Ending the message: poly(A) signals then and now

A Review of mRNA polyadenylation and a historical perspective of how the field has evolved since the discovery of poly(A) tails in the 1970s.

Abstract

Polyadenylation [poly(A)] signals (PAS) are a defining feature of eukaryotic protein-coding genes. The central sequence motif AAUAAA was identified in the mid-1970s and subsequently shown to require flanking, auxiliary elements for both 3′-end cleavage and polyadenylation of premessenger RNA (pre-mRNA) as well as to promote downstream transcriptional termination. More recent genomic analysis has established the generality of the PAS for eukaryotic mRNA. Evidence for the mechanism of mRNA 3′-end formation is outlined, as is the way this RNA processing reaction communicates with RNA polymerase II to terminate transcription. The widespread phenomenon of alternative poly(A) site usage and how this interrelates with pre-mRNA splicing is then reviewed. This shows that gene expression can be drastically affected by how the message is ended. A central theme of this review is that while genomic analysis provides generality for the importance of PAS selection, detailed mechanistic understanding still requires the direct analysis of specific genes by genetic and biochemical approaches.

The molecular biology of eukaryotic genes has been transformed in recent years from specific knowledge of how a few eukaryotic genes are expressed to a genome-wide perspective. This has been achieved by ingenious technological advances that afford the accumulation of enormous molecular detail. However, a theme of this review is that much of our current understanding of how gene expression is regulated was laid down in early experiments on specific genes, which is now being confirmed and extended by new genomic analysis. This particularly holds true for polyadenylation [poly(A)] signals (PAS) of eukaryotic protein-coding genes. This review charts our ever-increasing knowledge of the mechanism of formation of the ubiquitous 3′-terminal poly(A) tail, taken as a defining feature of translationally competent messenger RNA (mRNA). I begin by describing early experiments that revealed the presence of poly(A) tails on mRNA and the first clues as to how these nontemplated sequences are added to the right RNA 3′ ends. I also update our knowledge of how the placement of a poly(A) tail at the 3′ end of mRNA influences the expression of so many genes. This is revealed by current genomic data, where sequencing a whole genome now takes less time than it initially took to sequence the 3′ end of a single mRNA. Finally, I describe the host of experiments that place the polyadenylation process at its central point in the gene expression pathway, in particular by defining the extent of mRNA 3′ untranslated regions (UTRs).

Polyadenylation signals and 3′ noncoding RNA (ncRNA) sequences

The first clues that mRNA has a unique 3′-terminal tail came from early mammalian cell fractionation experiments that allowed the isolation of translationally active polysome-associated mRNA. RNase digestion (pancreatic RNase cuts at C and U residues, while T1 RNase cuts at G) of this mRNA preparation revealed a resistant fraction presumed to be poly(A) (Lim and Canellakis 1970; Edmonds et al. 1971; Adesnik et al. 1972; Mendecki et al. 1972; Birnboim et al. 1973). Since long poly(A) tracts were not thought to be DNA-templated (Birnboim et al. 1973; Jelinek et al. 1973), a poly(A) polymerase was sought and found that was subsequently shown to be responsible for poly(A) tail formation on mRNA (Winters and Edmonds 1973a,b). The function of mRNA poly(A) could only be guessed at in these initial studies. However, it was certainly very handy as a natural tag to allow isolation of mRNA by oligo(dT) affinity chromatography away from bulk ribosomal RNA that lacks poly(A) segments (Aviv and Leder 1972). These early experiments predated recombinant DNA technology. The only way to isolate individual mRNAs was to select tissue that had pronounced and selective gene expression so that particular mRNA is unusually abundant. Thus, globin mRNA was purified from mammalian red blood cells (Mathews et al. 1971), ovalbumin mRNA was purified from chicken oviduct cells (Rosen et al. 1975), and immunoglobulin mRNA was purified from murine B cells (Brownlee et al. 1973). My own research studies as a graduate student at Cambridge, UK, in the mid-1970s with George Brownlee and Fred Sanger began by using Escherichia coli DNA polymerase to partially reverse-transcribe mRNA (retroviral reverse transcriptase had yet to be purified). This enzyme inefficiently uses an RNA template when Mg²⁺ is replaced by Mn²⁺ in the reaction mix. Short stretches of complementary DNA (cDNA) were synthesized by use of oligo(dT) priming on the poly(A) tails of mRNA. In effect, these experiments were among the very first described examples of cDNA synthesis (Proudfoot 1976). I next used a now seemingly primitive, although quite effective, DNA sequencing technique developed in the Sanger laboratory (Brownlee and Sanger 1969; Sanger et al. 1973; Galibert et al. 1974) that involved two-dimensional (2D) chromatographic “fingerprinting.” By this and other nucleic acid analytic techniques (Fig. 1), I was able to piece together six separate mRNA sequences adjacent to the poly(A) tail and showed that each mRNA possessed the common sequence AAUAAA placed close (within 20–30 nucleotides [nt]) to the 3′-terminal poly(A) tail (Fig. 2A). We predicted from this small but then complete set of purified mRNA that AAUAAA was a signal for mRNA polyadenylation as well as a signal to terminate transcription (Proudfoot and Brownlee 1976). As described below, both of these predictions have turned out to be correct. What these early mRNA sequencing experiments also revealed was the fact that the stop codon of mRNA did not define the mRNA 3′ end. Rather, a 3′ ncRNA sequence, subsequently called the 3′ UTR, existed that at its 3′ end possessed the gene's PAS (Proudfoot and Longley 1976).

