Expression of an Oncogenic BARD1 Splice Variant Impairs Homologous Recombination and Predicts Response to PARP-1 Inhibitor Therapy in Colon Cancer

BRCA1-associated RING domain protein 1 (BARD1) stabilizes BRCA1 protein by forming a heterodimeric RING-RING complex, and impacts function of BRCA1, including homologous recombination (HR) repair. Although colon cancer cells usually express wild type BRCA1, presence of an oncogenic BARD1 splice variant (SV) in select cancers may render BRCA1 dysfunctional and allow cells to become sensitive to HR targeting therapies. We previously reported association of loss of full-length (FL) BARD1 with poor prognosis in colon cancer as well as expression of various BARD1 SVs with unknown function. Here we show that loss of BARD1 function through the expression of a BARD1 SV, BARD1β, results in a more malignant phenotype with decreased RAD51 foci formation, reduced BRCA1 E3 ubiquitin ligase activity, and decreased nuclear BRCA1 protein localization. BARD1β sensitizes colon cancer cells to poly ADP ribose polymerase 1 (PARP-1) inhibition even in a FL BRCA1 background. These results suggest that expression of BARD1β may serve as a future biomarker to assess suitability of colon cancers for HR targeting with PARP-1 inhibitors in treatment of advanced colon cancer.


Results
Human colon cancer cells exhibit differential sensitivity to PARP-1 inhibition. Increased sensitivity to PARP-1 inhibition is indicative of defective DNA double-strand break repair by homologous recombination 16,26 . To determine etiologies of PARPi sensitivities in colon cancer cells, we treated a panel of colon cancer cell lines representing a spectrum of genomic backgrounds with increasing concentrations of PARPi (Fig. 1a). Cell viability and dose response curve for each cell line was determined after 3 and 4 days of treatment and IC 50 s were , SW620, and Caco-2 were treated with PARPi for 3 (triangle) and 4 (square) days and IC 50 determined (**p < 0.01). The experiment was performed at least three independent trials with at least three technical replicates. (b) Untreated CRC cell lines were harvested, protein extracts were prepared, separated and immunoblotted with anti-BRCA1 antibody as described in Methods. The experiment was repeated at least three times. Quantification of protein band intensity was performed by ImageJ software (*p < 0.05 and **p < 0.01). calculated. None of the known genomic determinants of colon carcinogenesis, including microsatellite instability (MS), loss of SMAD4 or loss of p21, were associated with an increased sensitivity to PARPi. Yet the MS stable Caco-2 cell line was significantly more sensitive to PARPi, suggestive of an undefined distinct molecular feature in these cells (Table 1). Differential sensitivity to PARPi among colon cancer cell lines is not associated with BRCA1 mutation or loss of BRCA1 expression. BRCA1 is a major participant in HR, therefore cells deficient in BRCA1 show increased lethality with PARPi due to absence of DSB DNA repair by HR. All tested colon cancer cell lines are known to have wild type BRCA1 and all express FL BRCA1 protein. BRCA1 protein was in similar quantities among SW620, Caco-2, and SW480 cells and these were relatively higher than FET, HCT116 p21 −/− , HCT116 (Fig. 1b).
Alterations in BARD1 may affect BRCA1 function in a BRCA1 wild type setting. FL BARD1 is present in both PARPi-sensitive and -resistant cell lines. However, BARD1 SVs may have functions independent from its FL protein and may affect the function of BRCA1 independent of their effects on BARD1 expression. BARD1β has been reported to have poor prognosis in colon cancer 22 . Using real-time PCR, we compared the expression of BARD1β in a panel of colon cancer cell lines. Expression of BARD1β was significantly higher in PARPi sensitive Caco-2 cells compared to its expression in cell lines that were resistant to PARPi (Fig. 2a). BARD1 SV mRNA is associated with polysomes. To assess potential functions of the BARD1β , we determined whether the SVs mRNAs are translated. Real-time PCR of cDNA with primers targeting the first and last BARD1 exon revealed amplification of multiple SVs in the polysomal fractions, which is congruent with active translation (Fig. 2b).

