Mitotic homologous recombination maintains genomic stability and suppresses tumorigenesis

The central role of DNA strand exchange

The mechanism of strand exchange has recently been illuminated by crystal structure determinations of RecA filaments in complex with ssDNA and double-stranded DNA (dsDNA)¹⁸, which provided insight into how RecA and presumably RAD51 function (FIG. 2). As expected from previous studies, ssDNA within the RecA filament is stretched approximately 50% relative to B-form DNA; unexpectedly, the stretching is not uniform. Instead, the ssDNA has a repeating unit of three nucleotides, which maintains a B-form structure, whereas the DNA between every triplet is greatly stretched. The B-form structure of the triplets in the ssDNA is therefore well suited to pair through canonical Watson–Crick hydrogen bonds with complementary triplets in the donor duplex DNA. Although the structure of a RecA–ssDNA–dsDNA complex is not available, it is thought that the binding of duplex DNA to RecA disrupts duplex base stacking and base pairing to promote interaction with the invading ssDNA. Correct base pairing between the invading ssDNA and the complementary DNA provides the stable interaction between the two DNA strands, as RecA has few contacts with the complementary DNA to stabilize the interaction. Furthermore, the constrained B-form DNA in the filament excludes non-standard structures from forming with mismatched bases. This reliance on DNA–DNA interactions, rather than on protein–DNA interactions, works to ensure the fidelity of strand exchange. Finally, dissociation of the newly formed heteroduplex DNA and the displaced single strand is promoted by ATP hydrolysis.

End resection: a key first step in HR

The active HR intermediate for strand invasion is a RAD51–ssDNA nucleoprotein filament. The ssDNA used for strand invasion by RAD51 is generated by 5′ to 3′ DNA end resection, resulting in 3′ single-stranded tails (FIG. 1). In yeast, end resection occurs by a two-step mechanism: initial limited resection followed by processive resection. Limited resection involves the Mre11 complex and Sae2; more extensive resection involves either exodeoxyribonuclease 1 (Exo1; a 5′ to 3′ exonuclease) or the Sgs1 helicase (which unwinds duplex DNA) together with a nuclease to digest the 5′ strand^19,20. Experiments support a role for homologues of these proteins in end resection in mammalian cells as well. Human EXO1 resects DNA ends in vitro, and its activity is stimulated by Bloom’s syndrome protein (BLM), the human Sgs1 homologue; the resected ends can then be used by RAD51 in strand exchange reactions²¹. In addition, the nuclease activity of MRE11 promotes recruitment of the ssDNA-binding protein replication protein A (RPA) to sites of DSBs, presumably by promoting end resection to generate ssDNA²². Similarly, CtBP-interacting protein (CtIP; also known as RBBP8), which has limited homology to Sae2 and a binding partner of BRCA1 (REF. 23), also promotes RPA recruitment to DSBs²⁴. Studies in yeast have shown that DNA end resection is regulated during the cell cycle by cyclin-dependent kinase 1 (Cdk1; also known as Cdc28) (REFS 25,26). Sae2 and CtIP seem to be key targets for CDK phosphorylation in yeast and mammalian cells, respectively, limiting end resection to the S and G2 phases of the cell cycle^27,28.

RAD51 filament formation and beyond

Once DNA ends are resected, RPA efficiently binds ssDNA to melt DNA secondary structures that can interfere with the formation of RAD51 filaments. However, RPA also impedes RAD51 binding to ssDNA, such that RAD51 needs accessory factors — sometimes termed mediators — for filament formation²⁹. Several factors, including BRCA2 (see below), are involved in this process⁷. Once RAD51 nucleoprotein filaments form and strand invasion ensues, repair synthesis occurs from the invading DNA end, using the donor duplex as a template (FIG. 1). In the synthesis-dependent strand annealing (SDSA) pathway, the newly synthesized strand is displaced and anneals to the other DNA end to form a non-crossover. In the canonical DSB repair (DSBR) pathway, the second DNA end is ‘captured’ to form two Holliday junctions, which can either be dissolved to form a non-crossover (for example, by the BLM helicase³⁰) or resolved to form a crossover, for which several proteins have been implicated⁸. Crossovers are crucial in meiotic cells to generate haploid germ cells³¹; however, non-crossovers seem to be the more beneficial outcome of mitotic HR (see below).

