Splicing and the Spliceosome

A newly transcribed RNA polymerase II product, the pre-mRNA, contains both exons (sequences retained in the mature mRNA) and introns (sequences removed). Splicing is the process that excises introns and ligates exons together, and it happens almost entirely co-transcriptionally — while RNA polymerase II is still transcribing downstream sequence, the spliceosome is already assembling on and removing introns from RNA emerging behind it. This coupling depends on the phosphorylated carboxy-terminal domain (CTD) of RNA polymerase II, which acts as a mobile docking platform recruiting splicing factors as transcription proceeds. Each intron is defined by a small set of degenerate sequence elements, not by its length or its "junk" content. These are the 5′ splice site (donor site) at the intron's upstream boundary, the branch point sequence containing an invariant adenosine near the 3′ end of the intron, a polypyrimidine tract, and the 3′ splice site (acceptor site). None of these elements is individually sufficient or perfectly conserved; splice site selection is a probabilistic, combinatorial recognition problem, which is precisely why alternative splicing (covered separately) is so easy to generate evolutionarily — small changes in these weak signals shift which exons get included. The chemistry itself is deceptively simple: two transesterification reactions, each breaking one phosphodiester bond and forming another, with no net change in the number of phosphodiester bonds and no ATP consumed in the bond-breaking/bond-forming steps themselves. What requires massive energy input is not the chemistry but the machine — assembling and remodeling the spliceosome, which spends dozens of ATP molecules per intron on ordered conformational rearrangements carried out by DEAD/DEAH-box RNA helicases. The spliceosome is therefore a paradigm of a broader theme in molecular biology: a ribozyme active site (RNA doing the catalysis) embedded in and precisely controlled by a large protein/RNA scaffold (proteins providing fidelity, timing, and regulation).

Step 1 — The consensus signals. In vertebrates, the 5′ splice site consensus is approximately `AG|GURAGU` (the vertical bar marks the exon–intron boundary; R = purine). The branch point consensus in mammals is loose, roughly `YNYURAC`, where the branch adenosine (bold A) is invariant and bulges out of the helix. Just upstream of the 3′ splice site lies a polypyrimidine tract (a stretch of ~15–20 pyrimidines, mostly U and C), and the intron ends in an essentially invariant `AG` immediately before the exon. Nearly all introns begin with GU and end with AG at the DNA level (the "GU-AG rule," discovered by Richard Breathnach, Chambon and colleagues in 1978), though a rare minor class begins AU and ends AC and is processed by an entirely separate, structurally analogous "minor spliceosome" built from U11, U12, U4atac, and U6atac snRNPs.

Step 2 — snRNP recognition (spliceosome assembly, the E→A→B→C pathway). Splicing proceeds through an ordered series of complexes: 1. E complex (early/commitment complex): U1 snRNP base-pairs with the 5′ splice site via direct RNA–RNA complementarity between U1 snRNA and the intron's 5′ end. Simultaneously, SF1 (splicing factor 1) binds the branch point, and the U2AF heterodimer (U2AF65/U2AF35) binds the polypyrimidine tract and the 3′ AG, respectively. This step commits the transcript to splicing at these particular sites and can occur co-transcriptionally within seconds of the 5′ splice site emerging from the polymerase. 2. A complex: U2 snRNP displaces SF1 and base-pairs directly with the branch point sequence, bulging out the branch adenosine so that its 2′-OH is positioned to attack. This step requires ATP hydrolysis (by the helicase Prp5) even though the branch point-U2 duplex itself is just base pairing. 3. B complex: the pre-assembled U4/U6.U5 tri-snRNP joins. In this tri-snRNP, U4 and U6 snRNAs are extensively base-paired to each other and U6 is catalytically inert; U5 snRNP contacts exon sequences flanking both splice sites, helping align the two exons for eventual ligation. 4. Activation to Bact and B\*: U1 and U4 snRNPs are destabilized and released, driven by the ATP-dependent helicase Brr2. Freed from U4, U6 snRNA now base-pairs instead with U2 snRNA and with the 5′ splice site (displacing U1), forming the catalytic RNA core of the spliceosome. This U2/U6 helix, together with U6's internal stem-loop, coordinates two catalytic Mg²⁺ ions — the spliceosome, like the group II self-splicing introns it resembles, is a metalloribozyme. 5. Catalytic step 1 (branching): the 2′-OH of the branch point adenosine performs a nucleophilic attack on the phosphate at the 5′ splice site, forming a new 2′–5′ phosphodiester bond and releasing the 5′ exon, still tethered to the intron-lariat-3′exon intermediate, which now contains a lariat structure (a loop with a branched nucleotide). 6. Catalytic step 2 (exon ligation): the newly freed 3′-OH of the released 5′ exon attacks the phosphate at the 3′ splice site. This ligates the two exons together and releases the intron as a lariat (a loop-and-tail structure closed by that unusual 2′–5′ bond at the branch point). 7. C complex → disassembly: the spliceosome releases the mature mRNA (exon-exon junction now marked by a deposited exon junction complex, EJC) and the excised lariat intron, which is rapidly debranched by the enzyme Dbr1 and degraded. All snRNPs are recycled for another round.

