Splicing is a critical mRNA maturation process in eukaryotic cells in which intervening sequences (introns) are cut out from the nascent transcript and exon sequences are pieced together by the spliceosome to form uninterrupted coding DNA sequences (CDS) and untranslated region (UTR) sequences for proper protein translation. This process occurs co-transcriptionally and is dependent upon the actions and activities of RNA polymerase II (Pol II) (Fig. 1A).^10-12 In AS, certain exons, or part of certain exons, are alternatively included in the mature transcripts in different biological contexts. AS occurs in about 95% of multi-exon genes in human transcriptome.¹³ The regulation of AS is dependent upon the coordination of relevant trans-acting RNA binding proteins (RBPs) and the cis-acting elements surrounding the alternative exons. A series of RBPs such as serine arginine (SR) proteins and heterogeneous nuclear ribonucleoproteins (hnRNPs) that can affect the molecular actions of the spliceosomal complexes play important roles in the regulation of general RNA splicing and AS.^14,15 There are generally 5 basic types of AS^16,17: cassette type (or exon inclusion/skipping type) occurs when the 3′-splice site of the downstream intron is used for splicing, skipping over an entire exon; mutually exclusive type occurs when 2 cassette exons are mutually exclusively selected for splicing; alternative 5′-splice site type occurs when a different 5′-splice site is defined and used for splicing; alternative 3′-splice site type occurs when a different 3′-splice site is recognized; finally, intron retention occurs when neither the 5′-splice site nor the 3′-splice site of an intron is used for splicing, leaving that particular intron in the final mRNA transcript (Fig. 1B). Since AS alters the sequence of the resulting mature mRNA, it can impact the expression and function of the gene. For example, given that 5′-UTRs often contain sequence elements or secondary structures that can affect ribosome loading, AS of exons in the 5′-UTRs may alter the translation efficiency of the transcripts.^18,19 Moreover, AS of exons in the CDS may alter the function of certain domains or affect the folding of proteins, leading to changes in the characteristics of the protein products translated from the transcripts. These AS events increase the capacity of functional proteome.¹⁷ And more importantly, AS events can have significant physiological impacts without necessarily affecting the transcript expression level of genes. Indeed, multiple biological contexts and diseases, e.g. cancers, have been associated with transcriptome-wide AS events.^1,2 These demonstrate the importance of studying the roles of AS events in cellular gene expression regulations.

The termination of the transcription activity by RNA Pol II involves a sequence of molecular events. Towards the end of transcription, the nascent transcript undergoes endonucleolytic cleavage to be released and polyadenylated while Pol II continues with transcription. Then, Pol II is released from the DNA for recycling and allow for the next rounds of transcription (Fig. 1A). For cleavage and polyadenylation, namely the maturation of the 3′ end of mRNA molecules (sometimes simply referred to as polyadenylation), Pol II first reaches and transcribes sequence elements that recruit the formation of the 3′-end processing complex, which include a poly-A signal (PAS, most commonly AAUAAA, AUUAAA, and several other variants), and often a U-rich auxiliary upstream element (USE) and an U-rich or AU-rich downstream element (DSE). These signal the recruitment of trans-acting factors such as the PAS-binding cleavage and polyadenylation specificity factors (CPSFs) and the DSE-binding cleavage stimulation factors (CSTFs) to bind to the nascent transcript to catalyze an endonucleolytic reaction at a CA dinucleotide that is usually 15–30 nucleotides downstream of the PAS. Then, poly-A polymerase adds a stretch of untemplated adenosines, the poly-A tail, to the 3′-end of the transcript from the cleavage site. The poly-A tail is needed for downstream metabolism of the mature mRNA transcript including nuclear export of mRNA, translation, localization, and stability.^20-22 As for the release and the termination of the transcription reaction of Pol II, 5′ to 3′ exonucleases are recruited to attack the 5′-end generated by the cleavage during polyadenylation process, which is unprotected by a 5′-cap. The exonucleases (e.g. Xrn2, 5′-3′ exoribonuclease 2) then chase down Pol II along their substrate and finally displace Pol II from the transcription bubble to terminate the transcription reaction.^23-25

