Homologous Recombination Deficiency: Cancer Predispositions and Treatment Implications

Abstract

Homologous recombination (HR) is a highly accurate DNA repair mechanism. Several HR genes are established cancer susceptibility genes with clinically actionable pathogenic variants (PVs). Classically, BRCA1 and BRCA2 germline PVs are associated with significant breast and ovarian cancer risks. Patients with BRCA1 or BRCA2 PVs display worse clinical outcomes but respond better to platinum‐based chemotherapies and poly‐ADP ribose polymerase inhibitors, a trait termed “BRCAness.” With the advent of whole‐exome sequencing and multigene panels, PVs in other HR genes are increasingly identified among familial cancers. As such, several genes such as PALB2 are reclassified as cancer predisposition genes. But evidence for cancer risks remains unclear for many others. In this review, we will discuss cancer predispositions and treatment implications beyond BRCA1 and BRCA2, with a focus on 24 HR genes: 53BP1, ATM, ATR, ATRIP, BARD1, BLM, BRIP1, DMC1, MRE11A, NBN, PALB2, RAD50, RAD51, RAD51B, RAD51C, RAD51D, RIF1, RMI1, RMI2, RPA1, TOP3A, TOPBP1, XRCC2, and XRCC3.

Implications for Practice

This review provides a comprehensive reference for readers to quickly identify potential cancer predisposing homologous recombination (HR) genes, and to generate research questions for genes with inconclusive evidence. This review also evaluates the “BRCAness” of each HR member. Clinicians can refer to these discussions to identify potential candidates for future clinical trials.

Short abstract

This review discusses the cancer predispositions and treatment implications of homologous recombination genes, beyond BRCA1 and BRCA2.

Introduction

Essential to the repair of DNA double‐stranded breaks (DSBs), homologous recombination (HR) is a multistep DNA repair process involving several mediators, most notably BRCA1 and BRCA2. Alteration in HR genes is prevalent among many cancer types (13%–17%), especially breast, ovarian, and pancreatic cancers [1]. BRCA1 and BRCA2 are the most altered HR genes, with germline pathogenic variants (PVs) found in 2% to 3% of unselected tumors in several pan‐cancer analyses [1, 2, 3]. Both BRCA1 and BRCA2 have been extensively researched, and there are clear clinical guidelines on the testing and management of BRCA1 and BRCA2 PVs. Other HR genes also contribute to a significant proportion of familial cancers. In a recent analysis of 54,555 breast cancers, more than half of the germline PVs occurred in cancer predisposition genes other than BRCA1 and BRCA2 (BRCA1/BRCA2 PVs: 4.4% vs. the rest, e.g., ATM, BARD1, PALB2, RAD51C, RAD51D: 5.7%) [4]. Like those of BRCA1 and BRCA2, HR deficiency (HRD) secondary to PVs in the other HR genes has been associated with the treatment success in platinum‐based regimens and poly‐ADP ribose polymerase (PARP) inhibitors (PARPis). Ongoing studies also reveal that these non–BRCA1 or BRCA2 HR mediators may be potential targets for combination therapies with platinum agents or PARPis. In this review, we will discuss the role of other HR players, beyond BRCA1 and BRCA2, in cancer predisposition and treatment.

Materials and Methods

Homologous recombination genes were selected based on the authors’ literature search. References for this review were identified through searches of Google Scholar with the search terms “(gene name),” “germline/somatic variants,” “prevalence,” “spectrum,” “multigene panel,” “(cancer type),” “PARP inhibitors,” “platinum,” and “BRCAness” with an emphasis on articles published between 2016 and 2020. Papers published in journals not indexed in PubMed were excluded. Abstracts published in established conferences were also reviewed. Only papers published in English were reviewed. The final reference list was generated based on originality and relevance to the broad scope of this review.

The Homologous Recombination Pathway

HR provides accurate DSB repair during normal meiosis and after exposure to cytotoxic chemicals or ionizing radiation (Fig. 1). Members of the HR pathway can be categorized into sensors and effectors. DSBs are first sensed by the MRE11‐RAD50‐NBN (MRN) complex, which loads BLM helicase and EXO1 onto the breaks to start 5'–3' double‐stranded DNA (dsDNA) resection. The single‐stranded DNA (ssDNA) overhangs are coated by high‐affinity RPA, which, via negative feedback, suppresses further resections. ATR localizes to the RPA‐coated ssDNA via RPA‐ATRIP interaction and switches on the ATR‐Chk1 DNA damage checkpoint, which arrests cell cycle and protects stalled replication forks [5]. Working with ATR is ATM, a central regulator kinase responsible for phosphorylating and activating several downstream substrates including BRCA1.

Figure 1.

Homologous recombination repair pathway. Ionizing radiation or chemotherapy–induced double‐stranded breaks (DSBs) activate the 53BP1, which stabilizes the chromatin at the DSB site via a 53BP1‐RIF1‐shieldin‐CST complex. This inhibits DNA resection and promotes nonhomologous end joining. BRCA1 competes with 53BP1 for preferential activation of homologous recombination (HR). The DSBs are then sensed by the MRE11‐RAD50‐NBN complex, which loads BLM helicase and EXO1 nuclease onto the double‐stranded DNA (dsDNA) ends to commence 5'‐3' DNA resection. Once the single‐stranded DNA (ssDNA) overhangs are formed, they are coated by RPA, which suppresses further dsDNA resection. ATR then localizes to the ssDNA ends and switches on the ATR‐dependent checkpoint with the aid of ATRIP and TopBP1, arresting the cell cycle for HR to proceed. RAD51 assembles onto the ssDNA, a rate‐limiting step mediated by BRCA1/BARD1, BRCA2/PALB2, and the RAD51 paralogs. The presynaptic filament then undergoes homology search, strand invasion, and DNA synthesis and ligation. Finally, the BTRR complex resolves the Holliday junctions and completes HR.

