletion with the BARD1 (siBARD) or BRCA1 (siBRCA) 3'UTR-targeted siRNA. The Student’s t-test was done to examine the growth of variant and endogenous-only cell lines treated with BARD1 or BRCA1 siRNA relative to variant and endogenous-only cells treated with control siRNA (indicated by asterisks; * = p < 0.05, ** = p < 0.01). Cells expressing the four BARD1 variants, as well as endogenous-only cells depleted of BARD1 and BRCA1, were significantly different from variant and endogenous-only cells treated with control siRNA across most irradiation doses. (B) Results from panel A are shown for the BARD1 WT and variant cell lines treated with BARD1 3'UTR siRNA and endogenous-only cells treated with BARD1 or BRCA1 3’UTR siRNA. A Student’s t-test was done to compare the colony formation of BARD1 variant and BRCA1 or BARD1-depleted cell lines to BARD1 WT cells (indicated by asterisks). Colony counts of variants and BRCA1 or BARD1-depleted cell lines were significantly different (p < 0.01) from WT at all concentrations except siBARD 6 Gy (p < 0.05) and P707S 1 Gy, G753D 1 Gy, and G753D 2 Gy (p > 0.05). (C) An examination of the expression of endogenous BARD1 and BRCA1 in treated BARD1 variant-expressing and endogenous-only cell lines. Cells treated with BARD1 3'UTR siRNA had depleted endogenous BARD1 expression, while HBT BARD1 was unaffected. Cells treated with BRCA1 3’UTR siRNA showed depletion of endogenous BRCA1 and BARD1 expression. In all cases, the band representing the expression of the variant HBT BARD1 was denser than the faster migrating band from the endogenous BARD1 protein.
52
Continued
Figure 2-8. Sensitivity of BARD1 variants to cisplatin. (A) Cells expressing BARD1 WT, A460T, P707S, G753D, and V767fs cells, as well as endogenous-only cells, were treated with control, BARD1 3'UTR or BRCA1 3’UTR siRNA and treated with cisplatin at concentrations of 1.875, 3.75, 7.5, and 15 μM. Colonies were counted after 12 days of
53
Figure 2-8 Continued growth followed by staining with crystal violet. Results in each experiment were done in triplicate and converted to logarithmic scale (±S.E.M.). In each plot, the dashed line indicates control siRNA, and the solid line indicates depletion with the BARD1 or BRCA1 3'UTR-targeted siRNA. The Student’s t-test was done to examine the growth of variant and endogenous-only cell lines treated with BARD1 or BRCA1 siRNA relative to variant and endogenous-only cells treated with control siRNA (indicated by asterisks; * = p < 0.05, ** = p < 0.01). Cells expressing the four BARD1 variants, as well as endogenous- only cells depleted of BRCA1 and BARD1, were significantly different from variant and endogenous-only cells treated with control siRNA across most cisplatin concentrations. (B) Results from panel A are shown for BARD1 WT, variant and endogenous-only cell lines treated with BARD1 or BRCA1 3'UTR siRNA. A Student’s t-test was done to compare the colony formation of BARD1 variant cell lines, as well as endogenous-only cells depleted of BRCA1 or BARD1, to BARD1 WT cells. Colony counts from variants and endogenous-only cells were significantly different (p < 0.01) from BARD1 WT at most concentrations, excepting A460T 3.75 μM, siBRCA1 3.75 μM and A460T 15 μM (p < 0.05) and A460T 1.875 μM and siBRCA 15 μM (p > 0.05). (C) An examination of endogenous BARD1 and BRCA1 expression in treated BARD1-expressing and control cell lines. Cells treated with BARD1 3'UTR siRNA had depleted endogenous BARD1 expression, while HBT BARD1 was unaffected. Cells treated with BRCA1 3’UTR siRNA showed depletion of endogenous BRCA1 and BARD1 expression. The band representing the expression of variant HBT BARD1 was denser than the endogenous BARD1 protein band in all cases.
54
E/siBARD, E/siBRCA and Variant/siBARD cells formed significantly fewer colonies than WT/siBARD cells at most irradiation concentrations (Figure 2-7B). It was expected that increased HDR deficiency would result in increased sensitivity to DNA damage agents. For example, we expected that cells expressing BARD1 V767fs, the most HDR- deficient variant, would form the least number of colonies. Cells expressing BARD1
A460T, the variant with the most residual HDR activity, were expected to have the largest number of colonies compared with the other variants. Interestingly, this was not the trend observed in the results. Instead, HDR-defective BARD1 variants were all equally sensitive to IR, and as sensitive as non-rescued cells depleted of BARD1 or
BRCA1. Similarly, all four BARD1 variants were more sensitive than wild-type to treatment with cisplatin (Figure 2-8A). Variant/siBARD, E/siBARD and E/siBRCA cells formed significantly fewer colonies than Variant/siCON and E/siCON cells.
