RNA sequencing provides functional insights and diagnostic resolution in previously unsolved rare disease cases

Abstract

Exome and genome sequencing have greatly improved the diagnosis of rare genetic disorders but remain limited in their ability to identify and classify non-coding variants, including intronic variants, cryptic splice-site alterations, and disruptions in regulatory regions. RNA sequencing (RNA-seq) has emerged as a powerful tool to bridge this gap by providing functional insights into genomic variants that disrupt splicing or gene expression, thereby aiding in variant interpretation and classification. We retrospectively reviewed 30 cases from the Utah Penelope Program and the Undiagnosed Diseases Network over a three-year period, in which RNA-seq was performed on whole blood and/or fibroblasts following either negative DNA sequencing or the identification of candidate variants requiring functional assessment. In these cases, RNA-seq identified exon skipping, cryptic splice-site activation, and intron retention, leading to transcript disruption. Additionally, positional enrichment analysis clarified X-inactivation patterns and dosage effects, confirming the pathogenicity of copy number variants. By detecting these transcript-level alterations, RNA-seq provided functional evidence supporting the reclassification of multiple variants of uncertain significance, contributing to diagnostic resolution in selected cases. This study underscores the clinical utility of RNA-seq in detecting splicing and regulatory defects that DNA sequencing and predictive tools alone cannot resolve. Integrating RNA-seq into clinical workflows can support variant classification, aid in diagnostic resolution for selected cases, and provide mechanistic insights into genetic disorders, contributing to patient care and genetic counseling.

Introduction

Genetic testing through exome and genome sequencing has revolutionized the diagnosis of rare genetic disorders by enabling the identification of pathogenic variants across a broad spectrum of genes. These modalities have demonstrated higher diagnostic yields compared to targeted gene panels, reducing the number of missed diagnoses and improving the detection of novel disease-associated variants [1, 2]. Additionally, exome and genome sequencing report fewer variants of uncertain significance compared to gene panels due to their phenotype-driven variant interpretation and reporting [3]. This enhanced diagnostic capability can lead to earlier and more accurate diagnoses, facilitating timely medical management, informed reproductive decision-making, and improved access to condition-specific resources and clinical trials for patients and families. However, despite these advances, a significant proportion of patients with suspected genetic conditions remain without a definitive molecular resolution, limiting these opportunities.

One major limitation of exome and genome sequencing is their inability to detect or interpret certain classes of pathogenic variants. Up to 55% of diagnoses in previously unsolved rare disease cases involve causative variants that were not effectively interrogated by prior testing [4]. Exome sequencing primarily captures protein-coding regions, limiting its ability to identify intronic variants, variants in regulatory regions, and structural variations that affect gene expression. The probe capture step in exome sequencing defines its regions of interest, which varies by laboratory, but typically extends no further than ± 20 nucleotides from an exon boundary, leaving potentially pathogenic intronic and regulatory variants completely undetected [5]. In contrast, genome sequencing covers intronic and intergenic regions but often fails to classify rare noncoding variants as pathogenic due to a lack of functional evidence linking the variant to a genomic effect. Even with bioinformatic splice prediction tools which provide supporting evidence of pathogenicity, intronic variants detected by genome sequencing are frequently classified as variants of unknown significance (VUS) due to the absence of experimental validation demonstrating their impact on splicing which provides up to very strong evidence for pathogenicity [6]. These challenges highlight the need for complementary approaches to provide functional evidence and improve diagnostic rates.

RNA sequencing (RNA-seq) has emerged as a powerful tool in clinical genetics, addressing some of the limitations of exome and genome sequencing by providing direct functional evidence of transcript-level disruptions [7]. This approach enables the detection of aberrant splicing events such as exon skipping, cryptic splice-site activation, and intron retention that may arise from noncoding variants not captured or with uncertain consequences in DNA sequencing [8]. Additionally, RNA-seq can identify allele-specific expression imbalances, indicating loss of function due to nonsense-mediated decay or regulatory disruptions affecting gene expression [9]. Multiple studies have demonstrated that incorporating RNA-seq into clinical workflows increases the diagnostic yield of genome and exome sequencing while also reducing the number of VUS reported [10]. By providing functional evidence of variant pathogenicity, RNA-seq facilitates variant interpretation and enables the reclassification of VUS, ultimately improving outcomes for patients with rare genetic disorders [6].

Despite its potential, the clinical integration of RNA-seq remains limited to specialized research and diagnostic programs, often due to challenges related to tissue accessibility, standardization, and interpretation of RNA-based findings [11]. However, growing evidence supports its utility in resolving complex cases that would otherwise remain undiagnosed [12, 13]. In this study, we present a case series of participants evaluated at the University of Utah in whom RNA-seq contributed to the identification or confirmation of molecular diagnoses in cases where exome or genome sequencing results were negative, inconclusive, or required additional functional assessment. By highlighting these cases, we demonstrate the clinical value of RNA-seq in diagnostic workflows and provide insight into the types of variants and disease mechanisms that can be uncovered through transcriptomic analysis.

Methods

The cohort included two main categories of cases: (1) patients with candidate splice-altering variants identified by prior genome or exome sequencing where the molecular findings had partial or uncertain clinical overlap with the reported phenotype, and (2) patients without any diagnostic candidate variants but with a high suspicion of an underlying monogenic disorder based on clinical evaluation. The inclusion of patients with putative splice variants enriched the cohort for cases likely to benefit from transcript-level assessment, contributing to the relatively high overall diagnostic yield observed in this study.

RNA was extracted from whole blood and skin fibroblasts. For the Undiagnosed Diseases Network (UDN) cohort, libraries were prepared using a stranded, polyA-enriched kit (Illumina), while the Utah Penelope Program used a stranded Ribo-Zero depletion protocol (Illumina). Samples were multiplexed and sequenced with 150 bp paired-end reads. Sequencing depth averaged 30–50 million reads per sample for the UDN cohort and approximately 75 million reads per sample for the Utah Penelope Program cohort. FASTQ files were aligned to the GRCh37/hg19 reference genome using STAR v2.7.8a [14]. Gene-level quantification was performed using the featureCounts() function from the Rsubread package (v2.10.0), using strand-specific, paired-end settings. Reads were assigned to exons using the GENCODE v41lift37 annotation (lifted to hg19), with multi-mapping reads counted fractionally and overlapping features allowed. Count matrices were generated using the GenomicFeatures (v1.54.1) and GenomicAlignments (v1.38.0) R packages.

