The end of the message: multiple protein-RNA interactions define the mRNA polyadenylation site.

The key RNA sequence elements and protein factors necessary for 3' processing of polyadenylated mRNA precursors are well known. Recent studies, however, have significantly reshaped current models for the protein-RNA interactions involved in poly(A) site recognition, painting a picture more complex than previously envisioned and also providing new insights into regulation of this important step in gene expression. Here we review the recent advances in this area and provide a perspective for future studies.

Almost all eukaryotic mRNAs and primary microRNA transcripts and many long intergenic noncoding RNAs are polyadenylated at their 3′ ends. This involves a well-studied two-step reaction consisting of an endonucleolytic cleavage followed by the synthesis of the poly(A) tail (Colgan and Manley 1997; Zhao et al. 1999; Chan et al. 2011; Proudfoot 2011). Significantly, recent global analyses have revealed that ∼70% of eukaryotic genes produce multiple mRNA isoforms with distinct 3′ ends through the process of alternative polyadenylation (APA) (Di Giammartino et al. 2011; Shi 2012; Elkon et al. 2013; Tian and Manley 2013). APA isoforms from the same gene may encode different proteins and/or contain different 3′ untranslated regions (UTRs). As longer 3′ UTRs often harbor more binding sites for microRNAs and/or RNA-binding proteins than shorter 3′ UTRs, APA isoforms may acquire different stability, translation efficiency, and/or intracellular localization. APA is dynamically regulated in development and in response to environmental stimuli (Flavell et al. 2008; Sandberg et al. 2008; Ji et al. 2009; Shepard et al. 2011; Graber et al. 2013), and deregulation of APA has been associated with a number of human diseases (Mayr and Bartel 2009; Jenal et al. 2012; Masamha et al. 2014). Therefore, it is critical to understand how mRNA 3′ processing sites, often referred to as the poly(A) site (PAS), are recognized and how PAS selection is regulated.

Mammalian PASs are generally AU-rich and have a highly conserved nucleotide composition profile (Shi 2012; Tian and Graber 2012). A number of key cis elements have been identified in the mammalian PAS, including the AAUAAA (or its close variant AUUAAA) located 10∼30 nucleotides (nt) upstream of the cleavage site, which is frequently defined by a CA dinucleotide immediately 5′ to the site of endonucleolytic cleavage; a U/GU-rich downstream element (DSE) located within ∼40 nt downstream from the cleavage sites; and U-rich upstream auxiliary elements (USEs) and upstream sequences conforming to the consensus UGUA. PAS sequences are divergent, and many PASs lack one or more of these cis elements. For example, ∼30% of human PASs lack an A(A/U)UAAA hexamer (Beaudoing et al. 2000), and ∼20% of human PASs do not possess a U- or GU-rich DSE (Zarudnaya et al. 2003). Unlike splicing, which relies on RNA base-pairing for splice site recognition (Wahl et al. 2009), PAS recognition is accomplished solely by protein–RNA interactions (Colgan and Manley 1997; Chan et al. 2011). The mRNA 3′ processing factors CPSF (cleavage and polyadenylation specificity factor) and CstF synergistically bind to the AAUAAA hexamer and the DSE, respectively, while the CFI complex binds to the UGUA motifs (Hu et al. 2005). CPSF, CstF, and CFI directly bind to RNAs to form a core complex and in turn recruit other factors, including CFII, the scaffolding protein Symplekin, and the poly(A) polymerase (PAP), to assemble the active mRNA 3′ processing complex (Shi et al. 2009; Chan et al. 2011). The assembly of these factors on PASs occurs cotranscriptionally and is facilitated by the C-terminal domain of the RNA polymerase II (RNAP II) large subunit (Hirose and Manley 2000; Proudfoot et al. 2002; Bentley 2005).