DSB pathway choice: HR, NHEJ and SSA

HR is restricted to the S and G2 phases of the cell cycle by several factors, including the availability of sister chromatids, transcription of HR genes and CDK-mediated phosphorylation of HR proteins^28,32, whereas NHEJ functions throughout the cell cycle³³. A crucial determinant of DSB repair pathway choice during S and G2 phases is the requirement of a resected DNA end for HR. Binding of NHEJ components to DNA ends interferes with end resection³⁴; as a result of this competition for DNA ends, HR is increased in NHEJ mutants^35–37.

Efficient HR during S phase may have evolved to repair damage encountered during DNA replication^38,39. For example, replication of a nicked template gives rise to a DSB with just one end; this end is a substrate for repair by HR from the replicated unbroken strand. Because NHEJ requires two ends for joining, faithful repair of one-ended DSBs by NHEJ is not possible; instead, NHEJ of two one-ended DSBs would give rise to a genomic rearrangement. Thus, NHEJ has an important role only in maintaining genomic integrity in response to two-ended DSBs. Not surprisingly, HR and NHEJ are in competition with each other for the repair of two-ended, but not one-ended, DSBs³⁵.

In addition to HR and NHEJ, SSA provides a third DSB repair pathway in the context of sequence repeats. In this pathway, ssDNA that was formed after end resection at homologous repeats anneals, leading to the deletion of the intervening sequence (FIG. 1). Because SSA requires resected ends, it is also inhibited by NHEJ components^40,41. Unlike HR, SSA is a RAD51-independent pathway, and in fact is inhibited by HR components that work downstream of resection because the two pathways compete for the resected DNA ends⁴⁰.

Inter-sister versus inter-homologue HR

Accurately comparing the frequency of inter-sister and inter-homologue HR is challenging because inter-sister HR is genetically silent. Most estimates of HR rely on reporters consisting of homologous repeats that are located close by on the same chromosome (for inter-sister HR) or on the homologous chromosome (for inter-homologue HR), in which one repeat is targeted for DSB formation (BOX 1). Generally, HR between repeats on the same chromosome is efficient in both rodent and human cells, and can be as efficient as NHEJ^42–44. In mouse embryonic stem cells, HR between repeats on homologous chromosomes is approximately two orders of magnitude less efficient than that seen with tandem repeats^45,46. A similar 100-fold difference is also seen in a human lymphoblastoid cell line^47,48. It is perhaps not surprising that inter-homologue HR is so much less efficient, as sister chromatids are held in proximity by cohesion, whereas homologues are more distant from each other in the nuclear volume.

Non-crossover versus crossover HR

Inter-homologue HR does not lead to genomic rearrangement, but it has the potential to lead to genetic loss (that is, LOH) because information from one of the parental chromosomes is duplicated and information on the other parental chromosome is lost (FIG. 3b). The extent of genetic loss is minimal if HR results in a non-crossover gene conversion^46,49. By contrast, gene conversion associated with a crossover leads to LOH of the entire region of the chromosome that is distal to the HR event if recombinant sister chromatids segregate away from each other in mitosis.