PARPi causes DSBs in both sensitive Caco-2 and resistant SW480 cells. PARP-1 activity is crucial
in the recognition and repair of DNA single-strand breaks. Blocking PARP activity with the PARP-1 inhibitor ABT888 causes single strand breaks to accumulate, leading to formation of double-strand breaks 27,16,14 . We compared the formation of double-strand breaks in the PARPi-sensitive Caco-2 cells and the PARPi-resistant SW480 and HCT116 cells, using γ H2AX foci formation as an indicator (Fig. 2c). Untreated control cells displayed very few or no γ H2AX foci formation, consistent with the absence of double-strand breaks, while treatment with the PARPi induced double-strand breaks in both PARPi-sensitive (Caco-2) and PARPi-resistant cells (SW480 and HCT116). PARPi caused significantly more γ H2AX staining in the Caco-2 cells (Fig. 2d). This confirms PARPi causes double-strand breaks both in the PARPi-sensitive and PARPi-resistant cell lines further focusing on differences in DNA repair mechanism as the basis for PARPi sensitivity.
PARPi-sensitive colon cancer cells show impaired HR. We hypothesized that BARD1β expression in the PARPi-sensitive colon cancer cells impacts HR proficiency, which can be measured by the formation of RAD51 foci. RAD51 is a critical protein functioning in the HR DNA repair, which localizes to the nucleus to form foci upon DNA damage, and are detected by a florescent antibody and counted using a fluorescent microscope 15 . We determined the RAD51 foci formation in the PARPi-sensitive and -resistant colon cancer cells following treatment with PARPi. Each cell line was treated with a concentration of PARPi lower than its respective IC 50 (10 μ M) to avoid lethality. As expected, there were very few or no foci formed in untreated cells (Fig. 2e, left panels; 2f). After treatment with PARPi, PARPi-resistant cells (SW480) (Fig. 2e, right lower panel; 2f) displayed increased RAD51 foci formation compared to PARPi-sensitive colon cancer cells (Caco-2) which displayed a significantly weaker HR DNA repair response characterized by lower RAD51 foci formation (Fig. 2e, right upper panel; 2f). Therefore, impaired HR function may be due to BRCA1/BARD1 deficiency in a BRCA1 wild type setting.

Exogenous expression of SV BARD1β in PARPi-resistant cells results in impaired HR.
To test the hypothesis that expression of BARD1β leads to impaired HR, we expressed BARD1β in the SW480 PARPi-resistant cells, which have relatively low levels of endogenous BARD1β transcript and have very similar levels of BRCA1 as Caco-2 PARPi-sensitive cells have. Additionally, as reference controls, stable SW480 cells expressing V5 tagged FL BARD1, or both BARD1β and FL BARD1, or an empty vector were prepared ( Supplementary Fig. S1). Although HCT116 cells had considerably lower levels of endogenous BARD1β transcript relative to SW480 cells, we decided to prepare stable SW480 cells expressing BARD1β because HCT116 cells have significantly lower levels of endogenous BRCA1 protein, and are deficient in DNA mismatch repair. Exogenous overexpression of BARD1β in SW480 cells caused significantly more γ H2AX foci than FL + BARD1β cells in response to PARPi treatment (Fig. 3a,b) indicating increased sensitivity to PARPi. To evaluate the effect of BARD1β on HR function, RAD51 foci formation was compared among the PARPi-resistant SW480 colon cancer cell lines: SW480 expressing BARD1β , FL BARD1, or both BARD1β and FL BARD1 with and without PARPi treatment (Fig. 3c). In the absence of PARPi, no or only minimal RAD51 foci formation was observed in all cell lines (Fig. 3c, left panels; 3d). However, when the cells were exposed to PARPi,  Fig. 3c were quantified (*p < 0.05 and **p < 0.01). (e) Untreated Caco-2 and SW480 cells were stained with DAPI and anti-RAD51 antibody. 10 μ M PARPi treated Caco-2 and SW480 cells for 4 days were stained with DAPI and anti-RAD51 antibody. (f) Nuclear RAD51 foci formation in Fig. 2e were quantified (*p < 0.05 and **p < 0.01).

