Splicing: one gene, many messages

The splicing code in the sequence

The spliceosome recognises four short sequence elements in each intron:

The core motifs are short and degenerate, so they carry too little information on their own to define exons. Additional signals help: exonic and intronic splicing enhancers and silencers (ESE, ESS, ISE, ISS) recruit activating SR proteins (SRSF1–12) or repressing hnRNPs. Tissue-specific regulators (RBFOX, NOVA, PTBP1/2, MBNL, CELF, ESRP) create neural, muscle or epithelial splicing programmes. Chromatin and Pol II elongation speed matter too: slow elongation gives weak exons more time to be recognised.

The spliceosome's two chemical steps

1. Branching: the 2′-OH of the branch point adenosine attacks the 5′ splice site, freeing exon 1 and forming a lariat intron.
2. Exon ligation: the free 3′-OH of exon 1 attacks the 3′ splice site, joining the exons and releasing the lariat, which is debranched and degraded.

The spliceosome assembles stepwise from five small nuclear ribonucleoproteins (U1, U2, U4, U5, U6) and more than 150 proteins. Exon junction complexes (EJCs) are deposited ~20–24 nt upstream of each junction, which links splicing to mRNA export, translation and nonsense-mediated decay.

Types of alternative splicing

Consequences: nonsense-mediated decay and protein diversity

When an isoform contains a premature termination codon (PTC) more than ~50–55 nt upstream of the last exon–exon junction, a downstream EJC remains on the mRNA after the first round of translation and triggers nonsense-mediated decay (NMD). Cells use this deliberately: "poison exons" containing PTCs are included to downregulate genes such as splicing factors themselves (alternative splicing coupled to NMD, AS-NMD). In analysis, NMD means a frameshifting or exon-skipping variant often reduces RNA abundance rather than producing a truncated protein. This is why allele-specific expression in RNA-seq is used to confirm loss-of-function variants.

Isoforms that do get translated can differ in domains, localisation signals or interaction surfaces, sometimes with opposite functions (Bcl-xL vs Bcl-xS). How many alternative isoforms produce stable, functional proteins remains an open question; mass spectrometry detects far fewer protein isoforms than RNA-seq predicts, partly because of sensitivity and partly because many isoforms may be noise.

Why splicing mutations matter

Estimates vary by method, but a substantial fraction of disease-causing variants act through splicing, from roughly 10% when only canonical ±1/±2 sites are counted to well over 20% when deep intronic and exonic splicing variants are included. Mechanisms:

• Canonical site loss (GT→AT at +1): usually exon skipping or intron retention, often leading to NMD.
• Cryptic site activation: a variant creates a new donor or acceptor deep in an intron, adding a pseudo-exon (common in CFTR, DMD, ATM).
• Exonic variants that disrupt ESEs/ESSs: a "synonymous" change can alter splicing. Synonymous does not mean silent.
• Branch point and polypyrimidine tract variants, which are often missed by tools that only examine ±8 nt from exon boundaries.
• Trans-acting mutations in spliceosome genes (SF3B1, SRSF2, U2AF1, ZRSR2) in myelodysplastic syndromes and other cancers alter thousands of events at once.
• Repeat expansions that sequester splicing regulators, as CUG repeats in DMPK trap MBNL in myotonic dystrophy type 1.

Computational analysis of splicing

From RNA-seq

• Splice-aware alignment (STAR, HISAT2) finds reads that span junctions. Use two-pass mode for novel junctions. Long reads (minimap2 -ax splice) capture whole isoforms.
• Event-level quantification: rMATS, SUPPA2, MAJIQ, Whippet and VAST-TOOLS compute PSI and test for differential splicing.
• Annotation-free intron clustering: LeafCutter groups introns that share splice sites and tests intron-usage ratios; widely used for sQTL mapping in GTEx.
• Exon-bin or isoform-level: DEXSeq (differential exon usage), DRIMSeq, satuRn and IsoformSwitchAnalyzeR (functional consequences of isoform switches).
• Single-cell splicing: full-length protocols (Smart-seq2/3) or long-read single cell; 3′-tag 10x data are largely blind to internal splicing.

From DNA sequence (variant effect prediction)

• SpliceAI (deep residual CNN over 10 kb context) predicts donor/acceptor gain and loss. Delta score ≥ 0.2 is a common screening threshold and ≥ 0.5 is high-confidence.
• Pangolin (tissue-aware), MMSplice (modular), SQUIRLS, SPiP, MaxEntScan (classic position-weight/maximum-entropy scorer).
• Clinical laboratories increasingly validate predicted splice variants with RNA-seq from blood or fibroblasts, or with minigene assays, before classification.