Variant/siBARD, E/siBARD and E/siBRCA cells also formed significantly fewer colonies than WT/siBARD cells (Figure 2-8B). As seen with IR, all BARD1 variants and non-rescued cells depleted of BARD1 or BRCA1 were equally sensitive to cisplatin. In addition, BARD1 variant expression remained consistently equal to or greater than that of endogenous BARD1 in cells, indicating that decreased colony formation was not associated with decreased variant expression (Figure 2-8C). While decreased HDR function resulted in decreased colony formation, quantitative differences in the HDR activity did not correlate to quantifiable changes in sensitivity to cisplatin or IR.
55
Discussion
In this study, we found: 1) from 10,389 cancer samples across 33 cancer types, 76
BARD1 missense variants were identified as potentially pathogenic and were selected for functional analysis. 2) Fifteen of the 76 tested variants were defective for HDR, suggesting that these were potentially pathogenic variants. 3) Four of the 17 variants tested in the ankyrin repeat domain, for which there was no previously known DNA repair function, were deficient in homologous recombination. 4) Five of the 19 variants tested in the BRCT domain, which does have known DNA repair functions, were deficient in homologous recombination. 5) Variants that were deficient in HDR rendered the cells sensitive to treatment with DNA-damaging cisplatin or IR. 6) Quantitative differences in HDR deficiency among defective variants did not translate to quantitatively different sensitivity to DNA damage.
The BRCA1-BARD1 heterodimer is necessary for tumor suppressor function (Nandula et al., 2008; Wu et al., 1996). Variants that affect binding between BRCA1 and BARD1 have been linked to familial breast cancer or are non-functional in the HDR assay (Couch and Weber, 1996; Lee et al., 2015; Shattuck Eidens et al., 1995). Loss of BARD1 has been linked to increased susceptibility to hereditary breast and ovarian cancer (HBOC) and is associated with loss of tumor suppressor activity (Baer and Ludwig, 2002; Eyfjord et al., 2003; Hart et al., 2017; Levy-Lahad and Friedman, 2007; McCarthy et al., 2003;
Thai et al., 1998). The importance of BARD1 in cancer development indicates how significant it is to determine whether BARD1 VUS are benign or pathogenic.
A rise in the quantity of genomic data has led to an increasing number of VUS, uncertain 56 due to their low frequency and conflicting reports of pathogenicity. In this chapter, potentially pathogenic BARD1 variants were identified in a dataset of 10,389 cancer samples from 33 different cancers (Huang et al., 2018). From germline and somatic samples, 76 variants from across the entire BARD1 gene were identified as suggestive for being pathogenic. Analysis from the tumor sequencing data indicated BARD1 S339T,
T343I, V523A, N450H, G451fs, and L239Q had significantly increased LOH, suggesting that these variants were more likely to be pathogenic. However, with the exception of the
G451fs truncation variant, the other five variants were found to be functional in HDR.
Previous work (Lu et al., 2015) has proposed that HDR-deficient BRCA1 variants could be identified using filtering based on increased LOH. In contrast, we have found that increased LOH does not correlate with HDR deficiency for BARD1 variants. As we are unaware of other studies regarding the relationship between LOH and HDR-deficient
BRCA1 and BARD1 variants, these contrary results suggest that increased LOH is not a reliable indicator of non-functional variants. However, studies have shown that 60-90% of tumors with pathogenic germline BRCA1 or BRCA2 mutations have LOH (Johnson et al., 2002; Osorio et al., 2002). It is possible that the sequence analysis used to predict pathogenic BARD1 variants does not differentiate between driver and passenger mutations, making it difficult to filter for pathogenic variants using LOH. Data suggest that the HDR assay is a more effective method for identifying deficient variants. As the variant expression plasmids used only contain the mRNA coding sequence and we do not have access to patient samples, we cannot accurately examine the mRNA expression levels. However, if the mRNA expression levels of LOH mutants are lower than wild-
57 type BARD1, this could indicate an HDR deficiency that we do not observe in our protein expression experiments.