A detailed understanding of these key interactions is critical for delineating the mechanisms for PAS recognition. Additionally, an increasing number of regulatory factors have been shown to modulate PAS selection. There have been a number of recent reviews on APA and its physiological impact (Di Giammartino et al. 2011; Shi 2012; Elkon et al. 2013; Mueller et al. 2013; Tian and Manley 2013). Here we focus on the protein–RNA interactions that are responsible for PAS recognition and its regulation. An interesting and useful analogy to consider is the similarities between the core PAS and the core RNAP II promoter, specifically how multiple conserved sequence motifs are organized and recognized by multisubunit protein complexes, which in both cases function to recruit otherwise sequence-nonspecific polymerases.

CPSF and AAUAAA recognition

CPSF is a multisubunit protein complex consisting of CPSF160, Wdr33, CPSF100, CPSF73 (the cleavage endonuclease), Fip1, and CPSF30. CPSF specifically recognizes the most important cis element in the mammalian PAS, the AAUAAA hexamer (Fig. 1A; Bienroth et al. 1991; Murthy and Manley 1992; Shi et al. 2009). CPSF160 has long been implicated in directly binding to AAUAAA based on several lines of evidence. First, two CPSF subunits of ∼160 and 30 kDa, presumed to be CPSF160 and CPSF30, could be UV cross-linked specifically to AAUAAA-containing RNA oligos (Keller et al. 1991). Second, Yhh1/Cft1, the yeast homolog of CPSF160, binds to PAS-containing RNA near the cleavage site via an evolutionarily conserved central domain (Dichtl et al. 2002). Third, recombinant CPSF160 possesses RNA-binding activity and shows preference for AAUAAA-containing sequences (Murthy and Manley 1995). However, the affinity and specificity of the CPSF160–RNA interaction is significantly weaker than that of the CPSF complex (Murthy and Manley 1995), and more recent global mapping of CPSF160–RNA interactions failed to detect specific enrichment of its binding signals at or near the AAUAAA hexamers (Martin et al. 2012). Therefore, >20 years after the isolation of CPSF, the identities of its subunits that directly bind to AAUAAA remained to be determined.

Figure 1.

(A) Schematic models showing distinct modules within CPSF. CPSF73, CPSF100, and Symplekin form a module (which we refer to as mCF) that contains the endonuclease activity. mCF may cooperate with different RNA-binding modules, such as mammalian polyadenylation specificity factor (mPSF) (Wdr33, CPSF30, Fip1, and CPSF160), for cleavage/polyadenylation of most mRNAs or with SLBP and U7 snRNP for the cleavage of histone mRNAs. (B, bottom panel) Domain structures of the CPSF subunits involved or implicated in RNA binding. The known or putative RNA-binding regions are marked with red solid or dotted underlines.

Recent studies have provided unexpected new insight into AAUAAA recognition. Using intact CPSF complexes immunopurified from mammalian cell extracts or a CPSF subcomplex reconstituted with recombinant subunits, two recent studies provided direct evidence that Wdr33 and CPSF30 specifically recognize the AAUAAA hexamer in vitro (Chan et al. 2014; Schonemann et al. 2014). In vivo mapping of protein–RNA interactions using photoactivatable ribonucleoside-enhanced cross-linking and immunoprecipitation (PAR-CLIP) detected specific binding of both proteins at the AAUAAA regions of active PASs at the global level (Chan et al. 2014; Schonemann et al. 2014). Based on these data, a new model for CPSF–RNA interactions was proposed in which Wdr33 and CPSF30 simultaneously and synergistically recognize AAUAAA (Fig. 1A). The RNA-binding activity of CPSF30, which consists of five C3H zinc finger repeats and a putative RNA-binding zinc knuckle motif at the C terminus (Barabino et al. 1997), is mediated primarily by its zinc fingers 2 and 3 (Fig. 1B; Chan et al. 2014). Wdr33, which is a large protein similar in size to CPSF160, interacts with RNA at least in part via its N-terminal region, which includes a highly conserved WD40 repeat domain (Fig. 1B; Schonemann et al. 2014). Both zinc finger and WD40 repeat domains are known to mediate sequence-specific RNA interactions (Hudson et al. 2004; Lau et al. 2009). It remains to be determined, however, whether Wdr33, CPSF30, or both are required to confer sequence specificity to CPSF–RNA interactions. A combination of biochemical and structural approaches will be needed to delineate the specific contribution of Wdr33 and CPSF30 to CPSF–RNA interactions as well as determine the precise role of CPSF160 in PAS recognition. It is noteworthy that both CPSF30 (Nemeroff et al. 1998; Twu et al. 2006) and Wdr33 (Brass et al. 2009) have been implicated in host cell defense against influenza virus infection, likely reflecting a pathway for viral inhibition of host mRNA production.

