Most protein coding genes in eukaryotic cells are transcribed by RNA polymerase II (RNAPII). With the exception of the canonical, replication-dependent transcripts that encode histones in metazoans, the maturation of mRNA 3′ end involves endonucleolytic cleavage of nascent RNA followed by synthesis of a poly(A) tail at the 3′ terminus of the cleaved product by poly(A) polymerase (PAP) (Fig. 1A). Multiple cis-acting elements, including upstream UGUA elements, core AAUAAA motif, and downstream GU-rich sequences, collectively define the functional poly(A) site (PAS), which is recognized by the poly(A) machinery (Fig. 1B) to carry out the coupled cleavage and polyadenylation (CPA) reaction (Elkon et al., 2013; Gruber and Zavolan, 2019; Tian and Manley, 2017).

The poly(A) machinery has been biochemically elucidated (Shi and Manley, 2015; Shi et al., 2009). Briefly, besides PAP, the poly(A) machinery contains 4 subcomplexes: Cleavage factor I (CFI; in mammalian cells, CFIm) binds to the upstream UGUA element. Cleavage and polyadenylation specificity factor (CPSF) recognizes the core AAUAAA motif and CPSF also tightly couples with cleavage factor II (CFII) consisting of CLP1 and PCF11. Finally, cleavage stimulation factor (CSTF) binds the downstream GU-rich sequences. Within the CPSF complex, CPSF73 is the endonuclease, and the cleavage reaction is aided by CLP1 to cut RNA ~21 nt downstream of the AAUAAA motif followed by PAP-catalyzed poly(A) addition. The length of the poly(A) tail is controlled by the nuclear poly(A) binding protein PABPN1 (Kuhn et al., 2009; Wahle, 1995).

Multiple lines of evidence demonstrate that the CPA reaction takes place co-transcriptionally (Mitschka and Mayr, 2022; Tian and Manley, 2017). Most importantly, CPA is tied to transcriptional termination, because RNAPII does not have a defined transcription termination site. According to the popular torpedo model, the CPA reaction generates a free 5′ end to be attached by the exonuclease XRN2 (Fig. 1A), which tailgates the elongating RNAPII to cause eventual drop-off (Connelly and Manley, 1988; Proudfoot, 2004). The alternative allosteric model suggests that CPA coupled with certain cis-acting elements may cause transient RNAPII pausing near the PAS, which may induce conformational changes in the elongating RNAPII complex to become increasingly prone to drop-off (Proudfoot, 2016; Zhang et al., 2015). More recent studies suggest combined flavors of both models during the coupling between CPA and transcriptional termination and defects in this process cause global transcriptional readthrough (Eaton et al., 2020). Mechanistically, the poly(A) machinery is physically associated with RNAPII, which may even be loaded onto the elongating RNAPII complex at the beginning of transcription (Glover-Cutter et al., 2008).

An array of technologies has been developed to enable the global profiling of PASs (Zhou et al., 2014). Interestingly, over 70% of genes have multiple PASs in both Drosophila (Liu et al., 2017) and humans (Derti et al., 2012). Therefore, besides alternative promoters to generate mRNA isoforms with different 5′ ends and alternative splicing to produce mRNA isoforms with different sequences in gene bodies, alternative polyadenylation (APA) represents another key mechanism to produce mRNA isoforms with different 3′ ends from single genes. Growing evidence also suggests that many mRNA isoforms generated by APA may have distinct biological functions, which are subjected to regulation in development, differentiation, and disease (Gruber and Zavolan, 2019; Mitschka and Mayr, 2022; Tian and Manley, 2017). As APA that occur in an intronic region tends to truncate the C-terminus of the encoded protein, this type of APA is thus referred to as IPA (intronic polyadenylation) or CDS-APA (coding sequence-APA). A more frequent mode of APA is tandem APA with two or more PASs in the last exon, thereby generating mRNA isoforms with the same protein-coding capacity but with different lengths of 3′ untranslated region (3′UTR).