In addition, the variant A460T, which was non-functional in HDR, was identified as potentially pathogenic during the original analysis of exome sequencing data (Lu et al.,
2015) but was not considered pathogenic after updated analysis. These differences demonstrate the importance of empirical results, such as the HDR assay, for evaluating whether a given variant is predictive of cancer. Conversely, the results from functional assays may provide feedback for the improvement of the bioinformatic interpretation of genomic data. While large scale genomic analyses should be used to identify functionally significant variants, functional assays must still be utilized for more comprehensive characterization of these variants.
The 76 BARD1 missense variants were tested in the HDR assay to examine DNA repair function. Variants were considered non-functional if their HDR activity was significantly different from endogenous BARD1. Previous work in our lab (Lee et al., 2015) characterized fully non-functional variants as those with HDR activity significantly different from endogenous BARD1 but not empty vector, and intermediate variants as those significantly different from both endogenous BARD1 and empty vector. We have changed our classification standards based on the strength of the reproducibility in this study and for ease of analysis. We now interpret variants as functional or nonfunctional and do not include the intermediate phenotype. This interpretation is supported by the new data that the BARD1 A460T variant would have been ranked as intermediate using the previous interpretation, but assays measuring DNA damage sensitivity using IR and
58 cisplatin indicated that BARD1 A460T was just as sensitive as the V767fs variant, which scored similar to the empty vector in the HDR assay.
Based on BRCA1 variant function in HDR, it was anticipated that BARD1 variants with impact on HDR function would map to the RING and BRCT domains. In this study, we tested eight variants in the BARD1 RING domain and none of them were defective for
HDR. In a prior study (Lee et al., 2015), we had analyzed another nine variants in the
BARD1 RING domain; three of these were defective for DNA repair activity. In the current study, we tested 19 variants in the BARD1 BRCT domain, and five of these were defective. BARD1 T598I, S660R, G698D, P707S, and G753D were found to be non- functional in the BRCT domain. The BRCT domain has been shown to interact with the
HP1 protein in order to retain both the BRCA1-BARD1 complex and CtIP, which is involved in DNA end resection, at the damage site (Miyoshi et al., 2015). The BARD1-
HP1 interaction is also necessary for the accumulation of DNA helicase FANCJ at sites of DNA damage (Wu et al., 2016). BARD1 L570E/V571E and L570A/V571A variants have been shown to inhibit the interaction between BARD1 and HP1 (Miyoshi et al.,
2015). The BARD1 BRCT domain is also necessary for binding poly(ADP-ribose)
(PAR), allowing for rapid recruitment of the BRCA1-BARD1 complex to areas of DNA damage (Li and Yu, 2013). The variants K619A, C645R and V695L have been shown to disrupt BARD1-PAR interaction (Li and Yu, 2013). The T598 residue tested in this study
(T598I) is located on the surface of the protein next to K619, and could possibly affect
PAR binding. However, the relationship between BARD1-PAR binding and HDR is unclear since the BARD1 K619A and V695L variants, as well as several others that
59 disrupt PAR binding, have been shown to be functional in HDR (Billing et al., 2018;
Laufer et al., 2007). BRCA1 also binds to the BARD1 BRCT domain (Simons et al.,
2006), and previous work in our lab has identified that this binding is affected by the
BARD1 G623E variant (Lee et al., 2015). Most of the non-functional variants we identified in the BRCT domain, with the exception of T598I, are not located near known binding sites for proteins associated with DNA repair.
To our surprise, we found four variants, BARD1 A460T, L465F, L480S, and P530L, were identified as non-functional in the ankyrin repeat domain. Prior to this study, the ankyrin repeat domain had no known reported function in DNA repair. Previous work has shown that a large deletion of the ankyrin repeat domain results in chromosome instability and loss of HDR function (Laufer et al., 2007). In addition, the BARD1 ankyrin domain has been shown to interact with p53 to mediate apoptosis (Berardi et al.,
2005). The oncoprotein Bcl-3, which interacts with BARD1, is also involved in the regulation of NF-κB transcription via ankyrin repeat domain-associated protein interactions (Dechend et al., 1999). Our results indicate that the ankyrin repeat domain may have functions that are necessary for DNA repair. For both the ankyrin and BRCT domains, the non-functional variants identified may affect BARD1 folding and structure, which could also affect binding to proteins such as BRCA1 or HP1. For example, the
BARD1 P707 and G753 residues are located near one another on the surface of the protein. As coding substitutions at these amino acids result in loss of HDR they may be part of a binding pocket. The identified variants may also indicate binding sites for proteins whose interaction with BARD1 has not yet been discovered. The
60 characterization of non-functional BARD1 variants in areas that are not well-studied helps to further understand the roles of the ankyrin repeat and BRCT domains in homology- directed repair.
