BARD1 splice variants occur in primary colon cancer samples and matched normal colon tissue
To investigate the role of BARD1 splice variants in colon cancer, we analyzed RNA from 15 colon cancer samples and 15 matched normal colon samples for the occurrence of isoforms. RNA was reverse-transcribed and cDNA used as a template for PCR with primer targets in the first and last exon of the BARD1 transcript (). While the expected PCR product is 2362bp (, upper panel), we find several additional bands ranging between 2.3 and 0.6kb in size (, lower panel). Interestingly, we find these bands not only in the colon cancer samples, but also in adjacent matched normal colon tissue (, lower panel and Supplementary Figure S1).
Characterization of novel BARD1 splice variants
In order to characterize the PCR products, we purified and sequenced them by two different approaches: a) direct sequencing after gel purification and b) subcloning with subsequent sequencing. While direct sequencing was used in previous publications (13
), many splice variants are very similar in length and putative single bands may contain multiple isoforms which require subcloning for full characterization. We then analyzed the obtained sequences using the NCBI Spidey program, an mRNA to genomic alignment program, and aligned the sequences of the directly purified and subcloned splice variants against the full-length BARD1 mRNA (NM_000465.2). We characterized 19 distinct splice variants (), 11 of which have not been previously described. Eighteen of the 19 splice variants lack one or more complete exons and one variant lacks part of an exon, thus creating new and specific exon boundaries in each case.
As the designation of the BARD1 splice variants has thus far been inconsistent (), we created a systematic approach using the structure of the splice variant. For example, BARD1_1-7/9-11 is a splice variant that consists of exon 1 to exon 7, directly followed by exon 9 to exon 11, hence missing exon 8 (). Thus, a new and variant-specific exon boundary of exons 7 and 9 (7/9) is created, which we then exploited for quantification of splice variant expression by qPCR. summarizes the structure of the BARD1 isoforms in colon cancer.
Structural characterization of BARD1 splice variants
BARD1 splice variants account for a considerable amount of the total BARD1 transcript in colon cancer
In order to quantify the above-described BARD1 splice variants in normal colon and colon cancer, we performed SYBRGreen qPCR assays. We designed specific qPCR primer pairs, with one primer designed to target the new variant specific exon junction and the second primer positioned in one of the adjacent exons to yield a product of 50 to 150bp suitable for qPCR. Therefore, in the majority of cases, primers could not be designed as conventional exon spanning primers. Consequently, there was a risk that the primer that was supposed to target the new exon junction would anneal only with the primer portion lying within the same exon as the reverse primer, resulting in the amplification of an intraexonic fragment. To ensure that our primers specifically recognized the splice variants only, we performed exhaustive primer testing and optimization: We tested all primer pairs on cDNA and genomic DNA to rule out amplification of intraexonic fragments. We also tested varying primer concentrations to ensure saturation of all reactions and optimal performance and used No Template Controls to assure that the results were not affected by primer artifacts. Only primer pairs that fulfilled all above requirements were included in the analysis. With this stringent approach, we were not able to design primer pairs for three of the variants (BARD1_1-3/8-11, BARD1_1-3/10-11, BARD1_1-6/10-11) that would fulfill all these criteria, which is why we excluded them from the analysis. Since some of the exon junctions were found in more than one splice variant (), this led to a total of 13 primer pairs targeting a total of 16 splice variants the analysis (Supplementary Table S1). For example, primer pair 3/5 –targeting the new and splice variant specific exon boundary 3/5– detects splice variant BARD1_1-3/5-11 and splice variant BARD1_1-3/5-7/10-11. As exon junction 10/11 is the only exon junction that is found in all splice variants and in the full-length BARD1 transcript (), we used a primer pair spanning exon junction 10/11 to quantify the total BARD1 transcript.