The question that arises is what fraction of inter-homologue HR events is resolved as crossovers. In both mouse embryonic stem cells and human lymphoblastoid cells, non-crossover inter-homologue HR events predominate over crossovers, although the ratio of non-crossovers to crossovers ranges from 30 to 1 (REF. 46; J. Stark and M.J., unpublished observations) to 6 to1 (REF. 48), respectively. Thus, crossing over seems to be an infrequent outcome of mitotic HR. The low overall frequency of inter-homologue HR (1 in 100 HR events) coupled to a non-crossover bias reduces the probability that inter-homologue HR leads to genetic loss during DSB repair. Nonetheless, inter-homologue HR has an important role during the development of some tumours, the classic case being hereditary retinoblastoma, during which it is estimated that inter-homologue HR leads to loss of the wild-type retinoblastoma 1 (RB1) in about 40% of tumours⁵⁰. The contribution of inter-homologue HR to LOH of other tumour suppressor genes has not been studied in as much detail and is difficult to gauge in genetically unstable tumours.

HR in repeat-laden genomes

Repetitive elements and low copy number repeats that are present in different locations on the same chromosome or on different chromosomes can participate in non-allelic HR events (also known as ectopic HR events), with the potential to give rise to deletions, inversions and reciprocal translocations (FIG. 3c). Owing to the highly repetitive nature of mammalian genomes and the potential for genome rearrangements, investigators had initially considered it unlikely that HR could have an important role in DNA repair without scrambling the genome. However, sequence repeats have often highly diverged from each other, and even low levels of divergence substantially suppresses HR⁵¹. Moreover, when HR does occur between dispersed repeats, it rarely leads to genomic rearrangements because crossing over is a rare outcome of mitotic HR^52,53. More commonly, genomic rearrangements that arise from DSBs in the vicinity of two homologous repeats are repaired by SSA if the repeats are not highly divergent or by NHEJ if they are highly divergent⁵³ rather than HR. These mechanistic studies confirm the generally genome-stabilizing nature of HR. Ongoing genome instability, as would be found in HR mutants, may paradoxically lead to genetic reversion (BOX 2) in addition to deleterious mutations.

Studies of cancer genomes support the conclusion that NHEJ is more prone to giving rise to genomic rearrangements than HR^54,55. Recurrent chromosomal translocations are the initiating events in several tumour types, including several leukaemias, lymphomas, sarcomas and even prostate cancer. Some type of NHEJ is responsible for joining the two chromosome ends because translocation breakpoints have little or no sequence homology⁵⁴. More recent global analysis of rearrangements in lung cancer genomes using paired-end sequencing has also shown a predominance of non-homologous events⁵⁵.

Thus, restrictions on HR, including sequence heterology and crossover suppression, attenuate potentially deleterious HR events in the genomes of somatic cells, emphasizing the key role that HR has in promoting precise repair in lieu of mutagenic repair by other pathways. The question arises as to what benefit NHEJ has if it can be promiscuous in joining DNA ends. The canonical NHEJ pathway has an important role in maintaining genome integrity, especially in G1 or G0, when HR does not operate. Thus, loss of canonical NHEJ proteins, similarly to loss of HR proteins, results in genomic rearrangements⁵⁶. However, these rearrangements involve joining at non-homologous sequences, pointing to a non-canonical type of NHEJ for their formation^57–59, the normal role of which remains unclear.

PALB2 mutations and human disease

Several reports have provided evidence that PALB2 is a Fanconi anaemia protein and a tumour suppressor, although the frequency of PALB2 familial mutations is much lower than that of BRCA1 and BRCA2 (REFS 92–94). In studies of diverse populations, monoallelic truncating mutations of PALB2 have been identified in ~ 1% of familial breast cancer cases that do not have BRCA1 or BRCA2 mutations^94–96, and a founder mutation of PALB2 in the Finnish population occurs at a frequency of ~ 4% (REF. 97). Cancer risk is estimated to be increased fourfold in individuals with the Finnish founder mutation and approximately twofold in other populations with PALB2 mutations^94,96,98. The number of tumours analysed so far is too small to draw conclusions as to whether PALB2 mutations cause tumours that are similar to those seen in patients with BRCA2 mutations^94,97,99. However, surprisingly, out of the handful of tumours analysed only one PALB2 tumour has been reported with LOH of the wild-type PALB2 allele^94,97,100. PALB2 mutations have also been reported in male breast cancer and familial pancreatic cancer^96,100,101. Children with Fanconi anaemia N (which have a mutation in PALB2) have clinical similarities with children with Fanconi anaemia D1 in that they are at high risk for developing Wilms’ tumour and medulloblastoma within the first few years of life⁹³.