Many of the BARD1 variants tested have been recorded on ClinVar as having been isolated from patients with breast cancer or hereditary cancer-predisposing syndromes.
Variants S241C and E361D, which were functional in HDR, have been identified has likely benign, and truncation variants V154fs, S551*, and Q564* are likely pathogenic in
ClinVar. The variants V85L, R194K, I258T, N326S, R565H, and R641Q, which were functional in HDR, have conflicting reports of pathogenicity, with reports indicating they were VUS or likely benign. Since these variants were functional in the HDR assay, we would interpret such variants as likely benign. The non-functional truncating variant
V767fs also had conflicting reports of pathogenicity, with reports indicating that it was a
VUS or likely pathogenic. The trend observed in this chapter is supplemented by the 29 variants that were previously studied (Lee et al., 2015) —variants V507M and R658C were functional in the HDR assay and are considered benign in ClinVar, and several other functional variants are listed as having conflicting reports of pathogenicity because reports indicate they are VUS or likely benign. Previous work has shown that pathogenic
BRCA1 variants are non-functional in the HDR assay, and benign variants are functional
(Lu et al., 2015; Ransburgh et al., 2010; Towler et al., 2013). Based on the data from
ClinVar, this trend appears to be true for BARD1 variants as well, suggesting that the non-functional variants identified in this chapter would be likely pathogenic.
We also asked whether BARD1 HDR function affected cell sensitivity to DNA damage
61 agents. Cells expressing HDR-deficient variants A460T, P707S, G753D and V767fs, as well as endogenous-only, non-rescued cells depleted of BRCA1 or BARD1, were more sensitive to treatment with IR or cisplatin than cells expressing wild-type BARD1.
Testing non-rescued cells allowed us to set a standard for the effect of non-functional
DNA repair on damaged cells. We had hypothesized that the more HDR-deficient a variant was, the fewer colonies cells expressing that variant would form after damage, indicating a quantitative sensitivity to the DNA damage. The results did not support that expectation. Following treatment with IR or cisplatin, HDR-defective variants were in fact more sensitive to DNA damage, but they were as sensitive to DNA damage as cells depleted of BRCA1 or BARD1, suggesting that residual repair did not affect sensitivity.
The HDR assay indicates which variants are functional and non-functional, but it does not provide information on how this affects cell growth. This chapter has shown that loss of homologous recombination results in increased sensitivity to DNA damage, as indicated by decreased colony formation. The HDR results also reveal a correlation between BARD1 variants that cause a loss of DNA repair function with those that are likely cancer predisposing. While we examined in this study a large number of BARD1 variants across the length of the protein, including all three functional domains, many more BARD1 VUS exist. In future work, we hope to mutagenize the BARD1 functional domains on a larger scale, as we have previously done with the BRCA1 N-terminus
(Starita et al., 2018). Creating a library of all potential BARD1 variants in these functional domains and testing the HDR function of these variants would allow us to identify additional pathogenic variants and regions of interest. The work done in this study helps
62 better understand the role of BARD1 in DNA repair, and how loss of homology-directed repair affects cell growth and sensitivity.
Materials and Methods
Selection of BARD1 Variants: BARD1 missense variants conferring potential cancer predisposition were identified in a cohort of 4,034 samples from 12 cancer types (Lu et al., 2015) that was part of larger set of 10,389 TCGA samples from 33 cancer types
(Huang et al., 2018). Germline single nucleotide variants (SNVs) were identified with variant calling on whole exome sequencing data using GATK (McKenna et al., 2010)
(version 3.5, using its haplotype caller in single-sample mode with duplicate and unmapped reads removed and retaining calls with a minimum quality threshold of 10) and VarScan (Shen et al., 2012) (version 2.3.8 with default parameters, except where – min-var-freq 0.10,–p value 0.10,–min-coverage 3,–strand-filter 1) operating on a mpileup stream produced by SAMtools (version 1.2 with default parameters, except where -q 1 -Q
13). Germline indels were identified using VarScan and GATK (same parameters and version as above) in single-sample mode. Pindel (Ye et al., 2009) (version 0.2.5b8 with default parameters, except where -x 4, -I, -B 0, and -M 3 and excluded centromere regions (genome.ucsc.edu)) was also applied for indel prediction. For all analyses, the
GRCh37-lite reference was used and an insertion size of 500 was specified whenever this information was not provided in the BAM header.
