The five candidate breast cancer cell lines (HCC1599, BT-549, Hs578T, MDA-MB-231 and MDA-MB-453) were seeded in 6-well plates (4,000 cells/well) and incubated in Dulbecco's modified Eagle's medium (DMEM) or Roswell Park Memorial Institute-1640 (RPMI-1640) medium with 10% fetal bovine serum (FBS). Drug-containing medium (50 µM olaparib) was refreshed weekly (25). A total of 10⁶ cells were digested with trypsin after 24 h of culturing, followed by centrifugation at 110 × g for 5 min, and then cleaned twice and centrifuged at 110 × g for a further 5 min. The collected cells were fixed with 1 ml ethanol at 4°C overnight. Then, 100 µl of RNase A (Solarbio) was added to incubate the cells at 37°C in the dark for 30 min. Propidium iodide (PI) (50 µl) was then added. After further incubation in the dark for 1 h at 25°C, the cells were detected by flow cytometry (Becton, Dickinson and Company). Finally, the flow cytometric data were analyzed using FlowJo software version 10 (Tree Star, Inc.). All the cell culture media and related reagent were purchased from Thermo Fisher Scientific, Inc.

Five sgRNAs (sgRNA1-1, sgRNA1-2, sgRNA2-1, sgRNA2-2, and sgRNA2-3) were selected in the WALKER A and B Rad51 domains using an online predictor tool CCTOP (http://crispr.cos.uni-heidelberg.de) (26). The sgRNAs were then cloned into a human codon-optimized Streptococcus (S. pyogenous) Cas9-sgRNA vector (pX330-U6-Chimeric_BB-CBh-hSpCas9; Addgene plasmid #42230, Addgene). There were three reasons to choose this cell line. i) The function of Rad51 is closely linked with Brca1/2 whose gene products act upstream of Rad51 in HR (27). In order to eliminate interference influence of the BRCA gene status, the candidate cell lines do not carry the Brca1/2 mutations, as indicated by the American Type Culture Collection (ATCC) document (characterization data and other information relating to this material are available at www.atcc.org). ii) Given the high cost and off-target ratio of CRISPR/Cas9-based experiments, it is sensible to choose cells which have been used in similar experiments, as is the case for BT-549 cells (28). iii) Cells should have high sensitivity to olaparib, and in the pilot study (described above; see also Results and Fig. S1) BT-549 cells showed the highest apoptosis ratio induced by olaparib. Thus, BT-549 cells were selected as the in vitro cell model for this study. Five sgRNAs (the top sgRNAs scored by CCTOP) for four targeted sites were selected for each predicted locus and were cloned into the CRISPR/Cas9 vector (Table I). The constructed vector was packaged in lentiviruses, which were then transfected into BT-549 cells. One microgram of the Cas9-sgRNA vector was used to transfect BT-549 cells by electroporation. Briefly, the 2×10⁶ cells with Cas9-sgRNA vector complex per reaction were used for electroporation following the standard instruction of the Neon™ Transfection System (cat. no. MPK10025, Thermo Fisher Scientific, Inc.). The program (1600 V/10 msec/3 pulses) was adopted for electroporation at room temperature. After the electroporation, the cells were immediately transferred to a 24-well plate containing 0.5 ml of pre-warmed culture medium for recovery at 37°C. After 48 h of recovery, genomic DNA was extracted for polymerase chain reaction (PCR). The cleavage efficiency was evaluated using the T7E1 cleavage assay (NEB) as described previously (27). Potential off-target sites (OTS) were predicted using the CCTOP online prediction tool, amplified from the genomic DNA of the Rad51 KO BT_549, and sequenced. Bulk PCR products can also be sequenced using a next generation sequencing (NGS) approach followed by software analysis. The off-target ratios were indicated by the off-target/total reads count. Potential off-target sites were predicted using the CCTOP online predicting tool and then amplified from the genomic DNA extracted from Rad51-knockout (KO) BT-549 cells. The variants were identified by sequence alignment. The primers used to amplify the OTS are listed in Table SI.

