Alternative splicing allows one primary transcript to encode multiple protein isoforms
The basic mechanism of splicing was described. However, for the great majority of human genes, there is more than one way of splicing. Alternatively spliced transcripts (splice isoforms) can be identified for almost every human gene. These may skip one or more exons, include additional internal exons, or vary the length of an exon by repositioning the exon–intron junction (Figure 1). It may be that some of these isoforms just reflect imprecision in the complex splicing machinery and are not functionally significant—but numerous examples are known in which alternative splicing clearly is functional.
Fig1. Types of alternative splicing event. Red-colored boxes represent exons that are always included in the mature mRNA. (A) An intron (blue) is either retained or excluded. (B) Use of alternative splice donor sites results in the inclusion or exclusion of the sequence in blue. (C) Use of alternative splice acceptor sites results in the inclusion or exclusion of the sequence in blue. (D) The exon in blue may be either included or skipped (a cassette exon). (E) Alternative exons: the mature mRNA includes either the exon in yellow or the exon in blue, but not both or neither.
Functional alternative splicing can have a variety of results. Alternatively spliced exons may encode signals governing different intracellular localizations. Some proteins have soluble and membrane-bound isoforms, produced by the inclusion or omission of an exon that encodes a transmembrane domain. The two forms may compete. Competition for a ligand between soluble and membrane-bound isoforms can regulate cell surface receptors. The splicing may be tissue-specific, so that different tissues contain different variants. Alternatively spliced exons may introduce sites for important post-translational modifications, such as serine phosphorylation. Some genes are regulated by inclusion or exclusion of a poison exon that includes a premature stop codon.
As the most complex part of the human body, the central nervous system may need the most complex proteome, and alternative splicing is particularly marked in neurons. Many widely expressed genes have neuron-specific splice isoforms. As long ago as 1994 a compilation listed almost 100 examples of neuron-specific splicing, including every type shown in Figure 10.27. Many of the variant isoforms of ion channels and receptors are known to be functionally important. Some genes encode an extraordinary number of different transcripts. The example of protocadherins α and γ, with their batteries of alternative promoters, was mentioned earlier. Figure 10.28 shows the neurexin 3 (NRXN3) gene. This large gene on chromosome 14q encodes a cell adhesion and receptor molecule that is present at synapses in the nervous system. It has two promoters and 24 downstream exons. Seven of the exons can be each individually included or excluded in transcripts; one exon has alternative splice donor sites, and three have alternative splice acceptors. One of the alternatively spliced exons includes a stop codon, the use of which would produce a protein lacking transmembrane and cytoplasmic domains. Potentially this one gene could encode 1000 different proteins. Genes expressed in the central nervous system also often contain micro-exons, 6–30 nt exons that tend to be inefficiently spliced, thus adding to the variety of mature mRNAs.
Fig2. Alternative splicing of the neurexin 3 transcript in the nervous system. There are two alternative promoters, α and β (red bars). Exons 3, 4, 5, 12, 20, and 24 (blue) can each be either included or skipped. Exon 7 (light green) can be included, using either of two alternative 5′ splice acceptor sites, or it can be completely skipped. Exon 22 (purple) has two alternative 3′ splice donor sites. Exon 23 (yellow) has two alternative 5′ splice acceptor sites that use different reading frames, one of which leads to an in-frame stop codon within this exon. The protein produced from this variant lacks the transmembrane and cytoplasmic domains encoded by exon 24. Exon 24 (dark green) has three alternative 5′ splice acceptor sites. By using different combinations of variants, this single gene could potentially encode about 1000 different proteins.
What controls alternative splicing?
The basic splicing machinery was described in Chapter 1. However, not all splice sites are equal. The choice of where the spliceosome is assembled on the primary transcript depends on a balance of positive and negative factors. The sequence surrounding the invariant GU…AG sequences may be a better or worse fit to the optimum sequence. Nearby splicing enhancers bind SR (serine-arginine) proteins that help anchor the spliceosome in place, whereas splicing suppressors bind hnRNP (heterogeneous ribonucleoprotein) proteins that have the opposite effect. Thus, splice sites can be strong or weak. Weak sites may be skipped in favor of an alternative. Splicing patterns are often tissue-specific, presumably because enhancers or suppressors of splicing bind tissue-specific proteins. Attempts have been made to identify a splicing code that would predict this. A large study of 10,689 alternatively spliced human exons by Xiong and colleagues in 2015 (PMID 25525159; see Further Reading) used machine learning to identify combinations of 1393 sequence features (sizes of exons and introns, binding sites, structural features, and so on) that predicted the effect of sequence variants on splicing. Applying the predictor to 650,000 variants resulted in many successful predictions of pathogenicity. Evidently alternative splicing is predictable, but not by examination of just a few features.
Epigenetic marks can also affect splicing. Exons show epigenetic differences relative to introns, including increased CpG methylation and different histone modifications. Although these marks are present on the DNA and not on the primary transcript, they can affect the progress of the RNA polymerase along a gene, either directly or by binding proteins that block the polymerase. Splicing is co-transcriptional, and polymerase pausing can affect the splicing machinery, for example allowing more time to assemble a spliceosome on a weak splice site.