All variants were limited to limited to coding regions of full-length transcripts obtained from Ensembl release 70 plus the additional two base pairs flanking each exon that cover 63 splice donor/acceptor sites. SNVs were based on the union of raw GATK and VarScan calls, while indels were required to be called by at least two out of the three variant callers (GATK, VarScan, Pindel). High-confidence, Pindel-unique calls (at least 30x coverage and 20% VAF) were also included. Further, variants were required to have an
Allelic Depth (AD) ≥ 5 for the alternative allele. Readcount analyses for variants passing these filters were performed in both normal and tumor samples using bam-readcount
(version 0.8.0 commit 1b9c52c, with parameters -q 10, -b 15) in order to quantify the number of both reference and alternative alleles. Variants were required to have at least 5 counts of the alternative allele and an alternative allele frequency of at least 20%. Of these, rare variants were filtered for, with ≤ 0.05 % minor allele frequency in 1000
Genomes and ExAC (release r0.3.1).
Variants passing manual review, with low allele frequencies (MAF < 0.05%), and significant LOH were prioritized for characterization. These variants, their cancer type distributions and frequencies are shown in the latest data release of the 10,389 samples
(Huang et al., 2018) to the research community on NCI Genome Data Commons
(https://gdc.cancer.gov/about-data/publications/PanCanAtlas-Germline-AWG).
Cloning of BARD1 Variants: For transient expression, BARD1 (NCBI Reference
Sequence: NM 000465.3) wild-type and missense variants were cloned into a pcDNA3 vector backbone containing a rabbit β-globin intron upstream of the BARD1 translation initiation site to drive expression of the 777 amino acid human BARD1 transgene.
Variants were cloned using the New England BioLabs Q5 Site-Directed Mutagenesis kit.
For stable integrations, BARD1 wild-type and missense variants were cloned into a
64 pcDNA5/FRT/TO vector backbone containing the His-Biotin-Tobacco Etch Virus (TEV)
(HBT) tag (Tagwerker et al., 2006). PCR reactions were done using PfuUltra II Fusion
HS DNA Polymerase. Vectors and inserts were ligated together using Gibson assembly
(Smith et al., 2009). Colonies with successful ligations were fully sequenced to confirm expected variants. All variants were verified using Sanger sequencing services provided
OSU Comprehensive Cancer Center (OSUCCC) Genomics Shared Resource.
Homology-Directed Repair (HDR) Assay: For examining the HDR function of transiently expressed of BARD1 variants, HeLa-DR-13-9 (HeLa-DR) cells were utilized.
HeLa-DR cells contain two non-functional GFP coding sequences, one of which is interrupted by an I-SceI restriction endonuclease site. Cells were cultured in DMEM media containing 1% penicillin/streptomycin, 1% GlutaMAX, 10% bovine serum, and
1.5 μg/ml puromycin. Cells were seeded in a 24-well plate and transfected with siRNA to the BARD1 3’-UTR (5’-AGCUGAAUAUUAUACCAGAdTdT-3’) or control siRNA (5 pmol), and BARD1 wild-type, variant, or pcDNA3 empty vector (300 ng). All transfections were carried out using Lipofectamine 2000 per the manufacturer’s recommendations. Cells were moved to 6-well plates 24 hours later. 48 hours after the first transfection, cells were transfected with 25 pmol siRNA, 750 ng DNA, and 750 ng of expression plasmid containing the restriction endonuclease I-SceI to induce a double- strand break. If HDR is functional, the break is repaired by gene conversion with the second GFP allele, and cells become GFP-positive (Pierce et al., 2001; Ransburgh et al.,
2010). 72 hours after the second transfection, cells were collected and GFP-positive cells were counted using the FACSCalibur in the OSUCCC Analytical Cytometry Shared
65
Resource. 10,000 cells were counted, and the remaining cells were used for immunoblotting. Cells transfected with BARD1 siRNA and BARD1 wild-type plasmid
(wild-type rescue), and cells treated with control siRNA and empty vector, served as positive controls. Cells treated with BARD1 siRNA and empty vector were used as a negative control. HDR activity, as defined by the percentage of GFP-positive cells, was normalized to wild-type rescue control and set to 1.
For examining the HDR function of stably integrated BARD1 variants, HeLaDR-FRT cells (Starita et al., 2018) were used. Cells integrated with pcDNA5-FRT/TO-HBT- tagged BARD1 wild-type or variants (A460T, P707S, G753D and V767fs) were seeded in
24-well plates. Cells not integrated with BARD1 variants were used as a negative control.