Thus, termination of transcription reactions is dependent upon the occurrence and “strength” of the cis-acting elements that signal for polyadenylation (the USE-PAS-DSE pattern). Interestingly, at least 70% of human genes are predicted to possess two or more such cis-acting elements.²⁶ And, when more than one PASs in a gene are capable to be utilized for 3′-end processing, APA occurs. The regulation of APA, like AS, is determined by the coordination of various trans-acting factors and cis-acting elements surrounding the PAS and alternative PAS.^3,4 The expression levels and activities of a number of RBPs have been shown to be able to affect APA at a transcriptome-wide level due to their roles in interacting with 3′-end processing factors and/or the cis-acting elements near PASs; these factors include the components of the 3′-end processing complex such as CPSFs and CSTFs, as well as other polyadenylation-associated RBPs such as cytoplasmic polyadenylation element binding protein 1 (CPEB1) and poly(A) binding protein nuclear 1 (PABPN1).^27-32 Generally, there are 2 types of APA: UTR-APA and coding region (CR)-APA.³³ 3′-UTR of genes is often longer than CDS and contains binding sites for microRNAs and regulatory RBPs. It also contains alternative PASs. Most of the alternative PASs in 3′-UTRs are upstream of the canonical or annotated PASs (and in this scenario, the alternative PASs are termed proximal PASs and the canonical PASs are termed distal PASs, based on their relative distance to the stop codon). Thus, when these alternative PASs are utilized for polyadenylation, namely when UTR-APA occurs, in most cases, 3′-UTRs are shortened. This phenomenon is referred to as 3′-UTR shortening. As UTR-APA alters the availability of these regulatory cis-acting elements on the mature transcripts, it can affect the behaviors of mRNA transcripts, including translation efficiency, localization, and stability, without changing the coding capacity of the transcripts (Fig. 1C).^3,4,26,34,35 For example, PAX3, one of the master transcription regulators of the myogenic transcriptional network, has a miR-206 binding site in its 3′-UTR. It has been observed that among different muscle cell types, the ratios of PAX3 UTR-APA isoforms (thus the ratio of transcripts being able to be regulated by miR-206) differ. This results in varying degrees of protein translation efficiency for PAX3 transcripts in the different muscle cell types, which can help explain the varying differentiation patterns in these distinct muscle cell types.³⁶ It has also been shown that in neurons, where accurate localization of gene expression is crucial for proper cellular functions, the isoforms of hundreds of genes are differentially localized based on the UTR-APA events in their 3′-UTRs.³⁷ These highlight the crucial role of UTR-APA events in determining the fate of mRNA transcripts. Furthermore, the 3′-UTR of an mRNA transcript has also been shown to serve as a molecular scaffold for protein-protein interactions, particularly, immediate interactions between the nascent protein synthesized from the mRNA transcript and its binding partners. For example, the 3′-UTR of the membrane protein CD47, can recruit protein complexes including ELAV like RNA binding protein 1 (ELAVL1, or HuR) and SET nuclear proto-oncogene (SET), allowing immediate interaction of these proteins with the nascent CD47 protein. This molecular event leads to the efficient translocation of CD47 to the plasma membrane. Upon UTR-APA, CD47 transcript can no longer recruit the binding partners for plasma membrane localization. The CD47 protein produced from CD47 transcript with UTR-APA localizes to endothelium reticulum, instead.^38,39 As for CR-APA, it occurs when alternative PASs in the upstream intronic regions are utilized for polyadenylation. It is thus also sometimes referred to as intronic APA. Once a transcript is truncated, the transcript would lose the coding capacity of a chunk of polypeptide on the C-terminal end. The resulting protein product may thus function differently. Moreover, the truncated transcript would be differentially regulated compared to the full-length counterpart as it would possess a completely different 3′-UTR that originates from the intron region downstream of the alternative PAS. There are 2 types of CR-APA; the mechanism of both types of CR-APA involves the interplay between splicing and polyadenylation.^3,33,40 The first type of CR-APA occurs when the splicing of a PAS-containing intron is inhibited, and the 3′-end processing complex outcompetes the splicing machinery, leading to truncation of the transcript at that intron. The second type of CR-APA occurs when a cryptic exon that is followed by a PAS is utilized for splicing. Due to the presence of the PAS, this “alternative splicing” event leads to the truncation of the mRNA transcript (Fig. 1C). While the studies in the field focusing on CR-APA is relatively few, one of the first and most famous identified APA events belongs to this category. In resting B lymphocytes, immunoglobulin heavy constant mu (IGHM) gene is transcribed to full-length, which includes the last 2 exons that encode a transmembrane domain, leading to the expression of membrane-bound antibody chain. And upon B lymphocyte activation, CR-APA of IGHM transcript occurs, causing the two exons encoding the transmembrane domain to be excluded from the final mRNA transcript. This leads to the production of secreted form of immunoglobulin M antibody chain.⁴¹ Another example of CR-APA is the mitogen-activated protein kinase (MAPK) interacting serine/threonine kinase 1 (MKNK1) gene. The CR-APA event of MKNK1 gene results in the loss of 89 amino acids at the C-terminal end of the kinase and the addition of 12 amino acids encoded by the intronic sequence. It has been shown that this truncated version of MKNK1 has a higher kinase activity compared to its full-length counterpart.⁴² Moreover, intriguingly, a recent study found that in chronic lymphocytic leukemia cells, a number of tumor suppressor genes undergo CR-APA and consequently lose their tumor-suppressing activities. Some of the truncated proteins even seemed to act in an oncogenic manner.⁴³ Together, all these previous findings show that APA can drastically affect gene expression and steer cellular biology independent of the transcription regulation of genes.

