SPiP: Splicing Prediction Pipeline, a machine learning tool for massive detection of exonic and intronic variant effects on mRNA splicing.

Modeling splicing is essential for tackling the challenge of variant interpretation as each nucleotide variation can be pathogenic by affecting pre-mRNA splicing via disruption/creation of splicing motifs such as 5'/3' splice sites, branch sites, or splicing regulatory elements. Unfortunately, most in silico tools focus on a specific type of splicing motif, which is why we developed the Splicing Prediction Pipeline (SPiP) to perform, in one single bioinformatic analysis based on a machine learning approach, a comprehensive assessment of the variant effect on different splicing motifs. We gathered a curated set of 4616 variants scattered all along the sequence of 227 genes, with their corresponding splicing studies. The Bayesian analysis provided us with the number of control variants, that is, variants without impact on splicing, to mimic the deluge of variants from high-throughput sequencing data. Results show that SPiP can deal with the diversity of splicing alterations, with 83.13% sensitivity and 99% specificity to detect spliceogenic variants. Overall performance as measured by area under the receiving operator curve was 0.986, better than SpliceAI and SQUIRLS (0.965 and 0.766) for the same data set. SPiP lends itself to a unique suite for comprehensive prediction of spliceogenicity in the genomic medicine era. SPiP is available at: https://sourceforge.net/projects/splicing-prediction-pipeline/.

Assessment of branch point prediction tools to predict physiological branch points and their alteration by variants.

BACKGROUND: Branch points (BPs) map within short motifs upstream of acceptor splice sites (3'ss) and are essential for splicing of pre-mature mRNA. Several BP-dedicated bioinformatics tools, including HSF, SVM-BPfinder, BPP, Branchpointer, LaBranchoR and RNABPS were developed during the last decade. Here, we evaluated their capability to detect the position of BPs, and also to predict the impact on splicing of variants occurring upstream of 3'ss. RESULTS: We used a large set of constitutive and alternative human 3'ss collected from Ensembl (n = 264,787 3'ss) and from in-house RNAseq experiments (n = 51,986 3'ss). We also gathered an unprecedented collection of functional splicing data for 120 variants (62 unpublished) occurring in BP areas of disease-causing genes. Branchpointer showed the best performance to detect the relevant BPs upstream of constitutive and alternative 3'ss (99.48 and 65.84% accuracies, respectively). For variants occurring in a BP area, BPP emerged as having the best performance to predict effects on mRNA splicing, with an accuracy of 89.17%. CONCLUSIONS: Our investigations revealed that Branchpointer was optimal to detect BPs upstream of 3'ss, and that BPP was most relevant to predict splicing alteration due to variants in the BP area.

Novel diagnostic tool for prediction of variant spliceogenicity derived from a set of 395 combined in silico/in vitro studies: an international collaborative effort.

Variant interpretation is the key issue in molecular diagnosis. Spliceogenic variants exemplify this issue as each nucleotide variant can be deleterious via disruption or creation of splice site consensus sequences. Consequently, reliable in silico prediction of variant spliceogenicity would be a major improvement. Thanks to an international effort, a set of 395 variants studied at the mRNA level and occurring in 5' and 3' consensus regions (defined as the 11 and 14 bases surrounding the exon/intron junction, respectively) was collected for 11 different genes, including BRCA1, BRCA2, CFTR and RHD, and used to train and validate a new prediction protocol named Splicing Prediction in Consensus Elements (SPiCE). SPiCE combines in silico predictions from SpliceSiteFinder-like and MaxEntScan and uses logistic regression to define optimal decision thresholds. It revealed an unprecedented sensitivity and specificity of 99.5 and 95.2%, respectively, and the impact on splicing was correctly predicted for 98.8% of variants. We therefore propose SPiCE as the new tool for predicting variant spliceogenicity. It could be easily implemented in any diagnostic laboratory as a routine decision making tool to help geneticists to face the deluge of variants in the next-generation sequencing era. SPiCE is accessible at (https://sourceforge.net/projects/spicev2-1/).

Large scale multifactorial likelihood quantitative analysis of BRCA1 and BRCA2 variants: An ENIGMA resource to support clinical variant classification.