Cell line Characteristic
control and FL BARD1 expressing cells displayed significantly higher accumulation of RAD51 (Fig. 3c, left upper three panels; 3d), while both BARD1β expressing and the combined BARD1β and FL BARD1 expressing cells exhibited significantly fewer foci (Fig. 3c, left lower three panels; 3d) consistent with loss of BRCA1/BARD1 function with expression of BARD1β independent of FL BARD1 expression. The reduction of RAD51 foci formation (a) FL BARD1 expressing SW480 cells, both FL and BARD1β expressing SW480 cells, and BARD1β expressing SW480 cells were exposed to PARPi for 72 hours and stained with DAPI and phospho-histone γ H2AX DSB marker. Stably infected SW480 cells were treated with PARPi, and then fixed and stained with DAPI and phospho-histone γ H2AX antibody. (b) Nuclear γ H2AX foci formation in Fig. 3a were quantified (*p < 0.05 and **p < 0.01). (c) Control SW480 cells, FL BARD1 expressing SW480 cells, both FL and BARD1β expressing SW480 cells, and BARD1β expressing SW480 cells were stained with DAPI and anti-RAD51 antibody. Stably infected SW480 cells were treated with 10 μ M PARPi for 4 days, and then fixed and stained with DAPI and anti-RAD51 antibody. (d) Nuclear RAD51 foci formation in Fig. 3c were quantified (*p < 0.05 and **p < 0.01). (e) Control and FL, FL + BARD1β and BARD1β expressing SW480 cells were stained with PI, and monitored by flow cytometry to determine the percentage of G1 phase cells. Experiments were repeated at least three times (*p < 0.05 and **p < 0.01 relative to control).

Expression of BARD1β in PARPi-resistant cells increases cytoplasmic localized BRCA1. BARD1
plays an important nuclear chaperone role for BRCA1 [28][29][30][31][32][33][34] , suggesting that dimerization of BRCA1 with BARD1 is required for nuclear localization of BRCA1. Since BARD1β is missing most of the dimerization region, we examined the subcellular localization of BRCA1 in control and BARD1β stably expressing PARPi-resistant cells using immunofluorescence (Fig. 4a). In control or FL BARD1 overexpressing PARPi-resistant cells, BCRA1 was mainly localized to the nucleus, however, when BARD1β was overexpressed, we observed BRCA1, both, in the cytoplasm and nucleus implying BARD1β is deficient in nuclear chaperone activity (Fig. 4a). Cell fractionation to separate the nuclear and cytoplasmic fractions was performed to confirm differential localization of BRCA1 between FL and BARD1β expressing cells. FL BARD1 expressing cells displayed greater amount of nuclear BRCA1 than BARD1β expressing cells (Fig. 4b).
Impact of BARD1β on ubiquitin ligase activity of BRCA1. Mutations within the RING binding domain leads to impairment of E3 ubiquitin ligase activities of BRCA1 8,29-31 . BRCA1 has been reported to ubiquitinate cell cycle proteins including cyclin B; which ultimately contributes to regulation of G2/M cell cycle checkpoint 6 . We examined the impact of BARD1β expression in PARPi-resistant cells on E3 ubiquitin ligase activity of BRCA1, and ubiquitination of cyclin B. PARPi-resistant cells were co-transfected with His-tagged ubiquitin and MYC tagged cyclin B and treated with MG132, an inhibitor of the proteasome to reduce degradation of ubiquitinated molecules. Consistent with functional BARD1 in control and FL BARD1 overexpressing cells, we detected relatively high ubiquitination activity. In contrast, cells expressing BARD1β displayed little or no ubiquitinated cyclin B following MG132 treatment (Fig. 4c). Taken together, these findings suggest that colon cancer cells which express BARD1β display decreased ubiquitination and are more sensitive to PARPi due to impaired HR.
Expression of BARD1β is associated with a more invasive phenotype. We have previously observed that expression of BARD1β is associated with poor prognosis in colon cancer tumors 22 . To test if expression of BARD1β leads to a more transformed phenotype, we examined the differences in histology when BARD1β is over-expressed in PARPi-resistant cells (Fig. 5a). Under standard cell culture conditions, control and FL BARD1 expressing PARPi-resistant cells formed single cell monolayers, while BARD1β expressing cells formed clusters indicative of loss of cell-cell adhesion. The transition to metastatic disease includes epithelial to mesenchymal transitions (EMT) and is associated with an increase in cellular ability to form colonies. We examined BARD1β overexpressing PARPi-resistant cells for expression of EMT marker proteins by immunoblotting, and observed a decrease in E-cadherin and increases in β -catenin, Vimentin and Snail, all consistent with induction of EMT (Fig. 5b).
