RNA-based diagnostic studies in genetics: Review and guidance from a multidisciplinary French network

Abstract

The widespread use of high-throughput sequencing for genetic diagnosis has led to considerable advances in patient care, but interpretation of the variants identified remains a challenge and geneticists routinely face the question of variants of uncertain significance. The clinical interpretation of genomic variants requires a high level of expertise to ensure appropriate genetic counseling. Assessing the impact of variants on splicing is a key issue in order to determine their pathogenicity as each variant can impact pre-mRNA splicing by disruption of the splicing code. It is for this reason that a diverse group of French molecular and clinical genetics experts from different diagnostic laboratories nationwide was established to discuss splicing issues and elaborate diagnostic recommendations. We describe an update of these recommendations with the aim of highlighting the importance of transcript characterization for variant interpretation and facilitating the diagnostic implementation of transcript studies, an important source of new diagnostics in human genetics.

Background

Use of short-read high-throughput sequencing technologies has led to an unprecedented upheaval in our knowledge of human genetics. This rapidly evolving knowledge, with its major medical impact, enables the rapid diagnosis of genetic diseases in a growing number of patients, their early detection or even prevention, appropriate follow-up, i.e what is known as precision medicine. However, patients and families still face obstacles in obtaining a genetic diagnosis and one reason for this diagnostic impasse stems from our incomplete knowledge of the functional effects of some variants, known as variants of uncertain significance (VUS). In a clinical setting, this uncertainty has major implications which is why it is mandatory to assess the functional impact of VUS on the protein in the case of amino acid substitutions, and on splicing in all cases, in order to state on their pathogenicity and relevance for genetic counseling. With regards to splicing, Cheung and coworkers [1] showed that close to 4% of ExAC variants lead to a splice alteration and cryptic splice variants are estimated to comprise 9–11% of pathogenic variants in patients with rare genetic disorders [2]. Moreover, 10% of exonic pathogenic variants reported in Human Gene Mutation Database (HGMD) have been found to affect splicing [3]. In other words, there are good reasons to look for splicing variants in introns as well as exons, as a number of intronic and translationally silent variants can be wrongly dismissed. The pre-mRNA splicing uses a number of cis-acting motifs (i.e. consensus sequences) recognized by the core spliceosomal components, the splice donor site (5′ss), the splice acceptor site (3′ss), the branch point (BP), the polypyrimidine tract (PPT) located between the BP and the 3′ss. More discrete cis-regulatory elements embedded in the exon or in flanking intronic sequences are also recognized by numerous trans-acting splicing factors (such as RNA-binding proteins or RBPs) to modulate exon inclusion and produce alternative transcripts. It is the combined effects of these elements that constitute the “splicing code” [4]. As a result, each variant (either exonic or intronic) can potentially affect normal pre-mRNA splicing and be deleterious via disruption of consensus sequences, creation of de novo sequences, or alteration of splicing regulatory elements making their detection mandatory for the genetic diagnosis of hereditary and somatic diseases and more broadly for genomic medicine [2, 5].

The splice GGC-ANPGM-NGSdiag group (splice GANG): a coordinated national initiative to facilitate the deployment of diagnostic RNA studies

The increase in the number of diagnostic DNA tests has not been matched by a corresponding investment in mRNA testing, and laboratories are generally less well trained in RNA-based approaches, leaving patients without a definitive diagnosis. Since 2008, the network of French oncogenetics laboratories has had a group dedicated to splicing issues with experts in both wetlab and bioinformatics aspects (the GGC splice group https://www.unicancer.fr/en/), which discusses splicing variants and draws up recommendations in conjunction with the ENIGMA network [5, 6]. Thanks to the network of French molecular geneticists (Association Nationale des Praticiens de Génétique Moléculaire, ANPGM,https://anpgm.fr/) and the NGS-DIAG group (https://ffgh.net/index.php/presentation/les-reseaux-partenaires/reseau-ngs-diagnostic), we have expanded this expert panel beyond cancer [4–10], reviewed the currently used strategies and established consensus guidelines to facilitate and speed up RNA testing implementation. The aim of this document is to guide laboratories through this complex field of RNA diagnostic testing, helping them to navigate between all the possible options for laboratory organization, tissue sampling, analytical strategies i.e. in silico selection of variants, RNA analyses, bioinformatics tools, eventually leading to variant classification. These different aspects will be treated successively.