In addition to Wdr33 and CPSF30, and perhaps CPSF160, Fip1 also contributes CPSF–RNA interactions (Fig. 1A). Recombinant Fip1 binds to U-rich RNAs via its arginine-rich C-terminal region (Kaufmann et al. 2004). Within the CPSF complex, Fip1 binds to RNA sequences upstream of the AAUAAA hexamer in vitro (Chan et al. 2014). Consistent with this, in vivo mapping of Fip1–RNA interactions revealed that Fip1 binds to U-rich sequences in the AAUAAA hexamer region (Fig. 1B; Martin et al. 2012; Lackford et al. 2014). Fip1 was recently identified as a regulator of APA in embryonic stem cells (ESCs), and Fip1-mediated APA regulation is required for ESC self-renewal, further suggesting that Fip1–RNA interactions play an important role in PAS selection (Lackford et al. 2014). Finally, Fip1 and CPSF160 function directly in recruiting PAP to the mRNA 3′ processing site (Murthy and Manley 1995; Kaufmann et al. 2004).

What, then, is the function of CPSF160 in PAS recognition? Most surprisingly, CPSF160–RNA interactions were not detected in the two recent studies (Chan et al. 2014; Schonemann et al. 2014). Previous evidence for CPSF160–RNA interactions were mostly based on experiments using purified recombinant CPSF160 or its yeast homolog, Cft1 (Murthy and Manley 1995; Dichtl et al. 2002). In contrast, both recent studies used CPSF complexes (Chan et al. 2014; Schonemann et al. 2014). So it is possible that the RNA-binding activity of CPSF160 is inhibited within the CPSF complex. However, given the conserved nature of the CPSF160 RNA-binding activity, a more likely possibility is that CPSF160 participates in RNA interactions at a specific stage during mRNA 3′ processing or, intriguingly, at a specific subset of PASs.

Similarly complex protein–RNA interactions have been observed in yeast PAS recognition. The yeast mRNA 3′ processing factor CPF contains Mpe1 (Rbbp6 homolog) (see below), Pap1, Pta1 (Symplekin), and the homologs of all six subunits of the mammalian CPSF complex, including Cft1 (CPSF160), Cft2 (CPSF100), Ysh1/Brr5 (CPSF73), Fip1 (Fip1), Yth1 (CPSF30), and Pfs2 (Wdr33) (Zhao et al. 1999). Similar to CPSF, multiple CPF subunits have been shown to bind RNA, including Cft1, Cft2, Fip1, Yth1, and Mpe1 (Barabino et al. 1997; Zhao et al. 1997; Dichtl et al. 2002; Lee and Moore 2014).

We envision several scenarios to explain why such complex protein–RNA interactions are necessary. First, given the generally lower specificity of RNA–protein interactions compared with DNA–protein interactions or RNA base-pairing, combinatorial interactions may be necessary to achieve the RNA-binding specificity necessary for accurate PAS recognition. Second, due to the dynamic nature of mRNA 3′ processing (Chan et al. 2011), it is possible that different factors participate in PAS binding at different stages of 3′ processing. For example, Yth1 binds to RNA and Fip1 in a mutually exclusive manner, and it was proposed that Yth1 binds to PAS RNA during the cleavage step but releases RNA to associate with Fip1 during the poly(A) synthesis step (Tacahashi et al. 2003). To determine whether CPSF160 participates in RNA binding at specific stages of mRNA 3′ processing, it will be critical to characterize better the dynamics of the cleavage/polyadenylation machinery. Third, an attractive idea is that there may be distinct CPSF subcomplexes or alternative complexes that bind to PAS sequences through distinct protein–RNA interactions. As mentioned earlier, mammalian PASs are highly diverse and may require multiple alternative factors for their recognition.