Following cell cycle arrest, HR continues with the assembly of presynaptic filament, homology search, and strand invasion. In a rate‐limiting step, RAD51 displaces RPA from the 3' overhangs to form presynaptic filaments, a process mediated by several HR effectors. BRCA2 and PALB2 colocalize RAD51 into the nucleus, displace RPA, and load RAD51 onto the resected ssDNA ends [5]. Once formed, the RAD51 nucleofilament probes the sister chromatid for homology and invades the homologous dsDNA, giving rise to a displacement loop. Using the intact chromatid as a template, both resected ends undergo DNA synthesis and ligation. The Holliday junctions linking the sister chromatids are resolved by the BTRR dissolvasome, completing HR repair [5].

DNA Damage Sensors

Located most upstream in the HR pathway are ATM and ATR. Both are phosphatidylinositol‐3‐kinase‐like kinases (PIKKs) responsible for phosphorylating downstream HR effectors. ATM is a moderate penetrance gene. Individuals with biallelic loss of ATM can develop ataxia telangiectasia, an autosomal recessive neurodegenerative disease with increased cancer risks (particularly lymphoid malignancies) and increased radiation sensitivity. Monoallelic ATM germline PVs exist in approximately 1% to 2% of the adult population [6]. Over 80% of these PVs encode truncated proteins (with the loss of PIKK domain), whereas fewer than 15% are missense nonsynonymous variants [7]. Heterozygotes are two to four times more likely to develop cancers. ATM PVs account for up to 17% of non–BRCA1 or BRCA2 PVs in patients with breast cancer (see Table 1 for summarized risk estimates of ATM PVs) [8, 9, 10]. A breast cancer odds ratio (OR) of 11.0 (95% confidence interval [CI], 1.4–85.7) was even reported for c.7271T > G, with the caveat of a wide inferential error bar [11]. Individuals with ATM germline PVs are also more likely to develop contralateral breast cancer than noncarriers (relative risk [RR], 3.3; 95% CI, 1.4–8.0) [12]. The cumulative lifetime risk for ATM‐associated breast cancer increases with age, with 6% risk by age 50 and 33% risk by age 80 [13]. Other ATM‐related malignancies include pancreatic (OR, 5.7–9.0; p < .05) and prostate cancers (OR, 2.9; 95% CI, 1.6–5.2) [14, 15, 16]. Carriers should undergo early screening for breast and pancreatic cancers. The National Comprehensive Cancer Network recommends annual mammogram starting at age 40 and screening magnetic resonance cholangiopancreatography in carriers with a positive family history of pancreatic cancers [17, 18]. Early cancer detection may facilitate prompt treatment and improve overall survival (OS).

Table 1.