Cells were transfected with siRNA to the BARD1 3’-UTR or control siRNA. All transfections were carried out using Oligofectamine according to the manufacturer’s recommendations. Transfections were carried out on the same time pattern as detailed for transiently expressed BARD1. For 24-well transfections, 30 pmol siRNA was used, and for 6-well transfections, 50 pmol siRNA and 3 μg of I-SceI expression plasmid were used. Cells were collected and GFP-positive cells counted as detailed for transiently expressed BARD1. HDR activity, as defined by the percentage of GFP-positive cells, was normalized to cells treated with control siRNA for each individual cell line and set at 1.
Alignment of BARD1 Protein Sequences: Homo sapiens, Felis catus, Canis lupis familiaris, Mus musculus, Monodelphis domestica, and Ovis aries BARD1 protein sequences were aligned using Clustal Omega (Remmert et al., 2011) to examine conserved residues.
66
Integration of Variants into HeLaDR-FRT Cells: pcDNA5-FRT/TO-HBT-tagged
BARD1 wild-type and variants were co-transfected with plasmid expressing the flippase recombinase in a 1:2 ratio into HeLa-DR-FRT cells to induce integration at the flippase recognition target site (Craig, 1988; Sauer, 1994). Transfections were done according to
Lipofectamine 2000 manufacturer’s recommendations. 24 hours after transfection, cells were incubated at 30˚C for 24 hours and then moved back to 37˚C. Integrated cells were selected for with 550 μg/ml Hygromycin-B.
Immunoblotting: For BARD1 variants, replicates were combined, spun down at 1200 rpm for 5 minutes and resuspended in 150 μl of 1X LDS-PAGE dye. Samples were sonicated at 45% for 15 seconds three times. Sample was resolved on 6 or 8% SDS-
PAGE gels and transferred to PVDF membrane. Samples were probed with BARD1
(Bethyl, 1:1000) and BRCA1 (1:500) (Schlegel et al., 2003) antibodies. Antibodies to
RHA1 (1:20000) (Schlegel et al., 2003) and α-tubulin (Sigma, 1:20000) were used as loading controls. Membranes were incubated with fluorescent (LI-COR, 1:20000) or chemiluminescent (GE, 1:5000) rabbit and mouse secondary antibodies.
Clonogenic Assays: HeLaDR-FRT cells stably expressing pcDNA5-HBT-BARD1 WT,
A460T, P707S, G753D and V767fs, as well as control cells expressing only endogenous
BARD1, were seeded in 24-well dishes and transfected with 30 pmol of siRNA to the
BARD1 3’-UTR, BRCA1 3’-UTR or control siRNA. Transfections were done using
Oligofectamine as per the manufacturer’s protocol. Cells were transferred to 6-well dishes after 24 hours, and were treated with 50 pmol siRNA 48 hours after the first transfection. 48 hours after the second transfection, 1000 cells of each treatment
67 condition were plated in 10-cm dishes. Remaining cells were saved for immunoblotting to confirm knockdown. After 24 hours, cells were treated with either ionizing radiation
(IR) or cisplatin. Cells that were treated with IR were subjected to 0, 1, 2, 4, or 6 Gy using the RS 2000 X-Ray Irradiator. For cisplatin treatment, cells were treated with 0,
1,875, 3.75, 7.5, and 15 μM of cisplatin for 2 hours, after which cells were washed twice with 1X PBS and fresh media was added. Untreated cells were used as a control. After two weeks of growth at 37˚C, cells were fixed with cold methanol and stained with crystal violet. Dishes were coded to blind their treatment and cells were counted using
OpenCFU (Geissmann, 2013). The log value of the count was used for comparison.
Statistical Analysis: All BARD1 variants in the HDR and clonogenic assays were tested in triplicate. For HDR assays using transiently expressed BARD1 variants, HDR activity was normalized to wild-type rescue, which was set to 1. The Student’s t-test was applied to determine whether BARD1 variant HDR activity significantly differed (p < 0.01) from endogenous BARD1. Variants that were significantly different and below the cutoff of
0.6 were considered non-functional. For clonogenic assays, the Student’s t-test was carried out to examine whether BARD1 variant-expressing and endogenous-only cells treated with BARD1 or BRCA1 3’-UTR siRNA formed a significantly different number of colonies than variants treated with control siRNA and cells expressing BARD1 wild-type
(p < 0.05).
68
Chapter 3. Multiplex Homology-Directed Repair Analysis of BRCA1 C-terminus Missense Variants
Aleksandra I. Adamovich1, Mariame Diabate1, Tapahsama Banerjee1, Lea Starita2, and
Jeffrey D. Parvin1*
1 Department of Biomedical Informatics, Ohio State University Comprehensive Cancer
Center, Ohio State University, Columbus, OH
2 Department of Genome Sciences, University of Washington, Seattle, WA
Address correspondence to [email protected]
Author contributions: A.A. performed experiments and wrote the manuscript, A.A. and
M.D. performed the analysis, T.B. helped perform functional assays, L.S. and J.P. supervised experiments and J.P. edited the manuscript.