An intriguing similarity exists between CPSF and the general RNAP II transcription factor TFIID (Cler et al. 2009). Both are multisubunit complexes that not only recognize key promoter/PAS sequences, which are strikingly similar (TATAAA vs. AAUAAA), but also contain subunits that bind additional promoter/PAS elements. TFIID exists as tissue-specific and cell type-specific subcomplexes and can also contain distinct isoforms of different subunits (Muller et al. 2010). As we shall see, a similar picture of CPSF is beginning to emerge.

Several lines of evidence point to the existence of heterogeneous CPSF complexes. For example, Schonemann et al. (2014) showed that a relatively stable complex, called mammalian polyadenylation specificity factor (mPSF), can be reconstituted with Wdr33, CPSF30, Fip1, and CPSF160 and is active, with PAP, in AAUAAA-dependent poly(A) synthesis (Fig. 1A). A distinct subcomplex consisting of CPSF100, Symplekin, and CPSF73 functions in 3′ processing of nonpolyadenylated histone mRNA precursors (Sullivan et al. 2009). This factor, which we refer to as mCF, can either partner with SLBP, the RNA-binding protein that functions in recognition of the histone mRNA 3′ processing site, to carry out histone mRNA cleavage (Marzluff et al. 2008) or, we suggest, associate with mPSF to form CPSF (Fig. 1A). This modular design seems to be evolutionarily conserved, as mPSF shares most of its components with the yeast PFI, and mCF shares most of its components with CFII (Zhao et al. 1999). Another more speculative idea is that distinct CPSF complexes exist and may function at different PASs. For example, CPSF160 and Wdr33 are similar in size and domain structure, and perhaps both function in different CPSF complexes. Although there is currently no evidence in support of this, it is noteworthy that early size estimates of native CPSF were in the 200- to 300-kDa range (Christofori and Keller 1988; Takagaki et al. 1989), too small to include both of these ∼160-kDa proteins. Finally, another possibility is that distinct CPSF complexes can contain different subunit isoforms. For example, two splice isoforms of CPSF30 exist and can be found in CPSF complexes (Chan et al. 2014).

CstF and the DSE

CstF is a trimeric complex consisting of CstF77, CstF50, and CstF64 or its paralog, CstF64τ (Takagaki et al. 1990; Gilmartin and Nevins 1991; Wallace et al. 1999). CstF is believed to form a homodimer, as both CstF77 and CstF50 self-associate (Fig. 2A; Takagaki and Manley 2000; Bai et al. 2007; Moreno-Morcillo et al. 2011). CstF specifically recognizes the DSEs, which are generally characterized as U/GU-rich, and this interaction is mediated by CstF64 or CstF64τ (MacDonald et al. 1994; Takagaki and Manley 1997). It was originally suggested that CstF64τ is a testis-specific factor that has more distinct RNA-binding specificity than CstF64 and that CstF64τ may play a role in mediating testis-specific PAS selection (Wallace et al. 1999; Monarez et al. 2007). However, recent studies provided compelling evidence that CstF64τ, similar to CstF64, is widely expressed in mammalian tissues (Yao et al. 2013). Furthermore, in vitro and in vivo analyses demonstrated that CstF64 and CstF64τ have nearly identical RNA-binding specificities (Takagaki and Manley 1997; Yao et al. 2013). Finally, individual knockdowns of CstF64 or CstF64τ had little effect on global PAS selection, but double knockdowns led to significant APA changes (Yao et al. 2012). Together, these recent studies strongly suggest that CstF64 and CstF64τ play largely redundant roles in mediating CstF–RNA interactions. Another CstF64 isoform, called βCstF64, was detected predominantly in the CNS (Shankarling et al. 2009), but its role in PAS recognition has yet to be characterized.