Overview of BARD1 qPCR primers
In order to evaluate the relative abundance of each variant, we calculated the efficiency-corrected relative quantity of each splice variant and normalized it to the relative quantity of the total BARD1 transcript for each of the 30 samples, also corrected for efficiency. We find that BARD1 splice variants account for a considerable percentage of the total BARD1 transcript () either in colon cancer and matched normal colon samples. In both groups, BARD1_1-2/4-11 is the most abundant splice variant and accounts for 18.1% (8.1–34.0%) of the BARD1 transcript in colon cancer samples and 16.8% (10.6–26.8%) in the matched normal colon samples (). BARD1_1-4a/5-11 is the second most abundant splice variant with a relative expression of 9.8% (7.1–14.3%) in cancer samples and 14.7% (7.1–24.0%) in matched normal colon tissue. Splice variants detected by primer pair 3/5 (BARD1_1-3/5-11 and BARD1_1-3/5-7/10-11) account for an average of 5.0% and 7.3% respectively. Expression of splice variant BARD1_1/4-11 amounts to 4.4% of the total BARD1 transcript in cancer and 3.9% in normal colon. Splice variants detected by primer pair 1/5 (BARD1_1/5-11, BARD1_1/5-6/8-11, BARD1_1/5-7/9-11, BARD1_1/5-7/10-11) account for 2.4% of the transcript in cancer samples and 2.9% in normal tissue, BARD1_1-2/5-11 for 1.5% and 1.4%. The quantities of the seven remaining primer pairs (1/7, 7/10, 1/10, 2/7, 7/9, 6/8, 2/10) each comprise less than 1% of the total transcript for both cancer samples and matched normal tissue. The relative expression levels of the splice variants are quite similar among cancer and normal samples, with the exception of BARD1_1-4a/5-11 and the variants detected by primer pair 3/5 (BARD1_1-3/5-11 and BARD1_1-3/5-7/10-11) which are lower in cancer than in normal colon tissue (9.8% versus 14.7% and 5.0% versus 7.3%), while levels of BARD1_1-2/4-11 and BARD1_1/4-11 are slightly elevated (18.1% versus 16.8% and 4.4% versus 3.9% respectively). Comparing cancer and normal samples by t-test analysis shows a significant downregulation of BARD1_1-4a/5-11 and the variants detected by primer pair 3/5 (BARD1_1-3/5-11 and BARD1_1-3/5-7/10-11) (). As a control, we analyzed the expression of BARD1 splice variants in five colorectal mucosa samples from non-cancer patients. The percentage expression of the variants from non-cancer patients lies within the range of the matched normal samples from cancer patients suggesting a tissue specific expression profile that is altered in cancer. To further analyze the tissue specific expression pattern of the BARD1 variants, we performed quantitative assays on pooled cDNA from 16 human tissues and calculated the percentage expression for each variant in each tissue (). The splice variants can be detected at varying levels in different tissues creating tissue specific profile. This finding again shows that BARD1 splice variants are also a common feature of normal cells and suggests a complex function in health and disease.
Quantitative real-time PCR analysis of BARD1 splice variants
Distinct BARD1 splice variants show a cancer specific regulation pattern
To focus on the differences between cancer samples and normal control samples, we calculated the ratio of the normalized relative quantity for each matched pair (). We included all splice variants with a relative expression of more than 1%. As the results are greatly influenced by the way of normalization and as no single perfect normalization gene is available to compare colon cancer and normal colon tissue, we carefully selected five genes that were successfully used in expression studies of other cancer types (24
): L19, GAPDH, B2M, RPLPO, HPRT1. We performed normalization runs and calculated the geometric mean of the relative quantities of these reference genes for each sample, all corrected for efficiencies (25
). We normalized all samples to the geometric mean of the reference genes and calculated the ratio of the normalized relative quantity of each cancer sample and its corresponding normal colon tissue. We depicted the ratio using a logarithmic scale. Using this approach, we find two different regulation patterns: One group, consisting of 4a/5, 3/5 and 1/5, shows a tendency of downregulation, as for each of the three primer pairs only one of the 15 matched pairs shows a slight upregulation, while the other 14 matched pairs show similar expression levels or downregulation (). The second group consists of 2/4, 1/4 and 2/5. These variants show more varying levels: Some of the matched pairs indicate an upregulation of the splice variant in cancer versus normal, some have similar expression levels and some show a downregulation of the splice variant in cancer versus normal.
Interestingly, the second group contains isoforms BARD1_1-2/4-11 and BARD1_1/4-11, that show a higher percentage expression (when normalized to the total BARD1 transcript) in the cancer group than in the normal colon tissue group, even though the difference is rather small for BARD1_1/4-11 (4.4% in cancer versus 3.9% in normal tissue) (). BARD1_1-2/4-11 and BARD1_1/4-11 are also the splice variants that show the highest variation in percentage expression among different human tissues (), which may hint towards an important tissue specific function that could be altered in cancer. Notably, BARD1_1/4-11 was previously published as BARD1β and proposed to be associated with advanced stage of ovarian cancer and a negative prognostic factor in ovarian cancer (12
). Due to the lack of patient data associated with our samples and the limited number of good quality RNA samples, we could not test for a similar relationship in colon cancer. It would be interesting to see if the differential expression of BARD1_1/4-11 and/or BARD1_1-2/4-11 also correlates with advanced colon cancer in a future larger cohort.
