During genetic recombination and the recombinational repair of chromosome breaks, DNA molecules become linked at points of strand exchange. Branch migration and resolution of these crossovers, or Holliday junctions (HJs), complete the recombination process. Here, we show that extracts from cells carrying mutations in the recombination/repair genes RAD51C or XRCC3 have reduced levels of HJ resolvase activity. Moreover, depletion of RAD51C from fractionated human extracts caused a loss of branch migration and resolution activity, but these functions were restored by complementation with a variety of RAD51 paralog complexes containing RAD51C. We conclude that the RAD51 paralogs are involved in HJ processing in human cells.

The use of cultured mammalian cell lines selected for sensitivity to DNA-damaging agents to identify genes and gene functions has had a major impact on our understanding of DNA repair pathways. In particular, components of the nucleotide excision repair and DNA end joining pathways were identified and their cellular responses characterized using DNA damagesensitive cell lines (1, 2). Recently, the value of this approach has been reinforced by the identification of genes from the homologous recombination repair pathway, with the cloning of the XRCC2 and XRCC3 genes by their ability to complement sensitive mammalian cell lines (3, 4).The repair of DNA damage by homologous recombination is important for the maintenance of genetic stability in cells. The RAD51 protein is central to the recombination process, and this protein is highly conserved from yeast to humans (5). Using molecular recombination assays, yeast (Saccharomyces cerevisiae) and human RAD51 proteins have been shown to promote DNA strand exchange. In S. cerevisiae, two RAD51-like proteins, Rad55p and Rad57p, form a heterodimer and stimulate Rad51-mediated recombination reactions (6). Yeast mutants that lack either Rad51 or these Rad51-like recombination proteins are extremely sensitive to agents causing severe forms of damage to DNA, such as doublestrand breaks and interstrand cross-links (7). Rad55p and Rad57p also have counterparts in mammalian cells, including XRCC2, XRCC3, RAD51L1 1 (hREC2, RAD51B, R51H2), RAD51L2 (RAD51C), and RAD51L3 (R51H3, RAD51D) (8). The last three proteins were identified through data base searches using partial homologies to RAD51-like proteins (9-13), and have not as yet been found to be defective in DNA damage-sensitive mammalian cell lines. At present very little is known about the functions of these mammalian RAD51-like proteins, although specific protein-protein interactions have been described which suggest that they form heterodimers and larger complexes that may help recruit RAD51 to sites of DNA damage (14-22).A series of DNA damage-sensitive hamster cell lines, termed the irs mutants, were previously isolated in this laboratory (23). We and others used the irs1 cell line to clone the XRCC2 gene by its complementing ability for sensitivity to the potent DNA cross-linking agent, mitomycin-C (24 -26). The irs1 line is also sensitive to other DNA-damaging agents, including ionizing radiation, ultraviolet light, and alkylating agents (23). It shows spontaneous genetic instability, with increased frequencies of mutations (27), chromosomal aberrations (28), and chromosome non-disjunction (29). It has also been shown that the repair of a site-specific double-strand break by homologous recombination is severely reduced in irs1 compared with the paternal V79 cells (30). We have recently established a functional link between XRCC2 and RAD51 by showing that the irs1 line is defective in the formation of damage-dependent RAD51 focus formation (31).The irs3 cell line is another member of this series of damagesensitive cell line...

In germ line cells, recombination is required for gene reassortment and proper chromosome segregation at meiosis, whereas in somatic cells it provides an important mechanism for the repair of DNA double-strand breaks. Five proteins (RAD51B, RAD51C, RAD51D, XRCC2, and XRCC3) that share homology with RAD51 recombinase and are known as the RAD51 paralogs are important for recombinational repair, as paralog-defective cell lines exhibit spontaneous chromosomal aberrations, defective DNA repair, and reduced gene targeting. The paralogs form two distinct protein complexes, RAD51B-RAD51C-RAD51D-XRCC2 and RAD51C-XRCC3, but their precise cellular roles remain unknown. Here, we show that, like MLH1, RAD51C localized to mouse meiotic chromosomes at pachytene/diplotene. Using immunoprecipitation and gel filtration analyses, we found that Holliday junction resolvase activity associated tightly and co-eluted with the 80-kDa RAD51C-XRCC3 complex. Taken together, these data indicate that the RAD51C-XRCC3-associated Holliday junction resolvase complex associates with crossovers and may play an essential role in the resolution of recombination intermediates prior to chromosome segregation.

