Gene regulation through exon-junction complex (EJC) modularity

Abstract

The exon-junction complex (EJC) begins to assemble on the spliceosome, which deposits EJCs upstream of most exon–exon junctions during pre-messenger RNA (mRNA) splicing. EJCs acquire additional alternative modules that define heterogeneous EJCs during pre-mRNA processing to mRNA in the nucleus, and after mRNA export into the cytoplasm. In this Review, we discuss the mechanisms of EJC formation, the many roles of the EJC in pre-mRNA and mRNA regulation, and how these roles are influenced by EJC composition.

Introduction

The vast majority of mammalian-cell pre-mRNAs contain multiple introns that must be removed before the resulting mRNA, which consists only of exons, can produce the encoded protein having the appropriate amino acid sequence at the appropriate level and subcellular location. Otherwise, introns that disrupt or flank the coding sequence would preclude the production of full-length protein or alter regulatory sequences that control mRNA expression levels, stability, localization, and other aspects of mRNA metabolism, respectively. The spliceosome, a large and dynamic complex of proteins and small nuclear RNAs (snRNAs), is responsible for pre-mRNA splicing, i.e. removing introns and joining the resulting exons in a process crucial for the maturation of pre-mRNAs into mRNAs.¹

During pre-mRNA splicing in the nucleus of human cells, the spliceosome deposits an exon-junction complex (EJC) ~20–24-nucleotides (nts) upstream of an estimated 60–80% of exon–exon junctions, according to original approximations.^2–4 More recent crosslinking studies of the EJC core protein EIF4A3 center the EJC at a position that resides 27-nts upstream of most exon–exon junctions.⁵ Variability in EJC deposition at exon–exon junctions can be explained by factors such as RNA secondary structure influencing both the positioning and efficiency of EJC deposition (see below).^4,6

EJCs are comprised of the three core proteins EIF4A3, RBM8A (also called Y14), and MAGOH; usually either UPF3B or UPF3A (also called UPF3X and UPF3, respectively); and peripheral factors that we and others have defined as constituents of alternative EJCs (Fig. 1a).³ To date, the best-characterized alternative EJCs contain UPF2, RNPS1, either ACIN1 or PNN, and SAP18 (called here the RNPS1-EJC), or CASC3 and AKT (called here the CASC3-EJC) (Fig. 1a).^3,7 These designations appear to accurately reflect cell physiology since tagged RNPS1 and CASC3 fail to co-immunoprecipitate in cell lysates⁸, even though recombinantly expressed tagged RNPS1, ACIN1, or SAP18 pulls down CASC3 in vitro.⁹

Fig. 1 |. Diagrams of the core, RNPS1-EJC and CASC3-EJC, and models of RNPS1-EJC assembly on the spliceosome.

a, Alternative EJCs are characterized by a common three-protein core of EIF4A3, RBM8A, and MAGOH (leftmost diagram). For the EJCs discussed here, UPF3B (or in the absence of UPF3B, most likely, its paralog UPF3A) is also present. Peripheral EJC constituents define alternative EJC identity. This perspective focuses on the RNPS1-EJC, which in addition to the core contains UPF2, RNPS1, SAP18, and either ACIN1 or PNN (middle diagram), or the CASC3-EJC, which in addition to the core contains AKT and CASC3 (rightmost diagram). b, Two states of the B^act-spliceosome are shown. EJC constituents start to assemble on the B^act-spliceosome (leftmost diagram). Some but not other in vitro studies indicate that EIF4A3 is recruited to B^act by CWC22–CWC27. The splicing factor EFTUD2 also likely directly contacts EIF4A3 in B^act, as it does in the C-spliceosome. Subsequently (middle diagram), the RBM8A–MAGOH heterodimer is recruited by EIF4A3 once CWC27 is displaced. In one model of PSAP complex recruitment to the spliceosome, SNIP1 may directly recruit the PSAP complex (leftmost diagram). During the transition of the B^act-spliceosome to the C-spliceosome (rightmost diagram), both SNIP1 and PSAP depart (left “?”) or SNIP departs and the PSAP complex transitions to the core EJC (right “?”). c, In another model of PSAP complex recruitment to the spliceosome, PSAP is recruited directly to the core EJC on the B^act-spliceosome and, when at detained introns, promotes SNIP1-mediated spliceosome stalling via long-range interactions.