Summary of the HR genes, germline variants, and cancer predispositions

Gene	Role in HR	Distribution of germline PVs (obtained from ClinVar)	Cancer predisposition (pathogenic variant frequency; risk estimate of monoallelic germline PVs)	Hereditary cancer syndromes
53BP1	DSB repair pathway choice	Nonsense (n = 1/1)	Not known	Not known
ATM	Activate downstream effectors	Nonsense (n = 26/26)	Breast cancer (0.6%; 3.3) [9] (0.9%; 1.7) [80] (0.5%; 2.2a) [61] (1.0%; 2.8) [125] Colorectal cancer (0.7%; 2.8) [48] Ovarian cancer (0.9%; 1.7) [80] (0.9%; 2.3) [96] (0.2%; 0.9a) [126] (0.6%; 2.4) [127] Pancreatic cancer (2.3%; 5.7) [14] (3.4%; 9.0) [16] Prostate cancer (0.5%; 2.9) [15]	Ataxia telangiectasia
ATR	Activates cell cycle checkpoint	Nonsense (n = 2/2)	Breast cancer (0.2%; 3.0a) [61]	Seckel syndrome
ATRIP b	Mediates ATR functions	Nonsense (n = 2/2)	Not known	Seckel syndrome
BARD1	Mediates BRCA1 functions and retains BRCA1 in the nucleus	Nonsense (n = 4/4)	Breast cancer (0.2%; 3.2) [9] (0.3%; 1.9) [80] (0.4%; 2.2) [94] (0.2%; 3.0a) [61] (0.2%; 2.2) [125] Colorectal cancer (0.2%; 3.1a) [48] Ovarian cancer (0.1%; 0.6a) [80] (0.1%; 1.3a) [96] (0.2%; 4.2) [127]	Fanconi anemia
BLM b	Unwinds dsDNA and regulates RAD51 foci formation, part of BTRR complex	Nonsense (n = 2/2)	Breast cancer (0.3%; 1.7a) [61] (1.1%; 6.3) [128] Colorectal cancer (0.4%; 3.1a) [48] (1.6%; 8.7a) [129] (1.9%; 2.3) [130] Ovarian cancer (0.5%; 1.1a) [131] Prostate cancer (0.4%; 0.7a) [132] (0.6%; 24.3) [133]	Bloom syndrome
BRIP1 b	Mediates BRCA1 functions, facilitates ssDNA extension, RPA loading, and D‐loop unwinding	Nonsense (n = 9/9)	Breast cancer (0.3%; 1.2a) [80] (0.4%: 1.3a) [134] (0.3%; 1.6) [125] Colorectal cancer (0.3%; 1.9a) [48] Ovarian cancer (0.7%; 2.6) [80] (1.0%; 5.0) [96] (2.4%; 8.1) [134] (0.09%; 1.0a) [135] (1.0%; 6.4) [127] Pancreatic cancer (0.1%; 0.8a) [14] (1.2%; 2.7) [134] Prostate cancer (0.1%; 1.4a) [15]	Fanconi anemia
DMC1 b	Meiosis‐specific recombinase	Not known	Not known	Not known
MRE11A	Exonuclease, part of MRN complex	Not known	Breast cancer (0.2%; 9.0a) [61] (0.02%; 0.6a) [9] (0.1%; 0.9a) [125] Colorectal cancer (0.3%; 4.5a) [48] Ovarian cancer (0.1%; 1.2a) [96] (0.1%; 0.8a) [127] Pancreatic cancer (0.1%; 0.7a) [14]	Ataxia telangiectasia‐like disorder
NBN	Recruits ATM and ATR to ssDNA, part of MRN complex	Not known	Breast cancer (0.02%; 0.3a) [9] (2%; 3.2) [57] (0.2%; 1.3a) [80] (0.1%; 0.7a) [61] Lymphoid malignancies (0.6%; 1.4–1.8) [58] Ovarian cancer (0.3%; 1.9) [80] (0.5%; 2.3a) [127] Pancreatic cancer (0.1%; 0.9a) [14] Prostate cancer (0.1%; 1.2a) [15] (2.2%; 3.9) [59]	Nijmegen breakage syndrome
PALB2	Localize BRCA2 and recruits RAD51 to ssDNA	Nonsense (n = 17/17)	Breast cancer (1.0%; 3.4) [80] (3.2%; 7.2) [102] (1.3%; 6.6) [61] (0.9%; 7.5) [125] (NA; 9.5) [136] Colorectal cancer (0.4%; 4.9) [48] (0.04%; 1.0a) [102] Ovarian cancer (0.4%; 1.6a) [80] (0.4%; 3.1) [96] (0.2%; 2.9) [102] (0.6%; 4.4) [127] Pancreatic cancer (0.4%; 2.3) [14] (1.6%; 14.8) [16] (0.1%; 2.4) [102] Prostate cancer (0.1%; 1.6a) [15] (0.1%; 0.4) [102]	Fanconi anemia
RAD50	Part of MRN complex	Nonsense (n = 2/2)	Breast cancer (0.1%; 0.5a) [61] (0.2%; 0.8a) [125] Ovarian cancer (0.3%; 0.9a) [96] (0.2%; 0.7a) [127]	Nijmegen break syndrome‐like disorder
RAD51	Recombinase	Not known	Not known	Fanconi anemia
RAD51B	Part of BCDX2 complex	Not known	Breast cancer risk (0.2%; 1.0a) [83]	Not known
RAD51C	Part of BCDX2 and CX3 complexes	Nonsense (n = 2/2)	Breast cancer (0.02%; 0.4a) [9] (0.2%; 1.4a) [80] (0.1%: 0.8a) [125] Ovarian cancer (0.3%; 5.2) [79] (0.6%; 5.0) [80] (0.8%; 5.1) [96] (0.6%; 3.4) [127] Pancreatic cancer (0.1%; 1.1a) [14]	Fanconi anemia
RAD51D	Part of BCDX2 complex	Nonsense (n = 1/1)	Breast cancer (0.1%; 8.3) [9] (0.1%; 1.4a) [80] (0.1%; 3.1) [125] Ovarian cancer (0.3%; 6.3) [96] (0.9%; 6.3) [81] (0.6%; 4.8) [80] (0.6%; 10.9) [127]	Not known
RIF1 b	Part of 53BP1‐RIF1‐shieldin‐CST complex	Nonsense (n = 4/4)	Not known	Not known
RMI1 b	Part of BTRR complex	Not known	Not known	Not known
RMI2 b	Part of BTRR complex	Not known	Not known	Not known
RPA1 b	Stabilizes ssDNA	Not known	Not known	Not known
TOP3A b	Topoisomerase, part of BTRR complex	Not known	Not known	Bloom syndrome‐like disorder
TOPBP1	Topoisomerase, scaffold protein for ATR activation	Not known	Not known	Not known
XRCC2	Part of BCDX2 complex	Nonsense (n = 2/2)	Breast cancer (0.02%; 1.3a) [9] (0.07%; 0.9a) [86] (0.1%; 5.1a) [61]	Not known
XRCC3	Part of CX3 complex	Not known	Colorectal cancer (0.2%; 6.8a) [48]	Not known

^a Risk estimate is not statistically significant.

^b Refer to supplemental online data for an extended discussion of these HR genes (BLM, DMC1, TOP3A, RMI1, and RMI2).

Abbreviations: DSB, double‐stranded break; dsDNA, double‐stranded DNA; HR, homologous recombination; MRN complex, MRE11‐RAD50‐NBN complex; NA, not available; PV, pathologic variant; ssDNA, single‐stranded DNA.