Step 3 — Why it's a ribozyme, and how we know. U6 snRNA's catalytic core is strikingly similar in sequence and structure to domain V of self-splicing group II introns, which also excise themselves via two transesterifications through a branched lariat intermediate with no protein catalyst at all. Cryo-EM structures of the activated spliceosome (largely from yeast, resolved by groups including Kiyoshi Nagai's and Reinhard Lührmann's labs in the 2010s) show that U2/U6 RNA, not any protein side chain, directly coordinates the two catalytic magnesium ions at the active site, confirming that the spliceosome's chemistry is RNA-catalyzed even though it sits inside a shell of ~150 proteins.

The discovery that genes are not colinear with mRNA (1977). Before 1977, molecular biologists assumed — reasonably, given bacterial genetics — that a gene's DNA sequence corresponded directly, base for base, to its mRNA and protein product. This assumption was overturned almost simultaneously by two groups. Phillip Sharp's laboratory and Richard Roberts's laboratory (working independently, both using adenovirus, a system with abundant, well-characterized mRNAs) performed R-loop mapping and electron microscopy on hybrids formed between mature adenovirus mRNA and the viral genomic DNA from which it was transcribed. The experiment: when a single-stranded mRNA is annealed to double-stranded genomic DNA under conditions that favor RNA-DNA hybrid formation, the mRNA displaces one DNA strand wherever it is complementary, forming an "R-loop" visible directly under the electron microscope as a bubble of paired RNA-DNA duplex with a looped-out displaced single DNA strand. If mRNA were colinear with the gene, R-loop mapping should show one continuous duplex region. Instead, Louise Chow, Susan Berget, and colleagues in Sharp's lab (Berget, Chow, and Sharp, 1977) and Berget & Sharp separately, alongside Roberts's group with the hexon mRNA, saw something startling: the mRNA hybridized to the DNA in several separate segments, with large loops of unpaired, single-stranded DNA bulging out between them — loops that had no corresponding mRNA sequence at all. What this proved: the mature mRNA was assembled from discontinuous segments of the genomic template. The DNA sequences in the loops — later named introns by Walter Gilbert — were present in the gene and in the primary transcript but were absent from the final, mature mRNA. The segments that did hybridize continuously were the exons. This was the direct physical demonstration that a gene's coding information is interrupted, and that removing and splicing together the interrupted pieces is a required step of gene expression. Sharp and Roberts shared the 1993 Nobel Prize in Physiology or Medicine for this discovery.