Laboratory organization and equipment

Due to their structure, RNA molecules are more fragile than DNA molecules. Consequently, when handling them, it is very important to take measures to preserve their integrity. Ideally, they should be handled in a dedicated area (room or hood or bench), with consumables (including filter tips), reagents and dedicated equipment, all RNase-free. Handlers must wear gloves that must be changed regularly to avoid contamination with RNases, and a gown used exclusively for RNA handling. Work surfaces, pipettes, racks, etc. must be periodically cleaned with a solution that allows RNase decontamination. In the absence of a dedicated area, it is advisable to follow the good RNA handling practices listed above in order to establish an RNase-free environment. When the laboratory is required to work with human cells in culture, whether immortalized or not (primary cultures, stimulated lymphocytes or lymphoblastoid lines, for example), it must have a containment laboratory of a level appropriate to the risks involved. If genetically modified organisms are to be produced during the studies carried out (bacteria and cell cultures for minigene-type functional tests), the laboratory must comply with the laws of its country governing their use. Laboratories that do not have suitable premises can carry out part of their studies in research facilities, ensuring compliance with regulations and good handling practices.

Tissue sampling

RNA analysis may be performed on blood samples, biopsies, primary cells in culture (lymphocytes, skin fibroblasts, amniocytes, choriocytes, urothelial cells, etc.), lymphoblastoid cell lines or any other clinically accessible tissue (see “Expression and Tissue-Specificity Considerations” below). Immortalized cell lines are known to exhibit substantial transcriptomic changes compared to primary tissues, likely reflecting adaptation to in vitro culture and the immortalization process. Stabilizing RNA in a blood sample using a specialized collection tube (e.g. PAXgene) will allow a longer storage time before extraction compared to a sample taken on an EDTA tube. Correct homogenization of the blood and the stabilizer as well as compliance with the manufacturer’s recommendations for storage and turnaround time are essential to guarantee the integrity of the sample. The disadvantages of blood sampling include the variable proportion of blood cells, which hampers quantitative analysis, and the lack of nonsense mediated decay (NMD) inhibition [11]. The use of cultured cells (primary or immortalized) enables transcription to be studied in the presence of a NMD inhibitor (e.g. puromycin, emetine, cycloheximide or another compound with identical translation inhibition properties) that blocks the degradation of nonsense transcripts [12]. Inhibition of NMD prevents the degradation of the mutant transcript, which may otherwise be completely degraded and thus not detected, leading to false-negative results (see “RNA analyses” section). Given experimental variability and whenever possible, it is advisable to collect two blood tubes or to work on two different cell cultures in order to analyze independent duplicates. Analyses should always compare treated cells with untreated cells, to avoid any misinterpretation.

RNA extraction protocols

Many manual or automated protocols are available. Particular attention should be paid to extraction from tissues in order to guarantee the integrity of the extracted RNA [13]. Microelectrophoresis systems and fluorimetric assays have gradually replaced agarose gel deposit and absorbance measurement. Microelectrophoresis systems generate RNA integrity scores e.g. RNA Integrity Number, RIN and RNA Integrity and Quality number, RNA_IQ measured on a scale of 1 (RNA totally degraded) to 10 (RNA perfectly intact) [14]. The RIN score is based on a mathematical model that uses the ribosomal ratio. Below RIN 5, results of RNA studies should be interpreted cautiously as degraded transcripts and artefacts e.g. aberrant transcripts generated in vitro complicate the correct interpretation and could lead to inaccurate results [5]. In case of degraded RNA the DV200 metrics (percentage of RNA fragments > 200 nucleotides in size) is an interesting alternative. DNase treatment of extracted RNA is not essential for RT-PCR studies involving large introns but should be done for RNAseq applications.

Storage

Samples taken in the absence of stabilizer and requiring storage prior to extraction should be rapidly frozen at −80 °C. For other samples, storage must comply with the specific instructions for each stabilizer. After extraction, RNA should be aliquoted to avoid freeze/thaw cycles and stored at −80 °C. cDNA can be stored in aliquots at a minimum of −20 °C.