Tumor expression of ATM can influence the choice of treatment. ATM‐deficient patients have better response to radiotherapy, albeit with more severe adverse reactions [19]. Besides radiotherapy, platinum‐based chemotherapy and PARPi exhibit preferential responses in ATM‐deficient tumors [20]. In a randomized controlled trial of 124 patients with metastatic gastric cancers, the addition of olaparib to paclitaxel exhibited a trend toward a more pronounced OS benefit among ATM‐deficient patients versus the population (hazard ratio, 0.4 vs. 0.6; p value unavailable) [21]. Leveraging on the success of ATM deficiency, several research groups have developed ATM inhibitors with ongoing clinical trials [22].

Another major sensor of DNA damage is ATR. Hypomorphic ATR PVs can manifest as growth retardation and microcephaly in Seckel syndrome, an autosomal recessive neurodegenerative syndrome [23]. Despite ATR's importance in cell cycle arrest and HR repair, in vitro studies show only a modest cancer incidence in ATR/Chk1 mutants [24]. Human carriers of ATR PVs also lack distinct cancer predisposition [24]. Paradoxically, ATR underexpression confers resistance to tumorigenesis [24]. ATR‐deficient cancer cells destabilize and undergo p53‐independent cell death when forced into mitosis [24]. Exploiting this phenomenon, researchers inhibited ATR and resensitized chemoresistant cancer cells to platinum agents and PARPi [23, 25]. Many clinical trials are underway to confirm these in vitro observations [23].

53BP1 is a DNA damage sensor that works with BRCA1 to decide on the repair pathway of DSBs. Phosphorylated by ATM, 53BP1 stabilizes the DSBs via a 53BP1‐RIF1‐shieldin‐CST complex and inhibits DNA end‐resection. Nonhomologous end joining (NHEJ) then commences [26]. 53BP1 promotes NHEJ in the G1 phase, with antagonism from BRCA1 in the S/G2 phase [26]. 53BP1 has other roles in apoptosis, cell cycle arrest, and autophagy [27]. Despite 53BP1's ubiquitous role, its germline PVs do not cause cancer predisposition. Rapakko et al. screened 126 families with histories of breast or ovarian cancer for 53BP1 PVs and found 11 germline variants (five missense nonsynonymous and one truncating variant, c.1347–1352delTATCCC) [28]. These germline variants were not situated in functional domains, and none were associated with increased cancer risks. A meta‐analysis of c.1074C > G, a common 53BP1 polymorphism (allele frequency 0.474) with conflicting evidence for cancer predisposition, concluded that this variant did not influence cancer risks [29]. Aberrant tumor 53BP1 expression has both prognostic and predictive significance. Loss of 53BP1 is common in triple‐negative breast cancers (TNBCs) (90% TNBCs vs. 10% non‐TNBCs; p < .0001) [30]. It also correlates with poorer prognosis and higher metastasis [30]. In vitro, 53BP1 loss rescues HR defect in BRCA1‐deficient cells [30]. It is possible that tumors downregulate 53BP1 for better cell survival against chemotherapy. Recently, 53BP1 accumulation in circulating tumor cells identified patients with chemotherapy‐responsive metastatic breast cancer [30].

Proper functioning of the DNA damage sensors requires scaffold proteins such as TopBP1. TopBP1 contains eight BRCT domains for protein‐protein interactions. These are crucial to cell cycle checkpoint, p53 binding, and the ATR‐Chk1 pathway [31]. Overexpression of TopBP1 deregulates p53 function and inhibits both p53‐mediated apoptosis and G1/S arrest [32]. TopBP1‐overexpressed tumors show aggressive features such as high‐grade breast carcinomas (66.7% vs. 35.5% grade 3; p = .007), advanced sarcomas (hazard ratio for stage III–IV sarcomas, 3.4; 95% CI, 2.0–5.6), lung metastasis (hazard ratio, 5.1; 95% CI, 3.1–8.6), with reduced survival (OS in breast cancer, 40 vs. 165 months; p = .003 and hazard ratio in sarcoma, 2.2; 95% CI, 1.3–3.7) [32, 33, 34]. Excessive TopBP1 induces chemoresistance against platinum agents, whereas TopBP1 depletion enhances PARPi efficacy by inhibiting RAD51 loading to ssDNA ends [35, 36], although TopBP1 has both prognostic and predictive significance, but not cancer predisposition. Karppinen et al. identified 19 TopBP1 germline variants in 125 families with breast or ovarian cancers [37]. Only two were missense nonsynonymous, and neither caused functional impairment [37]. The rest were either intronic (n = 9), synonymous missense variants (n = 1), or single nucleotide polymorphisms (n = 7) [37]. No truncating variant was identified. We will need larger studies to delineate the cancer susceptibility of TopBP1 germline PVs.

The MRN complex is both a sensor and an effector of DNA damage [38]. Its MRE11 subunit is an exonuclease with DNA‐binding domains for localizing the complex to the ssDNA ends, whereas RAD50 contains an ATPase domain for ATP hydrolysis and MRE11 exposure. Finally, the NBN subunit binds and recruits both ATM and ATR onto the ssDNA ends [38]. Collectively, MRE11A, RAD50, and NBN are postulated cancer susceptibility genes with conflicting penetrance estimates [38].