Total proteins were extracted by using RIPA Lysis Buffer (cat. no. P0013E; Beyotime Institute of Biotechnology). The BCA Protein Assay kit (cat. no. P0012S; Beyotime Institute of Biotechnology) was used to determine the concentration of protein extracted from each sample. SDS-PAGE (on 10% gel) was then adopted to isolate the protein (40 µg) from each sample after which samples were removed to a polyvinylidene fluoride (PVDF) membrane. The film was then blocked for 1 h in 5% nonfat milk at 4°C. The membrane was then incubated in primary monoclonal and secondary antibodies (all from Abcam) against GAPDH (cat. no. ab8245, dilution 1:1,000), pro-caspase-3 (cat. no. ab179517, dilution 1:1,000), cleaved caspase 3 (cat. no. ab2302, dilution 1:500), Bcl-2 (cat. no. ab32124, dilution 1:500), Bax (cat. no. ab205718, dilution 1:500) for 2 h at 4°C. The secondary antibodies were anti-rabbit (cat. no. ab205718, dilution 1:1,000) conjugated with horseradish peroxidase for 1 h at room temperature. Lastly, the signals were visualized using an enhanced ECL Western blotting kit (Abcam), and the optical density of each band was analyzed by Gel-Pro-Analyze software (version 4.5; Media Cybernetics, Inc.).

Given the high number of variants in the homology-directed repair (HDR)-edited loci, evaluation of each variant in gDNA from a population of unedited cells would require a large amount of sequencing and sufficient cells. To improve the HDR efficiency, we introduced mutations between the protospacer adjacent motif (PAM) and the protospacer sequences, which would prevent Cas9 from recutting HDR-edited genomes. Based on a previous study, we designed PCR-selective sites in the sequence (17). An HDR library containing all 6,859 possible DNA residues substituted at positions +37 to +40 and +52 to 55 of Rad51 exon 5 and +17 to +19 of exon 8 was constructed using series of oligonucleotide. The oligonucleotide was then cloned through the In-Fusion reaction (Clontech; Takara Bio USA) into a pUC19 vector containing a preinserted 1000 bp fragment as homologous arms (Table II). Finally, the amplification products were purified and ligated in a pRNDM vector in frame as described above (28).

Thirteen million 293(T) cells (ATCC) were cultured in a 15-cm dish with 15 ml Opti-MEM (Thermo Fisher Scientific, Inc.) in a 5% CO₂ incubator at 37°C with 95% humidity. The third-generation packaging system was used in this experiment. Transfection of 16 µg genome-editing vectors (donor template library, lenti-Cas9) with 16 µg packaging plasmid psPAX2 and 6 µg envelope plasmid pMD2.G (Thermo Fisher Scientific, Inc.) were mixed in 1 ml Opti-MEM. About 76 µl of 1 mg/ml polyethylenimine (Polysciences Inc.) was mixed with the plasmids in 1 ml Opti-MEM for 15 min at room temperature. The DNA/PEI mixture was then added to the 293(T) cells in Opti-MEM. Forty-eight hours after transfection, cell culture medium was collected and centrifuged at 200 × g at 4°C for 10 min. The supernatant was then filtered through Φ 0.45-µm low-protein binding HV/PVDF membrane (Merck KGaA). The viral dose (multiplicity of infection ≈0.5) was identified and used to infect the cells. Aliquots of the supernatant containing viral particles were stored at −80°C until required.

The human breast cell line BT-549 (ATCC) was used as the cell model, as explained above. All cell lines were subjected to confirmation of cell line identity. Cells were cultured in a medium composed of 10% FBS and 90% RPMI-1640 medium in a 5% CO₂ incubator at 37°C with 95% humidity. Cells were harvested at 80% confluence and were used for subsequent experiments. All media were supplemented with 100 U/ml penicillin and 100 µg/ml streptomycin (Thermo Fisher Scientific, Inc.).