Genes were retained if they had ≥ 15 counts in at least two samples. The resulting count matrix was normalized using size factors estimated with DESeq2 (v1.40.1) [15]. Outlier analysis was performed using OUTRIDER (v1.14.0) [16], which models gene expression while correcting for hidden confounders using an autoencoder-based approach (controlForConfounders()). Genes with |log₂ fold change|> 0.1 and p-value < 0.05 were considered significantly dysregulated. Genes on the X chromosome and in enriched cytogenetic bands were highlighted as part of downstream visualization. The Database for Annotation, Visualization, and Integrated Discovery (DAVID, v2025q1) was used for chromosomal positional gene enrichment analyses. Expression outlier analyses included 28 fibroblast and 17 blood samples from participants in the University of Utah UDN, and 22 blood samples from the Utah Penelope Program. Student’s t-tests and graph generation were conducted using GraphPad Prism v10.4.1. Splicing analysis was manually performed using the Sashimi plot function in Integrative Genomics Viewer (IGV) v2.18.1.

Results

Overall study results

Combined DNA and RNA sequencing analysis was performed on 30 participants (11 males, 19 females) with diverse clinical presentations. The median age was 6 years for pediatric participants and 27 years for adults (Fig. 1A). All cases had negative or inconclusive results from prior clinical exome or genome sequencing before RNA-seq was pursued. RNA-seq contributed functional data that supported a definitive resolution in 10 cases (Table 1) and a likely resolution in 1 additional case (27%). This aligns with previous studies showing that adding RNA-seq alongside exome or genome sequencing increases diagnostic yield by 10–35%, depending on the cohort [10, 11, 19, 20]. Participants in the cohort whose cases were resolved were significantly younger than those who remained unresolved (Fig. 1B; two-tailed t-test; P = 0.0054). Among diagnosed cases, variant resolution was achieved using fibroblast-derived RNA in 3 of 11 cases (27%), blood-derived RNA in 6 of 11 cases (55%), and both tissues in 2 of 11 cases (18%) (Fig. 1C). Resolution varied by clinical category: 75% for multiple congenital anomalies (3/4), 67% for hematological or immune disorders (2/3), 50% for skeletal disorders (3/6), 50% for developmental disorders (2/4), and 17% for neurological disorders (1/6) (Fig. 1D). No case resolutions were made in cases categorized as metabolic (n = 1), neuromuscular (n = 3), or pulmonary (n = 3). The molecular mechanisms for each variant assessed (N = 13) was exon skipping (6 variants, 46%), intron retention (2 variants, 15%), cryptic splice-site activation (1 variant, 8%), positional enrichment (2 variants, 15%), and multiple splicing effects (2 variants, 15%) (Fig. 1E). The variants causing multiple splicing effects included one variant with both cryptic splice-site activation and exon skipping and another with cryptic splice-site activation and intron retention (Table 1). A total of 8 out of 13 variants (62%) were reclassified as pathogenic, including 5 VUS and 3 likely pathogenic variants. The median time from initial consultation to resolution was 185 weeks, whereas the median time from RNA-seq analysis to resolution was only 9 weeks (Fig. 1F), highlighting its potential to accelerate variant interpretation once incorporated into testing workflows.

Fig. 1.

Overview of cohort characteristics and case outcomes. A Age distribution of pediatric and adult participants who underwent combined DNA and RNA sequencing. Each dot represents an individual and horizontal lines represent median and 95% confidence intervals (CI). Blue circles represent males and red circles represent females. B Comparison of ages between participants with resolved and unresolved cases. Each dot represents an individual and the horizontal lines indicate the median and 95% CI. Participants with resolved cases were significantly younger than those with cases not resolved participants (two-tailed Student’s t-test, P = 0.0054). Blue circles represent males and red circles represent females. C Sample types used for RNA-seq in resolved and not resolved cases, represented as a bar chart. D Parts-of-a-whole chart illustrating clinical categories of participants, distinguishing between those with case resolution (Dx) and those without (No Dx). E Molecular mechanisms for each variant assessed, represented as a bar chart. F Time to case resolution in 10 participants, showing weeks from initial clinical consultation to final resolution and from RNA-seq analysis to resolution. Horizontal bars represent the median and 95% confidence interval. Case 9 was excluded (Table 1), as RNA-seq was performed concurrently with karyotyping, and the diagnosis was established prior to the return of RNA-seq results

Table 1.

Summary of rare disease cases resolved by RNA-sequencing

Case(s)	Sex	Age (yrs)	Gene	Variant	Inheritence	RNA sample type	SpliceAI	Variant effect	Disease association	ACMG criteria before RNA-seq	ACMG criteria after RNA-seq	Identified (citation)
1	F	6	Chromosome X	arr[GRCh37] Xq22.1q28(98385656_148818854) × 3, Xq28(148819148_152329158) × 4	de novo	Both	N/A	Positional Gene Enrichment	Xq duplication	Pathogenic	Pathogenic	this study
2	M	16	AIFM1	NM_004208.4:c.760G > A; p.Glu254Lys	matermal	Fibroblasts	Donor loss = 0.38	Exon Skipping	Spondyloepimetaphyseal dysplasia, X-linked, with hypomyelinating leukodystrophy	VUS: PM2_supporting, PP4	Pathogenic: PVS1 PS3_supporting PM2_supporting	this study
3	F	3	RAB18	NM_021252.5:c.125-23A > C	maternal & paternal	Blood	Acceptor loss = 0.47	Exon Skipping	Warburg micro syndrome 3	VUS: PM2_supporting, PP4	Pathogenic: PVS1 PS3_supporting PM2_supporting	this study
4	F	10	EFL1	NM_024580.6:c.932 + 2 T > G	maternal	Blood	Donor loss = 0.94	Exon Skipping	Shwachman-Diamond syndrome 2	Likely Pathogenic: PVS1, PM2_supporting	Pathogenic: PVS1, PS3_supporting PM2_supporting	this study
4				NM_024580.6:c.245-12A > G	paternal		Acceptor gain = 0.60	Intron Retention		VUS: PM2_supporting, PP4	Pathogenic: PVS1 PS3_supporting PM2_supporting
5		9	NIPBL	NM_133433.4:c.5329-15A > G	de novo	Blood	Acceptor loss = 0.54 Acceptor gain = 0.26	Exon Skipping Cryptic Splice Site	Cornelia de Lange syndrome 1	Pathogenic: PVS1 PS4 PS2_supporting PM2_supporting	Pathogenic: PVS1 PS4 PS2_supporting PM2_supporting	this study
5 and 6	F	5 (case 6)	TBCK	[GRCh38] chr4:106,170,052-106178330del	maternal	Blood	N/A	Exon Skipping	Hypotonia, infantile, with psychomotor retardation and characteristic facies 3	Pathogenic: PVS1 PS4 PM2_supporting	Pathogenic: PVS1 PS4 PS3_supporting PM2_supporting	this study
5 and 6				NM_001163435.3:c.193 + 2dup	paternal		Donor loss = 0.84	Exon Skipping		VUS: PM2_supporting, PP4	Pathogenic: PVS1 PM3 PS3_supporting PM2_supporting
7	F	1	RPS19	NM_001022.4:c.172 + 350 C > T	de novo	Blood	Donor gain = 0.98	Cryptic Splice Site	Diamond-Blackfan anemia 1	Likely Pathogenic: PS2 PM2_supporting PP4	Pathogenic: PVS1 PS3_supporting PM2_supporting	Wen T, et al. 2025 [17]
8	M	1	RPS7	NM_001011.4:c.−19G > C	maternal	Blood	Donor gain = 0.57	Intron Retention Cryptic Splice Site	Diamond-Blackfan anemia 8	Likely Pathogenic: PS4 PM2_supporting PP4	Pathogenic: PVS1 PS4 PS1_supporting PM2_supporting	Wen T, et al. 2025 [17]
9	F	2	Chromosome 4	Mosaic Trisomy 4	de novo	Fibroblasts	N/A	Positional Gene Enrichment	Mosaic trisomy 4	Pathogenic	Pathogenic	Teresa-Rodrigo et al., 2024 [18]
10	M	4	NFIX	[GRCh37] chr19:13,185,527-13190803dup	de novo	Fibroblasts	N/A	Exon Skipping	Malan syndrome	VUS: PS2 PM2_supporting	Pathogenic: PVS1 PS3_supporting PM2_supporting	Zhao J, et al. 2024 [8]