EJCs have many “active” regulatory roles that include promoting alternative pre-mRNA splicing, nuclear mRNA export, and the pioneer rounds of translation of newly made mRNAs that harbor the co-transcriptionally acquired CBP80–CBP20 nuclear cap-binding complex (CBC).¹⁰ Another active regulatory role of EJCs is in nonsense-mediated mRNA decay (NMD), which occurs during the pioneer rounds of translation for those mRNAs that contain features required for NMD (see below).³ EJCs also perform many “passive” regulatory roles by physically masking the region where they are bound to RNA. Through this mechanism, EJCs can prevent the use of cryptic splice sites,¹¹ mRNA re-splicing,¹² or recursive exon splicing that utilizes splice sites generated upon formation of an exon–exon junction.¹³ EJCs can additionally stabilize cytoplasmic mRNAs by physically precluding the nuclear deposition of N⁶-methyladenosine (m⁶A), which for the majority of internal exons would preclude modification over most if not all of their length.^14,15

As the known regulatory roles of EJCs in mRNA metabolism increase, it becomes important to better understand the determinants of EJC composition, deposition, and, ultimately, function. Here, we review what is currently known about EJC structures and functions, focusing largely on mammalian-cell EJCs, with the view that EJCs constitute mRNA hubs that contribute in multiple ways to regulated gene expression.

EJC formation on the spliceosome

When the various EJC constituents are recruited to the spliceosome is currently incompletely understood. One model, based on cryo-EM structures and some in vitro biochemical studies, indicates that the EJC core, i.e. EIF4A3, RBM8A, and MAGOH, forms on the spliceosome during the transition from the B^act-spliceosome to the catalytic C-spliceosome, which is the earliest spliceosome structure in which the EJC core is resolved.¹⁶ The B^act-spliceosome, which transitions through multiple states, comprises the activated but yet-to-be catalytically primed spliceosome, while the C-spliceosome is generated from the B*-spliceosome during the first lariat-forming branching step of the splicing reaction.¹ However, in support of a different model, other studies performed in vitro and in cell lysates demonstrate that EIF4A3 directly interacts with the CWC22–CWC27 heterodimer in the B^act-spliceosome and recruits EIF4A3 in the absence of RBM8A and MAGOH, as joining of RBM8A is incompatible with CWC27-bound EIF4A3 (Fig. 1b, left panel).¹⁷ Thus, this alternative model indicates that the EJC core likely starts assembling on the B^act-spliceosome and continues assembling in a step-wise fashion thereafter, rather than joining the C-spliceosome as a pre-formed trimer (Fig. 1b, left and middle panels). Supporting either model, in the C-spliceosome, the EJC core directly contacts the splicing factors CWC22 and EFTUD2.¹⁶ Consistent with this, perturbations of the CWC22–EIF4A3 interface, like depletion of CWC22, reduce the efficiency of EIF4A3 deposition on RNA.^18,19

EJC deposition also involves the splicing factor Aquarius (AQR, also called IBP160), given that AQR depletion reduces the co-immunoprecipitation (co-IP) of spliced RNAs with EIF4A3 and RBM8A.²⁰ Since AQR and the EJC core bind to distant sites on the spliceosome,¹⁶ long-range protein–protein interaction networks through the spliceosome may mediate EJC deposition via AQR. Alternatively, given the dynamic flexibility of spliceosome intermediates²¹, the EJC core and AQR may come in contact via structural rearrangements that have not been resolved by cryo-EM.

Peripheral RNPS1-EJC constituents can also be recruited prior to C-spliceosome formation. RNPS1, PNN, and SAP18, which form the PSAP complex,²² co-immunoprecipitate with the B^act-spliceosome constituent SNIP1 in a manner dependent on the serine-rich domain of RNPS1 (Fig. 1b, left panel).²³ It is unknown if, in the B^act-spliceosome, SNIP1 binds the PSAP complex independently of an EJC (Fig. 1b), given that SNIP1 resides far from where the EJC core binds,²⁴ or if SNIP1 engages in long-range interactions with a PSAP-bound EJC core (Fig. 1c). The association between the PSAP complex and SNIP1 is observed at “detained introns”, i.e. introns not removed co-transcriptionally but subsequently removed by regulated splicing dependent on B^act-spliceosome stalling²⁵, where depletion of either SNIP1, RNPS1, or PNN increases the abundance of the detained intron (Fig. 1c).²³ Consistent with RNPS1 functioning at introns other than detained introns (see below), IP and proximity labelling experiments indicate that RNPS1 exists in spliceosomes prior to C- spliceosome formation together with SF3B1, SF3B2, SF3B3, SF3B4, PRPF38A, and CWC27.²⁶