Construction of the donor libraries was as described in a previous study (29). Cell cultures were co-infected with equivalent amounts of amplified lentiviral library and CRISPR/Cas9-sg Rad51. Cells were seeded in 100-mm dishes with 3×10⁶ cells per dish. Puromycin (Merck KGaA) was added into the selection media after lentiviral infection and cultured at 2 µg/ml for one day. After selection, the surviving cells were pooled and used for amplification culture for 7 days. During the amplification culture, selection medium (containing 2 µg/ml puromycin) was replaced every 2–3 days. The surviving cells were then split into three individual groups that were treated with DMSO (Merck KGaA), 5 or 10 µM olaparib for 1 week. For each treatment, 1×10⁷ cells per replicate were used. Genomic DNA was extracted and real-time PCR (RT-PCR) was performed using the specific primers listed in Table III. PCR product was separated using agarose gel electrophoresis, then an QIAquick Gel Extraction Kit (cat. no. 28704, Qiagen, Inc.) was used to isolate DNA. The concentration of the DNA was quantified using a Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Inc.). DNA quality was determined by measuring 260/280 and 230/260 absorbance ratios on a spectrophotometer (Nanodrop ND-1000, Peqlab Biotechnologie). The DNA integrity number was determined using the Lab-on-a-Chip-System Bio-analyzer 2100 (Agilent Technologies, Inc.). The PCR products were used to prepare the NGS library. The library for sequencing was constructed using the NEB Next Ultra II DNA Library Prep Kit (cat. no. NEB#E7645L, NEB) in line with the manufacturer's instructions. The average insert fragment in the library (approximately 292 bp) was measured and controlled using the Bioanalyzer 2100 (Agilent Technologies, Inc.). The final library concentration was measured and adjusted to approximately 10 ng/µl using the Qubit 2.0 Fluorometer (Thermo Fisher Scientific, Inc.). The library was then sequenced using HiSeq 2500 (Illumina, Inc.) (PE=125; data are available at DOI 10.4121/19086689). Data analysis was performed using the MAGeCK analysis (http://www.bioconductor.org/packages/release/bioc/html/MAGeCKFlute.html) (29). The sequencing was performed using a paired-end 125 (PE125) sequencing model. HRR efficiencies were evaluated according to the selective PCR results. Sequencing reads containing the selective PCR site and at least one variant in the donor library were used to calculate HRR efficiency.

A total of 772 breast cancer patients were admitted to the Affiliated Hospital of Hebei University (China) from August 2012 to September 2019. Among these, 538 patients (females; age range, 28 to 69 years; median age ± SD, 51.4±3.5 years) diagnosed with breast cancer were enrolled in this study. The study was approved by the Review Board of the Affiliated Hospital of Hebei University (approval number AHHU20200930). The inclusion criteria included the following: i) newly diagnosed cases; ii) BRAC1/2 mutation detected; iii) completion of PAPRP inhibitor (olaparib) treatment and 5 years of follow-up. Information on patient treatment, recurrence, and survival status was well documented. Exclusion criteria included: i) recurrent breast cancer; ii) therapy performed prior to admission; and iii) comorbidities due to other clinical disorders. All patients were informed of the purpose of the experiment and signed informed consent. The clinical characteristics of the study subjects are shown in Table SII. Among the 538 patients who received the PARP inhibitor treatment, 68 were cases with a positive family history. The remaining 234 patients who failed to meet the inclusion criteria underwent surgical resection and adjuvant chemotherapy between August 2012 and September 2019. Of this group, those who did not receive PARP inhibitor treatment were also invited and accepted to participate as a population control. In the control group, the volunteers received standard medical and nursing care, such as vital sign observation, chemotherapy or surgical resection treatment and wound and pain management. Five patients withdrew voluntarily from follow-up in the next five years, but they agreed to our use of the data gathered during their follow-up.