Alternative sites for 3′ cleavage and polyadenylation produce additional variation
About half of human genes use alternative cleavage and polyadenylation to generate messenger RNA transcripts that differ in the length of their 3′ untranslated regions (3′ UTRs), while producing the same protein. The use of specific alternative sites often depends on the particular cell type and can change upon proliferation or differentiation. As described below, 3′ UTRs include recognition sites for RNA-binding regulatory proteins and small RNAs, particularly microRNAs. The alternative 3′ UTRs can affect the cellular location or stability of a transcript and the abundance of the encoded protein.
RNA editing can change the sequence of the mRNA after transcription
In contravention of the central dogma, there are examples in which the DNA sequence of a gene does not fully determine the sequence of its transcript. RNA editing involves the insertion, deletion, or modification of specific nucleotides in the primary transcript. In humans the main types of event are deamination of cytosine or adenine resulting in C>U or A>I conversions (I is inosine; see below).
A>I editing is performed by members of the ADAR (adenosine deaminase acting on RNA) family of deaminases (Figure 3). Inosine base-pairs with cytosine rather than thymine. Over 99% of A>I edits occur in Alu sequences. Editing in human protein- coding genes is particularly seen in the central nervous system. Often the editing converts CAG codons, encoding glutamine (Q), into CIG codons that, like CGG, encode arginine (R). So-called Q/R editing alters the function of several genes encoding neurotransmitter receptors or ion channels (GABRA3, GRIA2, and GRIK2). A>I editing sometimes makes other coding-sequence changes (I/V, Y/C, N/S, and so on). In the HTR2C serotonin receptor gene, A>I editing at splice sites modulates alternative splicing.
Fig3. Deamination of adenosine. Enzymes of the ADAR family deaminate the amino group at carbon 6 of adenosine to produce inosine. R, ribose.
C>U editing is performed by enzymes of the APOBEC family. The human apolipoprotein gene APOB encodes the large ApoB100 protein in the liver. In the intestine, however, C>U editing at nucleotide position 6666 of the mRNA causes replacement of the CAA glutamine codon by a UAA stop codon (Figure 4). The mRNA now encodes a shorter polypeptide, ApoB48. Uncontrolled APOBEC editing is a major source of mutations in tumors —for example, converting arginine codon 3916 to a stop codon in the NF1 (neurofibromin) gene.
Fig4. APOBEC RNA editing: the two products of the APOB gene. In the liver, the APOB mRNA encodes a 4536-residue protein, ApoB100. In the intestine, the APOBEC1 cytosine deaminase specifically converts cytosine 6666 in the mRNA to uridine, changing the CAA glutamine codon 2153 into a UAA stop codon. The mRNA now encodes ApoB48, consisting of just the first 2152 amino acids of ApoB100.
Regulation of translation
For protein-coding genes, the layers of regulation extend from transcription and maturation of the transcript through to translation. The 5′ and 3′ UTRs of mRNAs have important regulatory functions. As described below, the 3′ UTR is the main site of miRNA binding. Two features in the 5′ UTR can affect initiation of translation.
• Between 10 and 30% of human mRNAs have additional open reading frames (ORFs) upstream of the main coding sequence. These can inhibit translation of the main reading frame. For example, upstream ORFs of the TPO gene limit production of the encoded protein, thrombopoietin. Mutations that prevent recognition of the upstream ORF lead to excessive production of thrombopoietin, causing hereditary thrombocythemia. The HR (hairless) gene has four upstream ORFs in its 691 bp 5′ UTR (Figure 5). Loss-of-function mutations in the inhibitory U2HR cause over expression of the HR protein. HR is a regulator of Wnt signaling; the disturbed Wnt signaling affects the cycling of hair follicles, resulting in an autosomal dominant hair-loss syndrome, Marie Unna hypotrichosis (OMIM #146550).
Fig5. Upstream open reading frames in the 5ʹ untranslated region of the HR (hairless) gene control production of the HR protein. ORF open reading frame. (Reprinted from Wen Y et al. [2009] Nat Genet 41:228–233; PMID 19122663. With permission from Springer Nature. Copyright © 2009.)
• Stem-loop structures formed by the single-stranded mRNA can directly affect the level of the gene product by impeding the progress of ribosomes, and they act as recognition sites for RNA-binding regulatory proteins. Figure 6 shows how stem-loop iron-responsive elements in the ferritin and transferrin receptor genes orchestrate the response to dietary iron.
Fig6. Iron-response elements in the ferritin and transferrin receptor mRNAs. (A) Stem-loop structure of an iron-response element (IRE) in the 5′ untranslated region (UTR) of the ferritin heavy (H)-chain mRNA. (B) Low iron concentrations activate a specific IRE binding protein (IRE-BP), enabling it to bind the IRE in the ferritin heavy-chain gene and also IREs in the 3′ untranslated region of the transferrin receptor (TfR) mRNA. Binding inhibits the translation of ferritin but protects the transferrin receptor mRNA from degradation. When the iron concentration is high, IRE-BP is degraded, releasing translational repression of stored ferritin mRNA but inhibiting production of the transferrin receptor.