The multifactorial likelihood analysis method has demonstrated utility for quantitative assessment of variant pathogenicity for multiple cancer syndrome genes. Independent data types currently incorporated in the model for assessing BRCA1 and BRCA2 variants include clinically calibrated prior probability of pathogenicity based on variant location and bioinformatic prediction of variant effect, co-segregation, family cancer history profile, co-occurrence with a pathogenic variant in the same gene, breast tumor pathology, and case-control information. Research and clinical data for multifactorial likelihood analysis were collated for 1,395 BRCA1/2 predominantly intronic and missense variants, enabling classification based on posterior probability of pathogenicity for 734 variants: 447 variants were classified as (likely) benign, and 94 as (likely) pathogenic; and 248 classifications were new or considerably altered relative to ClinVar submissions. Classifications were compared with information not yet included in the likelihood model, and evidence strengths aligned to those recommended for ACMG/AMP classification codes. Altered mRNA splicing or function relative to known nonpathogenic variant controls were moderately to strongly predictive of variant pathogenicity. Variant absence in population datasets provided supporting evidence for variant pathogenicity. These findings have direct relevance for BRCA1 and BRCA2 variant evaluation, and justify the need for gene-specific calibration of evidence types used for variant classification.

Genomic hallmarks of homologous recombination deficiency in invasive breast carcinomas.

Therapeutic strategies targeting Homologous Recombination Deficiency (HRD) in breast cancer requires patient stratification. The LST (Large-scale State Transitions) genomic signature previously validated for triple-negative breast carcinomas (TNBC) was evaluated as biomarker of HRD in luminal (hormone receptor positive) and HER2-overexpressing (HER2+) tumors. The LST genomic signature related to the number of large-scale chromosomal breakpoints in SNP-array tumor profile was applied to identify HRD in in-house and TCGA sets of breast tumors, in which the status of BRCA1/2 and other genes was also investigated. In the in-house dataset, HRD was predicted in 5% (20/385) of sporadic tumors luminal or HER2+ by the LST genomic signature and the inactivation of BRCA1, BRCA2 or RAD51C confirmed this prediction in 75% (12/16) of the tested cases. In 14% (6/43) of tumors occurring in BRCA1/2 mutant carriers, the corresponding wild-type allele was retained emphasizing the importance of determining the tumor status. In the TCGA luminal and HER2+ subtypes HRD incidence was estimated at 5% (18/329, 95%CI: 5-8%) and 2% (1/59, 95%CI: 2-9%), respectively. In TNBC cisplatin-based neo-adjuvant clinical trials, HRD is shown to be a necessary condition for cisplatin sensitivity. This analysis demonstrates the high performance of the LST genomic signature for HRD detection in breast cancers, which suggests its potential as a biomarker for genetic testing and patient stratification for clinical trials evaluating platinum salts and PARP inhibitors.

Germline BAP1 mutations predispose to renal cell carcinomas.

The genetic cause of some familial nonsyndromic renal cell carcinomas (RCC) defined by at least two affected first-degree relatives is unknown. By combining whole-exome sequencing and tumor profiling in a family prone to cases of RCC, we identified a germline BAP1 mutation c.277A>G (p.Thr93Ala) as the probable genetic basis of RCC predisposition. This mutation segregated with all four RCC-affected relatives. Furthermore, BAP1 was found to be inactivated in RCC-affected individuals from this family. No BAP1 mutations were identified in 32 familial cases presenting with only RCC. We then screened for germline BAP1 deleterious mutations in familial aggregations of cancers within the spectrum of the recently described BAP1-associated tumor predisposition syndrome, including uveal melanoma, malignant pleural mesothelioma, and cutaneous melanoma. Among the 11 families that included individuals identified as carrying germline deleterious BAP1 mutations, 6 families presented with 9 RCC-affected individuals, demonstrating a significantly increased risk for RCC. This strongly argues that RCC belongs to the BAP1 syndrome and that BAP1 is a RCC-predisposition gene.

Loss of heterozygosity at 13q13 and 14q32 predicts BRCA2 inactivation in luminal breast carcinomas.