While a small CASC3 peptide has been modeled alongside the EJC core in cryo-EM structures of the C-spliceosome, the experimentally derived electron density maps do not provide strong evidence for the presence of CASC3.¹⁶ Whether other peripheral EJC constituents, e.g., UPF3B or UPF3A, which join the EJC by binding to the interface formed by EIF4A3, RBM8A, and MAGOH²⁷, are recruited to the spliceosome is also unknown. The same is true for AKT, which is proposed to associate with the EJC via UPF3B based on in-cell proximity ligation assays and the loss of the interaction between UPF1 and AKT in cells transfected with UPF3B siRNA. ²⁸ In theory, these factors could be assembled onto the EJC core after it has been deposited at an exon–exon junction, i.e. after splicing of the associated intron.

Alternative EJC target specificity

The existence of alternative EJCs raises the question of EJC “target specificity”, i.e., whether one type of alternative EJC is deposited preferentially on some or possibly all exon–exon junctions of certain mRNAs. There is also the possibility of one type of EJC being converted to another.⁸ When cellular transcripts are globally assayed, results from individual-nucleotide resolution crosslinking and IP (iCLIP) show that, in the steady-state, RNPS1, CASC3, and the core EJC are largely deposited ~20–24-nts upstream of exon–exon junctions without an apparent bias toward alternative EJC type.²⁹ Among the few exceptions, in addition to the enrichment of RNPS1-EJC constituents at detained introns, which may be specific to this class of introns or may simply reflect the slower splicing rate of stalled spliceosomes²³, CASC3-EJCs are enriched on alternatively spliced exons, suggesting that there can be some degree of target specificity by unknown mechanisms.²⁹ For certain mRNAs and by incompletely understood mechanisms, spliceosome-mediated EJC deposition appears to take place at nuclear speckles, where these mRNAs are preferentially localized.^30–32

It is also unclear why only ~60–80% of exon–exon junctions are occupied by an EJC.^2,4 Among mRNAs having fewer EJCs than expected relative to the number of exon–exon junctions are those encoding ribosomal and housekeeping proteins.^29,33 The fact that not all exon–exon junctions are occupied by an EJC could be explained by differences in EJC deposition efficiency as a result of RNA secondary structure at or around the EJC deposition site^4,6, as mentioned above, and/or also differences in EJC residence times.³³ Another explanation could derive from the number of exon–exon junctions greatly exceeding the number of EJC subunits, as has been shown for HeLa cells.³⁴ The lack of an EJC at all exon–exon junctions could also be due to an inefficiency in the method used to map the EJC. For example, given that EIF4A3 binds the RNA backbone in a sequence-independent manner³⁵, its variable crosslinking efficiency to exon–exon junctions appears to reflect variations in crosslinking chemistry.^36,37

Despite examples of the preferential deposition of alternative EJCs at subsets of exon–exon junctions, no sequence motifs are known to confer preference.²⁹ However, footprints of the SR-like protein ACIN1, which harbors a serine-and-arginine (SR)-rich region, are enriched in the intronic polypyrimidine tracts that reside upstream of 3′-splice sites, in SR protein-binding sequences that constitute exonic splicing-enhancer sequences, and in canonical EJC deposition sites.³⁸ Additionally, footprints of core and peripheral EJC constituents map at low levels within introns and exons at sites distinct from those sites that are upstream of exon–exon junctions.²⁹ Possibly, long-range interactions between alternative EJC constituents and sequence-specific RNA-binding proteins that are not captured in EJC CLIP data could promote deposition of specific alternative EJCs. Finally, since alternative EJC composition is changed by insulin stimulation,⁷ changes in the cellular environment may promote alternative EJC deposition, possibly on specific mRNAs and/or exon–exon junctions.