69
Introduction
Breast and ovarian cancers are two of the most prevalent cancers among women, with breast cancer posing a 13% lifetime risk and ovarian cancer a 1.3% lifetime risk according to the National Cancer Institute. Approximately 5-10% of breast cancer cases
(Larsen et al., 2014) and 23% of ovarian cancer cases (King et al., 2011) are hereditary breast and ovarian cancers (HBOCs) due to germline mutations in tumor suppressor genes. A plurality of HBOCs are caused by mutations in BRCA1 and BRCA2 (Jack et al.,
2002; Melchor and Benítez, 2013). Individuals with pathogenic BRCA1 variants have, on average, a 39% lifetime risk for ovarian cancer and 65% lifetime risk for breast cancer
(Eyfjord et al., 2003), which is much higher than the risk for the general population.
The 1863 amino acid BRCA1 protein comprises an amino-terminal RING domain, a
DNA binding domain (DBD), coiled coil (CC) domain, and two carboxy-terminal BRCT domains (Roy et al., 2012). BRCA1 forms an obligate heterodimer with BARD1, another cancer susceptibility gene (Couch et al., 2017), via their RING and BRCT domains to mediate homologous recombination (Moynahan et al., 1999; Shakya et al., 2011; Simons et al., 2006; Wu et al., 1996). Several phosphoproteins involved in DNA repair also interact with the BRCA1 BRCT domain. Binding to CtIP mediates DNA end resection
(Yun and Hiom, 2009), binding to BRIP1 mediates helicase activity in DNA repair
(Cantor et al., 2001; Clapperton et al., 2004), and Abraxas binding is associated with recruitment to DNA damage sites (Wang et al., 2007).
Although the genome aggregation database gnomAD (Lek et al., 2016) contains over four million missense variants, only 2% are assigned a clinical significance classification in 70 the ClinVar database (Malheiro et al., 2017). In addition, many ClinVar variants are considered variants of clinical significance (VUS) because there is not enough data to assign a clinical risk. Approximately 50% of known BRCA1 variants have conflicting reports of pathogenicity or are labeled as VUS on ClinVar (Landrum et al., 2018).
BRCA1 and BRCA2 VUS are also identified in 10-20% of cancer screenings (Eccles et al., 2015), signifying that many people remain uncertain of their cancer risk. As people with pathogenic BRCA1 mutations have a much higher chance of developing HBOC, individuals need to be accurately informed of their risk. This information would indicate whether women need to be more vigilant in their cancer screenings or perform preemptive and potentially invasive procedures such as mastectomies and salpingo- oophorectomies (Hartmann and Lindor, 2016) to reduce their risk.
One way to predict BRCA1-associated cancer risk is with functional assays, where loss of function is associated with pathogenicity. One such assay is the homology-directed repair (HDR) assay (Pierce et al., 2001; Ransburgh et al., 2010). The HDR assay utilizes a reporter cell line that contains non-functional GFP coding sequences, one of which is interrupted by a rare I-SceI restriction endonuclease site. Expression of the I-SceI expression plasmid results in a double-strand break in one of these GFP sequences. If the
BRCA1 variant expressed in the reporter cell is functional in homology-directed repair, the second GFP coding sequence will be used as a template to repair the double-strand break and the first GFP is rendered functional. After transfection of the I-SceI expression plasmid, 10–20% of the cells, quantified using flow cytometry, are converted to GFP- positive (Ransburgh et al., 2010). Previous work in our lab has examined over 130
71
BRCA1 variants in the HDR assay using a HeLa-derived cell clone known as HeLaDR
(Lu et al., 2015; Starita et al., 2015; Towler et al., 2013). With the exception of splicing variants, known pathogenic variants were non-functional in HDR while known benign variants were functional. Non-functional missense variants were located in the RING and
BRCT domains, suggesting that these two domains are the primary regions of BRCA1- mediated tumor suppression. Based on this information, we have concluded that BRCA1 pathogenicity correlates with loss of HDR function.