BRCA2 is the major high-penetrance predisposition gene for luminal (estrogen receptor [ER] positive) breast cancers. However, many BRCA2 mutant carriers lack family history of breast/ovarian cancers and do not benefit from genetic testing. Specific genomic features associated with BRCA2 inactivation in tumors could help identify patients for whom a genetic test for BRCA2 may be proposed. A series of ER-positive invasive ductal carcinomas (IDCs) including 30 carriers of BRCA2 mutations and 215 control cases was studied by single-nucleotide polymorphism (SNP) arrays. Cases and controls were stratified by grade and HER2 status. Independently, 7 BRCA2 and 51 control cases were used for validation. Absolute copy number and Loss of heterozygosity (LOH) profiles were obtained from SNP arrays by the genome alteration print (GAP) method. BRCA2 tumors were observed to display a discriminatively greater number of chromosomal breaks calculated after filtering out and smoothing <3 Mb variations. This argues for a BRCA2-associated genomic instability responsible for long-segment aberrations. Co-occurrence of two genomic features-LOH of 13q13 and 14q32-was found to predict BRCA2 status with 90% of sensitivity and 87% of specificity in discovery series of high-grade HER2-negative IDCs and 100% of sensitivity and 88% of specificity in an independent series of 58 IDCs. Estimated positive predictive value was 17.2% (confidence interval: 6.7-33.5) in the whole series. In conclusion, the simplified BRCA2 classifier based on the co-occurrence of LOH at 13q13 and 14q32 could provide an indication to test for BRCA2 mutation in patients with ER-positive IDC.

Germline mutation in the RAD51B gene confers predisposition to breast cancer.

BACKGROUND: Most currently known breast cancer predisposition genes play a role in DNA repair by homologous recombination. Recent studies conducted on RAD51 paralogs, involved in the same DNA repair pathway, have identified rare germline mutations conferring breast and/or ovarian cancer predisposition in the RAD51C, RAD51D and XRCC2 genes. The present study analysed the five RAD51 paralogs (RAD51B, RAD51C, RAD51D, XRCC2, XRCC3) to estimate their contribution to breast and ovarian cancer predisposition. METHODS: The study was conducted on 142 unrelated patients with breast and/or ovarian cancer either with early onset or with a breast/ovarian cancer family history. Patients were referred to a French family cancer clinic and had been previously tested negative for a BRCA1/2 mutation. Coding sequences of the five genes were analysed by EMMA (Enhanced Mismatch Mutation Analysis). Detected variants were characterized by Sanger sequencing analysis. RESULTS: Three splicing mutations and two likely deleterious missense variants were identified: RAD51B c.452 + 3A > G, RAD51C c.706-2A > G, RAD51C c.1026 + 5_1026 + 7del, RAD51B c.475C > T/p.Arg159Cys and XRCC3 c.448C > T/p.Arg150Cys. No RAD51D and XRCC2 gene mutations were detected. These mutations and variants were detected in families with both breast and ovarian cancers, except for the RAD51B c.475C > T/p.Arg159Cys variant that occurred in a family with 3 breast cancer cases. CONCLUSIONS: This study identified the first RAD51B mutation in a breast and ovarian cancer family and is the first report of XRCC3 mutation analysis in breast and ovarian cancer. It confirms that RAD51 paralog mutations confer breast and ovarian cancer predisposition and are rare events. In view of the low frequency of RAD51 paralog mutations, international collaboration of family cancer clinics will be required to more accurately estimate their penetrance and establish clinical guidelines in carrier individuals.

Guidelines for splicing analysis in molecular diagnosis derived from a set of 327 combined in silico/in vitro studies on BRCA1 and BRCA2 variants.