Biallelic MRE11A germline variants may give rise to ataxia telangiectasia‐like disorder (ATLD), an autosomal recessive neurodegenerative disease with cerebellar ataxia and radiation hypersensitivity. Unlike ataxia telangiectasia, patients with ATLD do not have telangiectasia, obvious cancer risks, or immunodeficiency [39]. A case report identified compound heterozygous MRE11A germline variants in two brothers with ATLD and lung adenocarcinoma: c.727T > C (nonsynonymous missense) and g.24994G > A (splicing variant). Patient‐derived cells of these patients showed reduced MRE11 expression, unlike those of their parents and sibling, who were carriers of a monoallelic MRE11A germline variant [40]. Beyond this case report, no data are available on the cancer predisposition of ATLD. Monoallelic MRE11A germline variants have been implicated in familial breast and ovarian cancers (Table 1) [41]. However, MRE11A germline PVs are rare and confer minimal and inconsistent cancer risks. Two well‐studied germline variants, c.442A > G and c.2501A > G, were genotyped in 315 patients with breast cancer, but neither variant was enriched in breast cancer [42]. Another study discovered an MRE11 polymorphism (rs569143), which raised breast cancer risk nonsignificantly by 30% [43]. In contrast, a large‐scale analysis of the MRN genes and breast cancer susceptibility identified 23 missense variants and one truncating variant (c.3852del4) with an increased breast cancer risk (OR, 3.2; p = .012) [44]. This discrepancy in breast cancer risk may be due to the different variants involved in the studies or to the fact that MRE11A variants are best interpreted using a multiplicative polygenic model for low‐to‐moderate penetrance genes. MRE11‐deficient tumors are more often TNBCs (30.6% vs. 12.3%; p < .0005), of higher grade (50.7% vs. 29.0%; p = .0004), and more likely to metastasize (70% vs. 0%; p = .026) than those with normal MRE11 expression [41, 45]. Excessive MRE11 expression is detrimental, as suggested by another in vitro study in which high MRE11 expression correlated with more advanced breast cancer stage and poorer OS [46]. To reconcile these contradictory observations, we hypothesize that MRE11 exists predominantly as the MRN complex at lower or physiological levels, beyond which it exists and functions independently. Beyond HR, MRE11 is involved in pathways such as signal transducer and activator of transcription (STAT) 3 signaling and matrix metalloproteinase (MMP) activation [46]. When overexpressed, MRE11 upregulates the STAT‐MMP axis, which promotes cell proliferation, extracellular matrix degradation, and metastasis [46]. Comparable risk estimates were seen in gastric (RR, 4.2; 95% CI, 2.0–9.8) and colorectal cancers (OR, 4.5; 95% CI, 0.8–19.4) [47, 48]. High nuclear MRE11 is predictive for poor chemotherapy response (progressive disease, 21.8% vs. 9.1%; p = .044), and MRE11‐depleted tumors are more sensitive to PARPi [47, 49].

Similar to MRE11A, biallelic RAD50 germline variants are related to an autosomal recessive genetic disorder, Nijmegen break syndrome (NBS)–like disorder [38, 50]. These patients share NBS features of microcephaly, growth retardation, short stature, and bird‐like facies, but their immune system is intact. One extensively studied RAD50 germline PV is c.687delT, which gave rise to RAD50's cancer predisposition. Yet, current evidence for its cancer predisposition role is weak. This truncating variant was sequenced in two patients with hereditary breast or ovarian cancers. However, it was also carried by the unaffected family members, and no tumor loss of heterozygosity (LOH) was seen in the affected patients [51]. To ascertain its pathogenicity, Tommiska et al. screened another 1,025 families with familial breast cancers and found it in three families (0.5%) with incomplete segregation. Breast tumors of patients with variant c.687delT displayed normal RAD50 immunostaining, bringing into question the pathogenicity of this variant. For now, the penetrance and clinical significance of RAD50 germline PVs appear to be very low, if any [52]. As opposed to germline variants, RAD50 somatic PVs are implicated in prognosis and sensitivity to platinum drugs and PARPi. Prognosis of patients with RAD50‐depleted ovarian tumors exceeds those of RAD50 wild‐type tumors (5‐year survival of 41% vs. 22%) [53]. Conversely, RAD50 overexpression correlates with poorer disease‐free survival in gastric cancers (RR, 3.73; 95% CI, 1.22–8.17), prostate cancers (hazard ratio, 9.8; p = .025), and melanoma. RAD50‐depleted ovarian tumors are also more sensitive to platinum drugs and PARPi. Silencing RAD50 in the presence of ATM deficiency achieves synthetic lethality; ATM/RAD50 double‐mutant tumors are 500‐fold more sensitive to topoisomerase I inhibition than either of the single mutants [53, 54].

Specific to eukaryotes, NBN serves as the protein recruitment module of the MRN complex. It comprises two BRCT domains for binding to MRE11/RAD50 and ATM/ATR. Hypomorphic NBN variants impede MRE11 nuclear localization and ATM/ATR activation [38, 55]. Patients with biallelic NBN PVs can develop NBS, an autosomal recessive disorder characterized by immunodeficiency and cancer predisposition with up to 40% of patients developing cancer before age 21 [38]. Almost 90% of all patients with NBS harbor the c.675del5 variant, which results in the loss of one BRCT domain [56]. This germline variant is also the most common monoallelic NBN variant associated with cancer risk, with a frequency of one in 100 to 200 among Slavic populations [57]. It predisposes patients to breast (OR, 3.2; p = .01), prostate cancers (OR, 3.9; p = .01), and various childhood hematological malignancies (OR, 1.4–1.8; p < .05) [57, 58, 59]. Prognosis is worse for carriers with prostate (5‐year survival of 49% vs. 72%) and breast cancers (hazard ratio, 1.5; p = .19) [57, 60]. A caveat to these associations is that they mostly stem from Slavic population studies and are not seen in other cohorts (Table 1); for instance, panel testing of an Australian cohort of patients with breast cancer found more NBN PVs among controls than cases (OR, 0.7; 95% CI, 0.1–4.0) [61]. High tumor NBN expression correlates with a lower OS in gastric (RR, 2.8; 95% CI, 1.2–8.0) and ovarian cancers (recurrence‐free survival of 30 for carriers vs. 78 months for noncarriers) [47, 62]. Like MRE11‐ and RAD50‐deficient cells, NBN‐deficient cancer cells are more sensitive to platinum drugs and PARPi [63, 64].