Figure 1.

(A) Schematic of cDNA synthesis by E. coli DNA polymerase (Klenow subfragment that lacks the 5′–3′exonuclease domain) using oligo(dT)₁₂ as primer (in red), base-paired to mRNA 3′ poly(A) in a reaction mix containing MnCl₂ in place of MgCl₂ and four deoxyribonucleotides, one α³²P-labeled. Short cDNAs produced by this process correspond to the 3′ end of the mRNA 3′ UTR. Purified cDNA was depurinated by formic acid treatment (selectively degrades G and A) releasing oligopyrimidies, especially oligo(dT) copied from the poly(A) sequence of the mRNA. Alternatively cDNA was digested with partially C-specific endonuclease IV, and the endonuclease IV digestion products were then purified and further degraded by partial digestion with venom 5′–3′ exonuclease (Proudfoot 1976). (B) cDNA depurination products were fractionated by denaturing polyacrylamide gel electrophoresis. A series of oligo(dT) products are visible, ranging from dT₁₃–dT₅₀, and correspond to oligo(dT)-primed cDNA generated by oligo(dT), which is base-paired at positions ranging from the 5′ to 3′ ends of the poly(A) tail (known to be about A₅₀ for globin mRNA). Marker (M) and sample (S) lanes are shown, with the positions of oligo(dT)₁₂ and unincorporated α³²P dNTP also indicated. (C) 2D fractionation of cDNA endonuclease IV products. The first dimension was by electrophoresis on cellulose acetate strips at pH 3.5. Base composition of oligonucleotide determines mobility. A cellulose acetate strip with fractionated oligonucleotides was blotted onto the bottom of a DEAE cellulose acetate, thin-layer plate using high-salt buffer. Homochromatography was then performed (second dimension) using a buffer (called a homomix) containing partially degraded crude yeast RNA that acts to displace endonuclease IV oligonucleotides from the base of the thin-layer plate. The extent of this displacement depends on molecular size (Brownlee and Sanger 1969). Separated ³²P oligonucleotides were then eluted from the thin-layer plate. The first and second dimensions are indicated by arrows. (D) 2D fractionation (as in C) of partial venom exonuclease digested spot 9. A series of products are evident, with adjacent ones varying by 1 nt. The angle of the 2D shift between adjacent spots is characteristic of a specific nucleotide loss. Purine nucleotides (G and A) give a larger shift in the second dimension. T and G give large and small rightward shifts (respectively), while C and A both give small leftward shifts in the first dimension. The sequence TTATT was deduced from observed mobility shifts and corresponds to the PAS AAUAAA. The first T of the PAS cDNA was inferred from depurination data (not shown). (B) Bromophenol blue marker. (E) Separated oligonucleotide sequences were aligned, as some sequences overlap due to partial endonuclease IV digestion. This gave the sequence of the β-globin cDNA adjacent to the poly(A) tail. Endonuclease IV spots not in the β-globin cDNA sequence were inferred to derive from the α-globin cDNA (Proudfoot 1976). († and ‡) Related oligonucleotides from varying 5′ and 3′ ends of cDNA.

Figure 2.

(A) Sequence alignment of the original six mRNA 3′ ends derived from sequencing technology as outlined in Figure 1. The positions of the conserved AAUAAA signals (boxed) and 3′-terminal nucleotides (underlined) originally noted to be conserved are indicated. (Proudfoot and Brownlee 1976). (B) Current general consensus sequences for the mammalian poly(A) signal. Distance variation between different parts of the PAS is indicated. (Red thunderbolt) Cleavage position.

Soon after these early cDNA sequencing experiments, the recombinant DNA era began, and coincidently Fred Sanger developed much more powerful gel-based DNA sequencing techniques (Sanger et al. 1977). Many more mRNAs were cloned and characterized from so-called libraries of cDNA plasmids that were made from various RNA preparations derived from tissues or cell culture sources of different eukaryotes (Maniatis et al. 1976, 1982). It was generally shown to be the case that mRNAs possess 3′ UTRs varying in length from ∼50 nt to several thousand nucleotides with a 3′ proximal AAUAAA PAS or close variant (AU/GUAAA or UAUAAA) (Wickens and Stephenson 1984; Zhao et al. 1999). Recombinant DNA technology also led to site-directed mutagenesis approaches. Thus, mutation analysis of the SV40 late PAS by limited exonuclease degradation from a closely positioned restriction site showed that AAUAAA did indeed form a required part of the mRNA PAS (Fitzgerald and Shenk 1981). Two rare forms of thalassaemia in humans were then characterized with point mutations in the AAUAAA sequence of the mutant gene (AAUAAG in the α2-globin gene and AACAAA in the β-globin gene). In both cases, subcloning of the mutant PAS revealed that it lost its poly(A) formation function (Higgs et al. 1983; Orkin et al. 1985).