Expression and tissue-specificity considerations

Blood is the most convenient tissue to be sampled for RNA analyses. However, the results obtained in blood may not always reflect the actual mechanisms in the tissue of interest. This is why, and whenever possible, transcripts should be analyzed in the tissue involved in the pathology (e.g. muscle biopsy for muscle diseases) or in a clinically accessible tissue (CAT), making sure that the transcript of interest i.e. the one that predominates in the affected tissue, and related isoforms are expressed in this CAT. This information can be obtained from the literature, by analyzing the isoforms expression in different tissues, or in public data from expression studies (e.g. GTEx projects https://gtexportal.org). GTEx is a useful resource, especially for laboratories without previous experience on a gene, but also suffers from some limitations e.g. does not use globin depletion and expression discrepancies can be observed due to differences in sequencing protocols. The best database to use is an in-house database built with known samples and identical protocols. According to Bournazos and coworkers [15], a minimum of 0.2 TPM (Transcripts Per Million) would be required for RT-PCR study and 5 TPM for RNAseq study. If in doubt, it may be useful to check the expression of the gene in the tissue of multiple control individuals before obtaining the patient’s sample. In addition to the level of expression, it is also advisable to check that the transcript is similarly spliced in studied and affected tissues [16] e.g. by consulting the online tool https://tools.biociphers.org/majiq-cat/. In an ideal world, alternative splicing should be verified in the same tissue in multiple age- and sex-matched control individuals. However, outside of cancers, this is difficult to achieve, particularly in paediatric patients with neurodevelopmental diseases, but it does underline the importance of an in-house database based on the laboratory’s diagnostic expertise.

Variant prioritization for diagnostic RNA testing using in silico predictions

Assuming that each intronic and exonic nucleotide variation can have an impact on normal pre-mRNA splicing, this means that all sequenced variations would have to be tested in RNA, which is currently complex from a practical point of view. For this reason, in silico prediction tools for splicing anomalies have been widely used for decades and the French splicing group published guidelines as soon as 2012 with a simple combination of two bioinformatics variation scores MaxEntScan (MES) and Splice Site Finder-like (SSF-like) [5]. In silico tools are based on the amount of available transcript sequencing data and our accumulated knowledge of the splicing mechanisms, our current knowledge of the consensus sequences of splice regulatory elements and the splicing mechanism in general. Experience and the literature show that while predictors have over the years become of high quality for 5′, 3′ splice sequences, and even the branch point, they can be faulty for distal auxiliary splicing regulatory elements. Despite these limitations, these in silico tools have become an indispensable part of the splicing diagnostic strategy to select variants that are worthy of expensive and time-consuming RNA analyses. They can be used either as stand-alone programs or as interfaces integrating multiple algorithms. There are a large number of in silico prediction tools available, and to list them all is beyond the scope of this paper. Each new tool published claims superior performance to existing tools, with sometimes varying levels of proof. For these reasons, it is interesting to refer to the latest edition of the CAGI challenge [17], which performed a blinded comparison of 8 teams (for a total of 14 prediction methods) volunteering for this interpretation challenge on a set of 56 splice variants disseminated between exons and deep introns. The popular SpliceAI tool was widely used across the different prediction strategies with different decision-making thresholds to determine positive prediction and/or weighting by other parameters such as MAF. Recently, a free online tool called Splice AI-visual, based on the SpliceAI algorithm, has been developed to improve the use of SpliceAI scores, notably adding a user-friendly graphical presentation [10]. Splice AI-visual is fully integrated in the free online variant interpretation tool, MobiDetails [18]. As expected, the accuracy with which splice effects of variants are predicted depends much on their distance to a 5′ or 3′ splice site and whether it is located in an exon or deep in an intron, see Supplementary Fig. 1 [17, 19]. Overall, SpliceAI weighted by allele frequency and Splicing Prediction Pipeline (SPiP) [4] were found to be the most accurate [17]. No predictor can or will ever achieve 100% sensitivity and specificity. A trade-off is therefore necessary, as high sensitivity (allowing a maximum of true positives) is often obtained at the cost of lower specificity (more false positives). Based on the existing literature and the group’s expertise [2, 4], we recommend performing RNA analysis if SpliceAI is above 0.2 and SPiP predicts an alteration. If SpliceAI is below 0.1 and SPiP predicts no alteration with a confidence interval below 5%, then RNA analysis should not be done, unless the clinical phenotype is highly specific or the intron is spliced with a minor spliceosome. MobiDetails (mentioned above) summarizes the scores of most frequently used free tools as MaxEntScan, SPiP, dbscSNV, SpliceAI, AbSplice Max and spliceAI-visual, thus allowing a quick and reliable overview of the effect of a variant on splicing. Other situations have to be discussed on a case by case basis (see “classification” below). Importantly, SPiP and Splice AI have completely different algorithms and can therefore be used in combination. This is not the case for all the tools available, so the user must check that the algorithms are not redundant in order to be sure of obtaining complementary predictions.