Case 1: Positional gene enrichment uncovers active Xq duplication despite skewed X-inactivation in a female with a de novo Xq22.1–q28 gain

Presentation

A 6-year-old female presented with multiple congenital anomalies, including hypoplastic right arm (Fig. 2A), short stature, and severe neurodevelopmental impairment. The main neurodevelopmental findings included shunted non-obstructive hydrocephalus, epilepsy, cortical blindness, global hypotonia, profound intellectual disability, and multiple joint contractures. She also had dysphagia requiring a gastrostomy tube and chronic constipation with gastroparesis.

Fig. 2.

Clinical and molecular findings in a participant with a de novo Xq22.1–Xq28 duplication and nested Xq28 triplication. A Clinical photograph of the participant demonstrating non-familial facial features and limb asymmetry. B Cytogenetic analysis showing the abnormal X chromosome (right) with additional material on the long arm compared to a normal X chromosome (left) which was described initially as 46,X,add(X)(q28) and was later clarified as 46,X,dup(X)(q22.1q28). C Cytogenomic microarray results demonstrating a 50.4 Mb duplication spanning Xq22.1–Xq28 and a nested cytogenetically cryptic 3.5 Mb triplication of Xq28. D Interphase and metaphase fluorescence in situ hybridization (FISH) using a red X centromere probe and a green Xq28 probe confirms the gain of the Xq28 region. E RNA-seq analysis from participant-derived fibroblasts. Left: Volcano plot of differentially expressed genes from the OUTRIDER analysis. Upregulated genes are shown in red, downregulated genes in blue, and genes located within the duplicated Xq22.1–Xq28 region in green. Middle: Positional enrichment analysis of upregulated genes by chromosome using DAVID. Right: Positional enrichment of upregulated genes by cytogenetic band, highlighting the duplicated interval. F RNA-seq analysis of whole blood. Left, middle, and right panels show the same volcano plot and positional enrichment analyses as in (E), illustrating consistent overexpression of genes within the duplicated region in whole blood

Genetic testing

Initial cytogenetic analysis revealed an abnormal female chromosome complement with additional material on the long arm of one X chromosome, initially described as 46,X,add(X)(q28) (Fig. 2B). Further characterization by chromosomal microarray (CMA) identified a 50.4 Mb duplication spanning Xq22.1 to Xq28, along with a nested 3.5 Mb triplication within Xq28. The copy number gain was defined as arr[GRCh37] Xq22.1q28(98385656_148818854) × 3, Xq28(148819148_152329158) × 4 (Fig. 2C). These findings clarified the initial karyotype as 46,X,dup(X)(q22.1q28) (Fig. 2B) and were corroborated by metaphase FISH which confirmed the presence of duplicated Xq material (Fig. 2D). Though the gain was not expected to include the MECP2 gene based on the breakpoints of the array, deletion/duplication testing by MLPA was performed and confirmed no copy number changes.

X-inactivation (XCI) studies showed complete skewing in peripheral blood (100:0) and moderate skewing in fibroblasts (84:16), suggesting preferential inactivation of one X chromosome. Although skewed XCI is sometimes associated with attenuation of clinical severity in females with X chromosome copy number variations, this correlation is not consistent and is known to be tissue-specific. In this case, the participant exhibited a markedly severe phenotype, prompting further investigation into whether the abnormal X chromosome was functionally silenced or if additional molecular mechanisms, including the presence of other pathogenic variants, may have contributed to the clinical presentation.

Trio exome sequencing did not identify any causative variants in disease genes with reported phenotypes. Trio genome sequencing confirmed the Xq22.1 to Xq28 rearrangement was de novo on the maternally inherited X chromosome and revealed a complex rearrangement mechanism involving two template switching events between opposite DNA strands. The proximal breakpoint was mapped to a strand switch between [GRCh38] Xq22.1(99154787) and Xq28(149738633). The distal breakpoint could not be precisely resolved due to the presence of highly homologous segmental duplications but is predicted to involve a strand switch between two regions at [GRCh38] chrX:153105761–153149741 and chrX:153250233–153294975. However, no additional clinically significant sequence variants or copy number changes were identified through genome analysis.

RNA-seq

RNA-seq was performed to evaluate the functional consequences of the Xq duplication, as conventional assays for assessing XCI, such as androgen receptor methylation profiling or BrdU-based replication timing studies, may not fully capture the transcriptional status of the entire X chromosome [21, 22]. We hypothesized that if the abnormal X chromosome remained active and the normal X was preferentially inactivated, then genes within the duplicated region would be disproportionately upregulated. Indeed, positional enrichment analysis revealed a significant overexpression of genes located within the duplicated Xq22.1 to Xq28 interval in both fibroblasts (Fig. 2E) and blood (Fig. 2F), including several known and predicted triplosensitive genes/regions. Upregulated genes were positionally enriched on the X chromosome and were specifically localized within the regions of the copy number gain (Xq22.1 to Xq28) in fibroblasts (Fig. 2E) and blood (Fig. 2F). These findings directly contradict the expectation that the duplicated X would be transcriptionally silenced due to skewed XCI and instead provide evidence that the abnormal X chromosome remained active. The resulting dosage imbalance offers a plausible explanation for the participant’s unexpectedly severe phenotype and establishes a mechanism of pathogenicity driven by gene overexpression from the structurally altered X chromosome.

Case 2: Exon skipping and AIFM1 downregulation reveal splicing-driven pathogenicity in X-linked spondyloepimetaphyseal dysplasia with hypomyelinating leukodystrophy

Presentation

The participant was referred at 16 years for short stature associated with radiographic findings suggestive of skeletal dysplasia, undiagnosed despite extensive evaluations and diagnostic testing. Skeletal findings included metaphyseal flaring and endplate irregularities (Fig. 3A). He also had a history of respiratory failure and subglottic stenosis during surgery that required temporary tracheostomy. His developmental milestones were normal and with no apparent neurologic findings.