Next, we compared the clonogenic capacity of PARPi-resistant cells expressing either FL BARD1 or BARD1β , or both using the colony formation assay as previously described 32 . BARD1β expressing cells produced significantly higher numbers of colonies relative to cells expressing either FL or both FL BARD1 and BARD1β , indicating that BARD1β expressing cells retain a higher capacity to produce colonies (Fig. 5c). Increase in migration of BARD1β expressing cells further supports the notion of enhanced tumorigenesis (Fig. 5d).

Exogenous expression of SV BARD1β in PARPi-resistant colon cancer cells imparts sensitivity to PARP inhibition.
To evaluate the effect of BARD1β expression on PARPi-mediated cell death, we exposed stable transfectants of PARPi-resistant cells, characterized above, to low dose PARPi. Compared to control and FL BARD1 expressing cells, expression of BARD1β sensitized cells to low concentrations of PARPi (Fig. 6a). When FL BARD1 and BARD1β are co-expressed, we observed an intermediate phenotype with respect to PARPi sensitization (Fig. 6a).
DNA damaging agents including Irinotecan and Oxaliplatin are commonly used in the treatment of metastatic colorectal cancers 33 . Irinotecan is a topoisomerase 1 poison leading to the formation of a potentially lethal DNA double-stranded break. Oxaliplatin, in contrast, is a platinum analogue, which causes the formation of interand intra-strand crosslinks in DNA that prevent transcription and replication. Ultimately this results in DNA damage and cell death. We investigated if PARPi can sensitize colon cancer cells expressing BARD1β in vitro to BARD1β ; ≠ BARD1β vs. FL + BARD1β ; * ,#,≠ p < 0.01). (b) Stable SW480 cells which were prepared as explained above were exposed to PARPi (20 μ M) for 84 h followed by either Oxaliplatin (50 μ M) or Irinotecan (5 μ M) or both supplementation for 36 h. Survival fraction was determined using cell viability assays (*control vs. BARD1β ; # FL vs. BARD1β ; ≠ BARD1β vs. FL + BARD1β ; * ,#,≠ p < 0.01). (c) SW480 cells stably infected with either FL BARD1, BARD1β , or the combination FL + BARD1β , were exposed to Olaparib PARPi (10 nM or 100 nM) for 72 hours. Survival fraction was determined using cell viability assays (*p < 0.05 and **p < 0.01). (d) Protein lysates (40 μ g) from each sub-line of SW480 cells were separated and immunoblotted with anti-PARP1 antibody as described in Methods. For normalization, blots were probed with α -tubulin and β -actin. This blot is representative of three independent experiments. (e) shRNAs targeting BARD1β were prepared, and BARD1β was stably knocked down in Caco-2 cells as described in Methods. Cells were exposed to 10 μ M ABT-888 for 72 hours. Survival fraction was determined using cell viability assays (*p < 0.05 and **p < 0.01). either Oxaliplatin and/or Irinotecan treatment. We established the optimal time of drug exposure and observed that Irinotecan and Oxaliplatin work more rapidly and achieve optimal action at 36 hours while PARPi requires 96 hours of incubation ( Supplementary Fig. S3). Similarly we established concentrations for each drug that are sub-lethal to allow for optimal action when used in combination (Supplementary Fig. S3). Expression of BARD1β in PARPi-resistant cells did not selectively sensitize the cells to either Oxaliplatin or Irinotecan alone relative to control cells (Fig. 6a). Due to the differential time course of drug response, we chose to treat cells sequentially, first with PARPi followed by Oxaliplatin or Irinotecan. Addition of Oxaliplatin or Irinotecan decreased the fraction of viable cells in all stable PARPi-resistant cells (Fig. 6b); however, addition of Irinotecan to PARPi pre-treated cells created higher specificity with respect to killing of BARD1β expressing PARPi-resistant cells. These data indicate that colon cancer patients with high expression of BARD1β may receive therapeutic benefit from a serial combination of PARP inhibition and Irinotecan treatment.
To test whether BARD1β expressing SW480 cells was exclusively sensitive to ABT-888 or other PARP inhibitors display a similar feature, we used a potent PARP inhibitor, Olaparib (AZD2281), which has the additional cytotoxic function due to its higher ability of PARP-trapping on chromatin. Consistent to our findings with ABT-888, BARD1β expression in SW480 cells increased sensitivity to Olaparib (Fig. 6c) as well.