DNA Damage Effectors

Central to HR, RAD51 encircles the ssDNA as a helical nucleoprotein filament, forming presynaptic filaments for the recruitment and activation of other DNA damage effectors. Structurally, RAD51 constitutes a linker peptide (for oligomerization) and a RecA homology domain (containing Walker A and B motifs for ATP‐dependent strand exchange, and DNA/protein‐binding surfaces) [65]. Biallelic RAD51 inactivation leads to chromosomal instability and early embryonic lethality, whereas dominant‐negative RAD51 variants accelerate tumorigenesis nearly twofold in vitro [66]. Only two RAD51 germline variants (R150Q and E258A) have been widely studied. Kato et al. identified R150Q in two patients with bilateral breast cancers but were unable to justify its cancer causative role [67]. Functional analysis of this variant revealed that both its ATPase and oligomerization functions were unchanged. Only a modest decrease in DNA‐binding ability was noted, casting doubt on its pathogenicity [68]. The other variant, E258A, exhibited defective DNA‐binding activity, ATPase activity, and strand exchange in vitro. This germline variant was sequenced in healthy populations (from the 1000 Genomes Project) and is absent from all tumors in the Cancer Genome Atlas (TCGA) database [65]. Twelve RAD51 germline variants were identified in a cohort of 1,330 patients with early‐onset breast cancer, but none were predicted to be pathogenic [69]. In contrast, RAD51 somatic variants are more commonly found to be pathogenic: F86L and G151D (in breast cancers), Q268P (in lung cancer), and Q272L (in kidney cancer) [65, 70, 71]. F86L occurs upstream of the RecA homology domain, G151D between the Walker motifs, and Q268P and Q272L within the RecA homology domain, and all four variants were defective in their ssDNA‐binding and ATPase functions [65, 70]. Other PVs (c.391A > C and A293T), arising de novo, have been described in patients with Fanconi anemia (FA)–like phenotypes and associated with aberrant ATPase activity [72, 73]. Evident from the above studies, RAD51 somatic PVs can be highly deleterious, and the lack of germline PVs may be due to the indispensability of RAD51 [69]. In terms of therapeutic implications, increased tumor RAD51 expression correlates with more aggressive cancer phenotypes and treatment resistance to platinum agents or PARPi [74]. More RAD51 foci were found in breast cancers of higher grades, size, and nodal invasion and predicted for a shorter OS (hazard ratio, 7.3; p = .013) [75, 76]. Downregulating RAD51 may sensitize BRCA1‐mutant tumors to PARPi and overcome the platinum chemoresistance; a comprehensive list of preclinical and clinical RAD51 inhibitors was summarized by Ward et al. [74].

Structurally resembling RAD51, the paralogs share a 20% to 30% amino acid homology, including the Walker A and B motifs [77]. These paralogs assemble into two main functional complexes, BCDX2 and CX3 [77]. Each complex serves a nonredundant role in mediating RAD51 functions [77]. Independent of RAD51, they maintain replication fork stability and repair interstrand crosslinks, without which would lead to FA [77]. Patients with FA present with congenital malformations or physical stigmata, bone marrow failure, and early cancer onset [77]. Likewise, monoallelic germline PVs are linked with cancer predisposition, in particular, breast and ovarian cancers. In a study of 2,649 patients with breast or ovarian cancers, 21 RAD51 paralog PVs were identified in 30 patients [78]. The majority occurred in RAD51C (n = 12); the rest were as follows: RAD51B (n = 4), RAD51D (n = 7), XRCC2 (n = 2), and XRCC3 (n = 5) [78]. RAD51C is the first paralog accepted as an ovarian cancer predisposition gene (OR, 5.2; 95%, CI, 1.1–24.0), and carriers may consider prophylactic salpingo‐oophorectomy [79]. Meanwhile, RAD51D germline PVs are associated with increased ovarian cancer risks (OR, 4.8–6.3; p < .05) and share the same management recommendations as those of RAD51C [80, 81]. Both RAD51C and RAD51D germline PVs have been assessed for breast cancer risk with promising but conflicting evidence (Table 1); for example, RAD51C germline PVs are strongly associated with TNBCs (OR, 27.33; 95% CI, 5.2–268.5), but this needs to be corroborated in future studies [82].