EJC-mediated regulation of pre-mRNA splicing

Once deposited upstream of exon–exon junctions, EJCs can actively regulate splicing in a manner that depends on EJC composition, as evidenced by ACIN1 and EIF4A3 co-regulating alternative splicing events.³⁸ As another example, RNPS1 regulates a subset of alternative splicing events²⁶ dependent on whether RNPS1 and SAP18 are in a complex with ACIN1 (the ASAP complex) or with PNN (the PSAP complex).^9,11 Since only a subset of these RNPS1-reliant alternative splicing events also rely on EIF4A3 and RBM8A, alternative splicing events may or may not be influenced by the core EJC.⁹ As an additional example, depletion of either PNN or SAP18 can promote exon skipping (Fig. 2a).^11,39 As a final example of EJCs regulating alternative splicing in a pre-mRNA sequence-dependent manner, knockdown of either EIF4A3 or RNPS1, each of which directly binds RNA, increases the skipping of exon 4 in MRPL3 pre-mRNA.¹¹

Fig. 2 |. Mechanisms by which EJCs can actively or passively regulate pre-mRNA metabolism.

a, EJC constituents can influence pre-mRNA splicing. For example, knockdown (KD) of either PNN or SAP18 can promote exon skipping, indicating that PNN or SAP18 can inhibit exon skipping. b, AKT may promote splicing in the context of SR protein networks, including in the context of the EJC, by phosphorylating either or both SR proteins or SR protein kinases (SRPKs), the latter of which phosphorylate SR proteins. Phosphorylated SR proteins are known to bind exonic splicing enhancers. c, RNPS1-KD increases cryptic 5′-splice-site usage, while KD of either eIF4A3 or RBM8A increase cryptic 3′-splice site usage, indicating that EJCs can inhibit cryptic splice-site usage by physically masking their accessibility. d, KD of either CASC3, eIF4A3, RBM8A, RNPS1, or PNN increases recursive 5′-splice-site usage, suggesting that EJCs (or a constituent acting apart from the EJC) can also physically prevent the splicing of recursively spliced (RS)-exons. e, KD of eIF4A3, RBM8A, or RNPS1 inhibits METTL3/METTL14-mediated m⁶A modification at DRACH motifs residing in the ~100-nt region around exon–exon junctions, indicating that EJCs physically protect this region from m⁶A modification.

The finding that AKT is a constituent of CASC3-EJCs⁷ potentially adds a new layer of active EJC-mediated splicing regulation by SR proteins.⁴⁰ This could occur via the AKT-mediated phosphorylation of SR proteins, as has been shown for SRSF1,^41,42 or via the AKT-mediated phosphorylation of SR protein kinases (SRPKs) that then phosphorylate SR proteins (Fig. 2b).^41–43 Given that EJCs can form large complexes with SR proteins, called “SR protein networks”⁴, the links between AKT-containing EJCs and SR protein function in alternative splicing are ripe for investigation. Possibly, increasing the level of AKT-containing EJCs, e.g, upon insulin stimulation⁷, promotes SR protein phosphorylation in the context of these networks, thereby altering splicing.

In addition to actively regulating alternative splicing patterns, EJCs can passively regulate splicing by physically blocking cryptic 5′- and/or 3′-splice sites (Fig. 2c). Cryptic splice sites resemble splicing consensus sequences but are generally deleterious to gene expression when utilized. For example, depletion of RNPS1 from PSAP-containing EJCs increases mis-splicing transcriptome-wide, including through increased utilization of cryptic 5′-splice sites.¹¹ Notably, even though RNPS1 and CASC3 largely CLIP to the same exon–exon junctions²⁹, RNPS1-depletion data suggest that CASC3 cannot functionally replace RNPS1 to block cryptic 5′-splice-site usage.¹¹ However, a block to cryptic 3′-splice-site usage remains upon RNPS1 depletion, but not EIF4A3 depletion (Fig. 2c).¹¹

EJCs can also passively prevent splicing at recursive 5′-splice sites.¹³ Knockdown of either CASC3, EIF4A3, RBM8A, RNPS1, or PNN increases the skipping of exons produced via recursive splicing (RS-exons), indicating that both CASC3- and RNPS1-EJCs can repress the utilization of recursive 5′-splice sites (Fig. 2d).¹³ RNPS1- and CASC3-EJCs may be preferentially deposited at certain exon–exon junctions by unknown mechanisms, possibly through the action of RNA-binding proteins such as alternative splicing factors residing at nearby RS-exons. Since different RS-exons manifest different sensitivities to the knockdown of different alternative EJC constituents¹³, the level of RS-exon inclusion may be regulated by which alternative EJC is present.