Although the AAUAAA sequence was shown to be absolutely required for mRNA 3′-end polyadenylation, other sequence elements were shown to be necessary to fully reconstitute a functional PAS. In particular, the GU-rich sequence (or downstream sequence element, DSE) present just past the mRNA 3′ end in the immediate gene 3′ flanking region was shown to enhance 3′-end formation (Gil and Proudfoot 1984, 1987; McLauchlan et al. 1985). Similarly, the sequence immediately upstream of AAUAAA (upstream sequence element, USE) (Carswell and Alwine 1989; DeZazzo et al. 1991; Valsamakis et al. 1991; Moreira et al. 1995; Brackenridge and Proudfoot 2000; Venkataraman et al. 2005; Danckwardt et al. 2007) can in some cases act as an enhancing element for 3′-end processing efficiency. Finally, the actual nucleotides at the site of 3′-end cleavage can also influence the efficiency of this process (Chen et al. 1995). This proved to be medically relevant for the human prothrombin gene, which normally carries a CG dinucleotide sequence just 5′ to the cleavage position. In ∼1% of Caucasians (myself included), this sequence is mutated to CA, causing a mild thrombophilia phenotype. This is due to a twofold increase in prothrombin gene expression as a result of the acquisition of a more efficient PAS (Gehring et al. 2001; Danckwardt et al. 2008). These defining experiments for mRNA PAS were focused on mammalian mRNA. However, other eukaryotic model systems revealed that the mammalian pattern of USE-AAUAAA-DSE (Fig. 2B) was generally conserved across eukaryotes, even though in budding yeast greater variation in these cis elements appears to be tolerated (Graber et al. 1999). One notable feature of PAS is that, depending on exactly how the AAUAAA and DSE signals are positioned and defined, the actual site of poly(A) addition can vary by several to tens of nucleotides (Zhang et al. 1986; Sheets et al. 1990; Tian et al. 2005). This is especially evident in yeast PAS, which, perhaps due to lower sequence conservation, do not accurately define the position of polyadenylation (Zhao et al. 1999). This is in marked contrast to pre-mRNA splicing, which must always occur at a precise nucleotide position to maintain the correct reading frame of the mRNA for translation.

Genomic analysis of poly(A) sites has strikingly illustrated the generality of the originally laboriously defined eukaryotic PAS. Initially, annotation of genes within complete genome sequences proved hard, as the highly intronic nature of higher eukaryotic genes meant that even defining genes based on their protein-coding capacity was difficult, let alone exactly where the gene transcript starts or stops. The standard approach has been to match transcriptomes to genomes by hybridizing oligo(dT)-primed cDNA from various RNA preparations to genomic microarrays, possible in smaller yeast genomes but still technically challenging for larger mammalian genomes. From these data, many potential mRNA poly(A) sites were assigned (Nam et al. 2002; Brockman et al. 2005). While a good number corresponded to canonical poly(A) sites as defined by the earlier, above-described mutagenesis experiments, many more potential poly(A) sites were apparently devoid of the expected RNA signals. However, several artifacts crept into these bioinformatic analyses. Firstly, oligo(dT) priming on RNA fractions can frequently occur on internal oligo(A) sequences (Nam et al. 2002), especially a problem for genomes such as Caenorhabditis elegans, which possess relatively A-rich nongenic sequence (Jan et al. 2011). Secondly, alternative poly(A) polymerases have been identified in most eukaryotes (Trf4 and Trf5 in Saccharomyces cerevisiae) that cooperate with the nuclear RNA degradation apparatus to promote degradation of unwanted or misprocessed transcripts (Schmid and Jensen 2008). RNA degradation-associated oligo(A) tailing (West et al. 2006; Slomovic et al. 2010) can similarly generate oligo(dT)-primed cDNA that is unrelated to authentic mRNA 3′-terminal poly(A) tails. Finally, reverse transcriptase can, in some cases, “misbehave” by template switching from the original RNA template to the newly synthesized cDNA (Houseley and Tollervey 2010). Consequently, antisense transcription, often a widely attributed feature of transcriptomes, may be less common than many studies have predicted. Such template switching can also result in the apparent fusion of separate RNA templates and may be miscategorized as trans splicing.

Recently, various genomic approaches using massive parallel DNA sequencing technology have been applied that avoid some or all of the above pitfalls. For example, an analysis of mRNA 3′ ends in C. elegans used a clever trick of ligating a biotin-tagged RNA:DNA duplex “splint” onto the 3′ end of poly(A) tails (Jan et al. 2011). The DNA component has a 3′-terminal oligo(dT) overhanging sequence that guides the complementary biotinylated RNA up against the poly(A) 3′ end. Following RNA ligation, partial RNase digestion, biotin selection, and reverse transcription with dTTP and RNase H digestion to remove the mRNA poly(A), authentic mRNA 3′-end fragments lacking their poly(A) tail were then amplified and sequenced. Using this clever “splint” 3′-end sequencing procedure, many new examples of alternate poly(A) sites were defined for C. elegans mRNA. Also, as predicted, a lot of mRNA 3′ ends were shown to be misassigned, presumably derived from priming of oligo(dT) on internal A-rich sequences prevalent in the worm genome. Another approach applied to both yeast and mammalian mRNA is to sequence the nucleotides adjacent to the poly(A) tail by direct sequencing using an approach not dissimilar to my original analysis of the mRNA PAS. Essentially, a bacterial DNA polymerase is used to reverse-transcribe the mRNA poly(A) tail using dTTP. Then, fluorescent, chain terminator-modified (VT) nucleotides (their chemistry is proprietary and therefore hard to fully comprehend) are added one by one, complementary to the mRNA sequence adjacent to the oligo(dT) copy of the poly(A) tail. This process is performed on a matrix so that millions of single mRNA 3′-end sequences can be simultaneously read for up to ∼50 nt from the poly(A) tail of the mRNA population (Ozsolak et al. 2009). By these massive sequencing procedures, the original sequence consensus for the mammalian PAS has been shown to be truly general (Ozsolak et al. 2010). Where deviations from the AAUAAA sequence occur, these can be ascribed to a weaker PAS that may have a particular regulatory purpose. Also VT nucleotide-derived massive sequencing identified new sequence motifs associated with polyadenylated ncRNA, especially a U₉ sequence motif just upstream of the polyadenylated 3′ end. It remains to be established by mutagenesis whether this U₉ element is actually required for RNA 3′-end formation of particular ncRNA classes.