An acquired genetic instability, resulting from the loss of some types of DNA repair, is an early event in the development of a subset of human cancers. The involvement of BRCA1 and BRCA2 in the homologous recombination repair (HRR) of double-strand breaks in DNA implicates this pathway in the suppression of breast cancer. A family of proteins related to human RAD51, including XRCC2, are essential components of this repair pathway. Using site-directed mutagenesis of XRCC2, we show that non-conservative substitution or deletion of amino acid 188 of XRCC2 can significantly affect cellular sensitivity to DNA damage, and that a polymorphic variant at this site (R188H ), present on 6% of chromosomes in the population, has a weak effect on damage sensitivity. We tested the hypothesis that the R188H polymorphism could be a low-penetrance susceptibility factor for breast cancer, by genotyping 521 women with breast cancer and a total of 895 control women. Carriage of the rare allele of XRCC2 R188H was associated with breast cancer overall [odds ratio 1.3; 95% confidence interval (CI) ¼ (1.0, 1.8)] and when younger-onset cases with a positive family history were compared with older controls with no family history [odds ratio 1.9; 95% CI ¼ (1.0, 3.8)]. These results support the hypothesis that subtle variation in DNA repair capacity may influence cancer susceptibility in the population.

The human XRCC2 gene was recently identified by its ability to complement a hamster cell line, irs1, which is sensitive to DNA-damaging agents and shows genetic instability. The XRCC2 protein is highly conserved in mammalian species and has structural features, including a putative ATP-binding domain (P-loop), consistent with membership of the RecA/RAD51 family of recombination-repair proteins. We show that a hybrid XRCC2-green fluorescent protein, which was found to be functional by complementation, localizes to the nucleus. We have established a functional link between XRCC2 and RAD51 by looking at damage-dependent RAD51 focus formation in the irs1 cell line. Little or no formation of RAD51 foci occurred in irs1. This effect was specific to the loss of XRCC2 because transfection of the gene into irs1 restored normal levels of focus formation. Surprisingly, XRCC2 genes carrying site-specific mutations in P-loop residues were found to be able to complement the XRCC2-deficient irs1 line for a number of different end points. We conclude that XRCC2 is important in the early stages of homologous recombination in mammalian cells to facilitate RAD51-dependent recombination repair but that it does not make use of ATP binding to promote this function.The repair of DNA damage by homologous recombination has an important function in maintaining genetic stability in cells. In bacteria, the RecA protein has a central role in the recombination process, and in the last decade RecA-like proteins have been discovered in eukaryotes. In particular the RAD51 protein is highly conserved from yeast to humans and has been shown to have similar attributes to RecA (1, 2). Mutations in both RecA and RAD51 cause severe defects in recombination and extreme sensitivity to DNA-damaging agents. RecA acts directly in recombination processes in which, in the presence of ATP, it forms a polymer on single-stranded DNA and promotes strand exchange with a homologous sequence (3). Using molecular recombination assays, yeast (Saccharomyces cerevisiae) and human RAD51 proteins have been shown to promote strand exchange similarly, although some of the biochemical properties of RAD51 differ from those of RecA.For example, purified RecA preferentially binds to singlestranded DNA and hydrolyzes ATP at a relatively high rate, whereas the yeast and human RAD51 proteins bind equally to single-and double-stranded DNA and show a much lower rate of ATP hydrolysis (1). All members of this family have a highly conserved sequence motif, first described by Walker et al. (4), that has been linked to ATP binding. The flexible loop of this motif (Walker box A) interacts with the phosphates of ATP and is therefore sometimes called the P-loop.In S. cerevisiae, two further members of the RecA/ RAD51family of proteins facilitate homologous recombination in mitotic cells; Rad55p and Rad57p form a heterodimer and stimulate RAD51-mediated recombination (5). Yeast mutants that lack these recombination proteins are also extremely sensitive to agents causing severe forms of da...

BioBERT and BioMegatron are Transformers models adapted for the biomedical domain based on publicly available biomedical corpora. As such, they have the potential to encode large-scale biological knowledge. We investigate the encoding and representation of biological knowledge in these models, and its potential utility to support inference in cancer precision medicine -namely, the interpretation of the clinical significance of genomic alterations. We compare the performance of different transformer baselines; we use probing to determine the consistency of encodings for distinct entities; and we use clustering methods to compare and contrast the internal properties of the embeddings for genes, variants, drugs and diseases. We show that these models do indeed encode biological knowledge, although some of this is lost in fine-tuning for specific tasks. Finally, we analyse how the models behave with regard to biases and imbalances in the dataset.

No abstract