EJC-mediated regulation of mRNA methylation

m⁶A regulates mRNA processing, export, translation efficiency, and decay.⁴⁴ m⁶A modifications are added in the nucleus by the METTL3/14 RNA methyltransferase complex at ~5% of DRACH motifs (D = A/G/U, R= A/G, and H = U/A/C) and reside largely in long internal exons and the last exon of ~25–60% of cellular transcripts.^44,45 EJCs inhibit mRNA methylation by physically blocking METTL3/14 access to a ~100-nt region around exon–exon junctions (Fig. 2e)^46,47, which explains why depletion of either EIF4A3 or RBM8A leads to mRNA hypermethylation around short internal exons, which normally harbor relatively few m⁶A modifications.^46–49 Generally, hypermethylation shortens mRNA half-lives transcriptome-wide, providing another link between EJC-regulated nuclear pre-mRNA processing and what is most likely cytoplasmic mRNA decay.^47,50 Notably, only ~45% of hypermethylated sites are in common when the level of RNPS1 is reduced versus when the level of either EIF4A3 or RBM8A is reduced⁴⁷, possibly reflecting differences in the efficiency of depletion of RNPS1 relative to EIF4A3 or RBM8A depletion and/or EJC heterogeneity, as not all exon–exon junctions occupied by the EJC core necessarily contain RNPS1.

EJC-mediated regulation of nuclear mRNA export into the cytoplasm

EJCs promote the nuclear export of spliced mRNAs into the cytoplasm via recruitment of the transcription-export complex (TREX).^51–53 In a current model, the 5′-cap-bound CBC recruits the mRNA export adaptor ALYREF to the 5′-end of nascent transcripts, which is followed by deposition of ALYREF immediately upstream of EJCs at exon–exon junctions in an EIF4A3-dependent manner (Fig. 3).⁵⁴ ALYREF then recruits the RNA helicase UAP56 (also called DDX39B) and the six-subunit THO complex to form TREX⁵². Once assembled, these mRNAs are remodeled by the joining of SAP Domain-Containing Ribonucleoprotein (SARNP), which changes UAP56 from an open to an mRNA-clamping conformation that displaces the THO complex. SARNP-bound mRNAs next engage the nuclear pore complex and the mRNA export factor TREX-2 (Fig. 3).⁵¹ UAP56 is released, and a heterodimer of NXF1–NXT1 then facilitates mRNA export through the nuclear pore complex (Fig. 3).^51,55,56 Depletion of EIF4A3⁵⁶ or ACIN1⁵⁷ increased polyadenylated mRNA abundance in the nucleus, and EIF4A3 depletion also reduced the deposition of ALYREF on transcripts.⁵⁶ TREX complexes immunoprecipitated from human-cell lysates are highly enriched in EJC core, ASAP, and PSAP complex proteins but are comparatively low in CASC3.⁵² However, the relative efficiencies with which CASC3-EJCs and RNPS1-EJCs promote mRNA export have yet to be assessed.

Fig. 3 |. EJCs promote the packaging and nuclear export of mRNAs.

ALYREF is recruited to the 5′-end of a newly made mRNA by the largely nuclear cap-binding heterodimer CBP80–CBP20 (CBC) and subsequently deposited on nucleotides that reside immediately upstream of an EJC. ALYREF recruits UAP56 and the THO complex, all three of which constitute the TREX complex. Interactions between TREX and the EJC via ALYREF compact the mRNA and its associated proteins. The THO complex is displaced by UAP56 clamping to the mRNA and recruitment of SARNP, followed by association of the export factor-bound mRNA with TREX-2 and the nuclear pore complex. UAP56 is released and the heterodimer NXT1–NXF1 is then recruited to facilitate export of the protein-bound mRNA through the nuclear pore complex and into the cytoplasm.