mRNA 3′-end processing: connections with transcriptional termination

A milestone in our understanding of mRNA 3′-end processing (cleavage and polyadenylation) came from biochemical characterization of the process in mammals, pioneered by the laboratories of J.L. Manley and W. Keller. Fractionation of HeLa cell nuclear extract resulted in the purification of two protein complexes: cleavage and polyadenylation specificity factor (CPSF) and cleavage stimulatory factor (CstF). These proteins together recognize AAUAAA and GU-rich DSEs on synthetic RNA substrates and act to promote RNA cleavage between these cis elements. Furthermore, poly(A) polymerase was shown to be recruited to this 3′ processing complex, resulting in polyadenylation of the 3′ end of the 5′ cleavage product. Hence, mRNA 3′-end processing was successfully reconstituted in a biochemically tractable in vitro system. Other factors were also shown to be required for in vitro 3′-end processing, such as cleavage factors I and II (CFI/II). For detailed accounts and a literature survey on this 3′-end processing mechanism, see many comprehensive reviews (Colgan and Manley 1997; Zhao et al. 1999; Edmonds 2002; Mandel et al. 2008; Millevoi and Vagner 2010). Slightly later, budding yeast biochemical characterization of mRNA 3′-end processing was achieved. This was greatly aided by genetic screens for 3′-end processing that allowed the identification of many 3′-end processing factors, some homologous to mammalian factors and others apparently unique to yeast. Whole-cell extracts isolated from yeast strains lacking particular factors (through growth of temperature-sensitive mutants at restrictive temperature) proved valuable in pinning down particular 3′-end processing functions. Adding back recombinant factors lacking particular protein domains further uncovered the surprising complexity of this process (Zhao et al. 1999; Mandel et al. 2008). Why upward of 50 polypeptides divided between multiple subcomplexes are required to simply cleave the pre-mRNA 3′ end and then couple poly(A) addition remains an enigma to this day. However, the pivotal role of 3′-end polyadenylation in gene expression is well emphasized by the fact that most 3′-end processing factors are encoded by essential genes. Mutation of any such gene is invariably lethal to the organism.

In more recent years, biochemists and structural biologists have got their teeth into the molecular characterization of many of the components of the mRNA 3′-end processing reaction, and details of these studies are reviewed elsewhere (e.g., Mandel et al. 2008). However, a key feature of this reaction is the actual endoribonuclease activity itself, and I outline here the experiments that have led to our current state of understanding of this process. It was originally presumed that the 3′ ends of mRNA are generated by direct transcriptional termination by RNA polymerase II (Pol II) rather than an RNA endonucleolytic cleavage reaction. Such a simple mechanism is well known to occur for prokaryotic polycistronic mRNA and also eukaryotic RNA polymerase III (Pol III) (Richard and Manley 2009). However, the fact that in vitro synthesized RNA, spanning a PAS, could be demonstrably cleaved and polyadenylated in vitro, which was first shown with mammalian nuclear extracts (Moore and Sharp 1985) and subsequently with yeast whole-cell extracts (Butler and Platt 1988), proved that mRNA 3′-end formation and termination were separate, albeit connected, molecular events. Critically, the cleavage and polyadenylation steps could be biochemically separated. For instance, blocking poly(A) polymerase activity by use of ATP inhibitors such as cordycepin (3′ deoxy ATP) or depletion of essential Mg²⁺ by EDTA treatment allows the visualization of the 3′ cleaved product without poly(A) addition. Similarly, S. cerevisiae temperature-sensitive mutants in PAP1 [encoding poly(A) polymerase] still generate 3′-end-cleaved RNAs (Butler and Platt 1988; Zhao et al. 1999).

A natural example of cleavage-only mRNA 3′-end formation is found with replication-dependent histone mRNAs (Gick et al. 1986). These mRNAs are formed by the recognition (through direct base-pairing) of a purine-rich sequence downstream from the conserved 3′-terminal hairpin by the small nuclear RNA (snRNA) U7 (Schaufele et al. 1986; Schumperli 1988). This RNA, in association with multiple protein components (forming U7snRNP), is recruited to histone mRNA 3′ ends by interaction with additional factors, including a zinc finger protein and hairpin-binding protein (Marzluff et al. 2008). Additionally, specific components of the cleavage/poly(A) complex are also recruited: CPSF-73, CPSF-100, and a scaffold-like protein, Symplekin (Kolev and Steitz 2005; Sullivan et al. 2009). Importantly, CPSF-73 has a particular metallo β lactamase domain (MLD) with the hallmarks of endoribonuclease activity (Dominski 2007). Both for PAS and histone 3′-end processing, CPSF-73 was shown to cross-link to the exact nucleotide that defines the mRNA 3′ end (Ryan et al. 2004; Dominski et al. 2005), and, at least for histone mRNA, CPSF-73 mediates direct and specific RNA 3′-end cleavage (Kolev et al. 2008). Curiously, CPSF-100 also possesses a clear MLD, although with inactivating amino acid replacements in its active site. From these comparative studies between histone and poly(A)⁺ mRNA 3′-end formation, it is clear that CPSF-73, aided and abetted by other factors generally specific to the two mRNA classes, mediates the enigmatic 3′ processing reaction to form the authentic mRNA 3′ ends, as shown in Figure 3 (Mandel et al. 2006, 2008).