Following up: proving the mechanism, not just the phenomenon. The R-loop experiments showed introns exist and are removed, but not how. Subsequent biochemical work through the 1980s (notably from the laboratories of Tom Maniatis, Phillip Sharp, Joan Steitz, and Christine Guthrie using both mammalian nuclear extracts and yeast genetics) established the details described above:

• In vitro splicing extracts (HeLa cell nuclear extracts capable of accurately splicing radiolabeled pre-mRNA substrates, developed around 1983–84) allowed the reaction to be reconstituted and its intermediates trapped and characterized biochemically.
• The lariat intermediate was identified directly: debranching enzyme-sensitive circular RNA species with an unusual 2′–5′ phosphodiester bond at the branch point were isolated and sequenced from spliced-out introns (Padgett, Konarska, Grabowski, Sharp 1984; Ruskin, Krainer, Maniatis, Green 1984), confirming the two-step transesterification mechanism rather than a simple cut-and-paste.
• U1 and U2 snRNA complementarity to the 5′ splice site and branch point, respectively, was demonstrated by direct sequence comparison and confirmed functionally: oligonucleotide-directed cleavage of U1 snRNA (using RNase H plus antisense oligos, Kramer, Appel, Maniatis, and colleagues in the early 1980s) blocked splicing, showing U1 base-pairing is functionally required, not incidental.
• Yeast genetics (Christine Guthrie's lab) used cold-sensitive and dominant-negative mutations in snRNA genes to demonstrate that compensatory mutations in U6 snRNA and the 5′ splice site could restore splicing when paired together — direct genetic proof of the base-pairing model, since restoring complementarity restored function even when the individual mutant sequences alone abolished it.
• Cryo-EM structures of the yeast spliceosome at near-atomic resolution (2015 onward) visualized the U2/U6 catalytic RNA fold and the two active-site Mg²⁺ ions directly, closing the loop between the biochemical and genetic evidence and the physical structure.

Example 1 — Reading a splicing genotype-phenotype cross (a point mutation at a 5′ splice site). Consider a hypothetical autosomal gene CFX1 with the wild-type 5′ splice site of intron 4 reading (exon)...AG | GUAAGU...(intron), matching the consensus `GURAGU` closely. A patient is homozygous for a single point mutation changing the invariant G immediately after the exon-intron boundary to an A (GT→AT at the DNA level, the single most common class of splice-site mutation in human disease). Predict the outcome. Because U1 snRNP recognition of the 5′ splice site depends on direct base pairing between U1 snRNA and this GU dinucleotide, and GU is essentially invariant (>99% of human introns), this mutation destroys U1 binding at intron 4's own boundary. Three outcomes are possible, and all three are observed for different real splice mutations depending on context:

If exon 4 is 150 nucleotides long (a multiple of 3), exon skipping alone would preserve the reading frame but delete 50 amino acids — an in-frame deletion, potentially producing a partially functional, internally truncated protein. If exon 4 is, say, 148 nucleotides, skipping shifts the reading frame downstream, almost always introducing a premature stop codon, which would trigger nonsense-mediated decay and produce a null allele. This is exactly the logic clinicians use when interpreting splice-site variants in genetic diagnostics (e.g., in DMD, CFTR, or BRCA1 variant classification): first check whether the disrupted exon's length is a multiple of 3, because that determines whether skipping is in-frame (often milder) or frameshifting (often severe).

Example 2 — A quantitative reasoning problem: how much of a genome is intron, and why splicing fidelity is a fidelity problem, not just a rate problem. Suppose a human gene has:

