TIBS 25 ��� AUGUST 2000 381 0968 ��� 0004/00/$ ��� See front matter �� 2000, Elsevier Science Ltd. All rights reserved. PII: S0968-0004(00)01604-2 EUKARYOTIC mRNAs ARE transcribed as precursors containing intervening se- quences (introns). These sequences are subsequently removed such that the flanking regions (exons) are spliced together to form mature mRNA. Alternative splicing pathways generate different mRNAs encoding distinct protein products, thus increasing the coding capacity of genes. Alternative splicing can also act as an on���off gene expression switch by the introduction of premature stop codons. Figure 1 illustrates examples of differ- ent modes of alternative splicing, which have dramatic biological consequences. Combinations of these basic events in complex transcription units can gener- ate a multitude of protein isoforms. A remarkable example is the cell-specific expression of a subset of 576 possible alternatively spliced forms of a K1 chan- nel mRNA. These are expressed in a gra- dient along the 10 000 sensory-receptor cells present in the inner ear of birds, which enables perception of different sound frequencies1. Alternative splicing is often tightly regulated in a cell-type- or developmental- stage-specific manner. Coordinated changes in alternative splicing patterns of multiple pre-mRNAs are an integral component of gene expression pro- grammes like those involved in nervous system differentiation2 and apoptotic cell death3. Splice-site choice must therefore be tightly regulated in time and space. The purpose of this article is to review recent evidence from a variety of mammalian and Drosophila genes which supports the notion that even simple decisions can result from a com- plex interplay between signals in the RNA and trans-acting factors that assem- ble on them. (Excellent reviews on more general aspects of splicing regulation can be found in Refs 4,5.) Needles in a haystack The question of splice-site choice is intimately connected to the problem of normal recognition of constitutive splice sites6. A feature shared by both regulatory sequences and splice-site signals is that they are usually short and often degenerate. The information con- tent of their primary sequence is there- fore rather limited (Fig. 2a). Even the best computer programs are only 50% accurate in predicting actual splice sites over multiple, equally good candidate sequences that are not used. Splice-site recognition represents a daunting prob- lem, underscored by the fact that 15% of human genetic diseases are caused by mutations that destroy functional splice sites or generate new ones7. Initial recognition of splice sites in- volves crosstalk between multiple, rela- tively weak interactions that contribute to establishing complexes that commit the pre-mRNA to splicing8. The splicing reaction occurs in the spliceosome, a complex composed of five small nuclear ribonucleoprotein particles (snRNPs) and 50���100 polypeptides, many of which are not associated with snRNPs (Refs 9,10). The early steps of spliceo- some assembly, which provide the main targets for regulation, involve recogni- tion of the consensus elements at both ends of the intron (Fig. 2a). U1 snRNP binds to the 59 splice site, splicing factor 1 (SF1, also known as branch-point-bind- ing protein or BBP) to the branch point11, whereas the 65 and 35 kDa subunits of U2 snRNP auxiliary factor (U2AF) recognize the polypyrimidine tract and 39 AG, respectively12 (Fig. 2b). Bridging interactions between U1 snRNP bound to the 59 splice site and SF1/U2AF bound to the 39 splice-site region necessarily exist, but their molecular nature has not yet been fully established. Candidate bridging factors in higher eukaryotes are members of the serine���arginine (SR) family of splicing factors13,14. These pro- teins contain N-terminal RNA recogni- tion motifs, which mediate binding to pre-mRNA. Their C-terminal arginine��� serine-rich (RS) domains mediate protein���protein interactions with simi- lar RS domains in U2AF35 and U1 snRNP 70K protein, thereby promoting U2AF and U1 snRNP binding to splice sites (Fig. 2b)4,5,14. One class of sequences bound by SR proteins are exon splicing enhancers (ESEs), which are often purine rich and play stimulatory roles in both constitutive and regulated splicing15. Different ESEs are recognized by specific subsets of SR proteins14. Vertebrate exons are usually short and separated by long introns that can be many kilobases in length. For the ma- jority of internal exons, which are be- tween 50 and 300 nucleotides in length, initial recognition of splice sites is en- hanced by interactions between the 39 and 59 splice sites across the exon in a process termed exon definition16 (Fig. 2b). How the pairing between splice sites across exons is subse- quently swapped to pairing of sites across the (usually) much longer in- trons is an important question this step is another potential target for regulating alternative splicing. A second question is how are the terminal 59 and 39 exons defined? The 7-methylguanosine tri- phosphate (m7Gppp) cap structure at the 59 end of the transcript promotes recognition of the first 59 splice site, and polyadenylation signals at the 39 end promote the use of the last 39 splice site17. These processing events all occur co-transcriptionally, and the functional coupling between them reflects the inte- gration of all steps of mRNA synthesis in mRNA ���factories���. The hyperphosphoryl- ated C-terminal domain (CTD) of the REVIEWS Alternative pre-mRNA splicing: the logic of combinatorial control Christopher W.J. Smith and Juan Valc��rcel Alternative splicing of mRNA precursors is a versatile mechanism of gene expression regulation that accounts for a considerable proportion of pro- teomic complexity in higher eukaryotes. Its modulation is achieved through the combinatorial interplay of positive and negative regulatory signals pre- sent in the RNA, which are recognized by complexes composed of mem- bers of the hnRNP and SR protein families. C.W.J. Smith is at the Dept of Biochemistry, University of Cambridge, 80 Tennis Court Road, Old Addenbrookes Site, Cambridge, UK CB2 1GA and J. Valc��rcel is at the Gene Expression Programme, European Molecular Biology Laboratory, Meyerhofstrasse 1, D-69117 Heidelberg, Germany. Emails: cwjs1@mole.bio.cam.ac.uk juan.valcarcel@embl-heidelberg.de