Figure 3.

Diagram comparing the mechanism of cotranscriptional 3′-end formation of mammalian polyadenylated mRNA versus unpolyadenylated histone mRNA (RNA is in red, and DNA is in black). For poly(A)⁺ mRNA, only factors conserved with histone pre-mRNA processing are shown. The rest of the poly(A) complex is depicted by a large grey shadow. Pol II (in light blue) is depicted with striped CTD, and the position of the ssDNA bubble is depicted with associated nascent RNA. Histone 3′ processing factors are indicated. The cap at the 5′ end of mRNA is depicted as a red ball. The histone mRNA 3′ hairpin is shown, as is the U7 RNA hairpin and the interaction with the histone downstream element (HDE). Positions of cis RNA 3′-end processing sequence elements are shown in white boxes. mRNA products of poly(A)⁺ and histone pre-mRNA 3′-end processing are shown (in boxes), as is the subsequent coupled termination of Pol II by Xrn2-mediated torpedo effects [established for poly(A)⁺ genes, but only inferred for histone genes].

The above-defined RNA processing reaction effectively separates mRNA 3′-end processing from actual Pol II transcriptional termination. Even so, it is long established that 3′-end processing is absolutely required for termination, raising interesting mechanistic questions about how these two processes are connected. Early on when studying the mutated PAS of the α2-globin gene, which is associated with α-thalassaemia, we showed that not only is the 3′ processing mechanism disrupted, but so too is termination (Whitelaw and Proudfoot 1986). In this mutant α2-globin allele, Pol II reads past the normal termination site (∼600 nt into the 3′ flanking region) and actually perturbs the activity of the downstream α1-globin gene by a transcriptional interference process, thus reducing its expression. Soon afterward, the mouse β-globin and SV40 early PAS were also shown to dictate Pol II termination (Logan et al. 1987; Connelly and Manley 1988). These results provided the first evidence that pre-mRNA processing is coupled to transcription. Such coupling has subsequently been shown to be a general feature of all pre-mRNA processing reactions (especially pre-mRNA capping and splicing) (Proudfoot et al. 2002).

The next stage in these experiments was to reveal that the unique heptad repeated C-terminal domain (CTD) of the Pol II large subunit Rpb1 acts to directly recruit cleavage/poly(A) factors to the elongating Pol II complex (McCracken et al. 1997; Zhao et al. 1999). Indeed, in vitro 3′ processing reactions revealed that Pol II CTD significantly enhances this reaction, leading to the view that Pol II CTD itself acts as a component of the 3′ processing machinery (Hirose and Manley 1998). Some cleavage/poly(A) factors, as well as mediating 3′-end processing, also aid subsequent Pol II termination (Birse et al. 1998). A clear example is Pcf11, a component of CFII in mammals and CF1A in yeast. This factor has a CTD-interacting domain (CID) with a preference for CTD Ser 2 phosphorylation, a feature of elongating Pol II (Licatalosi et al. 2002; Meinhart and Cramer 2004). Interestingly, a further 5′–3′ exonucleolytic RNA processing reaction is facilitated that acts to degrade the uncapped residual RNA still attached to elongating polymerase. In yeast, this exonuclease, called Rat1, is part of a complex (comprising Rai1 and Rtt103) that also possesses a CID, presumably explaining how this final RNA processing reaction is also transcriptionally coupled (Kim et al. 2004). In mammals, the Rat1 homolog Xrn2 similarly acts to degrade Pol II-associated RNA (West et al. 2004). In both systems, it is thought that the exonuclease is in kinetic competition with elongating Pol II so that when/if RNA degradation catches up with the elongating Pol II, this will promote termination by inducing conformational changes in the Pol II active site through loss of its associated nascent RNA (Connelly and Manley 1988; Proudfoot 1989; Kuehner et al. 2011). This mechanism is called the torpedo model, and likely acts in consort with Pol II-recruited cleavage/poly(A) factors to promote efficient termination at a distinct 3′ flanking region location, downstream from the gene, poly(A) site (Fig. 3; Richard and Manley 2009; Kuehner et al. 2011).

Two additional classes of termination elements can act to enhance the termination process. One class acts as a Pol II transcription pause site and, by slowing down elongation, allows the exonuclease more time to degrade the nascent RNA and so reach the Pol II complex (Plant et al. 2005; Gromak et al. 2006). These elements may be G-rich in sequence, and recent results show that such G-rich elements are associated with the formation of RNA:DNA hybrids. Indeed, resolution of these hybrids requires the action of dedicated helicases: Sen1 in yeast, and senataxin in mammals (Mischo et al. 2011; Skourti-Stathaki et al. 2011). The other class of terminator elements may occur at more distal 3′ flanking regions and is referred to as a cotranscriptional cleavage (CoTC) sequence (Dye and Proudfoot 2001). CoTC termination may operate when PAS-proximal pause sites are lacking, so that Pol II effectively escapes further into the 3′ flanking region. As the name implies, CoTC RNA sequences are highly unstable, so that as soon as they emanate from the Pol II RNA exit channel, RNA cleavage occurs (West et al. 2008). This will allow Xrn2 to directly degrade the nascent transcript at this rapidly formed uncapped RNA 5′ end. Since this occurs close to the still elongating Pol II, termination quickly ensues. Indeed, in this type of termination, Pol II is released from the chromatin template before cleavage at the PAS occurs. So, in this instance, 3′-end processing occurs post-transcriptionally, but still in association with Pol II. Interestingly, CoTC-mediated termination can greatly increase the yield of mRNA from a gene, as 3′-end processing can occur in the nucleoplasm away from the nuclear RNA degradation apparatus, which appears to be chromatin-associated (West and Proudfoot 2009). As discussed below, defining the actual mRNA 3′ end is a complex interplay between poly(A) site recognition and associated Pol II termination (Fig. 3).