Fig. 3.

Clinical, radiologic, and transcriptomic findings in a participant with AIFM1-related spondyloepimetaphyseal dysplasia with hypomyelination. A Skeletal survey demonstrating characteristic radiographic abnormalities including metaphyseal flaring and irregular endplates in the lower extremities. B Brain MRI at age 17 demonstrating extensive bilateral symmetric T2/FLAIR hyperintensities with involvement of the deep and subcortical supratentorial white matter. C IGV sashimi plot of RNA-seq data highlighting aberrant splicing of AIFM1 in the proband. Red box indicates exon 6/8 junction reads present in the proband but absent in controls, consistent with exon 7 skipping. D Volcano plot of differential gene expression from RNA-seq of fibroblasts. Upregulated genes are shown in red, downregulated genes in blue, and AIFM1 is labeled as the top expression outlier (P = 6.5*10^–6). E OUTRIDER expression outlier analysis showing normalized AIFM1 transcript counts in the proband (pink triangle) and controls (grey circles). Horizontal lines represent median and 95% CI

Genetic testing

Initial genetic testing, including a skeletal dysplasia panel, exome sequencing, and CMA, was negative. Given the complexity of his presentation, exome reanalysis at the University of Utah and ARUP Laboratories was performed which revealed a maternally inherited AIFM1 variant (NM_004208.4:c.760G > A; p.Glu254Lys). Pathogenic variants in AIFM1 are associated with X-linked spondyloepimetaphyseal dysplasia with hypomyelinating leukodystrophy (SEMDHL; OMIM:300,232). The identified variant was absent from gnomAD v4.1 and computational missense prediction tools indicated uncertain deleteriousness (REVEL: 0.598) while splicing prediction tools suggested a potential splice donor loss (SpliceAI: 0.38). A brain MRI was obtained which showed diffuse white matter changes on T2/FLAIR imaging involvement of both subcortical and deep white matter (Fig. 3B). These findings were consistent with white matter changes seen in SEMDHL, however, neurologic examination was largely normal except for hyperreflexia. Under ACMG/AMP criteria, this variant was classified as a VUS due to the lack of functional validation, despite supporting neuroimaging and computational predictions.

RNA-seq

Based on supportive imaging, the variant’s proximity to a splice donor site (21 bps away), and its moderately elevated SpliceAI score, RNA-seq was conducted on fibroblasts to evaluate its potential impact on splicing. RNA-seq revealed the presence of exon 6/8 junction reads by manual sashimi plot review in the proband but not in controls, confirming exon 7 skipping (Fig. 3C). This skipping event is predicted to result in an out-of-frame transcript leading to nonsense-mediated decay (NMD) or a truncated protein. Gene expression outlier analysis identified AIFM1 as the top differentially expressed gene (P = 6.5*10^–6) (Fig. 3D), with an approximate 50% reduction in transcript levels compared to controls (Fig. 3E), confirming loss of function due to NMD. Based on this functional validation, the AIFM1 c.760G > A variant was reclassified as pathogenic, leading to a resolution of this case.

Case 3: RNA-seq confirms branch point disruption and exon skipping in RAB18 in a case of Warburg micro syndrome 3

Presentation

A 3-year-old female with multiple congenital anomalies presented with extensive brain malformation, including diffuse, bilateral, symmetric polymicrogyria and dysgenesis of the corpus callosum (Fig. 4A). These abnormalities were associated with a seizure disorder, global developmental delay, hypotonia, and growth delay with weight, height, and BMI all below the 3rd percentile. Additionally, she exhibited congenital cataracts and coloboma (Fig. 4B).

Fig. 4.

Clinical, neuroimaging, and transcriptomic findings in a proband with Warburg Micro Syndrome 3 due to a homozygous intronic variant in RAB18. A Sagittal, coronal, and axial T1 brain MRI images (left to right) demonstrate diffuse, bilateral, symmetric cerebral polymicrogyria. B Clinical photograph of the proband showing non-familial features including broad nasal root and craniofacial asymmetry. C Sashimi plot from whole blood RNA-seq showing RAB18 exon 3 skipping in the proband and reduced exon 3 usage in both carrier parents. Red boxes highlight exon 2/4 junction reads in the proband and parents. The long red box spanning exon 3 demonstrates complete absence of exon 3 reads in the proband, consistent with exon 3 skipping. D OUTRIDER analysis of RAB18 transcript expression. The proband (pink triangle) shows significant underexpression compared to controls (P = 0.019;) with intermediate reduction observed in the father (blue circle; P = 0.049) and mother (red circle; P = 0.074). Horizontal lines represent median and 95% CI

Genetic testing

Initial trio exome sequencing performed was non-diagnostic and follow-up trio genome sequencing also failed to identify a causative variant. CMA also did not reveal any clinically significant deletions or duplications but did identify extended regions of homozygosity. Reanalysis of genome sequencing data at the University of Utah and ARUP Laboratories identified a homozygous intronic variant in RAB18 (NM_021252.5; c.125-23A > C). Pathogenic variants in RAB18 are associated with autosomal recessive Warburg micro syndrome 3 (WARBM3; OMIM 614222), which demonstrated significant phenotypic overlap with the proband. The variant was absent from gnomAD v4.1 and had a moderate prediction to disrupt the canonical splice acceptor site (SpliceAI: 0.38). Additionally, it was predicted to alter the branch point motif by substituting the conserved adenine within the heptamer consensus sequence, reducing branch point confidence by 10.53% (SPiP) [23]. This disruption is predicted to result in exon 3 skipping and generate an out-of-frame transcript. However, due to the absence of functional confirmation at the time of interpretation, the variant was initially classified as a VUS.

RNA-seq

Given the phenotypic overlap with WARBM3 and computational predictions of splicing disruption, RNA-seq was performed to assess the functional impact of the RAB18 c.125-23A > G variant. Whole blood RNA-seq analysis confirmed exon 3 skipping, as demonstrated by manual review of sashimi plots. Exon 2/4 junction reads were present in the proband and both carrier parents, but absent in control samples (Fig. 4C). Consistent with complete exon skipping, the proband showed a complete absence of exon 2/3 and exon 3/4 junction reads. Additionally, there was notable accumulation of intronic reads from introns 2 and 3 in the proband, and to a lesser extent in the carrier parents, compared to controls. Skipping of exon 3 is predicted to result in an out-of-frame transcript that would be subject to NMD or potentially result in a truncated protein.

Expression outlier analysis identified RAB18 as significantly underexpressed in the proband (P = 0.019; Fig. 3D). Interestingly, both carrier parents also showed reduced expression: the father’s reduction was statistically significant (P = 0.049), while the mother’s did not reach significance (P = 0.074). Taken together, these findings provided functional confirmation that the RAB18 c.125-23A > G variant disrupts normal splicing and reduces gene expression. The variant was therefore reclassified as pathogenic, allowing for this case to be considered resolved.