Figure 2.

(A) CstF-dependent and potentially CstF-independent PAS recognition. In the CstF-independent model, CstF is shown in a dotted boundary to indicate that CstF may or may not be present in this complex. The scaffolding protein Symplekin plays a role in bridging CPSF–CstF complexes. (B) CFI–RNA interactions. (Top panel) CFI binds to UGUA motifs often found upstream of the AAUAAA hexamer. (Bottom panel) The CFI dimer may bind to two UGUA sequences at different PASs, thus looping out the proximal PASs. (C) The role of Rbbp6 in PAS recognition. The question mark denotes the fact that Rbbp6 may or may not directly interact with upstream AU-rich elements (AREs) or may rely on an unknown factor.

The DSEs of mammalian PASs are highly variable, and thus an important question is how CstF64/τ recognize such diverse sequences. While properties of the CstF64 RNA recognition motif (RRM) allow it to recognize diverse GU-rich sequences (Perez Canadillas and Varani 2003), recent in vivo and in vitro analyses of CstF64/τ–RNA interactions provided evidence that these “general” 3′ processing factors only bind to a subset of PASs, which are characterized by GU-rich sequences in the first 20 nt downstream from the cleavage site followed by U-rich sequences (Fig. 2A; Yao et al. 2012, 2013). In contrast, PASs that are not stably bound by CstF64/τ contain more G-rich DSEs (Fig. 2A). These results raised the possibility that there may be distinct types of PASs that require different combinations of factors for their recognition. This conclusion is consistent with earlier studies. For example, it has been shown that CstF and mRNA 3′ processing activity is transiently inhibited following DNA damage due to an interaction between CstF50 and the BARD1–BRCA1 complex (Kleiman and Manley 2001). In the meantime, DNA damage response genes must be expressed, which would require mRNA 3′ processing activity (Gomes et al. 2006). Providing an answer to this paradox, a recent study revealed that the PAS of p53, a DNA damage-induced gene, contains a G-rich DSE. A unique G-quadruplex structure is formed within this DSE region and is recognized by hnRNP F/H to allow 3′ end formation of p53 mRNAs after DNA damage (Decorsiere et al. 2011). Together, these studies provide evidence for CstF-independent PAS recognition, but additional work is needed to delineate such mechanisms fully.

CFI–upstream UGUA interactions

CFI binds to UGUA motifs, which are typically located upstream of the AAUAAA hexamer (Fig. 2B; Brown and Gilmartin 2003; Hu et al. 2005). CFI consists of CFI25 and one of two related subunits, CFI59 or CFI68 (Ruegsegger et al. 1996), and appears to exist as a tetramer of two CFI25 subunits and two large subunits (Yang et al. 2011). All three CFI subunits participate in RNA binding, as they can be UV cross-linked to RNA (Ruegsegger et al. 1996). The small subunit, CFI25, contains a Nudix domain, which, in its canonical form, possesses pyrophosphohydrolase activities. The CFI25 Nudix domain, however, lacks such enzymatic activity due to substitutions at key residues and instead is involved in specific interactions with the UGUA motif (Yang et al. 2010b). Similar to CstF subunits, CFI25 forms a dimer and thus can bind and recognize two UGUA sequences simultaneously (Fig. 2B, bottom panel), a property that has been proposed to regulate PAS selection (Yang et al. 2010a,b). Extending these results, PAR-CLIP analyses detected robust CFI binding to UGUA motifs at the global level (Martin et al. 2012). Both CFI59 and CFI68 contain an N-terminal RRM, a proline-rich central region, and a C-terminal region rich in arginine–serine (RS) repeats (Ruegsegger et al. 1998). CFI59/68 share domain structures similar to SR proteins, which are well-known splicing regulators (Tacke and Manley 1996). Like SR proteins, CFI subunits have been detected in purified spliceosomes (Rappsilber et al. 2002; Zhou et al. 2002), implicating CFI in splicing itself or in mediating cross-talk between splicing and 3′ processing. In fact, CFI has been shown to interact with U2AF65, which is involved in the early steps of splicing, and this interaction helps to stimulate mRNA 3′ processing (Millevoi et al. 2002, 2006).