Alternative PAS (APA) define different mRNA 3′ UTRs

The characterization of particular eukaryotic mRNAs has often relied on the long-established Northern blotting technique, which provides a gel fractionation image of a specific gene's mRNA output, giving information relating to both mRNA size and quantity. This technique—unlike its modern replacement of quantitative (real-time) PCR-amplified cDNA (qRT–PCR)—allows visualization of the complete set of mRNA isoforms generated from a particular gene. Smaller genes often yield only one specific mRNA, which is the norm for simpler eukaryotes such as yeast. However, in higher eukaryotes, especially mammals, most of these larger and more complex genes (with multiple exons) generate multiple mRNA isoforms. These are frequently caused by complex alternative splicing patterns. Alternative splicing is now appreciated to regulate both the nature and complexity of mammalian proteomes and, as such, reflects a key aspect of the regulation of gene expression (Black 2003; Johnson et al. 2003; Wang et al. 2008; Chen and Manley 2009). However, a significant part of mRNA size variation derives not from alternative splicing, but rather from alternative PAS selection. Thus, it is calculated that well over half of all mRNAs have variable PAS selection, meaning that they will possess mRNA isoforms differing by the extent of their 3′ UTRs (Edwalds-Gilbert et al. 1997; Tian et al. 2005). Since mRNA 3′-end processing occurs cotranscriptionally and is stimulated by Pol II CTD (Proudfoot 2004), it is clear that once a particular PAS has been selected and mRNA 3′ cleavage occurs with consequent release from chromatin-associated Pol II, then further cleavage of more proximal PAS on the mRNA will not occur. Thus, mRNAs with extended 3′ UTRs are carried through into the cytoplasm, where particular 3′ UTR sequences act to regulate both the stability and translatability of mRNA as described below (for a recent review on APA, see Lutz and Moreira 2011).

The occurrence of alternative PAS selection for mRNA was initially considered to reflect a relatively random process. The failure of one PAS to fully end the mRNA resulted in further downstream cryptic PAS acting to end the rest of the transcripts. Thus, the insertion of multiple, identical PAS at the 3′ end of artificial gene constructs results in all of the PAS working to some degree, yielding mRNA with different-length 3′ UTRs (Denome and Cole 1988). In general, when the PAS is relatively strong (possessing both a canonical AAUAAA and clearly defined USE and/or DSE), then the first PAS dominates, with increasingly reduced usage of downstream PAS. However, with PAS lacking full consensus signals, a more even spread of PAS recognition is evident. However, it is generally the case that the earlier PAS in a series of multiple PAS is used more efficiently, implying a first-come, first-served pattern. Presumably, CTD-bound poly(A) factors will be sequestered onto earlier PAS, excluding the later usage of these factors on subsequent PAS. Another arrangement of PAS tested artificially was to have a weak PAS followed by a strong one. In this situation, the downstream PAS is selectively used. This type of duplicated PAS arrangement allowed the identification of transcription pause sites. When such elements are placed between a weak followed by a strong PAS, higher usage of the upstream weak PAS is promoted. This assay is referred to as a PAS competition assay (Ashfield et al. 1991). With the realization that endonucleolytic cleavage at a poly(A) site allows entry of the exonuclease torpedo (Rat1 or Xrn2) to promote Pol II termination (Kim et al. 2004; West et al. 2004), it is plausible that alternative PAS usage will result in equivalent alternative Pol II termination.

Bioinformatic analysis of PAS usage in higher eukaryotes has revealed the remarkable fact that well over 50% of genes display APA site usage (Tian et al. 2005). In general, where APA is clearly evident, then the downstream PAS appear to have sequence features that more closely match the canonical AAUAAA and DSE sequence elements (Legendre and Gautheret 2003). This sequence specificity is predicted from the above considerations that distal PAS will only be used if they can effectively outcompete proximal PAS. However, the relative usage of tandem PAS is invariably measured by levels of steady-state mRNA possessing particular lengths of 3′ UTRs. Consequently, the relative stability of these different mRNA isoforms will also affect the apparent usage of PAS (Moore 2005).

Several independent studies have revealed the remarkable fact that either levels of cell proliferation or developmental stage can cause a shift in APA from more distal to more proximal PAS selection (Sandberg et al. 2008; Ji et al. 2009; Mayr and Bartel 2009). In particular, rapidly dividing cells, as are often found in cancerous tissues, tend to use proximal PAS, while cells in later developmental stages tend to use more distal PAS. These striking results clearly correlate with the regulation of mRNA stability and translation by microRNAs. Thus, mRNAs possessing longer 3′ UTRs caused by distal PAS selection will have more potential microRNA-binding sites. This view is clearly confirmed by bioinformatics analyses. Even so, recent more comprehensive genomic analysis indicates that the relationship between 3′ UTR length and proliferation stage may be more complex than originally thought (Fu et al. 2011).