We cannot exclude that there are other splice variants using a different transcription initiation, as we focused in our study on the splice variants detected by primers in exon 1 and exon 11. In previous studies splice variants initiating in the 5’ end could not always be detected (12
). As the lack of detection in the 5’ end could also be explained by poor quality RNA and cDNA rather than the absence of splice variants initiating in the 5’ end, we performed careful quality testing of all samples by bioanalyzer and 3’:5’ assays, before we used them in our experiments. As for our study, we can detect splice variants initiating in exon 1 in all samples, cancer and normal.
Expression of the full-length BARD1 protein correlates with outcome in colon cancer samples
To quantify full-length BARD1 expression, we performed immunohistochemical stainings on 81 colorectal cancer samples using a BARD1 specific antibody. Quantitative PCR does not allow for quantification of full-length BARD1 expression, as there is no sequence, which is specific for full-length BARD1 only. All exon boundaries of full-length BARD1 are found in at least some of the splice variants. To quantify the full-length BARD1 expression, we performed immunohistochemistry with a mouse monoclonal antibody raised against the N-terminus of BARD1. We ascertained the specificity of the antibody by Western blot (Supplementary Figure S3). The antibody was used on multi-tissue arrays containing 81 colorectal cancer samples of different stages, detected with DAB and counterstained with hematoxylin. To strengthen the reliability of the expression analysis, we scanned the slides using the Aperio Slide Scanner and analyzed the staining intensity using the Aperio ImageScope software (). Three staining intensities were discriminated: 3 (strong nuclear staining), 2 (intermediate expression), 1 (weak or no staining). We assessed the relationship between BARD1 expression and overall survival by a log-rank test. Interestingly, we find that the expression levels of BARD1 correlate significantly with survival (P=0.0002). High expression of full-length BARD1 predicts a favorable outcome, while loss of expression is associated with a worse prognosis (). We also find that BARD1 expression significantly correlates with staging with over half of the patients with stage IV showing a weak expression (Expr.1) of full-length BARD1 (p=0.043614) (). This is in line with the expected finding that stage is also a good predictor of survival (p<0.0001) (). To control for possible effects of age, sex and staging, a multivariate model for BARD1 was estimated by Cox regression. Pairwise comparisons show a significant difference in survival probability for patients with low expression of BARD1 (Expr.1) versus patients with intermediate or high expression of BARD1 (Expr.2, Expr.3) supporting the potential usefulness of BARD1 as a predictive biomarker. In a second approach a prognostic risk model based on age, sex, staging, and BARD1 expression was derived by a conditional inference survival tree (). Stage IV versus stage II, III shows the highest correlation with outcome (p<0.001). As expected, stage IV patients have a worse prognosis than patients with stage II and III. There is no significant difference in survival between stage II and III patients. However, among patients with stage II and III disease, expression of full-length BARD1 protein shows a significant difference in survival comparing weak (Expr.1) and intermediate/strong expression (Expr.2, Expr.3) with weak expression predicting a worse prognosis. Interestingly, there is a significant difference in survival comparing stage II and III (p=0.022) within the group of intermediate/strong BARD1 expression ().
Full-length BARD1 protein expression predicts survival in colon cancer
These findings are consistent with the role of BARD1 as a tumor suppressor and identify full-length BARD1 as a novel and potentially useful biomarker for an improved risk stratification of colon cancer patients.
To further consolidate these findings and to test for a possible dependence on treatment we analyzed a second cohort of 43 patients from a previous study that determined the effect of 5-FU treatment on overall survival. This group consists only of stage II and stage III patients eliminating stage as a predictor of survival. The analyses were performed as with the first cohort and the multivariate analyses were also adjusted for treatment. Survival analysis again shows a significant correlation between survival and expression of full-length BARD1 (Supplementary Figure S4A), while neither 5-FU treatment (Supplementary Figure S4B) nor stage (Supplementary Figure S4C) is a significant predictor of survival in this cohort. Accordingly, there is no significant correlation between expression of BARD1 and stage (Supplementary Figure S4D) or 5-FU treatment (Supplementary Figure S4E), which strengthens the usefulness of BARD1 as additive marker for risk stratification. These findings are supported by the results of the Cox regression analysis and the conditional interference survival tree (Supplementary Figure S4F), that both show a statistical significance in survival between colon cancer patients with low expression of full-length BARD1 protein (Expr.1) and patients with intermediate or high expression of full-length BARD1 protein (Expr.2, Expr.3).