Alternative polyadenylation regulation: insights from sequential polyadenylation

ABSTRACT

The processing of the proximal and distal poly(A) sites in alternative polyadenylation (APA) has long been thought to independently occur on pre-mRNAs during transcription. However, a recent study by our groups demonstrated that the proximal sites for many genes could be activated sequentially following the distal ones, suggesting a multi-cleavage-same-transcript mode beyond the canonical one-cleavage-per-transcript view. Here, we review the established mechanisms for APA regulation and then discuss the additional insights into APA regulation from the perspective of sequential polyadenylation, resulting in a unified leverage model for understanding the mechanisms of regulated APA.

Almost all eukaryotic pre-mRNAs undergo cleavage and polyadenylation (3´ processing) when RNA polymerase II (Pol II) reaches their 3´end, which ensures not only the translation efficiency and stability of mRNA decorated with poly(A) tails [1] but also efficient transcription termination [2]. With the advent of high-throughput sequencing, it has become increasingly clear that the majority of mammalian genes contain multiple poly(A) sites (PASs) [3,4], and it is a general strategy to fine-tune gene expression through the differential choice of PAS in the process named alternative polyadenylation (APA) [5].

Pre-mRNA 3´ processing is orchestrated through the interaction between cleavage and polyadenylation complex (CPA complex) components and their cognate RNA elements. In general, the CPA complex can be classified into four subcomplexes: cleavage and polyadenylation specificity factor (CPSF), cleavage stimulation factor (CstF), and mammalian cleavage factors I and II (CFIm and CFIIm) [5–7]. The CPSF subcomplex contains CPSF160, WDR33, CPSF30, Fip1, CPSF100, CPSF73, and symplekin; while WDR33 and CPSF30 directly contact the polyadenylation signal AAUAAA (or its variants) upstream of the PAS [8,9], Fip1 can bind to U-rich elements flanking the AAUAAA motif [5,10]. The CstF, consisting of CstF77, CstF64, and CstF50, can bind to the U/GU-rich elements downstream of the poly(A) sites via CstF64 [11]. The CFIm, composed of CFIm25, CFIm59, and CFIm68, is responsible for recognizing the UGUA motif upstream of the polyadenylation signal [5,12]. Recent results from the 3´-end processing reconstitution assays in vitro elucidated that CPSF, CstF, CFIIm subcomplex (PCF11 and CLP1), and protein RBBP6 are essential, while CFIm is dispensable for 3´-end processing [6,7].

APA regulation through CPA factors

Given the critical role of these core polyadenylation factors in 3´-end formation [5–7], it is not surprising to see that alterations in their concentration or activity constitute a general mechanism for APA regulation. CstF64 upregulation during B cells activation, would lead to the increased usage of intronic proximal PAS (pPAS) in the IgM heavy chain pre-mRNA and hence the switch from expressing membrane-bound IgM to secreted IgM [13]. In line with this, co-depletion of CstF64 with its paralog CstF64τ in HeLa cells will globally favor the distal PAS (dPAS) utilization [11]. Analogous to CstF64, the depletion or mutation of other essential CPA factors such as Fip1[10], PCF11 [14–16], CLP1 [17,18], and RBBP6 [19], generally caused pronounced shifts to use dPASs in generating lengthened transcripts (Figure 1a). In contrast, knocking down CFIm, a strong bias of using pPAS was observed [12,20]. Additionally, it was found that the depletion of splicing factors U1-70 K or SF3b155 (the component of U1 snRNP and U2 snRNP, respectively) would lead to generally more usage of intronic poly(A) sites, suggesting that splicing activity may mainly suppress intronic polyadenylation [14].

Figure 1.