REVIEWS TIBS 25 ��� AUGUST 2000 382 large subunit of RNA polymerase II (Fig. 2b) binds capping, polyadenylation and SR splicing factors. In addition to delivering some processing factors to their sites of action, there is evidence that the CTD can play a more direct role in the processing reactions themselves18. Packing and remodelling RNA Primary transcripts form densely packed ribonucleoprotein complexes, known as heterogeneous nuclear (hn)RNPs, by associating with a family of polypeptides known as hnRNP pro- teins19,20. These are a diverse group of nuclear RNA-binding proteins that are involved in multiple functions. They contain various types of RNA-binding motifs, as well as domains rich in glycine and other amino acids (but not RS domains), which might serve in both RNA binding and protein���protein inter- actions. The set and arrangement of hnRNP proteins bound to an RNA influ- ences its fate within the nucleus. hnRNP proteins of the A, B and C families can assemble with nascent pre-mRNA into regular 40S structures, which have been compared with the nucleosomal packag- ing of DNA (Ref. 19). Although assembly of 40S structures has been argued to be nonspecific19, the individual hnRNP pro- teins do have preferred binding se- quences20. This could serve to ���phase��� the packaging of the pre-mRNA, and this, in turn, could antagonize or assist splice-site selection. Such packaging could compete with assembly of early- splicing complexes across exons and might explain why exons larger than 300 nucleotides are recognized ineffi- ciently8. hnRNP packaging can also bring together distant regions of the pre-mRNA and therefore assist splice- site pairing21. Several hnRNP and SR proteins accompany the mRNA from the nucleus and might influence mRNA nucleocytoplasmic transport, and even cytoplasmic translation, RNA localiz- ation and decay20,22. Many others are removed from the RNA before export. These changes in the hnRNP comple- ment of the RNA are achieved in part by the processing complexes themselves. But, by analogy with chromatin, it is possible that specific remodelling ma- chines might also play a role in altering the packaging of transcripts throughout their lifetime. Antagonism in splice-site selection Early biochemical studies indicated that hnRNP proteins could regulate splice-site choice13. SV40 virus large T and small t proteins are generated from the same pre-mRNA by the use of alter- native 59 splice sites. An excess of the SR protein alternative splicing factor (ASF, also known as splicing factor 2 or SF2) promoted the use of the proximal site. This observation was subsequently extended to other RNAs and SR pro- teins. A model that explains these obser- vations is that under limiting concen- trations of ASF/SF2, U1 snRNP binds only to functionally stronger splice sites. Thus, a weak 59 splice site might not be selected, despite its proximity to the 39 splice site. Higher levels of ASF/SF2 promote full occupancy of all 59 splice sites by U1 snRNP, and under these conditions the 59 splice site clos- est to the 39 splice site is selected4,13,23. By contrast, hnRNP A1 antagonizes this activity of SR proteins, causing a shift to selection of distal 59 splice sites. The molecular mechanisms underlying the effects of hnRNP A1, and whether these are related to RNA packaging (see above), are less well understood. Access to the proximal site by ASF/SF2 could be blocked by hnRNP A1. Indeed, binding of hnRNP A1 to certain exon splicing silencers (ESS) inhibits the use of adjacent 39 splice sites24,25. Inhibition can even be maintained after replace- ment of the ESS by a heterologous bind- ing site, if the cognate RNA-binding protein has been fused to the C-terminal 2 1 VASE 7 8 Embryonic brain Adult brain NCAM AUGF AUGM 2 3 (a) (b) (c) (e) (d) fruitless FGFR-2 Binds KGF Binds FGF 7 (IIIa) 10 IIIb IIIc Thyroid Neurons 3 4 5 6 A A msl-2 Calcitonin/CGRP Ti BS Figure 1 Different modes of alternative splicing and examples of its biological consequences. (a) Alternative 59 splice-site use in the Drosophila gene fruitless governs sexual orientation and behaviour. Male (green) and female (red) patterns of splicing, as well as translation ini- tiation codons giving rise to long open reading frames, are indicated. Red lines in exon 2 represent binding sites for Tra (for ���Transformer���) and Tra-2. (b) Alternative 39 splice-site usage, associated with differential use of polyadenylation sites (represented by A) in the vertebrate gene for calcitonin and calcitonin-gene-related peptide (CGRP) generates a cal- cium homeostatic hormone in the thyroid gland or a vasodilator neuropeptide in the nerv- ous system. Processing patterns in green are found in thyroid, those in red are found in neu- rons. (c) Differential inclusion or skipping of the variable alternatively spliced exon (VASE) in the gene for neural cell adhesion molecule (NCAM) in embryonic (green) versus adult (red) rat brain, represses or promotes axon outgrowth during development. (d) Mutually ex- clusive use of exons IIIb and IIIc in mammalian fibroblast growth factor receptor 2 (FGFR-2) changes its binding specificity for growth factors during prostate cancer progression. The pattern of splicing represented in green generates an mRNA encoding a receptor with high affinity for keratinocyte growth factor (KGF), whereas that in red generates a receptor with high affinity for FGF. (e) Female-specific retention of an intron at the 59 untranslated region (UTR) of the gene male-specific-lethal 2 (msl-2) allows export of the unspliced RNA to the cy- toplasm. The protein Sex-lethal facilitates both intron retention in the nucleus and transla- tional repression in the cytoplasm, thereby switching off msl-2 expression, which controls X- chromosome dosage compensation.