Characterization of cis-acting elements at proximal and distal PASs suggests a general rule underlying APA. Although the AAUAAA motif is the core signal for the CPA reaction, it is often insufficient to define a functional PAS in mammalian genomes. Both the upstream UGUA and downstream GU-rich sequences are critical, which together define the strength of a given PAS in a combinatorial fashion (Tian and Manley, 2017). In general, the proximal PAS is weaker compared to the downstream distal PAS (Gruber and Zavolan, 2019). Thus, the distal PAS may be considered a constitutive PAS while the upstream PAS a regulated PAS in APA, the latter of which may be specifically activated or repressed by a variety of mechanisms (see below). Conceptually, because the CPA reaction is mechanistically and functionally coupled with transcriptional termination, it has been generally thought that APA regulation also takes place co-transcriptionally. However, we wish to emphasize here that while there are multiple lines of evidence to support various co-transcriptional mechanisms, most studies published to date simply assume that is the case, especially those regulated by specific RNA binding proteins (listed in Mitschka and Mayr [2022]). However, in light with the recent finding of sequential APA (Tang et al., 2022), many data may be alternatively interpreted (see examples below), suggesting the need to revisit various regulatory paradigms previously thought to occur at the co-transcriptional levels.

Initial reporter-based studies show that specific transcription blockage between the proximal and distal PASs is sufficient to induce APA to favor the use of the proximal PAS (Gromak et al., 2006; Yonaha and Proudfoot, 1999). Such transcription blockage may be instituted with G-rich sequences (Beaudoin and Perreault, 2013), which play a key role in initiating R-loop formation (Skourti-Stathaki et al., 2011). This mechanism may account for the frequent pausing of RNAPII at PASs (Glover-Cutter et al., 2008), especially when coupled with a mutant RNAPII with reduced elongation rate (Pinto et al., 2011).

It has been widely accepted that transcription is intimately coupled with various mRNA processing steps (Glover-Cutter et al., 2008) and such coupling may be initiated even at the beginning of transcription, as indicated by gene promoters as the hotspots for the ChIP-seq signals of many RNA binding proteins involved in diverse RNA processing pathways (Xiao et al., 2019), including those involved in CPA. However, the specific coupling mechanism(s) has remained unclear because different transcription factors, including the C-terminal domain (CTD) of the largest subunit of RNAPII (McCracken et al., 1997), TFIID (transcription factor II D) (Dantonel et al., 1997), and the elongation factor PAF1c (Nagaike et al., 2011), etc. have been implicated in the recruitment of specific components of the poly(A) machinery. Such recruitment may be further enhanced by transcription enhancers, via a variety of transcription factors and co-activators (Kwon et al., 2022). These processes may also be modulated by various epigenetic mechanisms (Kaczmarek Michaels et al., 2020; Lin et al., 2020; Nanavaty et al., 2020; Soles and Shi, 2021; Wood et al., 2008).

Despite clearly recordable effects of transcription factors on APA, a major challenge in understanding the mechanism(s) is to determine how the proximal or distal PAS is differentially impacted to account for the observed switch in APA. Two recent studies shed critical light on this: One study characterized a pair of transcriptional anti-terminators SCAF4 and SCAF8 (Gregersen et al., 2019). The homologs of these SR protein-related RNA binding proteins have been genetically implicated in transcription termination in yeast (Yuryev et al., 1996). Importantly, depletion of these proteins in mammalian cells was found to enhance transcription termination at the proximal PAS of many genes, thereby favoring the use of the weak proximal PAS. Interestingly, this process appears to be coupled with the interaction of these RNA binding proteins with specific RNA elements in nascent RNAs, thus providing a mechanism for context-dependent effects. However, how exactly such RNA binding activity contributes to the anti-termination effect remains to be worked out. It also remains to be determined how broadly this mechanism may apply to the impacted APA events. In another study, inactivation of DNA methyltransferases was found to enhance CTCF binding, which in turn impacts APA (Nanavaty et al., 2020). While this effect on APA may enlist multiple direct or indirect mechanisms, on specific gene models examined, enhanced CTCF binding was found to induce RNAPII pausing between the proximal and distal PAS, thus favoring the use of the proximal PAS. Together, these studies illustrate the “first come, first served” model for the relatively weak proximal PAS to gain the advantage over the strong downstream PAS in recognition by the poly(A) machinery (Fig. 2A).