The EJC-dependent recruitment of TREX raises the question of how intron-less mRNAs are efficiently exported from the nucleus into the cytoplasm. EIF4A3 and ALYREF are relatively evenly distributed throughout the bodies of intron-less transcripts⁵⁶, indicating that splicing-independent mechanisms can recruit these factors. Possibly, these transcripts are trafficked through nuclear speckles, where EJCs and ALYREF may be deposited through a low-efficiency so-called “spliceosome-scanning” mechanism.^56,58 Other interactions, such as RBM33 binding to GC-rich sequences in intron-less transcripts, can recruit TREX independently of splicing.⁵⁹ Notably, the export of intron-less histone mRNAs relies on the direct recruitment of ALYREF by stem-loop-binding protein (SLBP)⁶⁰, and, for histone mRNAs and other intron-less RNAs, by specific RNA-sequence elements that recruit SRSF3 and SRSF7, both of which shuttle between the nucleus and cytoplasm.⁶¹

EJC-mediated mRNA localization

Although this review focuses on mammalian EJCs, the CASC3-EJC-mediated localization of oskar mRNA has long been appreciated for Drosophila melanogaster. During embryogenesis, EIF4A3, RBM8A, MAGOH, and CASC3 are required to properly localize oskar mRNA to the oocyte posterior pole.⁶² CASC3 may also facilitate what may be the Staufen-mediated localization of the noncoding brain cytoplasmic 1 (BC1) RNA in dendrites of rat hippocampal neurons since BC1 RNA, CASC3, and Staufen co-localize there, and BC1 RNA co-immunoprecipitates with both CASC3 and Staufen.^63,64

EJC core constituents in the cilium of mouse neuronal stem cells or human retinal pigment epithelial cells are required to localize Ninein (NIN) mRNA, which encodes a protein that nucleates and anchors microtubules, to the centrosome basal body.⁶⁵ Knockdown experiments indicate that while the EJC core is important for centrosome basal body integrity, CASC3 is not.⁶⁵ Whether any peripheral EJC constituents are required for NIN mRNA localization remains to be determined.

EJC-mediated translation regulation

EJCs actively link efficient pre-mRNA splicing in the nucleus to efficient mRNA translation in the cytoplasm. This was initially demonstrated in studies of NMD showing that EJCs on mRNAs that are not targeted for NMD also promote translation of their associated mRNAs.^66–69 Since EJCs are enriched on CBC-bound newly made mRNAs and not on eIF4E-bound mRNAs⁷⁰, their direct effects on translation would largely be on pioneer rounds of translation. The pioneer rounds of translation are defined as occurring on mRNAs that have yet to have 5′-cap-bound CBC replaced by 5′-cap-bound EIF4E.⁷¹ These rounds of translation can involve more than one translationally active ribosome depending on the length of the CDS, the efficiency with which the pioneer rounds of translation initiate and, in the case of an NMD target, the efficiency of NMD.⁷² This is evidenced by NMD requiring that phosphorylated UPF1 repress further translation initiation by preventing the 60S ribosomal subunit from joining a 48S pre-initiation complex, should one be poised at the AUG initiation codon.⁷³ Consistent with NMD occurring on mRNAs that may be undergoing translation by more than one 80S ribosome, single-molecule assays to visualize NMD dynamics in real time revealed that NMD is not restricted to the first ribosome that reaches a termination codon.⁷⁴

Interestingly, the efficiency of CBC-bound mRNA translation may predispose the efficiency of subsequent rounds of bulk translation on EIF4E-bound mRNAs, suggesting that EJC-mediated effects on translation efficiency may extend to mRNAs that no longer harbor EJCs.^70,72 As one possible example, the CASC3 speckle-localizer and RNA-binding (SELOR) domain, which is not present in other cellular proteins, directly interacts with the translation initiation factor EIF3 via EIF3A and EIF3D subunits (Fig. 4a), explaining in part why CASC3 depletion reduces global protein synthesis, which is a measure of CBC- and, largely, EIF4E-mediated translation.⁶⁸ As another example, EIF4A3 in the context of the EJC core also interacts with EIF3 via the EIF3G subunit to drive internal translation initiation of particular endogenous circular RNAs (circRNAs) (Fig. 4b).⁷⁵ Whether interactions of CASC3 and EIF4A3 with EIF3 subunits can occur simultaneously is unclear.