Based on clinical findings, imaging information, histopathological tests, and other laboratory indicators, patients (cases) were graded according to the 8th edition AJCC standards (30). There were 70, 258, and 210 cases at clinical stages I, II, and III–IV, respectively. Patients were treated with surgical resection accompanied by chemotherapy. From the day of admission, all patients were followed for five years by telephone call or outpatient visits. Patients (n=2) who died from causes unrelated to breast cancer were excluded from the survival analysis.

DNA was extracted from all blood taken from the patients using a GenElut™ Blood Genomic DNA Kit (Merck KGaA, Darmstadt, Germany). Five variants of Rad51 (SNPs and repeat polymorphisms) (Table IV) were genotyped using SNAPshot (Thermo Fisher Scientific, Inc.).

Haplotype frequencies of Rad51 variants (five alternative amino acids listed in Table IV) were estimated using the Expectation-Maximization algorithm in the haplo.stats R software package (http://cran.r-project.org/web/packages/haplo.stats). The three assembled haplotypes were estimated separately in breast cancer patients at the above clinical stages. The R haplo.stats package was used to estimate the posterior probabilities of the haplotypes. The STATA haplologit command was used to model the association between haplotypes and case status as in a previous study (31).

BT-549 cell suspension (1×10³ cells per well) with/without the selected haplotype were obtained by trypsinization and filtration through a 40-µm filter, after which the cells were seeded in a 6-well plate (4,000 cells/well) containing 0.35% soft agar. The plates were then incubated in DMEM with 10% FBS for 1 day at 37°C. Drug-containing medium (5 µM olaparib) was refreshed weekly. After two weeks, the cells were washed twice with PBS and fixed with methanol at 4°C for 15 min, following which the colonies were stained with 0.1% crystal violet for 10 min at 25°C.

All experiments were conducted in three technical replicates and three biological replicates, respectively. All data and survival curves were analyzed using GraphPad Prism (v.6) (GraphPad Software, Inc.) unless otherwise indicated. The quantitative data are presented as mean ± standard deviation (SD). All single nucleotide polymorphism strategies were employed to construct the haplotype. Haplotype analysis was performed using PHASE v2.1.1 software (31) and the frequencies of haplotypes were compared between all induced mutation cases vs. the wild-type. A logistic multivariate regression model was used to look for interactions between SNPs of the WALKER A domain and the genotype of WALKER B domain. Correlation analysis was carried out using Pearson's correlation coefficient. K-M plotter was used to plot survival curves, which were compared using a Mantel-Cox test. Log-rank test was performed to evaluate the hazard ratio (HR) and 95% confidence interval (CI). The non-parametric Mann-Whitney test was utilized for comparisons between two groups. For multiple groups, the Kruskal-Wallis test was utilized followed by Newman-Keuls test or Tukey's HSD. P<0.05 and P<0.01 were considered statistically significant and highly significant, respectively.

In the pilot study, using flow cytometry we found that BT-549 cells displayed the highest sensitivity to olaparib (Fig. S1A and B). The sgRNAs, which were then designed according to the WALKER A and B domain-related PAM sequence, were scored using the CCTOP online prediction tool. The HDR library containing all possible missense mutations was constructed using a vector library containing a 7-bp selective PCR site mutation site at the PAM position and the selective PCR products were then identified by sequencing (Fig. 1A and B). The knockdown efficiencies for sgRNA1-1, sgRNA1-2, sgRNA2-1, sgRNA2-2, and sgRNA2-3 were 73.2, 54.9, 33.65, 68.05, and 71.4%, respectively; the off-target ratios were 1.03, 3.55, 2.71, 0.88, and 2.53%, respectively. The sgRNA-1-1 and sgRNA-2-2, which displayed relatively high knock-out efficiency and the lowest off-target ratios (73.2, 1.03; 68.05, and 0.88%, respectively) were then chosen for the subsequent experiments (Fig. 1C and D). Eight and five amino acids were found in the WALKER A and B domains, respectively. The saturated mutation of this region was obtained by residue substitutions of all 6,859 possible mutations (Fig. 1). Fig. 1C shows the selective PCR product region and evaluation of the off-target ratio.