In addition to these survival assays, we evaluated whether exogenous expression of BARD1β in SW480 cells increased the overall apoptotic index. Expression of BARD1β increased PARP1 protein expression. Moreover, cleavage of PARP1 was further stimulated after 10 nM and 100 nM treatments of Olaparib which was an indication of increased apoptosis (Fig. 6d).
To further confirm the expression of BARD1β is associated with PARPi sensitivity, we knocked down the expression of BARD1β in PARPi sensitive Caco-2 cells using shRNA-mediated knockdown. Knockdown of BARD1β in Caco-2 cells rendered these cells less sensitive to PARPi relative to control Caco-2 cells (Fig. 6e).
In summary, we show that colon cancer cells which express BARD1β SV display impaired HR DNA repair and are more sensitive to PARP-1 inhibition. Overexpression of BARD1β in PARPi-resistant colon cancer cells rendered these cells sensitive to PARPi, and decreased E3 ligase activity as well as nuclear localization of BRCA1. This was associated with an increase in EMT, as well as colony formation and enhanced cell migration, supporting that oncogenic properties of BARD1β in colon cancer are at least in part, via targeting BRCA1 function.

Discussion
The tumor suppressor functions of BRCA1 are largely attributed to stabilization of BRCA1 by heterodimer formation with binding partners such as BARD1 within the BRCA1-associated RING domain protein. About twenty percent of clinically relevant BRCA1 mutations occur within the binding site of BRCA1 and BARD1, indicating that the binding of BARD1 is crucial for HR DNA repair function of BRCA1 5,34 . BRCA1 is typically wild type in colon cancer, but the presence of various BARD1 SVs of mostly unclear function has been reported 7,22,23 .
The loss of BARD1 in mice leads to embryonic lethal phenotype and chromosomal instability 9,35 . BARD1 spans 11 exons and we previously observed 19 distinct BARD1 SVs in colon cancer patients that account for over 40 percent of the total BARD1 transcript 22 . The individual functions of these SVs have not been assessed. However, BARD1β may have oncogenic activities. In cervical, breast, ovarian and endometrial cancer cell lines, BARD1 SVs missing the RING domain have been reported to be more abundant than the full length protein, and the knockdown of both FL and SVs decreased the rate of proliferation and led to growth arrest 25 .
Deficiencies in HR DNA repair sensitize cancer cells to PARPi 36 . PARPi leads to the accumulation of single-strand DNA breaks which are converted to double-strand DNA breaks. In normal cells, PARPi alone is not sufficient to induce apoptosis because the HR DNA repair mechanism repairs double-strand DNA breaks (Fig. 7) leading to cell survival. However, in cancer cells with HR deficiency, other error prone DNA repair mechanisms dominate which increases DNA errors and genomic instability, and ultimately induces apoptosis 14 . We found that BRCA1 protein expression in SW620, Caco-2, and SW480 cells was similar; however, Caco-2 cells had significantly higher levels of BARD1β transcript. This observation supports the idea of BARD1β expression level is a marker for PARPi sensitivity. Overexpression of BARD1β in colon cancer cell lines, such as in Caco-2 cells, impairs error free HR DNA repair by negatively influencing functions of BRCA1 thus sensitizing cells to PARPi. The high numbers of DNA breaks caused by HR DNA repair deficiency coupled with PARP inhibition leads to massive genomic instability resulting in apoptotic cancer cell death (Fig. 7). This increased sensitivity to the PARPi, suggests therapeutic applications for cancer patients expressing this BARD1 isoform. An effective ex vivo assay to determine the proficiency of HR in colon cancer patients will be useful for predicting which patient might obtain the most benefit from the therapeutic use of PARPi. Response to PARPi therapy could be predicted using the ex vivo RAD51 foci formation assay 37 . This would ultimately provide a biologic basis for therapeutic strategies such as PARPi alone or combination of PARPi with DNA damaging agents in the treatment of colon cancer patients.