Cancer susceptibility of the other RAD51 paralogs is less defined. Because of low variant frequencies, studies on these RAD51 paralogs were underpowered to establish risk estimates. In the study by Golmard et al., rates of RAD51B, XRCC2, and XRCC3 PVs among 2,649 patients with breast or ovarian cancers were 0.15%, 0.10%, and 0.19% [78]. A case‐control study on unselected ovarian cancer identified a single RAD51B splicing variant (c.854‐2A > G) in only two of 3,401 cases (0.06%) but was unable to conclude its cancer risk [79]. Another screening study detected only one RAD51B missense variant (c.541C > T) [83]. In silico algorithms predicted this variant as pathogenic, but it did not carry any breast cancer risk (OR, 1.0; 95% CI, 0.8–1.3) [83]. More recently, a large study on 9,287 patients with cancer identified RAD51B germline PVs in 11 patients (0.12%), of whom the majority (n = 8) had breast or ovarian cancers [84]. All the functionally tested breast and ovarian tumors possessed both tumor LOH and HRD features of high large‐scale state transition (LST) scores and a dominant mutation signature 3 [84]. This provides preliminary evidence of RAD51B as a breast and ovarian cancer predisposition gene [84]. Like RA51B, XRCC2 and XRCC3 germline PVs are uncommon and have uncertain clinical significance. An XRCC2 founder variant (c.96delT) reported among Polish patients with breast cancer did not affect breast cancer risks (OR, 0.8; 95% CI, 0.3–2.7) [85]. Another study of 13,087 patients with breast cancer also failed to find any statistically significant risk estimates for XRCC2 germline PVs (OR, 0.9; 95% CI, 0.3–4.2) [86]. Even XRCC2 Arg188His, a well‐studied variant once thought to increase ovarian cancer risk, is now considered a benign variant for its high allele frequency of 0.0636 and in silico predictions of a benign variant effect [87]. Several XRCC3 germline variants, including c.722C > T, have been reported in the literature, but none were associated with cancer risks [88, 89]. Although RAD51B, XRCC2, and XRCC3 are increasingly included in multigene panels, more functional and epidemiologic evidence is required to confirm their cancer predispositions.

Among the paralogs, RAD51C and RAD51D knockout mice incurred more RAD51 recruitment defects than XRCC2 and XRCC3 knockouts and were more sensitive to cross‐linking agents and cisplatin [90]. Conversely, secondary gain‐of‐function somatic variants can contribute to PARPi resistance. Kondrashova et al. found PARPi resistance in five of 12 tumor pairs (pretreatment and postprogression ovarian carcinomas) [91]. The tumors harbored secondary RAD51C and RAD51D somatic variants, restoring the reading frame in original truncating variants. The loss of any RAD51 paralog can sensitize tumors to platinum agents and PARPi and vice versa [90]. Hence, it might still be worthwhile to screen for germline and somatic variants of RAD51 paralogs, notwithstanding their rarity.

After RAD51 loading, strand exchange proceeds with the aid of BRCA1‐BARD1 heterodimer. Sharing a high degree of structural and functional homology with BRCA1, BARD1 heterodimerizes with BRCA1 to mask the BRCA1 nuclear export signal and retains BRCA1 in the nucleus [92]. BARD1 is a low‐to‐moderate penetrance breast cancer gene, and its PVs are rare in unselected breast cancers (<1%) [93]. BARD1 germline PVs increase breast cancer risk by two to threefold (OR, 2.2–3.2; p < .05), more pronounced among TNBCs (OR, 3.6–5.92; p < .05) [9, 94, 95]. Among BRCA1 and BRCA2–negative TNBCs, BARD1 germline PVs are the second most common after PALB2 variants with a prevalence ranging from 0.2% to 0.7% (Table 1) [95]. BARD1's role in ovarian cancer risk is still unconfirmed. Limited by the low variant prevalence in patients with ovarian cancer (0.12%–0.14%), several groups failed to demonstrate any significant ovarian cancer risk with BARD1 germline PVs [80, 94, 96]. BARD1 somatic variants are rare, with only 14 somatic variants recorded in the TCGA data set of 10,389 cancer samples. Six of 14 harbored tumor LOH, and a separate analysis of the COSMIC (Catalogue of Somatic Mutations in Cancer) database concluded that BARD1 somatic PVs were likely driver variants [97, 98]. Beyond its cancer predisposition, BARD1 may be a potential target for overcoming PARPi resistance. BARD1‐deficient tumors have increased PARPi‐induced DSBs via BRCA1 delocalization [99].

Responsible for loading RAD51 onto the ssDNA, PALB2 also stimulates RAD51‐mediated strand exchange and D‐loop formation via the BRCA1‐PALB2‐BRCA2 complex (Fig. 1). Depleting PALB2 can even cause a complete loss of RAD51 foci [100]. Cancers associated with PALB2 germline PVs display characteristic features of HRD such as mutational signature 3 and high LST score [101]. PALB2 has other roles in maintaining genome integrity, including stabilizing replication forks (known as FANCN). Carriers of biallelic PALB2 PVs can develop FA and the related malignancies. Monoallelic PALB2 PVs are associated with increased risk for many cancer types: breast (RR, 7.2; 95% CI, 5.8–8.9), ovarian (RR, 2.9; 95% CI, 1.4–6.0), pancreatic (RR, 2.4; 95% CI, 1.2–4.5), and potentially gastric cancers (no risk estimates are available) [102, 103]. Patients with PALB2 somatic PVs have longer OS than those without (24.6 vs. 18.8 months; p < .05) [104]. When treated early with platinum‐based therapies, these patients remained in remission for longer durations (progression‐free survival of PALB2 PV carriers vs. noncarriers: 21.1 vs. 7.9 months) [104]. Efficacy of PARPi has been demonstrated in vitro but has not been individually studied in clinical trials [105]. A randomized phase II trial evaluated the effect of veliparib on patients with pancreatic adenocarcinoma and BRCA1/BRCA2/PALB2 germline PVs and did not find any difference in OS when compared with the standard cisplatin/gemcitabine regimen (24.3 vs. 23.4 months; p = .60) [106]. There were too few patients with PALB2 germline PVs (n = 3) to perform subgroup analysis [106]. In a recent case report, a 37‐year‐old woman with recurrent hormone receptor–positive, HER2‐negative metastatic breast cancer, refractory to hormonal therapy and palbociclib, showed a response of at least 8 months to olaparib [107]. The PARPi sensitivity was attributed to her PALB2 germline PV (exon 1, c.18G > T), which led to a splicing defect and a truncated protein. Treatment response to olaparib was also observed in a 42‐year‐old woman with hormone receptor–positive, HER2‐negative metastatic breast cancer and a PALB2 germline PV (c.1653T > A) coding for a stop codon [108]. Larger trials with PALB2 subgroup analyses will be needed to clarify the role of PARPi in PALB2‐associated cancers.