Sequential polyadenylation and a leverage model for APA regulation. (a) Altered APA patterns upon loss of different CPA factors. The sizes of the circles represent the relative numbers of shortened (shifting to proximal PAS, red) and lengthened (shifting to distal PAS, blue) APA events upon losses of six individual factors, summarized from the literature. (b) The canonical and sequential polyadenylation views of APA regulation for genes with two or more PASs. In canonical view, one transcript is cleaved and polyadenylated once in a probabilistic manner dependent on the competition of the proximal PAS (pPAS) and distal PAS (dPAS) (left). In the sequential polyadenylation mode, the same transcript could undergo multiple cleavage-polyadenylation events from distal to proximal PASs (right). (c-e) A leverage model for APA regulation. The choice of PASs is dependent on the strengths of the PAS embedded in its cis-elements (threshold) and the relative usage depends on the competition between PASs, analogous to the leverage of a lever via its pivot (c). The loss of core CPA factor(s) would decrease the relative strengths of all PASs, analogous to moving down the pivot, and affect pPAS more severely due to a lowered pPAS strength below the threshold, resulting in a shift to the dPAS (d). Any factor that causes increased opportunity for recognizing pPAS by CPA factors in the nucleus would lead to a shift toward the pPAS, analogous to moving the pivot to the right (e). The reported factors include Pol II pausing, slowing Pol II elongation, increasing the distance between pPAS and dPAS, and CFIm depletion, which would extend the exposure time of pPAS to CPA factors leading to a higher probability to be used. Even the transcripts already processed at dPAS could be cleaved and polyadenylated again at pPAS via sequential polyadenylation in generating the shortened isoforms.

In mammals, Pol II traverses 1–4.5 kb/min and it takes seconds to several minutes for intron excision [21]. In contrast, early surveys on Ad2 RNA [22] and endogenous genes in CHO and HeLa cells [23] demonstrate the kinetics of 3´ processing is ultrafast, usually occurring no more than one minute after the poly(A) site is transcribed. In addition, a large body of evidence shows that transcription and 3´ processing are physically and functionally coupled in a reciprocal manner. On the one hand, the carboxy-terminal domain of the Pol II large subunit (Pol II CTD) can directly interact with the CPA factors [24,25], which is required for efficient 3´ processing both in vivo [24] and in vitro [25]. It is also elucidated that general transcription factor TFIID [26] and other transcription coactivators [27–30] are capable of recruiting the CPA factors and facilitating their association with Pol II preceding the entry of Pol II into productive transcription elongation, thereby enhancing 3´ processing. On the other hand, 3´ processing is essential for transcription termination; either poly(A) sites mutation [31–33] or CPA factors depletion [15,34,35] will result in the impairment of transcription termination. Together, these pieces of evidence point to the fact that 3´-end processing should occur co-transcriptionally, either at the pPASs or the dPASs (Figure 1b, left).

In the spirit of the facts above, there is a lack of direct evidence supporting the processing of pPAS and dPAS are co-transcriptional. A recent in vitro 3´ cleavage reconstitution study demonstrated that 3´-end cleavage could proceed without the Pol II CTD, and the addition of Pol II CTD into the reconstituted cleavage system only slightly impacts the reaction [7], which implies that 3´ processing may continue when RNA is detached from Pol II and DNA.

Sequential polyadenylation model for APA

In our recent study [36], we fractioned HeLa cells into the cytoplasm (CY), nucleoplasm (NP), chromatin, and nuclear matrix (NM). We found almost no RNA in the chromatin; the so-called chromatin-associated RNA in the previous study is actually tightly associated with NM. We further sequenced the polyadenylated RNA in the CY, NP, and NM, and the results showed that the long 3´ UTR isoforms for many genes are uniquely retained in the NM. In contrast, their short 3´ UTR isoforms are released into NP and exported into the CY. In our subsequent attempt to identify the putative RNA binding proteins (RBPs), which are responsible for the NM retention of the long 3´ UTR isoforms, we unexpectedly found that many of these identified RBPs are the components of the 3´-end processing machinery [37], including the core CPA factors Fip1, CstF64 and CFIm68 that can directly contact with the poly(A) sites [5,10,11]. These results inspired us to hypothesize that for many genes, the dPAS is firstly processed to generate the long 3´ UTR isoform, which can further serve as the substrates to produce the short 3´ UTR isoform through further cleavage and polyadenylation at the pPAS, named as sequential polyadenylation (Figure 1b, right).