Case 4: RNA-seq identifies intron retention in EFL1 supporting pathogenicity in Shwachman-diamond syndrome 2

Presentation

A 10-year-old female presented with skeletal dysplasia, short stature with normal head circumference, and mild restrictive respiratory insufficiency (Fig. 5A). Radiographic findings were consistent with an atypical form of spondyloepimetaphyseal dysplasia (SEMD, with abnormal vertebral endplates and metaphyseal cupping) (Fig. 5B-D). She was hypotonic at birth with a bell-shaped rib cage, requiring NICU admission, G-tube feeding, and nasal oxygen support during infancy. Her motor and intellectual development were within normal limits.

Fig. 5.

Clinical, radiologic, and transcriptomic findings in a proband with Shwachman-Diamond Syndrome 2 (SDS2) due to compound heterozygous splicing variants in EFL1. A Frontal (left) and lateral (right) clinical photographs of the proband. B Radiograph demonstrating metaphyseal flaring and marked knee abnormalities characteristic of spondyloepimetaphyseal dysplasia (SEMD). C Radiographic image showing abnormally shaped vertebral bodies. D Pelvic radiograph highlighting lacy ilia. E Sashimi plot of EFL1 exon 8 skipping due to the maternally inherited c.932 + 2 T > G variant. Red boxes highlight exon 7/9 junction reads observed in both the proband and mother, absent in the father and controls. F Read coverage plot of EFL1 intron 4 retention associated with the paternally inherited c.245-12A > G variant. Increased read coverage across intron 4 is observed in the proband and father (red boxes) but not in the mother or controls. G OUTRIDER expression outlier analysis showing significantly decreased EFL1 expression in the proband (pink triangle) compared to controls (P = 0.032), consistent with a loss-of-function effect from biallelic splicing disruption. The mother is represented by a red circle and the father by a blue circle. Horizontal lines indicate the median and 95% confidence interval

Genetic testing

Initial genetic testing, including a skeletal dysplasia panel and exome sequencing, was non-diagnostic. Clinical genome sequencing failed to identify any reportable sequence variants or copy number changes related to the participant’s condition. Subsequently, genome sequencing data was reanalyzed at the University of Utah and ARUP Laboratories, which identified compound heterozygous splicing variants in EFL1: a maternally inherited likely pathogenic variant (c.932 + 2 T > G) and a paternally inherited variant of uncertain significance (c.245-12A > G). Pathogenic variants in EFL1 are associated with autosomal recessive Shwachman-Diamond syndrome 2 (SDS2; OMIM 617941), which is characterized by skeletal abnormalities and bone marrow dysfunction. The c.932 + 2 T > G variant was absent from population databases and predicted to abolish the canonical splice donor site (SpliceAI score: 0.94), supporting its classification as likely pathogenic according to ACMG/AMP guidelines. The c.245-12A > G variant was also absent from population databases and predicted to generate a novel splice acceptor site (SpliceAI: 0.60); however, in the absence of supporting functional evidence, the c.245-12A > G variant was initially classified as a VUS.

RNA-seq

Given the proband’s clinical overlap with SDS2 and computational predictions of splicing disruption, RNA-seq was performed to evaluate the functional impact of the EFL1 variants. Whole blood RNA-seq confirmed that the maternally inherited c.932 + 2 T > G variant led to exon 8 skipping, as demonstrated by the presence of exon 7/9 junction reads in both the proband and her mother, based on manual review of sashimi plots (Fig. 5E). This splicing event is predicted to produce an out-of-frame transcript subject to NMD. Unexpectedly, analysis of the paternally inherited c.245-12A > G variant located in intron 4 revealed increased intronic read coverage in both the proband and her father in IGV, consistent with intron retention (Fig. 5F). This retained intron, which appears to be the entire intron, is likewise predicted to lead to NMD or generate an abnormal protein product. Expression outlier analysis identified EFL1 as significantly underexpressed in the proband (P = 0.032), further supporting a loss-of-function effect. Collectively, these transcriptomic findings confirmed the pathogenic consequence of both EFL1 variants. Based on this functional evidence, the c.932 + 2 T > G and c.245-12A > G variants were reclassified as pathogenic, allowing for this case to be considered resolved.

Cases 5 and 6: RNA-seq reveals exon skipping and confirms biallelic TBCK disruption in infantile hypotonia with psychomotor retardation and characteristic facies 3 and expands on a NIPBL splicing defect in Cornelia de Lange syndrome

Presentation

A 10-year-old female (sister A) presented with developmental delay, epilepsy, soft cleft palate, aberrant right subclavian artery, and non-familial traits (synophrys, an upturned nose, cutis marmorata, and brachydactyly) (Fig. 6A). Growth was delayed in early childhood and normalized over time (50th percentile). Additional clinical concerns included chronic constipation and recurrent vomiting without distress, and gastroesophageal reflux. Brain CT imaging was normal.

Fig. 6.

Facial features and transcriptomic analysis of biallelic TBCK variants in two sisters with IHPRF3 and a concurrent NIPBL variant in Sister A. A Front facial photograph of Sister A showing features including synophrys and upturned nose. B Front facial photograph of Sister B with macrocephaly, increased forehead height, low nasal root, and prominent cupid’s bow. C Left: Sashimi plot showing NIPBL splicing in the family and unrelated controls. In Sister A, red box highlights exon 27/29 junction reads consistent with exon 28 skipping; purple box indicates use of an alternative acceptor site. Right: Zoomed-in view of NIPBL exon 28 demonstrating alternative acceptor site usage with increased intronic reads (purple box); position of c.5329-15A > G variant is labeled in red. D Sashimi plot showing skipping of TBCK exon 23 due to the maternally inherited deletion. Red boxes mark exon 22/24 junction reads, while the purple box highlights the deletion breakpoints observed in both sisters and their mother but absent in the father and control samples. E Sashimi plot showing exon 2 skipping due to the paternally inherited TBCK c.193 + 2dup variant. Red boxes highlight exon 1/3 junction reads observed in both sisters (A and B) and their father (F), but not in the mother (M) or controls. F OUTRIDER expression outlier plot showing reduced NIPBL transcript levels in Sister A (green inverted triangle) compared to controls (P = 0.017). Horizontal lines indicate the median and 95% CI. G OUTRIDER expression outlier plot showing significantly reduced TBCK expression in sister A (green inverted triangle; P = 0.049) and sister B (pink triangle; P = 0.046), with intermediate reduction in the mother (red circle; P = 0.095). The father is represented by a blue circle. Horizontal lines indicate the median and 95% CI

Her 6-year-old sister (sister B) also presented with global developmental delay, absent speech, autism, a history of a single grand mal seizure, exotropia (post-surgical repair) and large head (OFC at the 90th percentile). She had non-familial traits different from her sister’s, including a prominent forehead, a low nasal root, and a prominent cupid’s bow (Fig. 6B). Her height and weight were normal, and she was otherwise healthy.