RNA analyses

RT PCR and sanger sequencing

Although the detection limit for Sanger sequencing is generally estimated at 10%, RT-PCR followed by Sanger sequencing of the PCR products is a simple and cost-effective targeted RNA-approach to prioritize when the gene of interest is expressed in a tissue that is easily accessible and relevant to the disease [20]. RT-PCR can be performed in a one-step or a two-step assay [21]. One-step RT-PCR utilizes sequence-specific primers and combines the reverse transcription and the subsequent PCR in a single assay. In the two-step RT-PCR, RT and PCR reactions are performed separately allowing more flexibility for RT priming (oligo(dT) primers, random primers, and gene-specific primers) and PCR optimization. Hence one-step may be advisable for large sample quantities and high throughput applications while two-steps is suitable for low target abundance and difficult PCR experiments. In any case, both approaches have advantages and choosing one or the other will also depend on local practices and expertise of the laboratory. Several key points must be addressed to avoid technical pitfalls and misinterpretation of results. First, particular attention must be paid to primer design to avoid false-negative/positive PCR results and ensure detection of the various possible splicing alterations (Supplementary Fig. 2). This requires a thorough in silico analysis before designing primers, ensuring the designed amplicon covers wild type splice sites, potential cryptic splice sites and all splicing alterations predicted in the variant context (see “Variant prioritization for diagnostic RNA testing, in silico predictions”, above). If an allele drop-out is suspected, e.g. because of a large intronic retention exceeding the polymerase range, an additional pair of primers targeting the most likely splicing alteration should be designed to promote its amplification. Primers should be designed at exon junctions or large introns included in the design, to prevent amplification from contaminating DNA. Whenever possible, heterozygous SNPs should be included in the amplicon to check for transcriptional heterozygosity (i.e. that both alleles are expressed and to discern complete from partial mis-splicing) [5, 15]. Second, whenever possible, the following good practices should apply to ensure diagnostic grade experiments. Two independent RT-PCRs using different primer pairs should be performed with multiple tissue-matched controls and affected cases or heterozygotes to demonstrate reproducible variant-induced mis-splicing. When these conditions cannot be met, the experiment should at least be duplicated. If no splicing anomaly is observed by RT-PCR and no heterozygous variant is present in the amplified region (e.g. for an intronic variant), additional quantitative analysis such as RT-qPCR may be necessary to check that both alleles have been amplified before concluding that there is no effect on splicing. In addition, a simple RT-PCR analysis targeting an exonic SNP is relevant to test if a candidate variant affects transcript expression (e.g. a variant located in regulatory regions) or transcript degradation (e.g. a truncating variant located before the last exon that is expected to induce NMD). Interestingly, this indirect approach could also be used in the absence of a candidate variant, in cases of strong unexplained clinical suspicion, in order to provide arguments in favor of the gene’s involvement in the disease.

RNA-seq

Thanks to the democratization of high-throughput short-read sequencing, RNA-seq is increasingly moving from the field of research to that of diagnostics. RNA-seq analysis can be used in two ways for the molecular diagnosis of genetic diseases, depending on whether or not a genomic anomaly is first suspected. First, it can be used to validate a candidate variant by finely assessing its impact on the transcripts of the gene of interest (hereinafter referred to as targeted analysis). Second, RNA-seq analysis can be carried out on samples from undiagnosed patients/families in order to search for transcriptional anomalies that would explain the phenotype and indirectly lead to the genomic cause (hereinafter referred to as pantranscriptomic analysis). These two strategies will be presented successively.

Messenger RNAs represent only 3–7% of total RNAs compared to 80–90% ribosomal RNA and 10–15% transfer RNA [22]. For RNA-seq applications, it is therefore necessary to enrich in RNA of interest prior to NGS library preparation. This enrichment can be achieved either by depletion (e.g. total RNA with ribosomal RNA depletion), by amplification of RNAs carrying a poly-A tail [23, 24], or by hybridization with DNA or RNA probes targeting restricted exonic regions or all exons (panel and exome capture, respectively). The enrichment method chosen depends on the quality of the initial RNA. In the case of medium-quality RNA (RIN < 7), poly-A enrichment does not provide good coverage, particularly of 5′ regions. In this case, a capture method is recommended. In fact, gene capture panels are increasingly used in routine diagnostics, and are particularly interesting for weakly expressed genes [25, 26]. This approach enables sequencing of a few genes to over a hundred transcripts of interest. Targeted analysis approaches are very effective to detect an impact on splicing, but quantifying this impact remains challenging, even if the significance of the effect can be estimated as a function of the number of reads with the anomaly. These methods remain semi-quantitative, probably due to the biases introduced by capture and PCR during the various stages of library preparation.