Truncating mutations found in tumours and in children with Fanconi anaemia N are located throughout the PALB2 coding sequence^92,93. The crystal structure of the BRCA2–PALB2 complex provides an explanation as to why even small deletions in the PALB2 C terminus are deleterious: they disrupt the β-propeller, rendering the protein susceptible to proteolysis⁸⁸. For example, the shortest deletion Tyr1183X (in which X denotes a stop codon) deletes only the last four amino acids, but this deletion disrupts the interaction between two β-strands that seal the last blade of the β-propeller (red–blue strand interface in FIG. 4) and prevents closure of the ring. Tumour-promoting missense mutations in BRCA1 and BRCA2 that abrogate PALB2 binding have been identified^87,102. Mutation of BRCA2 Trp31 disrupts the interaction of BRCA2 with a hydrophobic pocket on the outside of the β-propeller of PALB2 (REF. 88) (FIG. 4). Similarly, mutations in the coiled-coil domain of BRCA1 (Met1400Val, Leu1407Pro and Met1411Thr) abolish its interaction with PALB2 (REF. 102).

Abraxas — localization of BRCA1 to DNA damage

Post-translational modification of proteins by ubiquitylation of Lys residues is a dynamic regulatory process in various biological pathways, similar to phosphorylation. In DNA damage and repair, monoubiquitylation and Lys63 polyubiquitylation conjugates predominate¹¹¹. In a search for additional BRCA1 BRCT-interacting proteins, several groups identified new proteins, the binding of which depends on ubiquitylation. Abraxas binds directly to the BRCT domain, whereas RAP80 interacts with BRCA1 indirectly through abraxas¹⁰⁸. RAP80 binds to polyubiquitylated histone H2AX, modifications of which are important in the DNA damage response¹¹², thereby bringing BRCA1 to damaged DNA^108,113,114. The heterodimer UBC13–RNF8 (ubiquitin-conjugating enzyme 13–RING finger protein 8) carries out Lys63 polyubiquitylation of H2AX^115–118. UBC13-deficient cells have severe HR defects¹¹⁹, whereas cells depleted of abraxas or RAP80 exhibit mild HR defects¹⁰⁸, implicating additional functions of ubiquitylation in HR¹²⁰. Therefore, abraxas-mediated localization of BRCA1 has a role in the DNA damage response, although the less severe HR phenotype associated with abraxas knockdown suggests it is not essential for this pathway.

BRIP1 — a DNA helicase

BRIP1, a DNA helicase originally identified as BACH1 and later as FANCJ, binds to the BRCA1 BRCT repeats in S phase following its phosphorylation^121–123. Human BRIP1 interacts with BRCA1 at its C terminus, whereas the ATP-dependent helicase domain is in the N-terminal two thirds of the protein¹²⁴. BRIP1 depletion in human cells results in defective HR and hypersensitivity and chromosome instability following exposure to ICL-inducing agents¹²². Notably, in contrast to depletion of BRCA1, RAD51 focus formation is intact following exposure to both hydroxyurea and ICL-inducing agents^102,122. Interestingly, BRIP1-depleted cells have a greater degree of ICL hypersensitivity compared with BRCA1-depleted cells. BRIP1 is also required for timely progression through S phase; this and other functions of BRIP1 depend on its helicase activity^121,125,126.

Surprisingly, the worm and avian homologues of BRIP1 lack the C-terminal domain or the C-terminal pSerXXPhe residues, respectively, that are required for BRCA1 BRCT binding^125,127. Consistent with this, a physical interaction between the worm BRIP1 and BRCA1 orthologues is not detected. Despite the notable divergence, the worm and chicken BRIP1 mutants are hypersensitive to ICL-inducing agents, incur genetic instability after damage and monoubiquitylate FANCD2 (REFS 122,127). However, unlike human cells, chicken BRIP1 mutants show no decrease in HR and arrest in G2, a phenotype that is similar to that of Fanconi anaemia core complex mutants after ICL damage¹²⁵.