Fig. 4 |. EJCs promote translation by interacting with the translation initiation machinery.

a, mRNA translation can be promoted by interactions between CASC3 and EIF3 subunits eIF3a and eIF3d. b, Circular RNA (circ RNA) translation can be promoted by interactions between eIF4A3 and eIF3g. c, The translation of a reporter mRNA that derives from an intron-containing gene can be promoted by interactions between the 40S ribosomal subunit, PYM, and both subunits of RBM8A–MAGOH. d, Upon insulin-mediated phosphorylation of S6K1, S6K1 phosphorylates SKAR, and both phosphorylated proteins can then be recruited to an EJC. This is correlated with the increased phosphorylation of unidentified mRNA-associated proteins at RXRXXS/T consensus sequences targeted by S6K1, AKT or both, and the promotion of mRNA translation.

EJC-dependent translation regulation is also mediated by PYM, which serves a second role as an EJC disassembly factor (see below)⁶⁷, and S6K1 via SKAR. PYM bridges the interaction between the EJC core and the 48S preinitiation complex, which includes EIF3 and the 40S ribosomal subunit (Fig. 4c). This may explain in part why PYM-depleted cells manifest reduced translation of mRNAs derived from intron-containing reporter genes.⁶⁷ Upon insulin-stimulated mTOR signaling, the S6K1 ALYREF-like substrate (SKAR) associates with the EJCs of CBC-bound mRNAs and recruits activated (i.e., mTOR-phosphorylated⁷⁶) S6K1 (Fig. 4d).^77,78 The resulting increased phosphorylation of yet-to-be-identified proteins that constitute CBC-bound mRNAs was attributed to S6K1-mediated phosphorylation.⁷⁸ However, AKT also phosphorylates RXRXXS/T sequences⁷⁹. Given that AKT is now known to stably associate with EJCs in an insulin-promoted manner and directly phosphorylates mTORC1, which then phosphorylates and activates S6K1 (refs. ^7,76,80), it seems likely that the SKAR-mediated increase in the translation of CBC-bound mRNAs involves AKT that may or may not be associated with EJCs.

EJC-mediated mRNA decay pathways

The best-described role for EJCs in mRNA metabolism is in 3′ UTR EJC-dependent NMD, a topic covered extensively elsewhere.^3,81 Briefly, NMD is triggered by an EJC at an exon–exon junction that resides ≥50–55-nts downstream of a termination codon and, therefore, cannot be removed by translating ribosomes, including the terminating ribosome.⁸¹ This is a quality control mechanism that destroys many mutated or mis-spliced mRNAs containing a premature termination codon. Together with long and/or structured 3′UTR-mediated NMD, 3′UTR EJC-dependent NMD is also responsible for regulating the expression levels of ~5–10% of physiologic cellular mRNAs. Targets of 3′UTR EJC-dependent NMD include mRNAs that, e.g., undergo regulated alternative splicing or harbor an upstream open reading frame, should an EJC exist sufficiently far downstream during translation termination of that reading frame.⁸¹ The relatively poorly understood long and/or structured 3′UTR-mediated NMD pathway has been reviewed elsewhere.⁸²

3′UTR EJC-dependent NMD consists of multiple branches that are defined by EJC composition. NMD can depend on the UPF1-activating factor UPF2, which constitutes RNPS1-EJCs. UPF2 activates the key NMD factor UPF1 by binding to the UPF1 cysteine-histidine-rich (CH) domain, which promotes a structural rearrangement that activates UPF1 helicase activity and, therefore, NMD (Fig. 5a).^8,83,84 UPF1 also recruits, either directly or indirectly, cellular nucleases to NMD target mRNAs to facilitate their decay. These include the endonuclease SMG6, which directly binds the UPF1 CH domain in a manner that is mutually exclusive with the binding of UPF2 to this domain⁸⁵ and also contacts the EJC core directly by binding to EIF4A3.⁸⁶ While UPF2-independent NMD was known to exist^87,88, its mechanism was not understood until it was found that AKT, a constituent of CASC3-EJCs, likewise activates UPF1 helicase by phosphorylating the CH domain (Fig. 5b).⁷

Fig. 5 |. Mechanisms by which alternative EJC constituents promote UPF1 helicase activity.

a, In RNPS1-EJCs, UPF2 binds to the UPF1 cysteine-histidine-rich (CH) domain to activate UPF1 helicase activity, thereby promoting NMD. b, In CASC3-EJCs, AKT phosphorylates the UPF1 CH domain to activate UPF1 helicase activity, thereby promoting NMD.