Genetic testing

For sister A, multiple prior genetic and imaging evaluations were non-diagnostic. Trio exome sequencing was negative, as were SMC1A sequencing, a Cornelia de Lange syndrome (CdLS) gene panel including NIPBL and SMC1A, and CMA. However, trio genome sequencing performed identified a de novo pathogenic intronic variant in NIPBL (c.5329-15A > G). This variant, located in intron 27, is found in one individual in gnomAD v4.1 and has been previously reported in multiple individuals with CdLS. Functional studies had shown that this variant causes exon 28 skipping, supporting its pathogenic classification [18, 24].

For sister B, genetic testing which included CMA, fragile X testing, and exome sequencing was also non-diagnostic, and her clinical team suspected a different etiology from her sibling. Quad genome sequencing did not resolve this case; however, reanalysis of the genome sequencing data at the University of Utah and ARUP Laboratories identified compound heterozygous variants in the TBCK gene. This included a maternally inherited 8.28 kb intragenic deletion in TBCK affecting exon 23 in both sisters. This deletion was present in four alleles in gnomAD v4.1 and is predicted to result in an out-of -frame transcript. Prior studies have demonstrated reduced TBCK expression in homozygous individuals with Infantile Hypotonia with Psychomotor Retardation and Characteristic Facies 3 (IHPRF3; OMIM 616900), supporting its pathogenicity [25, 26]. In addition, both sisters were found to carry a paternally inherited TBCK variant (c.193 + 2dup) located in intron 2. This variant was absent from gnomAD v4.1 and predicted to result in donor site loss (SpliceAI score: 0.84). Skipping of exon 2 was expected to disrupt the start codon, potentially resulting in loss of translation or production of a truncated protein. However, in the absence of functional confirmation, this variant was initially classified as a VUS.

RNA-seq

RNA sequencing was performed on both sisters and their parents to evaluate the functional impact of the TBCK variants and to confirm the splicing effect of the NIPBL variant. In sister A, RNA-seq confirmed that the NIPBL c.5329-15A > G variant resulted in both exon 28 skipping and, in addition, activated an alternative acceptor site not previously reported (Fig. 6C), with both events predicted to produce out-of-frame transcripts usage of an alternative acceptor site.

Analysis of the maternally inherited TBCK exon 23 deletion revealed exon 23 skipping, demonstrated by exon 22/24 junction reads on sashimi plots in both sisters and their mother, but not in their father or control samples (Fig. 6D), further supporting the pathogenicity of the deletion. Interestingly, the sashimi plots also provided visual evidence of the deletion, highlighted by sporadic junction read calls within the affected region and at its breakpoints (Fig. 6D). The paternally inherited TBCK c.193 + 2dup variant was associated with exon 2 skipping, as demonstrated by the presence of exon 1/3 junction reads observed in both sisters and their father, but absent in their mother or controls (Fig. 6E). Importantly, exon 2 of TBCK contains the canonical start codon, and the next possible AUG is located in exon 4, suggesting that the resulting transcript may fail to initiate translation or could lead to a truncated protein.

Expression outlier analysis further supported the impact of these variants. Sister A was identified as a significant outlier for NIPBL compared to controls (P = 0.017). Both sisters were identified as significant expression outliers for TBCK, with sister A showing a p-value of 0.049 and sister B a p-value of 0.046. The mother also demonstrated reduced TBCK expression, although this did not reach statistical significance (P = 0.095) whereas the father had expression similar to the non-familial controls (Fig. 5F), suggesting that the exon 23 deletion may have a more substantial effect on transcript stability or expression than the splice donor variant. Together, these transcriptomic findings confirmed the pathogenicity of both TBCK variants. The c.193 + 2dup variant was reclassified as pathogenic, resulting in molecular resolution of IHPRF3 in sister B and dual molecular resolution of IHPRF3 and Cornelia de Lange syndrome in sister A.

Discussion

This study demonstrates the utility of RNA sequencing in providing functional evidence that supported the molecular resolution of previously unresolved cases after negative or inconclusive exome or genome sequencing. Among 30 cases, RNA-seq contributed to resolution or strong supporting evidence in over one-third of cases, facilitating variant reclassification and revealing transcript-level disruptions missed by DNA-based methods. Identified mechanisms included exon skipping, intron retention, and cryptic splice-site activation, each capable of causing loss of function via nonsense-mediated decay (NMD) or altered protein coding. Notably, RNA-seq enabled resolution on a significantly shorter timescale than traditional workflows, supporting its use as a reflex or concurrent test.

A key consideration in RNA-seq–based variant interpretation is the integration of splicing and expression data. While in silico splicing predictors such as SpliceAI offer valuable suggestions and ACMG/AMP-supporting evidence, confirmation of aberrant splicing requires both visual validation and quantitative analysis [6]. In this study, abnormal splicing events were confirmed by identifying novel splice junctions in manual sashimi plot review. In this study, abnormal splicing was confirmed by novel splice junctions identified via sashimi plot review. We recommend a threshold of three unique split reads to confirm splicing events, though lower thresholds may be acceptable when reads are entirely absent in controls. Current guidelines describe splicing alterations as 'complete,' 'near complete,' or 'incomplete' based on the proportion of altered transcripts produced [6]. However, this framework may be misleading in the context of NMD, where rapidly degraded transcripts can result in low junction read counts despite reflecting a fully penetrant splicing defect. Additionally, the presence or absence of canonical junctions and the coexistence of both canonical and aberrant junctions, may indicate partial or leaky splicing defects and should be considered during interpretation. This underscores the need for a more mechanistically informed approach and supports the integration of quantitative expression data to improve the interpretation of splicing outcomes. Our proposed threshold aims to balance the risk of overcalling artifacts with the need to detect low-abundance but biologically meaningful events.

Transcript stability further complicates splicing interpretation. In instances where variants induce NMD, the timing and efficiency of transcript degradation in a particular tissue type may limit the observable evidence of splicing disruption [27–29]. Transcripts subject to rapid decay may yield very few detectable junction reads, while transcripts with slower degradation kinetics or those exhibiting incomplete NMD may retain sufficient expression for detection. Therefore, high read depth, tissue-specific expression, and overall coverage are critical to enhancing detection sensitivity. Complementary findings, such as increased intronic read coverage flanking the affected exon or altered expression that is lower or absent in controls, can further corroborate splicing effects and increase interpretive confidence. Library preparation also plays a crucial role in this context, as poly-A selection specifically enriches mature transcripts that have already undergone splicing. Consequently, pre-mRNA and transcripts retaining introns may be underrepresented or entirely absent in poly-A selected libraries. To ensure the inclusion of these transcripts, an alternative approach such as total RNA sequencing or rRNA depletion, may be considered.

