Why SDHB splice variants stall at VUS
Pathogenic germline SDHB variants predispose to pheochromocytoma and paraganglioma, and SDHB-associated disease is frequently aggressive, making accurate classification clinically urgent [1][2]. Yet approximately 40% of SDHB variants in ClinVar remain variants of uncertain significance, and intronic or synonymous variants are especially difficult because functional readouts and patient samples are rarely available [1]. Hereditary PPGL surveillance guidelines already differentiate surveillance by genotype, so a variant left as VUS can directly affect whether a family receives SDHB-style screening [2]. Foundational work has also shown that SDHB loss can be screened by immunohistochemistry, but weakly positive staining still requires gene-level resolution, leaving the same interpretive gap [10].
What the minigene assay changed
The anchor study built an SDHB minigene spanning exons 2–5 with flanking intronic sequence, introduced 48 SpliceAI-prioritized single-nucleotide variants plus two negative controls, expressed them in HEK293T cells, and quantified transcripts by targeted RNA sequencing [1]. Nineteen variants (38%) showed ≥90% wild-type splicing, 17 (34%) showed ≥90% aberrant splicing, and 73 aberrant transcripts were observed across all variants, averaging 2.3 per variant and collapsing to 22 unique transcripts [1]. RNA-based evidence strengths were assigned to 64 of 73 aberrant transcripts (88%), and among 26 classified variants, 10 received PVS1_Strong (RNA), two received PVS1_Moderate (RNA), and 14 received BP7_Strong (RNA) [1]. The practical consequence was a mean ACMG score shift of 2.7 points and reclassification of 13 variants (50%), dominated by 12 VUS-to-likely-benign downgrades and one likely-pathogenic-to-VUS downgrade [1].
SpliceAI prediction versus functional readout
SpliceAI is benchmarked and widely used, but this dataset shows its limits: among variants with unmasked SpliceAI Δ ≥0.42, 40% still showed ≥90% wild-type expression, and for five of those the masked SpliceAI score fell below 0.2 [1]. The authors note that masked scores would have reduced false-positive predictions to 29%, still leaving a substantial specificity problem for non-canonical intronic and exonic variants [1]. Competing computational work supports the broader point that prediction performance in real clinical settings is more modest than developer-reported benchmarks: a benchmark of ten tools on ABCA4 and MYBPC3 functional datasets found SpliceAI best for deep intronic ABCA4 variants but SpliceRover or consensus approaches better for other variant classes [7]. Newer transformer-based models report improved splice-site detection over SpliceAI-10k, but these gains are measured on annotation and RNA-seq benchmarks, not on SDHB-specific functional reclassification [4]. CI-SpliceAI and Splam similarly report incremental accuracy improvements, yet none of these tools resolves the need for transcript-level confirmation in a gene like SDHB [8][9].
How minigenes compare with RNA diagnostics and tumor evidence
Minigene assays are not new to variant classification. A BRCA2 exon 16 minigene study functionally tested 12 likely spliceogenic variants and found eight altered splicing, using fluorescent capillary electrophoresis to detect minor transcripts that agarose gels missed [3]. That precursor established minigenes as robust functional tools for clinical classification, but it also showed that transcript interpretation can be complex when full-length and aberrant transcripts coexist [3]. The SDHB study extends this lineage by using targeted NGS instead of gel-based detection: only 47% of transcripts identified by NGS were detectable by gel electrophoresis and Sanger sequencing, with small splicing alterations and unexpected effects particularly missed [1]. Parallel DNA/RNA NGS frameworks such as paRDal now operationalize PVS1(RNA) in hereditary cancer diagnostics, but they rely on patient RNA and can be confounded by nonsense-mediated decay, loss of heterozygosity, and low aberrant transcript expression [5]. The SDHB authors explicitly acknowledge this boundary: tumor RNA from two patients confirmed variant-consistent aberrant splicing but did not permit reliable quantification, whereas the minigene provided allele-independent measurement [1].