Bioinformatics analyses are an important part of RNA-seq and require specific resources and expertise which can be a potential obstacle to its implementation in a diagnostic setting. They consist of three main steps. First, the sequencing data are aligned to a reference sequence, the genome rather than the transcriptome, in order to optimize the quality of the analysis. The reason is that pseudo-exons or partial or total intronic retentions would be incorrectly detected if the sequencing data are aligned to a transcriptome. A number of alignment tools exist, of which STAR [27] is the most widely used. At this stage, the alignment files can be used to view the locus of interest in a browser, such as Integrative Genomics Viewer, to examine transcripts and “visually” detect splicing anomalies. Second, transcripts and splice junctions are detected. Third, statistical analysis of transcript expression and splice junctions is performed, a crucial step for data interpretation. To date, there are tools for detecting several types of anomalies (splice junctions, expression levels, allelic balance), such as SpliceLauncher [28] or DROP [29]. It is important to evaluate these tools according to the type of data to be explored (e.g. gene expression studies, alternative junctions …), as well as according to the various previously published benchmarks [30]. It is also crucial to network with experts in order to maintain a technological watch on these constantly evolving tools, and to standardize analyses between laboratories, notably through expert groups and networks (e.g. BioInfoDiag in France https://bioinfo-diag.fr/). Definition of thresholds or parameters to be used with these tools is beyond the scope of this review, however, beyond the choice of bioinformatics tools, certain recommendations apply to guarantee the quality of results. Thus, a minimum coverage depth of 50x for regions of interest must be obtained, particularly for the study of alternative splicing. Due to the low diversity of alignment files, deduplication is also strongly recommended by the use of UMIs (Unique Molecular Indexes) see Supplementary Fig. 3. An identity verification step should be performed to ensure that previous DNA sequencing and the corresponding transcriptome analysis have been indeed performed on samples from the same individual. Variants identified by DNA sequencing (panel, exome, genome) can be searched for in the RNAseq data. Please note that given the limitations of current bioinformatics tools for detecting loss of allelic expression, we recommend objectifying expression variations by searching for exonic single nucleotide variants. Ideally, the variants considered should be as close as possible to the region impacted by the VUS. If the splicing anomaly identified by RNAseq is clearly characterized, confirmation by RT PCR is not necessary. On the other hand, if RNAseq interpretation is tricky or if there is no effect on splicing for a given variant, it is necessary to confirm that both alleles are expressed by examining the allelic ratio for closely located variants and/or by using a NMD inhibitor to restore a possible out-of-frame transcript (see above).

Finally, exploratory, pantranscriptomic RNAseq-short read approaches, in the absence of identified VUS, can be used to search for splice junction or global gene expression abnormalities. They make it possible to identify anomalies caused by deep intronic variants not detected by panel or exome sequencing of exonic regions and thus, indirectly, to identify these so-called cryptic variants [23, 24]. The most widely used analysis pipeline for this purpose is DROP [29], which uses FRASER to detect splicing anomalies and OUTRIDER to detect expression anomalies. At the time of writing, these approaches are not part of the routine diagnostic landscape. To reach diagnostic routine, at least positive control samples with various splicing and expression anomalies characterized by other techniques should be used to determine the sensitivity of the tools to the set thresholds for p-values, false detection rates, log2 fold changes or deltaPsi. For these tools, at least 50 to 100 samples, either healthy controls or affected samples with heterogeneous genetic defects are recommended in order to reach statistical significance. Users will need to master the version of each tool using these controls, ideally stored in an image (e.g. SINGULARITY or DOCKER container) and perform a non-regression test.