The molecular basis of how APA may be differentially regulated remains largely unknown. In general, cleavage/poly(A) factors appear to be constitutively expressed, and little evidence exists for the selective use of factors for one PAS versus another. However, it has been observed that some cleavage/poly(A) factors may be present at lower, limiting levels in some cells, such as CstF-64 in pre-B cells (Takagaki and Manley 1998; Ji et al. 2009). In this situation, stronger PAS will have a significant kinetic advantage (see below). In the case of alternative splicing, regulation of this RNA processing mechanism is often achieved by the enhancement of weak splice sites by so-called splicing enhancers (often in adjacent exons) that are recognized by splicing regulatory factors (usually SR proteins) (Black 2003). It seems plausible that a weaker PAS—particularly proximal PAS—may similarly be enhanced by PAS enhancers, even though evidence for such elements and their associated factors is still lacking.

One mechanism that may regulate APA is through more direct effects of Pol II transcription. Gene promoters may play a role in recruiting factors that subsequently enhance PAS recognition. Thus, CPSF has been shown to associate with the general transcription factor TFIID at gene promoters. Subsequent transfer of CPSF to the CTD may promote 3′-end processing (Dantonel et al. 1997). Communication between the promoter and terminator through gene loop formation (O'Sullivan et al. 2004; Perkins et al. 2008) may also afford efficient transfer of cleavage/poly(A) factors from the 3′ end of the gene back to new Pol II initiation complexes (Glover-Cutter et al. 2008; Mapendano et al. 2010). Also, very recently, specific transcription activators have been shown to enhance 3′-end processing through recruitment of a specific elongation factor complex (PAF1c) that, in turn, enhances PAS recognition (Nagaike et al. 2011). Whether these promoter effects on 3′-end processing factor recruitment play a role in APA remains to be established.

Gene promoters can also determine transcription elongation rates by setting up more or less processive Pol II elongation complexes (Cramer et al. 1999). Alternatively, as mentioned above, in the context of Pol II termination, it seems plausible that specific pause sites positioned within genes may be more or less active in a different cellular context. These could cause localized changes in elongation rate that might favor the use of upstream splice sites or PAS (Roberts et al. 1998; Gromak et al. 2006). In the case of alternative splicing, it is now well known that Pol II processivity set up by specific promoters, by modification of Pol II activity (such as UV-induced hyperphosphorylation of CTD), or by use of an artificial mutant (so-called slow Pol II) can influence alternative splicing patterns (de la Mata et al. 2003; Munoz et al. 2009). Similarly, loss of elongation factors in yeast has been shown to correlate with increased usage of upstream, cryptic PAS present within genes (Cui and Denis 2003). Very recently, similar mechanisms have been shown to exist for Drosophila APA. In the case of the gene Polo, with its two well-defined PAS, a weak proximal and stronger distal PAS are regulated such that the distal PAS is required for higher levels of Polo gene expression. Interestingly, flies expressing a slow Pol II mutant show a clear shift to usage of the proximal Polo PAS as well as several other tested examples of fly APA (Pinto et al. 2011).

Much remains to be learned about APA in eukaryotes. However, it is clear that this is a key regulatory process in eukaryotic gene expression. The ability of APA to generate mRNA with different 3′ UTRs that contain different regulatory cis elements represents a potentially major form of gene regulation. Such regulatory elements may act as targets for microRNAs that regulate mRNA stability or translation (Bartel 2009). Alternatively, they may act as mRNA stability or instability elements recognized by RNA-binding factors such as HuR or TPP. Finally, 3′ UTRs have been shown to contain complex RNA signals (with particular RNA secondary structures) for factors that mediate specific cytoplasmic localization during early development (St Johnston 2005; Lutz and Moreira 2011).

Interplay between poly(A) site selection and splicing

Pre-mRNA processing acts in a highly coordinated manner during transcription. Consequently, mRNA 3′-end processing is closely coordinated with splicing. In particular, the relatively low sequence complexity and redundancy of the PAS argues that inappropriate, premature poly(A) site selection must be prevented. Otherwise, incorrectly shortened mRNAs would form that could translate into truncated proteins with dominant-negative effects on cellular function.

One such example of these interconnections relates to the phenomenon of terminal exon definition. It has been shown that recognition of the 3′ splice site (3′SS) of a gene's last intron strongly enhances the efficiency of the downstream PAS (Niwa and Berget 1991; Niwa et al. 1992; Dye and Proudfoot 1999). In particular, the 3′SS-associated factor U2AF was shown to enhance PAS function by direct molecular contacts with poly(A) polymerase (Vagner et al. 2000). Similarly, protein components of U2snRNP that associate with the 3′SS and nearby lariat branch point help enhance downstream 3′-end processing through interactions with CPSF (Kyburz et al. 2006). This enhancement of terminal exon definition has several other consequences, as PAS recognition also enhances terminal intron splicing, and, furthermore, 3′SS recognition is required for Pol II termination, as this in turn depends on PAS recognition. In eukaryotic genes that lack introns, PAS recognition may require additional as-yet-undetermined selection mechanisms. Interestingly, for mammalian intronless genes, different PAS elements may operate, containing extra DSE (Dalziel et al. 2007; Nunes et al. 2010).