The worm BRIP1 homologue is known as DOG-1 (deletion of guanine-rich DNA 1) based on the observation that its loss results in deletions at poly-G tracts that have the ability to form non-canonical DNA structures which can impede replication¹²⁸. HR (and translesion synthesis) reduce the number of deletions, which supports the genome stabilizing effect of HR when unstable DNA structures are encountered¹²⁹. Importantly, resolution of four-stranded structures (G quadraplexes or G4 DNA) by purified BRIP1 was identified in human cell extracts¹³⁰. More recently, additional genome destabilizing phenotypes have been observed in worms, including chromosome rearrangements, large complex deletions, duplications and translocations¹³¹. In humans, array comparative genomic hybridization have identified large deletions with a bias towards adjacent G4 DNA regions in Fanconi anaemia J cells compared with Fanconi anaemia D2 and control cells¹³². BRIP1 may function to regulate HR by both DNA unwinding and displacement of proteins at the damaged site. Preliminary biochemical data suggest that excess BRIP1 may inhibit RAD51-mediated D-loop formation¹³³. The S phase-dependent, BRCA1-mediated association of BRIP1 with chromatin along with its DNA helicase substrate specificity for secondary DNA structures, including ICL damage, supports a role for BRIP1 in maintaining genome stability during replication-dependent HR.

BRIP1 has been identified as the deficient protein in Fanconi anaemia J subtype^{122,125,134,135}. Unlike the markedly severe clinical syndromes of patients with Fanconi anaemia D1 and Fanconi anaemia N described above, patients with Fanconi anaemia J typically present with growth and developmental anomalies and bone marrow failure occurring with an average latency of 4.5 years (ranging between 2–6 years)¹³⁵. One patient developed leukaemia at 13.5 years; however, no solid tumours have been observed. Notably, several early deaths occurred in utero and within the first days and weeks of life¹³⁵. Both truncating and missense mutations are identified in patients with Fanconi anaemia J, including the recurrent homozygous Arg798X truncation, which deletes part of the helicase domain and the BRCA1 BRCT-interacting domain. Clinical phenotypic variation is observed even with the recurrent homozygous Arg798X truncation. Similar to PALB2, BRIP1 has been implicated as a human breast tumour suppressor, and inheritance of a mutated BRIP1 allele is estimated to result in a twofold increased risk of breast cancer¹³⁶. So far, BRIP1 mutations have been identified in 12 patients with breast cancer. Because the mutation frequency is low (0.4% for the recurring Arg798X mutation in selected populations and 0.05% in unselected populations), there are limited data confirming the role of BRIP1 in tumour suppression^136,137.

CtIP — DNA end resection promotes HR

A key determinant for repair pathway choice between HR and NHEJ is the requirement for RAD51 to bind ssDNA, as discussed above. Resection of the DNA strand is regulated during the S and G2 phases of the cell cycle by CDK phosphorylation, which then allows binding of phosphorylated CtIP to the BRCA1 BRCT domain and subsequent 5′ to 3′ end resection of a DNA strand^23,24. This reaction also involves the MRE11–RAD50–NBS1 (MRN) complex^24,138. CtIP genetic loss is associated with early embryonic lethality in mice¹³⁹, and cells deficient in CtIP show HR defects^102,138,140. Further analysis of repair defects with single and combined RNA interference depletion experiments confirmed a mild HR repair defect with loss of BRIP1 or CtIP alone, moderately severe HR defects with depletion of BRCA1 or PALB2, and markedly severe defect with depletion of RAD51. Notably, combined knockdown of any two proteins resulted in severe HR defects, including knockdown of CtIP and BRIP1 (REF. 102). No inactivating CTIP mutations have been reported in human diseases.