To test this hypothesis, we firstly compared the relative usage of pPAS and dPAS between the newly synthesized RNA and steady-state RNA and found that dPAS is usually more used in the newly synthesized RNA. Furthermore, we validated that the degradation of the long 3´ UTR isoform by ribozyme insertion would synchronously lead to the reduction of the short 3´ UTR isoform. Meanwhile, we developed the Cleave-seq assay, in which purified polyadenylated nuclear RNA was firstly ligated to a 5´ linker with 3´-OH, then random primers were used to generate the cDNA libraries for deep sequencing. We analyzed the 5´-most signals of the libraries and found a strong and selective accumulation of the signals in the pPAS downstream region compared to the dPAS downstream region when the activity of nuclear 5´–3´ exonuclease XRN2 was inhibited. The results demonstrated the presence of 3´ cleavage intermediates with 5´ mono-phosphorylate and 3´ polyA tail, supporting the precursor-product relationship anticipated from sequential polyadenylation. Together, we concluded that the short 3´ UTR isoforms of many genes could be generated through sequential polyadenylation, owing to that the dPAS is inherently stronger than the pPAS [36,38,39].

During the transcription cycle, the Pol II CTD is dynamically phosphorylated, among which threonine 4 phosphorylation (T4ph) is specific to the transcription termination region [40]. T4ph was found to be more closely associated with the dPAS, even in the genes where the pPAS is more frequently used [15]. We reasoned that sequential polyadenylation may contribute to this phenomenon: dPAS processing takes place preceding pPAS processing and induces T4ph in the region downstream of the dPAS, although the final transcript uses the pPAS after sequential polyadenylation.

APA regulation: additional insights from sequential polyadenylation

The foundation of “sequential polyadenylation” indicates that the processing of a proximal PAS could lag far behind its synthesis, even after dPAS is transcribed or polyadenylated, which is different from what we thought previously and can deepen our understanding of APA regulation. Here, we propose a lever model to summarize potential rules in regulating the APA process.

It was reported that CstF64 can equivalently bind to their pPASs and dPASs for the 3´ UTR lengthening genes in the absence of CstF64 and CstF64τ. In contrast, for the 3´ UTR shortening genes, CstF64 preferentially binds to their dPAS [11]. Therefore, we propose that when CstF64 binds equally to the pPAS and dPAS, the processing efficiencies of both pPAS and dPAS are decreased with the loss of CstF64 and CstF64τ. As dPASs are usually more associated with other polyadenylation enhancing elements besides the U/GU-rich elements [5,38,39], which may compensate for the loss of CstF64 and CstF64τ to some extent, the effect is less profound on dPASs and ultimately lead to the lengthening of 3´ UTR. However, when CstF64 specifically binds to the dPAS, its loss would solely interfere with the processing efficiency of dPAS, thus facilitating the pPAS usage.

Based on these facts, we propose that the choice of PASs depends on the relative strengths of the PASs determined by the embedded cis-elements and the availability of their recognition factors (threshold). The relative usage depends on the competition between PASs, analogous to the leverage of a lever via its pivot (Figure 1c). The loss of core CPA factor(s) would decrease the relative strengths of all PASs, analogous to moving down the pivot, but the pPAS are more severely influenced due to the lack of compensation from other elements, resulting in a shift to the dPAS (Figure 1d bottom). We hypothesize that the same principle may also be shared by other core CPA factors (Figure 1d top).