In this study, we demonstrate that overexpression of BARD1β stimulated a decrease in BRCA1 functional capabilities with decreased ubiquitination, decreased nuclear translocation, and decreased HR repair. These results indicate that expression of BARD1 SVs lacking the BRCA1 binding domain could predispose patients to colon cancer similar to the predisposition for breast cancer by BRCA1 mutations. We have previously reported that BARD1β is overexpressed in a subset of colorectal cancer patients. Due to the lack of the ring binding domain, BARD1β does not contain the binding site necessary for interaction with BRCA1 and may display oncogenic effect via affecting BARD1/BRCA1 tumor suppressor function either independent or dependent on FL BARD1 protein. For example, Ryser et al. report that FL BARD1 binds to BRCA1 leading to ubiquitination and degradation of Aurora B during mitosis. In contrast BARD1 SV can bind the Aurora B/BRCA2 complex stabilizing Aurora B leading to enhanced cell proliferation 38 . Additional binding partners, such as FL BARD1 and/ or PARP, may exist for this BARD1 isoform which contribute to its oncogenic effect. We found that BARD1β expressing SW480 cells had higher levels of PARP1 protein expression relative to FL or both FL + BARD1β expressing SW480 cells. It has been previously reported that PARP1 also has a role in repair of double strand breaks, such as non-homologous end joining (NHEJ) and microhomology-mediated end joining (MMEJ), which are error prone DNA repair pathways 39,40 , therefore overexpression of PARP1 in BARD1β expressing cells might also increase genomic stability and cancer.
It has been previously reported that while overexpression of BARD1 SVs results in cell transformation, its repression leads to growth arrest of cancer cells 25 . The exogenous expression of BARD1β isoform in PARPi-resistant SW480 colon cells leads to a more transformed phenotype with enhanced EMT. E-cadherin, a mediator of cell-cell adhesion 41 , is reduced by BARD1β overexpression which can support invasion and growth of tumor cells. In contrast, β -catenin expression was markedly higher in BARD1β expressing cells. β -catenin overexpression has been linked to many epithelial cancers, including colorectal cancer 42 . BARD1β expressing cells exhibited increased levels of the EMT regulator, SNAIL, and mesenchymal marker, vimentin, which could contribute to the observed histological difference between control and BARD1β expressing cells.
In conclusion, BARD1β displays oncogenic activities through negatively affecting the tumor suppressor functions of BRCA1. Our findings indicate therapeutic targeting of deficient HR with PARPi therapy may be beneficial to the colon cancer patients who express this SV. Future development of fast, novel detection methods to identify colon cancer patients with HR deficiency is needed and detection of BARD1β may provide direction. were maintained in 10% Dulbecco's modified Eagle medium (DMEM), supplemented with 10% fetal bovine serum and 1% antibiotic-antimycotic agent in a 37 °C incubator with 5% CO 2 . Cells were validated by 9 STR (short tandem repeat) profiling, using CellCheck 9 Plus, tested for mycoplasma, and found to be mycoplasma free (both assays performed by IDEXX, Columbia, MO) in February 2015. BARD1β and FL BARD1-expressing Immunofluorescence. Cells seeded on glass coverslips were fixed in 4% paraformaldehyde, followed by blocking with 3% BSA in PBS. Then, cells were incubated with anti-RAD51 (Santa Cruz), anti-P-histone H2AX (Cell Signaling), anti-BRCA1 (Abcam), or anti-BARD1 (Santa Cruz) antibody in PBS, followed by incubation with goat anti-rabbit and anti-mouse IgG conjugated with Alexa Fluor 488 and 594 (Invitrogen) in PBS. Cells were washed in PBS, mounted, and then imaged with a fluorescence microscope (Nikon Eclipse E600 microscope, SPOT Imaging Software, Diagnostic Instruments, MI). γ H2AX and RAD51 immunofluorescence assays were performed as previously described 36,43,44 . An average of 100 nuclei were counted for γ H2AX and RAD51 immunofluorescence assays, and experiments performed in triplicates.

Measurement of proliferation rates and cell survival.
FET, HCT116 p21 −/− , HCT116, SW480, SW620 and Caco-2 cells were seeded into 96-well plate at a concentration of 20,000 cells with 100 μ L/well DMEM supplemented with 10% fetal bovine serum and 1% antibiotic-antimycotic agent in a 37 °C incubator with 5% CO 2 and incubated for 24 hr. A range of concentrations (0, 1, 5, 10, 20, 40 μ M) of PARPi were added to the seeded cells and incubated for 72 or 96 hr. Cell proliferation and cell survival were determined according to manufacturer's instructions using either Cell Counting Kit-8 from Dojindo Laboratories (Rockville, MD) and/or the BioRad automated cell counter (Bio-Rad). The half maximal inhibitory concentrations (IC 50 ) of PARPi were determined by constructing a dose-response curve for each cell line. The experiment was performed at least three independent trials with at least three technical replicates.