As BRCA1 mutations can cause cancer, it is very important to comprehensively identify which variants are cancer-associated. This would require testing all potential missense variants in the protein or the functional domains, which consists of hundreds or thousands of variants. The time and resources necessary to examine all these variants would be impossible in the singleton HDR assay, where all the cells in a population express one variant. We have been able to establish a multiplex version of the HDR assay in order to examine thousands of variants at once, and have identified the HDR function of more than 1000 variants in the 192 amino acids covering the amino-terminus of BRCA1, including the RING domain (Starita et al., 2018).
In this chapter, we generated high-coverage libraries of potential amino acid substitutions in the region encompassing BRCA1 amino acids 1577-1863, including the BRCT domain. We examined the HDR function of variants from residues 1577-1768 using the multiplex assay. We were able to identify with confidence the HDR function of over 500 variants from a population of cells integrated with over 2000 variants. Multiplex HDR assay variants that have also been tested in the singleton HDR assay had identical
72 functional classifications. Variant classifications from the multiplex assay were also compared to those from functional assays examining transcription activation (TA) and proliferation via saturation genome editing (SGE). The same variants were functional in all three assays, but several variants that were functional in the multiplex HDR assay were non-functional in the TA assay, SGE assay or both. Variants identified in the multiplex HDR assay were present in the ClinVar database as well. In addition to thirty
VUS, one benign variant and eleven pathogenic or likely pathogenic variants were identified. However, three known pathogenic variants – one splicing variant and two truncation variants – were functional in the HDR assay. Despite these inconsistencies, a majority of multiplex HDR variant classifications correlated with TA and SGE assay classifications and clinical data.
Generation of BRCA1 C-terminus Variant Libraries
In order to perform multiplex HDR on all potential amino acid substitutions in the
BRCA1 BRCT domain, variant library pools covering the last 287 codons of BRCA1
(amino acids 1577-1863) were generated. The BRCA1 C-terminus was cloned into the pUC19 vector and mutagenic inverse PCR was performed for each codon to generate all potential missense and nonsense variants for the region of interest (Jain and Varadarajan,
2014). The first 3 nucleotides of each codon-specific forward primer consisted of a hand- mixed NNN (amino acids 1577-1768) or NNK (amino acids 1769-1863) (N = A, T, C or
G; K = G or T) sequence to introduce all potential nucleotide variations at the codon site.
PCR products were pooled together and ligated back to themselves to produce a fragment 73 variant library. The mutagenized fragment was ligated back into full-length BRCA1 expression plasmid along with a degenerate 16-base barcode to create the final BRCA1 variant library pools. Libraries were sequenced in order to link variant sequences to barcodes and determine variant library diversity. Libraries had uniform variant abundance and had a minimum of 85% variant coverage and 11,000 barcodes (Figure 3-
1).
74
Figure 3-1. BRCA1 variant library diversity maps. The PacBio Sequel System was used to perform long read sequencing. Variants with zero or one amino acid changes were mapped to determine the diversity of the library. All potential amino acid changes, as well as nonsense mutations (as indicated by an asterisk at the top), are labeled on the y- axis. Variant libraries covered amino acids 1577-1672 (top), 1673-1768 (middle) and 1769-1863 (bottom). Variant abundance is indicated by color, with lighter colors representing less abundant variants and darker colors representing more abundant variants. Wild type amino acids are labeled in dark purple. Variants that were not present are shown in gray. Wild type sequence and position are indicated on the x-axis. The pool covering amino acids 1577-1672 contained 1769 unique variants (88% coverage) and 11,000 barcodes, the pool covering amino acids 1673-1768 contained 1712 unique variants (85% coverage) and 12,000 barcodes, and the pool covering amino acids 1769- 1863 had 1830 unique variants (92% coverage) and 18,000 barcodes.
75
Function of BRCA1 Variant Libraries in HDR
Variant libraries were integrated into HeLaDR-FRT cells (Starita et al., 2018) via FRT recombination (Craig, 1988) to create cell libraries with consistent variant expression at a single site in the genome. Variant cell libraries were treated with two rounds of transfection 48 hours apart with control siRNA or siRNA targeting the BRCA1 3′UTR to deplete endogenous BRCA1. Cells were also treated with siRNA targeting the BRCA1 coding sequence to deplete total BRCA1 expression and identify baseline GFP expression. The I-SceI restriction enzyme was also expressed during the second round of transfection to induce a double-strand break (Table 3-1). Four days after the first transfection, cells were sorted into GFP-positive and GFP-negative populations (Table 3-
2). Four HDR sorting replicates were performed for each variant cell library. The barcodes in the sorted cell populations were PCR-amplified and sequenced to count the number of times a barcode, and therefore a variant, occurred in the population. The variant frequency in a GFP-positive or GFP-negative sample was calculated using the