Besides the co-transcriptional regulation of APA, several mechanisms have been suggested to represent the post-transcriptional mechanisms for APA regulation, largely through differential degradation of mRNA isoforms with different 3′UTRs (Mitschka and Mayr, 2022; Tian and Manley, 2017). For example, mRNA isoforms with different 3′ ends may be differentially exported out of the nucleus, localized in different cytoplasmic compartments, or recruited to the translation machinery. As a result, different mRNA isoforms may be selectively targeted by microRNAs or other nucleases due to isoform-specific cellular compartmentalization mechanisms. These mechanisms may contribute to the global APA regulation during tumorigenesis, as cancer cells are in general associated with 3′UTR shortening (Mayr and Bartel, 2009; Xia et al., 2014) and the opposite appears to be the case with cell differentiation (Cheng et al., 2020; Ji and Tian, 2009). Here, we would like to argue that these modes of regulation cannot be considered post-transcriptional APA because alternative polyadenylation is supposed to take place during the CPA reaction to select one PAS over the other. Selective degradation of one mRNA isoform relative to the other can thus only be regarded as “apparent” switch in APA.

It has been documented that mRNAs with longer 3′UTR tend to accumulate in the nucleus compared to their isoforms with shorter 3′UTR (Djebali et al., 2012), likely due to various nuclear retention mechanisms granted by additional cis-elements in those longer mRNA isoforms, such as the hairpins for microRNA processing (Neve et al., 2016). Interestingly, a recent study further pursued this phenomenon, providing multiple lines of evidence that those nuclear detained long APA isoforms ended in the distal PASs can in many cases serve as the precursors to generate short APA isoforms through CPA at the proximal PASs (Tang et al., 2022). Because distal PASs are normally stronger than proximal PASs, the distal ones would be preferentially used unless the proximal ones employ certain gene-specific mechanism(s) to enhance their selection. Thus, according to the “first come, first served” model, the selection of the proximal PAS would be independent of the selection of the distal PAS. This proves not to be the case. By engineering a ribozyme between the proximal and distal PAS, the authors demonstrated that the selection of the proximal PAS for CPA depended on the presence of the distally polyadenylated sequence. Furthermore, a novel Cleave-seq was developed to provide direct evidence that the longer APA isoform is the intermediate of the shorter APA isoform. In this assay, the exonuclease XRN2 was first inactivated, thus protecting the 5′ end of the downstream RNA fragment after the cleavage reaction at the proximal PAS. By oligo(dT) selection and then ligating an unphosphorylated DNA linker with 3′-OH followed by deep sequencing, this captured the cleaved products.

This study suggests a general model for sequential polyadenylation as a key mechanism for post-transcriptional APA (Fig. 2B). In this model, the CPA reaction at the distal PAS is first carried out co-transcriptionally by the poly(A) machinery, which is coupled with transcriptional termination. The resulting products may either be exported out of the nucleus or transiently detained in the nucleus. This transient period depends on the combined strength of nuclear retention elements, such as pseudo-splice sites in individual transcripts. These detained RNAs thus serve as precursors for CPA at the proximal PAS. Interestingly, this mechanism may also allow an additional slow splicing reaction(s) to occur, which may also contribute to the nuclear retention of the precursor mRNA (Tang et al., 2022). It is conceivable that this mechanism enables various RNA binding proteins to influence APA at either co- or post-transcriptional levels. Thus, future studies need to revisit this widely presumed assertion instead of assuming that regulated APA by RNA binding proteins is always a co-transcriptional event. This resembles the mechanism for regulated alternative splicing, which predominately takes place post-transcriptionally, and as proposed earlier (Han et al., 2011), this mechanism may even also allow co-transcriptional commitment but post-transcriptional catalytic reaction.