Migration assay. Migration assays were performed as previously described 45  Ubiquitination assay. E3 ubiquitination ligase activity of BARD1 proteins were compared in control, FL BARD1, and BARD1β expressing SW480 cells as previously described 46 . These cells were co-transfected with HA-tagged ubiquitin and MYC-tagged cyclin B using Fugene HD in the presence and absence of 10 μ M proteasome inhibitor MG132 for 3 hr. After 48 hr, Myc-tagged cyclin B proteins were immunoprecitated with anti-Myc antibody, and resolved in 10% SDS PAGE. PVDF membrane was immunoblotted with HA antibody.
Polysomal fraction assay. To confirm translation of BARD1 SVs, we isolated polysomal fractions from Caco-2 cells. Cells were treated with 0.1 mg/ml cycloheximide (CHX) for 3 min at 37 °C, and harvested into hypotonic lysis buffer [0.5% Triton X-100, 0.5% sodium deoxycholate, 2.5 mM MgCl 2 , 5 mM Tris (pH 7.5), 1.5mM KCl, 0.1 mg/ml CHX, 100 units RNAse-Inhibitor (Ambion-Life Technologies, Grand Island, NY)] at 4 °C. Nuclei were removed, 200 μ g/ml heparin was added and residual debris was removed by centrifugation. The lysate was layered on a 10 ml continuous sucrose gradient (10-50% sucrose in 5 mM MgCl 2 , 20 mM HEPES (pH 7.6), 100 mM KCl). After 110 min of centrifugation at 35.000 rpm at 4 °C, the absorbance at 254 nm was measured continuously as a function of gradient depth using the Teledyne Isco UA-6 Absorbance Detector. Prepared polysomal fractions were pooled, and total RNA from polysomal fractions was isolated with the Allprep RNA/Protein Kit from Qiagen (Hilden, Germany). Reverse transcription-PCR of isolated RNA was performed with the primer pair targeting the first and last exon of the BARD1 transcript, and the PCR products were visualized on a 1% w/v agarose gel stained with ethidium bromide as previously described 22 .
Quantitative real-time PCR. To quantify the expression of BARD1β , SYBR green quantitative real-time PCR reaction was performed as previously described 22 with the following modifications. Superscript ™ III First-Strand Synthesis SuperMix and Oligo(dT)20 primers by Invitrogen (Carlsbad, CA) was used for cDNA synthesis. The expression of BARD1β was normalized to GAPDH and the results were analyzed by the comparative Ct method. Quantitative PCR was performed at least three times from three different RNA extractions for each cell line. Statistical comparisons were performed on mean expression of the independent samples and fold differences among colon cancer cell lines were calculated.
Cell cycle analysis. The cell cycle distribution of stable SW480 cells was analyzed by using propidium iodide (PI) staining. Briefly, 10 6 cells were trypsinized and washed with PBS, centrifuged and fixed in ice-cold 70% ethanol and washed with cold PBS, and supernatant was removed. For measurement of DNA content, cells were washed with PBS and then stained with 100 μ g/mL RNase A in PBS (Sigma), 50 μ g/mL PI (Sigma) and 0.1% (v/v) Triton X-100 for 30 min at room temperature in dark, and analyzed by flow cytometry (Becton-Dickinson, Franklin Lakes, NJ, USA) with Cell Quest software. Three independent experiments were performed for each cell line.
Gene knockdown. We conducted shRNA-mediated knockdown of BARD1β gene in Caco-2 cells and studied the effects on cell viability. The BARD1b target sequence was: (5′ -GCTAGCCACTGCTCAGTAATG). We constructed short hairpin RNA (shRNA) oligos against BARD1β and cloned into pLKO.1 puro lentiviral backbone according to Addgene's pLKO.1 protocol. Lentiviral particles were generated in 293T cells and transduced into Caco-2cells. Statistical analysis. All data presented as mean + /− SD. One-way ANOVA and Tukey's multiple comparison test were used as appropriate comparison among groups. The criterion for statistical significance was set at *p < 0.05 or **p < 0.01.