Another situation in which splicing and 3′-end processing must interact is in the few characterized cases in which APA results in the formation of mRNAs with different terminal coding sequences. For the immunoglobulin antibody heavy chain gene, alternative membrane-bound or secreted protein isoforms differ by the presence or absence of a specific hydrophobic C terminus required for membrane retention. An intron-located PAS, if used, must outcompete with the upstream 5′SS, which results in the shorter secreted antibody form found in mature B cells. Alternatively, in pre-B cells, splicing wins so that the intronic PAS is suppressed, and instead a downstream PAS is used with additional terminal exons included in the heavy chain mRNA. This process is regulated by levels of CStF-64, which are more limited in pre-B cells than in mature B cells (Takagaki et al. 1996; Takagaki and Manley 1998). Also, Pol II elongation factors may act to modulate this PAS switch (Martincic et al. 2009). A similar type of APA regulation exists for the calcitonin gene. Here, again, the selective recognition of an intronic PAS results in the formation of a truncated mRNA encoding calcitonin gene-related peptide (CGRP) rather than full-sized calcitonin (Amara et al. 1982; Zhao et al. 1999). Interestingly, genome-wide analysis of this type of APA suggests that the regulated formation of mRNAs encoding different C-terminal protein sequences may be quite widespread (Tian et al. 2005; Wang et al. 2008).

Finally, it has become apparent that PAS commonly present within the body of genes, especially in the much longer introns of higher eukaryotes, must be tightly repressed. Interestingly, this is achieved by the dominant role of the 5′SS and its recognition by U1snRNP, which act to block adjacent PAS recognition. Again, this phenomenon was first described in gene-specific studies (Levitt et al. 1989; Zhao et al. 1999) but has now achieved genome-wide status (Kaida et al. 2010). Early examples of such regulation were found in mammalian viruses, which often need to maximize their gene expression output by selective use of PAS. In bovine papilloma virus, viral gene expression programs require that a particular PAS is only used in late stages of viral infection (Furth et al. 1994). Interestingly, this late PAS was shown to be blocked from use in earlier stages of infection by the presence of a closely positioned, upstream 5′SS. Recognition of this 5′SS by U1snRNP blocks poly(A) polymerase activity at the late PAS by direct interaction with the 70K protein component of U1snRNP (Gunderson et al. 1998). This ability of closely placed 5′SS to repress PAS function has been cleverly manipulated experimentally to allow the specific repression of any gene. Thus, U1snRNA with its 5′ end (that normally binds 5′SS) modified to base-pair with sequence upstream of a particular PAS is expressed in mammalian cells resulting in inhibition of the target gene by blocking PAS function (Fortes et al. 2003). HIV-1 provirus also uses PAS regulation to maximize its gene expression. Like other retroviruses, its duplicated 5′ and 3′ long terminal repeats (LTR) possess identical PAS. While 3′ LTR PAS usage is essential for viral gene expression, use of the 5′ LTR PAS would preclude viral gene expression. Interestingly, cleavage at the 5′ LTR PAS was shown to be repressed by a major viral 5′SS positioned close by, ∼200 base pairs (bp) into the proviral gene sequence. Point mutation of this 5′SS activates the 5′ LTR PAS, causing the provirus to exclusively produce very short 5′ LTR-specific transcripts (Ashe et al. 1995, 1997).

These viral examples of PAS regulation by 5′SS repression have now been generalized to the whole genome by a recent study that tested the effect of blocking U1snRNA by antisense RNA analogs (morpholino) that were electroporated into mammalian cells (Kaida et al. 2010). In addition to blocking the splicing of many gene introns and causing the accumulation of unspliced mRNAs, truncated mRNAs were widely observed caused by the activation of intronic PAS. In other words, these studies imply a second major role of U1snRNA beyond its critical function in 5′SS recognition as a prelude to splicing. This is in the blockage of intronic PAS as first described for viruses. Indeed, a further variation of the mechanism of intronic PAS regulation is found in C. elegans, where cleavage of intronic PAS triggers the process of trans splicing by the small RNA leader sequence SL2 (Blumenthal 2005; Haenni et al. 2009). Overall, it is abundantly clear that PAS recognition is a widely regulated process that dictates multiple regulatory features of eukaryotic genes (Fig. 4).

Figure 4.

Diagram summarizing the selective use of PAS along a Pol II transcribed gene. PAS in introns are recognized by the poly(A) complex, but this is modulated or blocked by U1snRNP binding to adjacent 5′SS. Recognition of PAS at the end of genes is enhanced by splicing factors bound to the terminal intron 3′SS. Different 3′-terminal PAS are competitively and mutually exclusively used, resulting in APA.

Conclusions

This review aims to put current advances in our understanding of how the 3′ ends of mRNAs are selected into historical perspective. This field has accumulated massive detail since the genome-wide era began at the beginning of the second millennium. However, it is striking that many of the mechanistic principles behind PAS selection were already in place long before this time. Indeed, these earlier studies provided actual experimental evidence for the role of specific sequences by measuring the effect of their mutation on biological function. It is abundantly clear that bioinformatic analysis of genomic data has provided invaluable generality to our understanding of PAS function in gene expression. However, current genome-wide analyses often only provide bioinformatic correlations and lack direct functional experimentation. Genomic analysis will only achieve its full potential when bioinformatics can be matched by hypothesis-driven experimental approaches. In spite of the above concerns, it is the case that where and how to end the eukaryotic message is a central regulatory point in the elaborate process of gene expression.