Assessing the impact of variants of unknown significance (VUS) on splicing is a key issue in molecular diagnosis. This impact can be predicted by in silico tools, but proper evaluation and user guidelines are lacking. To fill this gap, we embarked upon the largest BRCA1 and BRCA2 splice study to date by testing 272 VUSs (327 analyses) within the BRCA splice network of Unicancer. All these VUSs were analyzed by using six tools (splice site prediction by neural network, splice site finder (SSF), MaxEntScan (MES), ESE finder, relative enhancer and silencer classification by unanimous enrichment, and human splicing finder) and the predictions obtained were compared with transcript analysis results. Combining MES and SSF gave 96% sensitivity and 83% specificity for VUSs occurring in the vicinity of consensus splice sites, that is, the surrounding 11 and 14 bases for the 5' and 3' sites, respectively. This study was also an opportunity to define guidelines for transcript analysis along with a tentative classification of splice variants. The guidelines drawn from this large series should be useful for the whole community, particularly in the context of growing sequencing capacities that require robust pipelines for variant interpretation.

EMMA, a cost- and time-effective diagnostic method for simultaneous detection of point mutations and large-scale genomic rearrangements: application to BRCA1 and BRCA2 in 1,525 patients.

The detection of unknown mutations remains a serious challenge and, despite the expected benefits for the patient's health, a large number of genes are not screened on a routine basis. We present the diagnostic application of EMMA (Enhanced Mismatch Mutation Analysis(®) , Fluigent, Paris, France), a novel method based on heteroduplex analysis by capillary electrophoresis using innovative matrices. BRCA1 and BRCA2 were screened for point mutations and large rearrangements in 1,525 unrelated patients (372 for the validation step and 1,153 in routine diagnosis) using a single analytical condition. Seven working days were needed for complete BRCA1/2 screening in 30 patients by one technician (excluding DNA extraction and sequencing). A total of 137 mutations were found, including a BRCA2 duplication of exons 19 and 20, previously missed by Comprehensive BRACAnalysis(®) . The mutation detection rate was 11.9%, which is consistent with patient inclusions. This study therefore suggests that EMMA represents a valuable short-term and midterm option for many diagnostic laboratories looking for an easy, reliable, and affordable strategy, enabling fast and sensitive analysis for a large number of genes.

Skipping Nonsense to Maintain Function: The Paradigm of BRCA2 Exon 12.

Germline nonsense and canonical splice site variants identified in disease-causing genes are generally considered as loss-of-function (LoF) alleles and classified as pathogenic. However, a fraction of such variants could maintain function through their impact on RNA splicing. To test this hypothesis, we used the alternatively spliced BRCA2 exon 12 (E12) as a model system because its in-frame skipping leads to a potentially functional protein. All E12 variants corresponding to putative LoF variants or predicted to alter splicing (n = 40) were selected from human variation databases and characterized for their impact on splicing in minigene assays and, when available, in patient lymphoblastoid cell lines. Moreover, a selection of variants was analyzed in a mouse embryonic stem cell-based functional assay. Using these complementary approaches, we demonstrate that a subset of variants, including nonsense variants, induced in-frame E12 skipping through the modification of splice sites or regulatory elements and, consequently, led to an internally deleted but partially functional protein. These data provide evidence, for the first time in a cancer-predisposition gene, that certain presumed null variants can retain function due to their impact on splicing. Further studies are required to estimate cancer risk associated with these hypomorphic variants. More generally, our findings highlight the need to exercise caution in the interpretation of putative LoF variants susceptible to induce in-frame splicing modifications. SIGNIFICANCE: This study presents evidence that certain presumed loss-of-function variants in a cancer predisposition gene can retain function due to their direct impact on RNA splicing.

Calibration of Pathogenicity Due to Variant-Induced Leaky Splicing Defects by Using BRCA2 Exon 3 as a Model System.