Defining and Targeting “BRCAness”

Any defects in HR can lead to genome instability (i.e., DNA deletions and chromosomal rearrangements) and trigger cell apoptosis. Tumors with inherited or sporadic HRDs are more susceptible to DNA damage from platinum agents and PARPi, a trait known as “BRCAness” [109]. There are several ways to measure “BRCAness”: HRD, LST scores, tumor mutational signatures, and immunohistochemical detection of nuclear RAD51 [109, 110]. Besides RAD51, we have introduced in this review several candidate genes whose aberrant expression can result in similar “BRCAness” phenotypes. Although some candidates are predictive of cancer risks and “BRCAness,” the others are still of uncertain significance (refer to supplemental online data).

Outstanding Questions

Several questions are still unanswered: Do PVs in these seemingly nonrelevant HR genes exert additive effects on the cancer risks or “BRCAness” of PVs in known cancer predisposition genes? Do they express cancer risks or “BRCAness” in a digenic or oligogenic pattern? If so, is there any difference between the HR sensors and effectors? Digenic or oligogenic inheritance has been demonstrated in hereditary breast and colorectal cancers [111]. Morak et al. reported a case of early‐onset metastatic colorectal cancer in a patient who inherited a maternal MUTYH PV and a paternal OGG1 PV. Both MUTYH and OGG1 are involved in the mismatch repair pathway, and the lack of cancer risks in the parents raises the possibility of digenic inheritance in colorectal cancer [112]. Perhaps the clinical relevance of PVs in the low‐to‐moderate penetrance HR genes may be better elucidated when studied as digenic or oligogenic inheritance.

Besides germline and somatic PVs, external factors such as tumor hypoxia and oncogenic viral or bacterial infections can induce contextual HRD. Intratumoral hypoxia induces the downregulation of several HR effectors and accelerates mutagenesis [113]. In solid cancers, a high burden of tumor hypoxia is prognostic for poorer survival outcomes and aggressive tumor behavior (e.g., metastasis) [114]. Oncogenic strains of human papillomavirus have also been shown to impede the HR pathway both by mislocalizing RAD51and by mistiming the repair process outside of S/G2 phases, causing cancers of the cervix and oropharynx [115]. Helicobacter pylori is another carcinogenic microorganism responsible for gastric cancers; it impairs HR by creating DSBs and downregulating HR genes [116]. Incidentally, platinum‐based therapies are well‐established first‐line regimens for metastatic cervical and oropharyngeal cancers [117, 118]. The questions then arise: Does the clinical outcome vary with the degree of contextual HRD in these cancers? Is there a role for PARPi in these cancers? Several trials are ongoing to investigate the efficacy of PARPi in metastatic cervical (NCT03795272, NCT03476798, NCT03644342) and gastric cancers (NCT03008278; lack of OS benefit in the GOLD trial, NCT01924533) [119, 120].

Targeted therapy with PARPi revolves around the concept of synthetic lethality, which explains how cell death occurs from combined defects in two genes with minimal individual effect on cell survival [121]. PARPi traps PARP1 at sites of DNA damage and stalls the replication fork. In the absence of HR, NHEJ is activated, and the resultant genomic instability will trigger a cell death response. More than a decade passed from the discovery of PARPi's synthetic lethality to its approval for cancer treatment. Unfortunately, the success of PARPi was interrupted by increasing reports of PARPi resistance. To overcome the resistance, PARPis have been used in combination with other targeted inhibitors such as ATR inhibitors; refer to Slade [122] and Pilié et al. [123] for the ongoing clinical trials of PARPi and combination therapies. Altering the tumor microenvironment (e.g., inducing tumor hypoxia with prodrugs such as tirapazamine) might also restore the synthetic lethality of PARPi [124].

Conclusion

Historically, cancer management is largely based on the clinicopathological features of the tumor and organ involved. With the advent of genomic analysis and targeted therapies, the underlying genetic alteration of the tumor is becoming relevant to its classification and treatment. Tumor HRD serves as a double‐edged sword, correlating with a worse prognosis but a better response to DNA damaging agents. Although most inheritable HRDs involve BRCA1 and BRCA2, PVs in non‐BRCA HR genes contribute a significant part of tumor “BRCAness.” As our mechanistic understanding of HRD and synthetic lethality evolves, there will be more “druggable” HR targets for a wider spectrum of HR‐deficient tumors.

Author Contributions

Data analysis and interpretation: Ming Ren Toh, Joanne Ngeow

Manuscript writing: Ming Ren Toh, Joanne Ngeow

Final approval of manuscript: Ming Ren Toh, Joanne Ngeow

Disclosures

The authors indicated no financial relationships.

Supporting information

See http://www.TheOncologist.com for supplemental material available online.

Appendix S1. Supporting Information.