With the advent of high-resolution deep sequencing technologies and numerous bioinformatics pipelines for data analyses, transcriptome-wide gene expression can be understood and alternative processing of gene products can be profiled. Transcriptome-wide studies on AS and APA have thus been conducted routinely in numerous biological systems and in various physiological contexts. A number of studies reporting mTORC1's role in transcriptome-wide post-transcriptional regulations have also emerged.

In mouse embryonic fibroblast cell (MEF) model, a genetic activation of mTOR by knocking out Tsc1 (Tsc1^−/−) showed an increase of transcriptome-wide UTR-APA events. Interestingly, the majority of the transcripts with shortened 3′-UTR by UTR-APA did not show significant overall transcript level changes. Importantly, polysome profiling for actively translating transcripts indicated that 3′-UTR shortening by APA in a transcript promotes the efficiency of protein synthesis. Thus, this transcriptome-wide 3′-UTR shortening phenomenon in the mTOR-activated transcriptome highlights the fact that traditional RNA-Seq analyses that only focus on gene expression level profiling can miss a significant portion of expression-regulating events in the transcriptome by mTORC1 (Fig. 3).⁵⁸

In general, most of the bioinformatics tools for UTR-APA profiling, including the one used in the above study, take advantage of the highly quantitative nature of RNA-Seq for the measurement of UTR-APA events by comparing the ratios of the read density of the long form-only regions and the read density of the regions common to both long and short transcripts between 2 samples. Changes of the ratios between 2 samples indicate the occurrence of an UTR-APA event. A major issue with this RNA-Seq-based approach for UTR-APA analysis is that the measurement relies on the accurate information on the locations of the 3′-ends of the short transcripts. UTR-APA events involving unannotated short transcripts can be missed, while UTR-APA events involving annotated yet biologically irrelevant (not expressed in the cell line or tissue under study) short transcript isoforms can be falsely identified.^33,59,60

An attempt to overcome this existing issue on UTR-APA analysis was made by integrating two sequencing technologies together. In that follow-up report, poly-A site sequencing (PAS-Seq) was performed on 2 MEF models with basal (WT) and hyper mTOR (Tsc1^−/−) activity. PAS-Seq is a sequencing technique that focuses the sequencing depth of RNA-Seq onto the polyadenylated sites of mRNA transcripts for a more confident and experimentally-proven profiling of PASs used for polyadenylation in the biological system. By integrating the PAS-Seq data sets and the previous RNA-Seq results, the confidence and accuracy of identification and quantification of biologically relevant UTR-APA events in the mTORC1-activated transcriptome were both elevated. Although this new method requires that the same samples must be sequenced by both RNA-Seq and PAS-Seq, which can be costly and prevents this method to be applied to old RNA-Seq data that lack the corresponding PAS-Seq information, this method did allow the identification of new biological pathways regulated by mTORC1-mediated UTR-APA events.⁶¹ Therefore, with ever-improving sequencing technologies and bioinformatics tools, the role and impact of mTORC1 in transcriptome-wide UTR-APA regulations have become clearer.