BRCA2 is a clinically actionable gene implicated in breast and ovarian cancer predisposition that has become a high priority target for improving the classification of variants of unknown significance (VUS). Among all BRCA2 VUS, those causing partial/leaky splicing defects are the most challenging to classify because the minimal level of full-length (FL) transcripts required for normal function remains to be established. Here, we explored BRCA2 exon 3 (BRCA2e3) as a model for calibrating variant-induced spliceogenicity and estimating thresholds for BRCA2 haploinsufficiency. In silico predictions, minigene splicing assays, patients' RNA analyses, a mouse embryonic stem cell (mESC) complementation assay and retrieval of patient-related information were combined to determine the minimal requirement of FL BRCA2 transcripts. Of 100 BRCA2e3 variants tested in the minigene assay, 64 were found to be spliceogenic, causing mild to severe RNA defects. Splicing defects were also confirmed in patients' RNA when available. Analysis of a neutral leaky variant (c.231T>G) showed that a reduction of approximately 60% of FL BRCA2 transcripts from a mutant allele does not cause any increase in cancer risk. Moreover, data obtained from mESCs suggest that variants causing a decline in FL BRCA2 with approximately 30% of wild-type are not pathogenic, given that mESCs are fully viable and resistant to DNA-damaging agents in those conditions. In contrast, mESCs producing lower relative amounts of FL BRCA2 exhibited either null or hypomorphic phenotypes. Overall, our findings are likely to have broader implications on the interpretation of BRCA2 variants affecting the splicing pattern of other essential exons. SIGNIFICANCE: These findings demonstrate that BRCA2 tumor suppressor function tolerates substantial reduction in full-length transcripts, helping to determine the pathogenicity of BRCA2 leaky splicing variants, some of which may not increase cancer risk.

Evaluation of in silico splice tools for decision-making in molecular diagnosis.

It appears that all types of genomic nucleotide variations can be deleterious by affecting normal pre-mRNA splicing via disruption/creation of splice site consensus sequences. As it is neither pertinent nor realistic to perform functional testing for all of these variants, it is important to identify those that could lead to a splice defect in order to restrict transcript analyses to the most appropriate cases. Web-based tools designed to provide such predictions are available. We evaluated the performance of six of these tools (Splice Site Prediction by Neural Network [NNSplice], Splice-Site Finder [SSF], MaxEntScan [MES], Automated Splice-Site Analyses [ASSA], Exonic Splicing Enhancer [ESE] Finder, and Relative Enhancer and Silencer Classification by Unanimous Enrichment [RESCUE]-ESE) using 39 unrelated retinoblastoma patients carrying different RB1 variants (31 intronic and eight exonic). These 39 patients were screened for abnormal splicing using puromycin-treated cell lines and the results were compared to the predictions. As expected, 17 variants impacting canonical AG/GT splice sites were correctly predicted as deleterious. A total of 22 variations occurring at loosely defined positions (+/-60 nucleotides from an AG/GT site) led to a splice defect in 19 cases and 16 of them were classified as deleterious by at least one tool (84% sensitivity). In other words, three variants escaped in silico detection and the remaining three were correctly predicted as neutral. Overall our results suggest that a combination of complementary in silico tools is necessary to guide molecular geneticists (balance between the time and cost required by RNA analysis and the risk of missing a deleterious mutation) because the weaknesses of one in silico tool may be overcome by the results of another tool.

A deep intronic mutation in the RB1 gene leads to intronic sequence exonisation.

Familial forms of retinoblastoma, an embryonic neoplasm of retinal origin, are caused by constitutional mutations of the RB1 gene. In this paper, we describe a family with retinoblastoma affecting two brothers with no previous family history of cancer. Complete RB1 mutational screening including point mutation and large rearrangement screening failed to demonstrate any mutation. The whole coding sequence was therefore investigated at the cDNA level, demonstrating a 103 bp intronic insertion between exons 23 and 24, leading to subsequent frameshift and premature termination of translation. This intronic exonisation was caused by a deep intronic mutation in intron 23 generating a cryptic 3' splice site. This is the first report of a deep intronic mutation in RB1 and is a proof of concept that some undetected RB1 mutations should be investigated at the cDNA level, particularly in hereditary forms of retinoblastoma.

Detecting splicing patterns in genes involved in hereditary breast and ovarian cancer.