Minigene assays

The most effective and rapid approaches to determine whether a genetic variant is responsible for mis-splicing are based on the analysis (RT-PCR, RNA sequencing) of RNA molecules extracted from a biological sample. However, patient samples are often not available or not relevant (e.g. absent/low expression of certain genes). Another approach is to test the effects of the variant on splicing using a minigene vector. This technique is time-consuming (average delay of two weeks) and requires dedicated laboratory environments (Fig. 1A–D and Supplementary Fig. 4), but it does have the advantage of allowing the variant to be tested in a perfectly defined mono-allelic context (i.e. the wild type and the mutated sequences are tested separately). Moreover, it isolates the variant of interest from other variants present in cis in the patient’s genomic sequence, which in turn may facilitate the interpretation of RNA splicing profiles and the demonstration of a causal link. Pros and cons of the minigene approach are summarized Table 1. Detailed strategies for the use of different minigene reporters and different assays to reveal splicing abnormalities or study splicing regulation have been described [31–33]. Here, we would like to highlight key points to be taken into account in the design process. First, the minigene vector must contain a 5′-promoter sequence that will enable transcription in a large variety of mammalian cell lines, and a 3′-AAUAAA sequence motif that will permit polyadenylation of the generated transcripts and promote their export to the cytosol. The backbone is typically derived from a commercial plasmid, containing a viral promoter sequence (i.e. CMV, RSV, SV40) and the bovine growth hormone (BGH) polyadenylation signal [32–34]. Some vectors have been optimized to: (i) reduce transcriptional activity and allow a more physiological coupling with spliceosome assembly and splicing catalysis, (ii) facilitate the analysis of first or last exons [35], or (iii) prevent the selective degradation of transcripts containing premature termination codons through the NMD pathway. Second, the genomic fragment to be cloned into the minigene reporter must contain all the primary elements (canonical donor and acceptor splice sites, branch point) that are required for the splicing of the exon in which, or near which, the variant to be tested is located. A segment encompassing the target exon and ~150–200 bp of the corresponding flanking intronic sequences is a reasonable starting point. A larger genomic fragment may be preferred if: (i) the size of the flanking intronic regions are naturally small, (ii) the physiological alternative splicing pattern in the region of interest implicates several exons (i.e. if an alternative exon is identified in an intronic sequence of the reference pre-mRNA by comparing the different transcripts referenced in human databases), (iii) the flanking intronic sequences are bioinformatically predicted to contain distant cryptic splicing donor and/or acceptor sequences (see in silico predictions). Assessing the suspected splice variant in a broadly genetic context can help identify variants that may promote the retention of a large intronic sequence or the aberrant inclusion of a poison exon, and thus avoid erroneous interpretations about effects at the protein level. For example, a variant affecting the first base of an intron, thereby inactivating the canonical donor splice site, could lead to the use of a pre-existing proximal cryptic site that is suboptimal compared to another deeper intronic site not included in the construct. Although the negative impact on splicing of the variant would be confirmed, the interpretation could be that of the insertion or deletion of a few amino acids (with a possible moderate effect on protein activity) rather than the production of a truncated protein.

Fig. 1. The splicing minigene assay.

A Cloning of the genomic segment of interest into the minigene by using PCR product from an heterozygote patient gDNA, or direct introduction of the variation in the wild-type construct by mutagenesis B Transient transfection of the wild-type and mutant minigenes in parallel into mammalian cells C Analysis of the splicing patterns. RNA extraction from the transfected cells, RT-PCR, gel electrophoresis, Sanger sequencing D Relative quantification of the transcripts by fluorescent RT-PCR and capillary electrophoresis.

Table 1.

Pros and cons of minigene assays.

Advantages (pros)	Limits (cons)
Suitable to all type of genes and exons.	Limitation of the size of the cloned genomic fragment; may not be compatible with deep intronic variants. Based on an artificial genomic context, with hybrid intronic sequences that combine the canonical splice sites of different genes.
Makes it possible to assess the effect of the variant in a mono-allelic context.	No comprehensive assessment of the effect of the variant (on full endogenous transcript splicing); possible discrepancies between minigenes and patient RNA results.
Enables the study of weakly transcribed genes in patient samples.	Does not respect physiological context; based on cell lines that may lack tissue-specific splicing regulation.
Allows multiple mutated sequences to be evaluated and compared with the wild-type sequence in a fully controlled genomic and cellular context.	Does not maintain physiological and genomic contexts.
Enables NMD system inhibitors to be tested.	No comprehensive assessment of the effect of the variant on a full endogenous transcript.
Cost effective.	Time consuming.