Gene expression outlier analysis provides an important additional layer of evidence in variant interpretation. In the AIFM1, RAB18, EFL1, and TBCK cases (cases 2–6), reduced transcript expression was consistent with NMD and supported the classification of the associated variants as pathogenic. In contrast, the Xq duplication case demonstrated the utility of expression analysis in a different context: positional enrichment revealed regional transcript overexpression, supporting a dosage-based pathogenic mechanism. This finding suggests that the duplicated X chromosome was transcriptionally active despite skewed X-inactivation detected by conventional methods. Notably, the same positional gene enrichment approach was applied to a fibroblasts on a separate case of mosaic trisomy 4 in our cohort (case 9) [30] and similarly revealed an enrichment of upregulated genes solely on chromosome 4 (P = 0.00087), further validating the utility of this method for detecting dosage effects. While most RNA-seq-based diagnoses are likely to involve reduced gene expression due to loss-of-function mechanisms, it is equally important to recognize that upregulation, particularly of triplosensitive genes, can also be pathogenic, especially in cases involving regulatory variants or structural alterations affecting gene dosage.

Currently, there is no standardized framework for integrating splicing and gene expression data into clinical variant interpretation. When a variant results in a confirmed splicing defect but the affected gene is not flagged as an expression outlier, it is unclear how this should influence classification. We propose that expression analysis be treated as an independent but complementary line of functional evidence. In cases where splicing analysis confirms an out-of-frame event consistent with loss of function, the PVS1 criterion may be applied. If expression analysis also shows reduced transcript levels consistent with NMD, this additional functional support may justify the use of PS3_supporting. This approach could also be applicable to canonical splice site variants. For example, the TBCK c.932 + 2 T > G variant was initially classified as likely pathogenic based on PVS1 and PM2_supporting criteria. RNA-seq subsequently confirmed exon skipping and revealed decreased expression in TBCK. Although this new data aligns with the original classification and does not change clinical implications, it provides increased confidence in the pathogenicity of the variant and justifies the additional application of PS3_supporting. In such cases, RNA-seq not only reinforces existing criteria but may also support the reclassification of variants, including canonical splice site variants, to pathogenic.

Importantly, the absence of an expression change should not be interpreted as evidence against pathogenicity when a splicing defect is confirmed. Rather, it may simply preclude the use of PS3_supporting. Several biological and technical factors can explain this discrepancy, including leaky or partial splicing, inefficient or tissue-specific NMD, alternative isoform usage, compensatory regulation, or limited sensitivity of expression outlier tools in genes with low expression or high variability across controls. These differences are well illustrated by the AIFM1 case, where controls cluster tightly (Fig. 3E) and reduced expression in the proband is more clearly outlying, versus the EFL1 case (Fig. 5G) where control variability could complicate detection of an expression outlier despite a confirmed splicing defect.

RNA-seq in the constitutional clinical setting is most often performed as a reflex test following the identification of candidate splicing variants through DNA-based methods. As analytical pipelines and integrated workflows continue to advance, concurrent DNA and RNA sequencing may become more widely adopted, potentially improving turnaround time and facilitating more streamlined variant interpretation. However, this shift also introduces new challenges particularly in cases where gene expression abnormalities are detected without an identifiable splicing or regulatory defect. For example, missense variants associated with altered expression levels may reflect compensatory cellular mechanisms [31]: upregulation could suggest reduced protein function, while downregulation may be a cellular response to toxic gain-of-function effects. The Xq copy number gain case in this study illustrates this complexity. Although the triplicated region of Xq28 did not include the MECP2 gene, RNA-seq revealed increased MECP2 expression in both blood and fibroblasts (Fig. 1E and F). This raises the important question of how regulatory disruptions, including variants in trans-acting elements or within regulatory networks of dosage-sensitive genes [32, 33], should be interpreted within current variant classification frameworks. Further research and consensus are needed to determine how such findings should be incorporated into clinical interpretation and variant classification.

Discrepancies between splicing and expression data, such as those observed in the RAB18 and TBCK cases, underscore the complexity of interpreting expression changes in isolation. In both instances, probands with biallelic pathogenic variants exhibited clear splicing defects confirmed by RNA-seq; however, overall gene expression was only modestly decreased and not significantly different from that of heterozygous carrier parents. This contrasts with the expected pattern, where carriers would typically exhibit approximately 50% expression due to one allele undergoing NMD, and probands with both alleles affected would show an even greater reduction. Similarly, in the AIFM1 case, although transcript expression was significantly reduced compared to controls, it was not near zero, despite strong evidence of an out-of-frame transcript subject to NMD. These discrepancies may reflect a combination of biological and technical factors, including variability in NMD kinetics, compensatory transcriptional regulation, alternative isoform expression, or limitations in the sensitivity of expression outlier detection tools. Notably, in a separate case previously reported by our group [34], a pathogenic exon duplication in the NFIX gene resulted in a clear 50% reduction in transcript expression, suggesting that the correlation between splicing disruption and expression change is both gene- and context-specific.

Interpreting intron retention and other cryptic splice site activation in clinical RNA-seq remains challenging, as it can be difficult to distinguish from alternative splicing or background transcription, particularly in genes with complex isoforms, low expression, and higher read depth in a sample. In our cohort, the EFL1 c.245-12A > G variant exemplifies a clinically significant intron retention event. Although located outside the canonical splice site and initially classified as a VUS, RNA-seq revealed abnormal retention of intron 4 in both the proband and her carrier father, absent in controls. This retained intron introduces a premature stop codon, likely triggering NMD or generates a truncated protein. Expression analysis showed reduced EFL1 expression in the proband, supporting a loss-of-function mechanism and prompting variant reclassification. A similar phenomenon was observed in our recent report of Diamond-Blackfan Anemia 8 syndrome due to an RPS7 c.−19G > C variant, where widespread intron 1 read retention occurred in both the proband and a mildly affected mother, although there was a notable drop in coverage downstream of the variant within the intron [17]. Similarly, cryptic splice site activation can lead to the formation of novel exons within intronic regions, as observed in our additional report of Diamond-Blackfan Anemia syndrome 1 due to an RPS19 c.172 + 350 C > T variant [17], where aberrant exon inclusion within intron 4 was detected. This highlights the need to consider the potential consequences of splice-altering variants, even when located distally from canonical splice sites. These cases also highlight a technical consideration: when evaluating expression changes in the context of intron retention, intronic reads should be excluded from count matrices to avoid overestimating gene expression. Collectively, these findings emphasize the importance of combining splicing predictions, visual RNA review, and expression analysis when interpreting non canonical variants that result in intron retention.

Initial genome sequencing for this cohort was performed at multiple outside clinical laboratories, each using distinct bioinformatic pipelines, variant prioritization strategies, and gene-disease annotation practices. The lack of reportable findings in several cases likely reflected these differences as well as limited or incomplete phenotypic information provided at the time of testing. Retrospective re-analysis, informed by updated and more detailed clinical data, enabled the prioritization of variants that had previously been overlooked or considered non-reportable and allowed their functional impact to be assessed using RNA-seq. These findings highlight the dynamic nature of genomic interpretation and emphasize the importance of comprehensive phenotypic data. Detailed patient histories or full clinical charts, rather than only a limited set of HPO terms, may be critical for optimizing variant detection and interpretation in clinical testing.