Interpretation of variants of unknown significance (VUS) is a major challenge for laboratories performing molecular diagnosis of hereditary breast and ovarian cancer (HBOC), especially considering that many genes are now known to be involved in this syndrome. One important way these VUS can have a functional impact is through their effects on RNA splicing. Here we present a custom RNA-Seq assay plus bioinformatics and biostatistics pipeline to analyse specifically alternative and abnormal splicing junctions in 11 targeted HBOC genes. Our pipeline identified 14 new alternative splices in BRCA1 and BRCA2 in addition to detecting the majority of known alternative spliced transcripts therein. We provide here the first global splicing pattern analysis for the other nine genes, which will enable a comprehensive interpretation of splicing defects caused by VUS in HBOC. Previously known splicing alterations were consistently detected, occasionally with a more complex splicing pattern than expected. We also found that splicing in the 11 genes is similar in blood and breast tissue, supporting the utility and simplicity of blood splicing assays. Our pipeline is ready to be integrated into standard molecular diagnosis for HBOC, but it could equally be adapted for an integrative analysis of any multigene disorder.

Fidelity of DNA double-strand break repair in heterozygous cell lines harbouring BRCA1 missense mutations.

Germ-line mutations of the BRCA1 and BRCA2 genes, when they lead to a truncated protein, confer a high risk of breast and ovarian cancer. However, the role of BRCA1 missense mutations in cancer predisposition is unclear. Functional assays may be very helpful to more clearly define the biological effect of these mutations, and could therefore be useful in clinical practice. A recent study using a Host Cell End-Joining assay showed that a truncating mutation results in impaired fidelity of DSB repair by DNA end-joining. In the present study, we examined the fidelity of DSB repair in four lymphoblastoid cell lines with BRCA1 missense mutations. The fidelity of DNA end-joining was impaired in the four cell lines studied compared to the normal control cell line. The fidelity of end-joining was similar to that of a truncated mutation control cell line for one cell line and slightly higher for the other cell lines.

Significant contribution of large BRCA1 gene rearrangements in 120 French breast and ovarian cancer families.

Genetic linkage data have shown that alterations of the BRCA1 gene are responsible for the majority of hereditary breast-ovarian cancers. However, BRCA1 germline mutations are found much less frequently than expected, especially as standard PCR-based mutation detection approaches focus on point and small gene alterations. In order to estimate the contribution of large gene rearrangements to the BRCA1 mutation spectrum, we have extensively analysed a series of 120 French breast-ovarian cancer cases. Thirty-eight were previously found carrier of a BRCA1 point mutation, 14 of a BRCA2 point mutation and one case has previously been reported as carrier of a large BRCA1 deletion. The remaining 67 cases were studied using the BRCA1 bar code approach on combed DNA which allows a panoramic view of the BRCA1 region. Three additional rearrangements were detected: a recurrent 23.8 kb deletion of exons 8-13, a 17.2 kb duplication of exons 3-8 and a 8.6 kb duplication of exons 18-20. Thus, in our series, BRCA1 large rearrangements accounted for 3.3% (4/120) of breast-ovarian cancer cases and 9.5% (4/42) of the BRCA1 gene mutation spectrum, suggesting that their screening is an important step that should be now systematically included in genetic testing surveys.

Ploidy and large-scale genomic instability consistently identify basal-like breast carcinomas with BRCA1/2 inactivation.

BRCA1 inactivation is a frequent event in basal-like breast carcinomas (BLC). However, BRCA1 can be inactivated by multiple mechanisms and determining its status is not a trivial issue. As an alternate approach, we profiled 65 BLC cases using single-nucleotide polymorphism arrays to define a signature of BRCA1-associated genomic instability. Large-scale state transitions (LST), defined as chromosomal break between adjacent regions of at least 10 Mb, were found to be a robust indicator of BRCA1 status in this setting. Two major ploidy-specific cutoffs in LST distributions were sufficient to distinguish highly rearranged BLCs with 85% of proven BRCA1-inactivated cases from less rearranged BLCs devoid of proven BRCA1-inactivated cases. The genomic signature we defined was validated in a second independent series of 55 primary BLC cases and 17 BLC-derived tumor cell lines. High numbers of LSTs resembling BRCA1-inactivated BLC were observed in 4 primary BLC cases and 2 BLC cell lines that harbored BRCA2 mutations. Overall, the genomic signature we defined predicted BRCA1/2 inactivation in BLCs with 100% sensitivity and 90% specificity (97% accuracy). This assay may ease the challenge of selecting patients for genetic testing or recruitment to clinical trials of novel emerging therapies that target DNA repair deficiencies in cancer.