CFI plays an important role in PAS recognition. Initial studies established that, like CstF, it functions early in the process, functioning to help stabilize CPSF binding to the PAS (Ruegsegger et al. 1996, 1998). This again suggests an analogy with core RNAP II promoter recognition, in which two general factors, TFIIA and TFIIB, function in part by stabilizing TFIID binding to promoter DNA (Thomas and Chiang 2006). Given the importance of CFI in not only PAS recognition but also APA (Gruber et al. 2012; Martin et al. 2012; Masamha et al. 2014), it will be important to delineate the detailed mechanisms by which CFI–RNA interactions contribute to PAS recognition.

Rbbp6 and PAS recognition

Rbbp6 is a newly identified 3′ processing factor that also functions in part by contacting RNA. Rbbp6 was originally identified in a screen for Rb-binding proteins (Saijo et al. 1995; Sakai et al. 1995), hence the name, and was later identified as a p53-binding protein as well (Simons et al. 1997). Although Rbbp6 was suggested to be the putative mammalian homolog of the yeast 3′ processing factor Mpe1 based on limited sequence similarity (Vo et al. 2001), it was not functionally implicated in mRNA 3′ processing until a proteomic analysis detected Rbbp6 in the purified mRNA 3′ processing complex (Shi et al. 2009), and direct evidence for its involvement in 3′ processing was provided only recently (Di Giammartino et al. 2014). Rbbp6 was found to associate with CstF via its evolutionarily conserved N-terminal regions, which contain a DWNN (domain with no name)/ubiquitin-like domain, a zinc knuckle, and a RING finger. This N-terminal region, which is the region of similarity with its yeast homolog, Mpe1, is sufficient for 3′ processing activity in vitro. In both proteins, the zinc knuckle and RING domains function in RNA binding, although in neither case does binding appear sequence-specific (Di Giammartino et al. 2014; Lee and Moore 2014). Interestingly, knockdown of Rbbp6 in mammalian cells preferentially inhibits the 3′ processing of mRNAs that contain AU-rich elements (AREs) in their 3′ UTRs (Di Giammartino et al. 2014). It remains unknown, however, whether Rbbp6 directly interacts with AREs or whether additional factors are involved in determining this apparent specificity (Fig. 2C).

Rbbp6 has several other intriguing properties. For example, Mpe1 has been implicated in regulating ubiquitination of PAP and potentially other factors (Lee and Moore 2014). Rbbp6 has also been shown to possess E3 ligase activity and function as an activator of Mdm2-mediated ubiquitination of p53 (Li et al. 2007; Miotto et al. 2014). Finally, a splice isoform of Rbbp6, called iso3 and consisting solely of the DWNN, competes with the full-length protein to modulate cleavage efficiency (Di Giammartino et al. 2014). As with a number of 3′ processing factors, levels of Rbbp6 are elevated in many types of cancers (Chen et al. 2013). Coupled with the fact that iso3 levels are reduced (Mbita et al. 2012), this likely contributes to the changes in processing efficiency that modulate APA in cancer cells (Mayr and Bartel 2009). It will be important to determine how all of these properties of Rbbp6 function in modulating 3′ processing and in tumorigenesis.