Since 3′-UTR of genes contain many microRNA and RBP-binding sites, 3′-UTR shortening would allow transcripts to escape the regulation of these trans-acting factors. In a similar context, siRNAs targeting 3′-UTR have been shown to be ineffective upon mTOR-driven shortening of select transcripts, suggesting that APA modulates diverse regulatory activities that may happen through 3′-UTR.⁵⁸ As such, 3′-UTR shortening has been shown to alter the stability, localization, and translation efficiency of mRNA transcripts.^3,4,26 In select cases of the mTORC1-activated 3′-UTR shortening events, a significant increase in protein synthesis without significant changes in their mRNA transcript levels was observed. Pathway enrichment analysis on these 3′-UTR-shortened genes revealed that ubiquitin-mediated proteolysis pathway is the most targeted pathway by mTORC1-mediated 3′-UTR shortening (Fig. 3).⁵⁸ Thus, with UTR-APA analysis on mTORC1-activated transcriptome, a pathway activated by mTORC1 that was previously not associated with mTORC1 activity was revealed. Moreover, with the integration of PAS-Seq in our UTR-APA analysis, it was further revealed that mTORC1 regulates the 3′-UTR shortening of several transcription factors, suggesting that mTORC1 may have a role in controlling various transcription networks through its regulation of UTR-APA. For example, it was demonstrated that mTORC1 upregulates CCAAT/enhancer binding protein gamma (CEBPG) through 3′-UTR shortening; this upregulation of CEBPG by mTORC1 is critical in protecting cells against endothelial reticulum stress (Fig. 3).⁶¹

Passacantilli et al.⁶² reported that upon chemical mTOR inhibition, extensive changes in the transcriptome were observed in Ewing sarcoma cells using microarray analysis. Particularly, 1,440 AS events were detected in 918 genes.⁶² These data were obtained by microarray technology, which can only identify previously annotated AS events. This suggests that we could expect to observe more mTOR-mediated AS events if unbiased sequencing technologies such as RNA-Seq are used for analyzing these biological samples. Nonetheless, their work demonstrates that mTOR activity indeed regulate transcriptome-wide AS events. Furthermore, Lee et al.⁶³ also reported that chemical inhibition of mTORC1 leads to malfunction in the splicing of select genes. With improper splicing, introns are retained in these genes, activating nonsense-mediated pathway for the degradation of the transcripts (Fig. 3).⁶³

Lee et al.⁶³ demonstrated that, through S6K1 phosphorylation, mTORC1 regulates the activity of serine and arginine rich splicing factor protein kinase 2 (SRPK2), a key regulator of a series of splicing factors. When mTORC1 is downregulated, SRPK2 activity is inhibited, preventing the proper function of downstream splicing factors, causing splicing dysregulation and thus downregulation of a number of lipogenic genes (Fig. 3).⁶³ This study showed that mTORC1, through regulating the activities of splicing regulators, can modulate gene expression at the level of splicing regulation. Passacantilli et al.⁶² also demonstrated that the transcriptome-wide AS events they observed in embryonic stem cells upon mTOR inhibition contribute to drug resistance in that particular cancer cell line.⁶² Moreover, recently, it was shown that cellular mTOR activity regulates the expression of U2AF1 isoforms (U2AF1a v. U2AF1b), a critical splicing factor determining 3′-splice site. One striking molecular outcome of this differential U2AF1 isoform expression is the changes of AS in 5′-UTR of many genes that significantly affect translation efficiency.⁶⁴ Thus, these reports not only reveal mTORC1's role in splicing regulation, but that these mTORC1-mediated splicing regulations have significant physiological impacts.

Taken together, these studies add post-transcriptional regulation as another layer of gene expression regulation to our understanding of the multi-faceted functions of mTORC1 (Fig. 3). Moreover, analyses of mTORC1-mediated post-transcriptional regulations in various cellular contexts have enabled researchers to not only gain new mechanistic insights into mTORC1's control over previously associated biological processes, but also to discover new cellular signaling pathways that are regulated by mTORC1 and make mechanistic connections that were previously masked due to a lack of analyses focusing on post-transcriptional regulations.