While RNA sequencing was beneficial in this cohort, the majority of cases remained unresolved due to several potential factors. RNA-seq is limited to genes expressed in the sampled tissue (blood or fibroblasts in this study) which may not include disease-relevant transcripts expressed only in specific tissues or developmental stages. Technical limitations such as low expression levels, allelic imbalance, or incomplete exon coverage can further hinder the detection of splicing or expression abnormalities. Additionally, some variants may only exert functional effects under specific physiological conditions not reflected in the sampled tissues. In other cases, the underlying pathogenic mechanism may not be transcript-detectable at all. For example, variants affecting regulatory elements, deep intronic regions with subtle splicing effects, or mechanisms involving protein stability, post-translational modification, or non-coding RNAs. Moreover, current computational tools for detecting aberrant splicing and expression outliers, though improving, may miss subtle or complex transcriptional changes. These limitations underscore the need for complementary approaches such as tissue-specific transcriptomics, long-read RNA sequencing, and targeted functional assays to further improve diagnostic yield.

While variant-specific RNA assays can confirm splicing effects, they are not routinely available and often require custom design and optimization, increasing cost and turnaround time. In contrast, RNA-seq is increasingly accessible and offers a transcriptome-wide approach that does not rely on prior variant selection. It is particularly valuable in cases with multiple candidate variants or atypical presentations, allowing comprehensive assessment of splicing, expression, and allele-specific expression in a single test. RNA-seq data can also be reanalyzed as new variants emerge. Although some cases could be addressed with targeted RNA studies, the scalability and efficiency of RNA-seq make it a more practical follow-up to inconclusive DNA testing.

RNA-seq is not only valuable for resolving VUS but also plays a key role in reclassifying likely pathogenic variants to pathogenic, an important step in increasing diagnostic confidence across the clinical continuum. For laboratories, this strengthens the evidence base for interpretation and reporting; for clinicians, it supports medical decision-making and counseling; and for patients, it offers more definitive conclusions. This is especially critical in recessive conditions where a likely pathogenic variant is found in trans with a VUS where confirmation of pathogenicity can directly impact classification of the second variant and enable resolution. Accurate classification ensures that if the same variant is encountered in a future case with a novel variant in trans, the new finding can be more reliably interpreted, improving consistency and diagnostic efficiency.

In addition to supporting variant classification, RNA-seq can uncover previously unrecognized molecular consequences of known pathogenic variants. For example, the NIPBL variant in our cohort had been previously shown to cause exon 28 skipping [18], but RNA-seq also revealed activation of a cryptic splice acceptor site (Fig. 6C), resulting in an alternative aberrant transcript. This highlights the power of RNA-seq not only to aid in molecular diagnostics but also to deepen our understanding of disease mechanisms. Its unbiased, transcriptome-wide assessment can reveal complex or multiple splicing outcomes from a single variant, providing insights into gene regulation, transcript diversity, and potential phenotype variability.

Together, these findings highlight the need for refined frameworks to integrate both splicing and expression data into variant classification and clinical reporting. While expression data can provide powerful mechanistic insight and support pathogenic classification, it must be interpreted within the broader genomic, transcriptomic, and clinical context. Ultimately, RNA-seq can enhance variant interpretation and provide mechanistic insights into rare diseases. Its broader integration into clinical workflows, alongside evolving interpretation guidelines, may help improve diagnostic yield and patient care.

Abbreviations

CdLS

Cornelia de Lange syndrome

CMA

Chromosomal microarray

IHPRF3

Infantile Hypotonia with Psychomotor Retardation and Characteristic Facies 3

NMD

Nonsense-mediated decay

RNA-seq

RNA sequencing

SDS2

Shwachman-Diamond Syndrome 2

SEMDHL

Spondyloepimetaphyseal dysplasia with hypomyelinating leukodystrophy

VUS

Variant of uncertain significance

WARBM3

Warburg micro syndrome 3

XCI

X chromosome inactivation

Authors’ contributions

Conceptualization: RGL, LDB, RM, PBT; Formal analysis: RGL; Funding acquisition: LDB; Investigation: RGL, JOS, LP, TW, MLF, JZ, JV, CZ, MV, TJN, SEB, AW, JHR, MM, EEB, AA, DV, JCC, SBB, RJB, VT, LDB, RM, PBT; Project administration: EEB; Supervision: RM, PBT, LDB; Visualization: RGL; Writing-original draft: RGL; Writing-review & editing: RGL, JOS, LP, TW, MLF, JZ, JV, CZ, MV, TJN, SEB, AW, JHR, MM,EEB, AA, DV, JCC, SBB, RJB, VT, LDB, RM, PBT.

Funding

Research reported in this publication was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health under Award Number U01HG010217. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Genome sequence alignment and variant calling were performed by the Utah Center for Genetic Discovery Core, part of the Health Sciences Center Cores at the University of Utah. This work utilized resources and support from the Center for High Performance Computing at the University of Utah, partially funded by National Institutes of Health Shared Instrumentation Grant 1S10OD021644-01A1.

We are grateful to the Miller Genomic Fund for their generous gift that enables the Penelope Program to push boundaries of genomic medicine and provide advanced diagnostic services to those in need, including underserved families and communities.

Data availability

A portion of the data used in this study are available from the Undiagnosed Diseases Network, but restrictions apply to the availability of these data. Individuals interested in accessing data through should submit a data access request to the Undiagnosed Diseases Network. For instructions on accessing this data see: https://sharing.nih.gov/accessing-data/accessing-genomic-data/how-to-request-and-access-datasets-from-dbgap. Data from the University of Utah Penelope Program are not publicly available due to privacy restrictions.

Declarations

Ethics approval and consent to participate

This study adhered to the Declaration of Helsinki. This study was reviewed and approved by the Institutional Review Board at the Undiagnosed Diseases Network and the University of Utah. IRB identifiers: 15HG0130 (UDN) and IRB_00129735 (Utah Penelope Program). Clinical and genomic data used in this study were obtained through participation in the Undiagnosed Diseases Network and the Utah Penelope Program. All protocols were reviewed and approved by the appropriate Institutional Review Boards, and informed consent was obtained from all participants or their legal guardians in accordance with institutional and national ethical guidelines.

Consent for publication

Written informed consent has been obtained from all the participants and the parents/legal guardians of minors for their personal or clinical details along with any identifying images to be published in this study. Participants were fully briefed on the purpose and use of their contributions and voluntarily agreed to their inclusion. Participant images were reviewed and have written approval by all participants and/or their legal guardians.

Competing interests

The authors declare no competing interests.