• 40 introns, averaging 3,500 nt each
• 41 exons, averaging 145 nt each

Total intron length ≈ 40 × 3,500 = 140,000 nt Total exon length ≈ 41 × 145 = 5,945 nt Total pre-mRNA length ≈ 145,945 nt Fraction of the primary transcript that is intron ≈ 140,000 / 145,945 ≈ 95.9% Now suppose the spliceosome recognizes 5′ splice sites with 99.5% per-site accuracy (correct site chosen over any single cryptic near-consensus alternative). Across 40 introns, the probability that every single 5′ splice site in this transcript is used correctly is: 0.995⁴⁰ ≈ e^(40 × ln0.995) ≈ e^(−0.2006) ≈ 0.818, i.e., about 82%. This shows why splice site recognition cannot rely on any single weak signal in isolation — with 40 introns per transcript and per-site error rates even as low as 0.5%, roughly 1 in 5 transcripts would carry at least one splicing error if sites were read independently and without redundancy. In reality, measured missplicing rates are far lower than this naive calculation because recognition is combinatorial: exonic and intronic splicing enhancers/silencers (bound by SR proteins and hnRNP proteins), the polypyrimidine tract, the branch point, and cross-exon/cross-intron "definition" by U1 and U2AF acting cooperatively across short exons all reinforce correct site choice, pushing real accuracy far above what any isolated consensus match alone could achieve. This is also precisely the regulatory substrate that alternative splicing exploits.

Phillip Allen Sharp trained as a chemist before turning to molecular biology, joining Cold Spring Harbor Laboratory and then MIT, where his laboratory worked on adenovirus, a DNA tumor virus whose late transcripts were abundant, well mapped, and biochemically tractable — ideal for physically visualizing RNA-DNA hybrids. Sharp's group, alongside Richard Roberts's independent team at Cold Spring Harbor, was searching for a rigorous physical method to test whether viral mRNA really corresponded, base for base, to the DNA template predicted by the then-standard colinear model of gene expression. The R-loop electron microscopy experiments were chosen precisely because they made an invisible molecular question visible: you could photograph, under the electron microscope, whether the DNA underneath an mRNA was one continuous duplex or broken into segments with intervening loops. What they saw contradicted the field's working assumption so cleanly that it forced an immediate revision of the definition of a gene.

Human genetic disease. A large fraction of disease-causing mutations — commonly cited estimates suggest up to roughly a third of inherited disease mutations — disrupt splicing rather than the protein-coding sequence directly, by damaging splice sites, branch points, or exonic splicing enhancers. Diagnostic pipelines for diseases such as Duchenne muscular dystrophy (DMD), cystic fibrosis (CFTR), and hereditary breast/ovarian cancer (BRCA1/2) routinely flag intronic variants precisely at the GU/AG dinucleotides and branch points as likely pathogenic, using the logic in Worked Example 1.

Splice-modulating therapeutics. Spinal muscular atrophy is caused by loss of functional SMN1, but a near-identical paralog, SMN2, is present and would compensate if only its exon 7 were included more often — a single nucleotide difference weakens an exonic splicing enhancer in SMN2, causing exon 7 skipping in most transcripts. Nusinersen (Spinraza), an antisense oligonucleotide, base-pairs to a nearby intronic splicing silencer in SMN2 pre-mRNA, blocking a repressive splicing factor and forcing inclusion of exon 7. This directly exploits the mechanism described above and was a landmark proof that splicing itself can be pharmacologically redirected, not just studied.

Exon-skipping drugs for muscular dystrophy. Antisense oligonucleotides such as eteplirsen are designed to mask specific exons' splice sites in DMD pre-mRNA, forcing the spliceosome to skip a mutated exon and restore an in-frame (if truncated) dystrophin protein, converting a severe frameshift-causing mutation into a milder in-frame deletion — the same reasoning as Worked Example 1, applied therapeutically.

Cancer biology. Recurrent somatic mutations in core spliceosomal components (SF3B1, U2AF1, SRSF2) are among the most common mutations in myelodysplastic syndromes and certain leukemias, altering branch point and 3′ splice site recognition genome-wide and producing aberrant isoforms that are now being explored as both biomarkers and drug targets (e.g., SF3B1-modulating small molecules in clinical trials).

Biotechnology and synthetic biology. Understanding splice site consensus sequences underlies computational gene prediction algorithms, intron-based reporter constructs used to boost transgene expression (introns can enhance mRNA export and stability), and engineered minigene splicing assays used routinely to functionally test candidate pathogenic variants in a research or diagnostic laboratory before committing to a clinical interpretation.

Ask the tutor: How do exonic splicing enhancers and silencers tip the balance toward including or skipping a weak exon, and how does this set up alternative splicing?