The discovery of many small RNAs that regulate gene expression caused a paradigm shift in cell biology
Two seemingly disparate lines of research in the 1990s, both involving the C. elegans worm, alerted biologists to the unsuspected importance of very small RNAs. Andrew Fire and Craig Mello received the 2006 Nobel Prize in Physiology or Medicine for their roles in understanding RNAi—the specific inhibition of gene expression by short double-stranded RNA molecules. Their Nobel lectures, describing the process of discovery, can be read at http://nobelprize.org/nobel _prizes/medicine/laureates/2006/. Meanwhile Victor Ambros and Gary Ruvkun, among others, opened up the world of micro RNAs (miRNAs) in development. Their accounts of how they came to make these discoveries can be read in two Commentaries published in a Cell supplement.
Not all small RNAs are rare. An adult C. elegans cell contains more than 50,000 molecules each of miRNAs 2, 52, and 58. Why were they not discovered earlier? When researchers ran RNA gels, they assumed—usually with good reason—that the smear of very-low-molecular-weight RNA at the bottom of the gel consisted of degradation products and was of no interest. In addition, such short molecules are difficult to study: standard bioinformatics approaches overlook them, and such tiny targets were seldom hit in mutagenesis experiments. Thus, although they must have been observed in many laboratory experiments, they were not recognized.
MicroRNAs as regulators of translation
The various categories of small RNAs were listed in Table 9.6. MicroRNAs, siRNAs, and piRNAs are all short RNAs that act in related ways to repress expression of their targets. MicroRNAs are processed from capped and polyadenylated precursor pri-miRNAs through the Drosha and Dicer ribonucleases (Figure 7). One strand, the guide strand, of the resulting 21–22 nt RNA complexes with Argonaute and GW182 proteins to form the RISC (RNA-induced silencing complex). miRNAs hybridize to target sequences in mRNAs, primarily in the 3′ UTRs. Hybridization depends on imperfect base pairing, with nucleotides 2–8 (the seed region) providing the main specificity. Normally miRNAs initially repress translation of an intact mRNA, although subsequently the target mRNA is often deadenylated and degraded. In some cases miRNAs have been reported to activate, rather than repress, gene expression.
Fig7. Maturation of a miRNA. miRNA26a1 is initially synthesized as a capped and polyadenylated pri-miRNA. Inverted repeats in the pri-miRNA cause it to adopt a hairpin structure. Asymmetric cleavage by the Drosha (RNASEN) and Dicer enzymes (green arrows) produces a short double-stranded RNA with 3′ overhangs. Argonaute proteins complex with the guide strand (red) to form the RNA-induced silencing complex.
The miRBase database (www.miRBase.org; consulted January 2017) lists 1881 precursor and 2588 mature miRNAs in the human genome. The genes occur singly or in clusters in a variety of genomic contexts (Figure 8).
Fig8. MicroRNA genes. (A) Stand-alone primary transcripts. (B) miRNA genes within exons or introns of a long noncoding RNA. Expression will depend on expression of the host RNA. (C) miRNA genes within the 3′ UTR or introns of a host protein-coding mRNA.
There is no one-to-one relation between miRNAs and their targets. Each miRNA can affect translation of many targets, and many messenger RNAs are targeted by multiple microRNAs. Thus miRNAs form a very broad-based regulatory system. However, individual miRNA effects are usually modest. Few proteins show an increase as great as two fold when an inhibitory miRNA is knocked down, or repression greater than 50% when the cell is transfected with an miRNA. The general picture that emerges is of very wide ranging but relatively small effects compared to controls on transcription.
Because an individual miRNA can bind to many different target mRNAs, the different targets compete for the miRNA. Unless the miRNA is very abundant in a cell, higher transcription of target A would mop up more of the miRNA, leading to less binding to target B. Since miRNA effects are almost always repressive, this would increase translation of target B. The result is a dense network of regulatory interactions. Regulation by competing endogenous RNAs may be an important general phenomenon; there can also be competition for RNA-binding proteins as well as for miRNAs. Two examples illustrate these effects:
• Cells are very sensitive to the level of the PTEN tumor suppressor protein. PTEN has a pseudogene, PTENP1, which is transcribed although not translated. The PTENP1 transcript competes with PTEN mRNA for binding of miRNAs 17, 21, 214, 19, and 26. Expression of the pseudogene suppresses cell growth by reducing miRNA-based repression of PTEN and hence up-regulating the level of the growth-inhibitory PTEN protein;
• Circular RNAs had generally escaped detection, but were recently shown by Salzman and colleagues to be the predominant transcript isoform of hundreds of human genes (Salzman et al. [2012], PMID 22319583; see Further Reading). They are produced by abnormal splicing of multiexon transcripts, where a downstream splice donor site is spliced on to an upstream acceptor. One well-studied circular RNA, CDR1as, contains 77 binding sites for miR-7. It is a stable and abundant miR-7 sponge, and hence is a powerful up-regulator of other miR-7 target transcripts.