Limits of the reclassification claim
The reclassification result is real but bounded. The assay was performed in HEK293T cells, a non-chromaffin, non-neural-crest model, so it cannot fully represent tissue-specific splice regulation in paraganglia or pheochromocytoma [1]. The minigene contains only exons 2–5 and limited flanking intronic sequence, so three aberrant transcripts representing whole intron retention at low abundance (<5%) could not be reliably evaluated, and five variants near minigene boundaries had no functional assessment [1]. The authors applied a conservative maximum evidence weight of strong for functional splicing assays, following BRCA1/2 expert recommendations, and only about 50% of variants met conservative cutoffs for combined-strength aggregation [1]. An in-frame isoform (r.370_423del, p.Val124_Pro141del) increased above 10% abundance in eight variants but could not receive an RNA-based ACMG code because its protein-level impact is undefined [1]. Super-minigene work in SMN1/SMN2 shows that including native promoters, all exons, full introns, and UTRs can capture transcription and splicing regulation that truncated constructs miss, reinforcing that the SDHB exon 2–5 minigene is a targeted assay rather than a full gene model [6]. Finally, the clinical actionability of downgrading 12 VUS to likely benign depends on whether those variants are later observed in affected families, and the study's tumor comparison is limited to n = 2 [1].
About These Sources
This research page is built on 10 peer-reviewed studies — published from 2018 to 2026, 5 from 2024 or later, 1 in Q1 journals, collectively cited 217 times — selected as the most relevant from 13 studies that passed quality screening, drawn from 95 papers retrieved from a database of over 500 million.
Sources used in this answer
Minigene-based characterization and classification of splice-associated variants in succinate dehydrogenase B
The anchor SDHB minigene study functionally characterized 48 SpliceAI-prioritized variants in HEK293T cells, found 40% with high prediction scores showed ≥90% wild-type splicing, and reclassified 50% of classified variants, mostly VUS to likely benign, after integrating RNA evidence into ACMG scoring [1].
Update on Tumor Surveillance for Children with Hereditary Pheochromocytoma/Paraganglioma Syndromes.
This foundational review establishes that hereditary pheochromocytoma/paraganglioma syndromes require genotype-tailored tumor surveillance and that SDHB-associated disease can be aggressive, making accurate variant classification clinically consequential [2].
Identification of eight spliceogenic variants in BRCA2 exon 16 by minigene assays
This precursor BRCA2 exon 16 minigene study functionally tested 12 likely spliceogenic variants, found eight altered splicing, and demonstrated that fluorescent capillary electrophoresis detects minor transcripts missed by agarose gels, establishing minigenes as robust tools for clinical variant classification [3].
Transformers significantly improve splice site prediction
This competing computational study reports that a transformer-based model using 45,000-nucleotide context outperforms SpliceAI-10k on splice-site detection and pathogenic splice variant classification, but its gains are benchmarked on annotation and RNA-seq data rather than SDHB functional reclassification [4].
paRDal: Bioinformatics Framework for a Parallel RNA and DNA NGS Analysis in Hereditary Cancer Diagnostics
This validation framework integrates matched DNA and RNA NGS for hereditary cancer diagnostics, supports PVS1(RNA) under ACMG/AMP, and shows that patient RNA interpretation remains limited by nonsense-mediated decay, loss of heterozygosity, and low aberrant transcript expression [5].
A super minigene with a short promoter and truncated introns recapitulates essential features of transcription and splicing regulation of the <i>SMN1</i> and <i>SMN2</i> genes
This limitation study shows that an SMN2 super minigene with its own promoter, all exons, full introns, and 3′-UTR recapitulates transcription and splicing regulation better than truncated constructs, indicating that the SDHB exon 2–5 minigene cannot capture all native regulatory context [6].
Benchmarking deep learning splice prediction tools using functional splice assays
This competing benchmark of ten splice prediction tools on ABCA4 and MYBPC3 functional datasets found that the best-performing tool varied by variant class and that real-world clinical performance is more modest than developer-reported accuracy [8].
CI-SpliceAI—Improving machine learning predictions of disease causing splicing variants using curated alternative splice sites
This competing study retrained SpliceAI on curated GENCODE splice sites and reported a modest accuracy improvement over original SpliceAI on 1,316 functionally validated variants, but did not test SDHB variants [9].
Splam: a deep-learning-based splice site predictor that improves spliced alignments
This competing deep-learning study reports that Splam, a residual convolutional network using 400-bp windows and paired donor-acceptor training, is more accurate than SpliceAI at predicting human splice junctions and improves spliced alignment accuracy [10].
Efficacy of Immunohistochemistry for SDHB in the Screening of Hereditary Pheochromocytoma–Paraganglioma
This foundational immunohistochemistry study shows that SDHB-negative staining reliably predicts SDHx mutations in pheochromocytoma/paraganglioma, but weakly positive staining requires additional gene testing, leaving a diagnostic gap that functional splice assays could help address [12].