Fig. 2A shows that the frequency of the genomic region mutation was highly correlated with the donor template library. Five amino acid mutations (128P, 129C, 129L, 133V, and 221E) were associated with high resistance to olaparib at the amino acid level (Fig. 2B). The alternated wild-type amino acids for these residues are shown in Table IV. The missense mutation frequencies of the five alternative amino acids were determined. According to the targeted NGS results, these SNPs were also determined at the RNA level using RT-PCR. A good biomarker for clinical diagnosis should have high signal to noise ratio. Single missense mutations are rare and have low signal to noise ratio (32). Several contradictory studies for the association of Rad51 genetic polymorphisms with cancer development have supported this viewpoint (33–35), although haplotyping can significantly overcome this drawback and improve the accuracy of the genetic polymorphisms (32,36). Therefore we assembled the SNPs into haplotypes using frequently co-occurring SNPs.

The following three haplotypes were assembled and identified in the olaparib-resistant cells (number indicates location of substitutions in cDNA): Haplotype 1 (387–388;664):CC>AC, T>A; Haplotype 2 (397–398):AA>GU; Haplotype 3 (382–383, 386–387):GA>CC, UC>GU (Fig. 2 and Table IV). Ten single clones were screened out by olaparib and evaluated by sequencing. All clones harbored at least one of the three haplotypes. Compared with the SNP results, it appeared that these haplotypes had greater discrimination ability in terms of cell survival, and cells with haplotype 2 and 3 showed the lowest sensitivity to olaparib (Fig. 3A and B) (P<0.05). Since RAD51 was a key HRR regulator which directly results in cell cycle arrest, we then investigated the effect of mutations in this gene on the cell proliferation and colony formation. The effects of haplotypes 1, 2 and 3 on the proliferation and cell viability of BT-549 cells were analyzed using the CCK-8 assay and colony formation assays. The former showed that harboring of the three haplotypes led to an increased proliferation of breast cancer cells (P<0.05) (Fig. 3C). In order to detect the tumorigenesis of tumor cells with/without the selected haplotype, colony formation assays was performed. Similarly, colony formation capacity was also enhanced in cells with the three haplotypes (Fig. 3D, P<0.05). Since the function of RAD51 is not directly related to apoptosis and cell migration, we did not investigate its effect on those cell behaviors. Furthermore, the cells were screened out in the olaparib-resistant experiment, indicating that the apoptosis ratio of this type of cell is very low and that the apoptosis difference would be difficult to determine

The association between Rad51 haplotypes and olaparib treatment was evaluated in 538 cases treated with PARP inhibitors and 234 controls (Table V). The distribution of the three Rad51 haplotypes was predicted and identified among the patient samples. The results indicated that the positive ratio of the three haplotypes was significantly higher in the olaparib treatment group compared to the controls (Table V). Haplotypes 2 and 3 showed a lower positive ratio in both groups compared to haplotype 1. As shown in Fig. 3A and B, haplotypes 2 and 3 were likely more sensitive and specific to the resistance of PARP inhibitors. To further identify the potential drug resistance of haplotypes 1, 2 and 3 in clinical samples, we performed five-year survival analyses to compare 245 patient cases carrying any of these haplotypes (Fig. 4). Two haplotypes (haplotypes 2 and 3) were closely associated with breast cancer survival (HR=1.303, 95% CI=0.9922-1.711, for haplotype-2 VS control; HR=1.494, 95% CI=1.063-2.100 for haplotype-3 VS control; HR=0.6197, 95% CI=0.4211–0.9120 for haplotype 2 VS haplotype 3). Haplotype-1 was not associated with survival (HR=0.8478, 95% CI=0.5848-1.077). The results indicated poorer prognosis with haplotype 3 than with haplotype 2 or 1.