It was well documented that transcription activities are actively involved in APA regulation. Initially, a study on a transcription pausing site downstream of the pPAS of the human α2 globin gene demonstrated the role of Pol II pausing in enhancing the usage of pPAS [41]. Recently, another study documented that CTCF can bind and recruit the cohesion complex to un-methylated regions downstream of the pPAS in some genes, which, in turn, will impede the traversal of Pol II and facilitate pPAS utilization [42]. Apart from transcription pausing, alterations in the transcription elongation rate, caused by the defects of transcriptional elongation associated factors [43,44] or the mutation of Pol II itself [45,46], will globally cause a shift to pPAS usage.

Analogous to Pol II pausing and slow Pol II speed, we suspect that any extended exposure to the CPA complex in the nucleus may alter the APA pattern. It is the case in which the pPAS will be more used with increasing the distance between two identical poly(A) sites [47]. “Sequential polyadenylation” also provides a new clue for the mechanism of CFIm25-mediated APA regulation. CFIm25, one subunit of CFIm, can bind to the UGUA motif upstream of the AAUAAA of dPAS and enhance its processing, and CFIm25 knockdown will facilitate the usage of pPAS [12]. As the depletion of the CPA factors usually caused transcription readthrough [15,34,35], we envision that CFIm25 decrease will lead to a reduction of dPAS processing efficiency and concurrent transcription readthrough. Together, the tethering of the unprocessed RNA to Pol II can restrict the RNA in the nucleus and lengthen the time for pPAS processing no matter whether the dPAS will be processed or not [43–46], hence leading to an APA pattern switching from dPAS to pPAS (Figure 1e).

Further study using CRISPR to delete or mutate the UGUA motif can help clarify this. A recently published paper from Wei Chen’s Lab described a method named CRISPR-iPAS, which can be effectively used to perturb the usage of specific poly(A) sites [48]. Their study found that perturbation of pPAS would generally reduce the total mRNA abundance, while perturbation of dPAS has nearly no effect on the total mRNA abundance. We guess this may be caused by the exact mechanism as CFIm25 knockdown and can indirectly corroborate our hypothesis.

Further perspective

Despite CPSF and CstF are tightly associated with Pol II [24,25], RBBP6, which is required for activating the cleavage activity of the CPA complex, is not a stable subunit of them [6,7]. Therefore, where and when RBBP6 will join the CPA complex remains mysterious. We suspect that chromatin immunoprecipitation coupled with high-throughput sequencing (ChIP-seq) on RBBP6 will provide the chromatin proximity information of pPAS and dPAS to infer whether dPAS processing happens co-transcriptionally and pPAS processing occurs post-transcriptionally.

Although we demonstrated that 3´ processing is not obligated to be co-transcriptional, we are not sure whether partially assembly of the CPA complex on the pPAS during transcription is required for sequential polyadenylation. Further study may clarify this requirement, such as injecting the in vitro transcribed long 3´ UTR isoform into the cells to check its pPAS processing. On the other hand, what proportion of genes or transcripts per gene will undergo sequential polyadenylation is also a technical challenge to solve.

More importantly, what are the biological functions and the physiological relevance of sequential polyadenylation? Can sequential polyadenylation be triggered by any biological signaling? As our and others’ studies show that the longer 3´ UTR isoforms of many APA genes tend to reside in the nucleus [36,44,49], one attractive hypothesis is that further processing at the pPAS could be an important regulatory mechanism to quickly respond to specific stimuli without de novo transcription when sequential polyadenylation is induced. Given the fact that global 3´ UTR shortening extensively occurs during virus infection [50], cellular stress [51], secretory cell differentiation [52], as well as mTORC1 activation [53], this hypothesis is merited to be tested in those systems in the future.

Funding Statement

This work was supported by grants from the China NSFC projects [31922039 and 31871316] and the Hubei Provincial Natural Science Foundation of China [2020CFA057] to Y.Z.

Disclosure statement

No potential conflict of interest was reported by the author(s).