Each minigene vector should ideally be validated before being used to characterize a VUS. This can be done by comparing minigene results with RT-PCR or RNA-seq data from patients or carriers with well-defined variants (with or without splicing defects). Another possibility is to draw on literature data and integrate variants with a well-defined effect on splicing as positive and negative controls. A third option is to select unreported variants that affect a critical base of a donor or acceptor splice site as positive controls. In any case, the methods used to reveal the effects of the variants will have to be adapted to the size of the constructs and of the alternative transcripts likely to be detected. Finally, Sanger sequencing will have to be performed to accurately characterize aberrant transcripts. It is worth mentioning that different strategies derived from minigenes are now described in the literature. Ultimately, this may consist in cloning an entire gene sequence i.e. all exonic and intronic sequences. The advantage is also that several variants can be tested using a single construct [36], but it can only work for small genes (<8–10 kb), which are actually quite rare in the human genome. Another possibility is to shorten the intronic sequences. This maintains the possibility of testing each variant in the context of a physiological association between donor and acceptor sites, in contrast to the conventional exon-trapping method, which is based on the use of unpaired canonical splicing sequences [37]. It could also be useful to focus on exon clusters, with cloning large blocks of whole exonic and intronic sequences. Sangermano et al. coined the term “midigene” to designate this type of splicing reporters [38]. The benefits of these alternative strategies have now been demonstrated by various groups [35, 38–40]. Difficulties in interpreting the results may however arise when alternative splicing patterns are observed [40]. Finally, high-throughput splicing assays have been developed [41]. These approaches offer the advantage of testing a large number of VUS in parallel and comparing their functional effects within the same experiments. They can also be used to perform saturation-scale analyses of variant effects on sequences deemed important for the splicing process and its regulation. These datasets may be generated within the frame of a call for a community activity to eventually help address variant classification questions more comprehensively and potentially enhance the predictions of in silico tools.

To conclude, the minigene approach can help to answer the question of the effect of a variant of unknown significance on splicing and thus aid its classification, when patient samples are not available or are irrelevant.

Variant classification

Several scientific societies, including ACMG/AMP (The American College of Medical Genetics and Genomics & the Association for Molecular Pathology) [15, 42, 43], ESHG-Eurogentest (The European Society of Human Genetics - EuroGentest Committee) [44], have issued recommendations aimed at homogenizing and standardizing the interpretation of variants identified in genetic diagnostic laboratories. These are based on a scoring system for clinical and biological data (objective or predicted) likely to contribute to the clinical interpretation of variants. Today, this scoring system constitutes a standard for the interpretation of variants identified by high-throughput sequencing. We now propose several adjustments to the currently used ACMG/AMP codes and the recently proposed rules for splice-affecting variants [45] in order to better interpret the variants with a predicted or confirmed effect on splicing (Figs. 2 and 3). The confidence placed in direct patient RNA studies is generally higher than that placed in experimental studies using mini-/midi-genes or simple in silico predictions. The arguments used to interpret variants are therefore weighted accordingly to these study designs or to these predictions. Similarly, PVS1 (RNA) should be used for RNA splicing studies, leaving PS3 for functional studies [45].

Fig. 2. Decision tree and classification options based on in silico predictions and before RNA analyses.

The PVS1 code is only valid for genes with a well-established “gene-disease” link, and for which loss of function is a known pathogenic mechanism. PVS1 should not be used in combination with PP3. The weight of the PVS1 code should be modulated according to context: altered region forming part of the physiological transcript and known to be critical for protein function or of unknown role; NMD predicted or not; possible use of a cryptic splice site, existence of alternative transcript(s) likely to exert rescue [40]. Possibility of adding PS1 if the variant evaluated has the same predicted +/- confirmed effect as another variant at the same nucleotide already classified as pathogenic (i.e. class 5) (or PS1_Moderate if same effect as another variant already classified as probably pathogenic (i.e. class 4); or a PS1_Moderate if the variant has the same effect (predicted or confirmed) as another variant located in the same splicing donor/acceptor motif already classified as pathogenic (or PS1_Supporting if same effect as another variant already classified as probably pathogenic).

Fig. 3. Decision tree and classification options based on RNA analyses.