Regulation of PAS recognition

A growing number of factors have been shown to modulate PAS recognition by the core 3′ processing machinery. In many cases, the effect of these regulatory factors, most of which were initially described as functioning in pre-mRNA splicing, depends on the positions of their binding sites relative to a PAS (Fig. 3). For example, U1 snRNP suppresses polyadenylation at downstream PAS by U1A-70K-mediated and U1-70K-mediated inhibition of PAP activity (Gunderson et al. 1997, 1998). More recent global analyses revealed that U1 snRNP plays a critical role in protecting pre-mRNAs from premature cleavage and polyadenylation (Kaida et al. 2010). Additionally, U1 snRNP levels relative to the cellular demand has been proposed to control mRNA length and isoform expression by modulating PAS selection (Berg et al. 2012). On the other hand, U2 snRNP promotes the usage of downstream PASs through multiple protein–protein interactions (Fig. 3; Kyburz et al. 2006; Millevoi et al. 2006). In addition to splicing factors, the m7G cap at the 5′ ends of mRNAs can stimulate mRNA 3′ processing (Cooke and Alwine 1996; Flaherty et al. 1997), possibly through interactions between the cap-binding complex and 3′ processing factors (Yang et al. 2011).

Figure 3.

Context-dependent regulation of PAS recognition. Regulatory factors bound at different locations relative to the core PAS sequence have different effects on PAS recognition by the mRNA 3′ processing factors. Positive effects are indicated by an arrow, and negative effects are indicated by a vertical line.

Many RNA-binding proteins, also initially described as splicing regulators, have been implicated in control of PAS recognition (Fig. 3). Some of these factors directly compete with core 3′ processing factors to block PAS recognition (Fig. 3). For example, PTBP1, TDP-43, sex-lethal (Sxl), and ELAV/HuR have all been shown to directly compete with CstF for binding to DSEs and in turn inhibit PAS usage (Castelo-Branco et al. 2004; Gawande et al. 2006; Zhu et al. 2007; Avendano-Vazquez et al. 2012). PABPN1 was suggested to compete with CPSF for binding to the AAUAAA hexamer region, especially at many proximal PASs (Jenal et al. 2012).

Notably, the effects of some regulatory factors on PAS recognition are position-dependent. For example, NOVA proteins, which are known splicing regulators in the brain (Jensen et al. 2000), were shown to regulate APA by binding to sites in 3′ UTRs and repress or enhance PAS usage in a position-dependent manner (Licatalosi et al. 2008). Binding near the PASs was found to be repressive, likely by interfering with core factor binding, while binding to more distant sites tended to result in activation of PAS utilization by currently unknown mechanisms. Likewise, when muscleblind (Mbnl) proteins bind to core PAS sequences, especially between the cleavage sites and the DSEs, they tend to inhibit PAS recognition, presumably through steric hindrance (Batra et al. 2014). In contrast, when they bind to sequences upstream of the core PASs, Mbnl proteins stimulate PAS usage. CPEB1 has been suggested to promote PAS recognition when it binds to either upstream or downstream sequences (Bava et al. 2013). Similar position-specific effects of RNA-binding proteins on splice site recognition have also been reported (Wang et al. 2012; Erkelenz et al. 2013; Fu and Ares 2014). It remains an important question for future studies as to how RNA-binding proteins bound to neighboring sequences regulate the interactions between the core mRNA 3′ processing factors and the PASs. Furthermore, recent studies have provided evidence that the effect of some RNA-binding proteins, such as ELAV, on PAS selection is linked to the promoter and RNAP II pausing (Oktaba et al. 2015).

Conclusions and future perspectives

Recent studies suggest that the protein–RNA interaction network involved in PAS recognition is more complex than previously thought, which raises many important questions for future studies. For example, which factors—Wdr33, CPSF30, or both—provide specificity for the CPSF–AAUAAA interaction? Does CPSF160 participate in PAS recognition? If so, when and where? How does RNA binding by other CPSF subunits contribute to PAS definition? How do alternative protein isoforms and distinct core complexes contribute to PAS recognition? How do regulatory proteins modulate PAS usage in a position-dependent manner? Solving these important problems will undoubtedly bring our understanding of mRNA 3′ processing and its regulation to a new height.