CFI-25 (also known as NUTD21) and CFI-68 (also known as CPSF6) are key components of the CFI complex. Because the CF1 complex is responsible for recognizing the UGUA element, which is more prevalently associated with distal PASs, knockdown of these factors has been shown to induce a global switch of APA to proximal PASs (Li et al., 2015). It has remained elusive how such switch occurred. Unaltered proximal PASs and compromised distal PASs would contribute to “apparent” APA switch. Alternatively, many proximal PASs may become reactivated. How would this occur, especially with those PASs without any sequences that resemble the UGUA motif? According to the sequential CPA model (Fig. 3A), impaired CPA at distal PASs in CFI knockdown cells would cause transcriptional readthrough as documented earlier (Zhu et al., 2018). Such aberrant transcripts (red lines in Fig. 3A) are more likely detained in the nucleus, which may then serve as the precursors for CPA at proximal sites. Future studies are needed to directly test this model in cells depleted of CFI-25 or CFI-68.

A recent study revealed a key role of NXF1, a critical adaptor for mRNA nuclear export, in the regulation of APA (Chen et al., 2019). NXF1 depletion switched APA to proximal PASs, which may result from its functional interaction with CFI-68. However, CFI-68 knockdown does not always phenocopy the effect of NXF1 knockdown, suggesting additional mechanisms in operation for both factors. NXF1 depletion-induced APA switch was proposed as a co-transcriptional event based on the ability of NXF1 to co-IP with RNAPII, and more importantly, on increased RNAPII pausing at PASs in response to NXF1 depletion. Unfortunately, it has not been determined whether NXF1 depletion selectively induced RNAPII pausing at the proximal PASs, as with SCAF4/SCAF8 (Gregersen et al., 2019) and CTCF (Nanavaty et al., 2020). Alternatively, NXF1 deficiency is known to induce R-loop formation, which is tied to RNAPII pausing (Niehrs and Luke, 2020; Santos-Pereira and Aguilera, 2015). This may account for increased RNAPII near PASs, as observed in NXF1 knockdown cells (Chen et al., 2019). Impaired NXF1 could detain distal isoforms in the nucleus, thus providing precursors for sequential CPA at proximal PASs (Fig. 3B). Future work will test this alternative interpretation on NXF1 deficiency-induced APA switch.

One may argue that many components of different types of RNA processing machinery, including the poly(A) machinery, are associated with RNAPII, which has been considered as key evidence for co-transcriptional RNA processing. In these processes, the most prevailing model is the loading of various RNA processing factors to RNAPII CTD. However, to date, no studies directly tested the functional requirement of RNAPII CTD for the observed interactions in cells. Given the ability of RNAPII CTD to form condensates and nascent RNAs likely play key roles in such condensate formation to create transcription hubs where transcription and RNA processing are functionally integrated (Boehning et al., 2018; Guo et al., 2019). In this regard, the CTD of RNAPII may play a general role as a key organizer of such gene expression hubs, rather than directly involved in binding and loading specific RNA processing factors and such hubs may not only enable co-transcriptional RNA processing but also provide tight links between transcription and post-transcriptional RNA processing events. Thus, co-IP with RNAPII does not necessarily reflect co-transcriptional RNA processing.

APA is a key strategy to regulate gene expression by creating mRNA isoforms from single genes that have the potential to be selectively regulated at the levels of RNA stability, export, location, and translation. Because of tight link of the CPA reaction to transcription termination, it has been generally thought that most regulated APA events also take place in a co-transcriptional fashion. However, a recent study suggests that the CPA reaction at the distal PAS is co-transcriptional, but the CPA reaction at the alternative proximal PAS(s) can be either co- or post-transcriptional. If post-transcriptional, the proposed sequential polyadenylation model provides a new framework to understand specific regulatory mechanisms. It is our hope that this review will serve as a catalyst to further enrich this new conceptual framework.

Y.H., T.C., C.L., and X.Z. were supported by NIH grants HG004659 and DK098808.

Y.H. and X.-D.F. wrote the manuscript. T.C., C.L., and X.Z. gave valuable inputs for improving manuscript modification.

The authors have no potential conflicts of interest to disclose.