The PVS1 code is only valid for genes with a well-established “gene-disease” link, and for which loss of function is a known pathogenic mechanism. PVS1 should not be used in combination with PP3. The weight of the PVS1 code should be modulated according to context: altered region forming part of the physiological transcript and known to be critical for protein function or of unknown role; NMD predicted or not; possible use of a cryptic splice site, existence of alternative transcript(s) likely to exert rescue [40]. PVS1 (RNA) for mini/midi genes studies should only be used for constructs validated for the exon of interest as some exons may be under or overexpressed depending on the construct used. Evidence weighting should be used when good practices are met (see text).

The weight of these arguments will be modulated according to the level of knowledge/expertise concerning the gene and associated disease (Figs. 2, 3), since standard approaches can lead to misinterpretation (Supplementary Fig. 5) [46]. Moreover, some naturally occurring alternate gene transcripts may skip truncating variants, encoding functional or partially functional proteins, and resulting in reduced or no haplo-insufficiency (i.e. rescue transcripts) [47]. This underlines the importance of gene/disease expert groups to discuss complex cases e.g. isoform level modifications, leaky variants with partial defects (see below).

The recommendations described herein are applicable only to genes clearly implicated in the disease for which the analysis has been carried out. If a variant is identified in a gene for which the level of evidence of its involvement in the concerned disease is insufficient, the maximum classification attributable to this variant is “VUS”[48, 49]. The gene-disease link may be established by an international panel of experts e.g. ClinGen (https://search.clinicalgenome.org/kb/curations/), PanelApp (https://panelapp.genomicsengland.co.uk/; https://panelapp-aus.org/), DDG2P (https://decipher.sanger.ac.uk/ddd/ddgenes) or on the basis of a consensus decision by a national panel of experts.

The case of leaky variants and isoform levels

Straightforward splicing or expression anomalies, such as the extinction of an allele due to an out-of-frame defect, can be easily interpreted as a loss of function and lead to an unambiguous diagnosis. While these cases do exist, the observed transcripts often include a complex mixture of normal and abnormal splicing isoforms, caused by “leaky” variants. These variants cause incomplete splicing defects and allow residual production of wild type transcripts (at a possibly non-pathological level), raising many clinical questions, whatever the method used to identify them [50]. Facing a partial defect, it may be difficult to decide between an anomaly responsible for the phenotype observed in the patient and therefore usable in diagnosis, a modifying factor, or even an anomaly without any repercussions. Moreover apparent leaky splicing may actually result from technical limitations and/or incomplete NMD (see above). The situation is further complicated by the fact that the partial effect observed in a clinically accessible tissue or a minigene does not necessarily reflect the impact in the tissue of interest, as splicing is tissue-specific. Variations in the expression level of physiological isoforms are also difficult to interpret in a clinical context, especially as some isoforms are known to rescue the phenotype [51]. A thorough knowledge of the gene of interest is therefore required. More generally, all these complex cases require a cross-interpretation of the data obtained with the literature, functional data and, above all, clinical data. We strongly recommend joining/consulting with clinical expert groups to reach collegial decisions. It is also desirable to refine the ACMG/AMP specifications for these cases, but this requires shared methodologies and well-defined quantification methods on a gene by gene basis. Overall it underlines the fact that these complex situations, not addressed by the current ACMG/AMP codes, need to be discussed on a case-by-case basis and also highlights the need for expertise.

Concluding remarks

We encourage teams to develop transcript studies because of the clinical benefit: not only new diagnoses, but also reclassification of VUSs, the latter being very important for the diagnostic community. To this end, we have reviewed the various highlights of these studies. The easiest and effective thing to do for a diagnostic laboratory without prior experience in RNA analysis, is to set up RT-PCR Sanger analyses, following the precautions described herein. Then, larger analyses requiring bioinformatics expertise (RNAseq) or cell culture (minigene) can be implemented. RNA analysis is a fast-growing arena of research that is fueling the diagnostic field. New minigene methods are constantly being developed to limit interpretation issues and enable the evaluation of a larger number of variants. Similarly, exploratory approaches using long-read RNA sequencing are under development. They have the particular advantage of identifying the phase of abnormal transcripts but require dedicated bioinformatics analysis pipelines. Very recently, a promising transactivation and transdifferentiation assay was described to circumvent insufficient gene expression in clinically accessible tissues [52]. All these advances offer hope for patients and their families, but require laboratories to be able to transfer them to diagnostics.

Author contributions

All authors took part in the regular meetings required to revise the guidelines; all authors participated in writing and/or reviewed, amended and approved the final text. CH compiled all the recommendations into the final text.

Funding

No specific funding.

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41431-025-01881-2.