The importance of EJC modularity to NMD is exemplified by the finding that insulin promotes NMD efficiency by increasing the level of AKT-containing EJCs at the expense of UPF2-containing EJCs.⁷ Conceivably, augmentation of NMD could be mediated by CASC3 via its interaction with EIF3.⁶⁸ However, as of this writing, how EJC modularity regulates NMD remains incompletely understood. Notably, downregulating either AKT or UPF2 inhibited 3′ UTR EJC function in NMD for all assayed mRNAs that terminate translation upstream of a single NMD-sensitive exon–exon junction, indicating that the same exon–exon junction in different mRNA molecules can be occupied by either a CASC3-EJC or an RNPS1-EJC.⁷ It is possible that the same exon–exon junction could be first bound by an RNPS1-EJC that is later replaced by a CASC3-EJC, should a precursor–product relationship exist.⁸

The extent to which each peripheral EJC constituent promotes efficient NMD remains unclear. While CASC3-KO Flp-In 293 T-REx cells manifest transcriptome-wide stabilization of hundreds of NMD-target mRNAs⁸⁹, other studies using siRNA-meditated depletion of either CASC3 or RNPS1 in Flp-In 293 T-REx cells demonstrated that, for those NMD-target mRNAs tested, RNPS1 depletion inhibits NMD significantly more than does depletion of CASC3.⁸ Experimental differences, i.e., using KO-cell lines versus siRNA-mediated downregulation, may account for the discrepancies. Nevertheless, the preferential recruitment of one alternative EJC over another to a given mRNA and/or exon–exon junction cannot be ruled out.

Studies of 3′UTR EJC-mediated NMD have focused mostly on EJCs containing UPF3B. However, EJCs can also contain UPF3A that, relative to UPF3B, is encoded by a different gene and binds free UPF2 more efficiently but the EJC less efficiently, thereby sequestering UPF2 from EJCs in a free UPF2–UPF3A complex and reducing NMD efficiency, possibly to fine-tune NMD activity.^90,91 In the absence of UPF3B, UPF3A can support some NMD activity, and some NMD targets are degraded independently of either UPF3B or UPF3A.⁹² These cases may be explained by, e.g., the UPF2- and UPF3B-independent mechanisms involving UPF1 recruitment downstream of a termination codon by the nucleo-cytoplasmic shuttling splicing factor SRSF1 (which may also be UPF3A-dependent, though this was not tested).⁹³ The full scope and mechanistic details of how the two UPF3 paralogs contribute to so-called “branched” NMD pathways remain an area of active investigation.

Mechanisms of EJC disassembly

PYM is both an EJC-associated translation stimulation factor⁶⁷ and an EJC disassembly factor, the latter by promoting EJC removal from spliced mRNAs in the cytoplasm.³⁴ Ribosome-bound PYM, by competing with EIF4A3 for binding to the EJC core proteins RBM8A and MAGOH, dissociates EJC factors from the mRNA and from each other, possibly also preventing EJC reassembly.³⁴ The association of PYM and ribosomes largely restricts PYM-mediated EJC disassembly to mRNAs that are being actively translated.³⁴ Accordingly, an estimated 85% of EJCs are removed dependent on translation.³² This is consistent with the failure of translating ribosomes to remove EJCs that reside at exon–exon junctions situated ≥50–55-nts downstream of a termination codon.³⁴ However, precluding the EJC–PYM interaction by over-expressing a MAGOH variant failed to increase CASC3-EJC footprints, including those residing in CDSs, indicating that these EJCs may be removed in both EJC–PYM-dependent and EJC–PYM-independent mechanisms.⁹⁴ Since PYM interacts with the EJC core, its EJC disassembly activity is unlikely to discriminate between alternative EJCs unless accessibility to the EJC core varies with EJC modularity.

EJC functions during cell signaling, development, and disease

Regulated changes in alternative EJC composition, e.g., in response to insulin signaling⁷, are believed to augment how cells adapt to a changing environment. As another example of signaling-mediated regulation of EJC function, after EJC disassembly in the cytoplasm, the cell cyclin-dependent kinases CDK1/2 phosphorylate EIF4A3.⁹⁵ This inhibits EIF4A3 binding to RNA and other EJC constituents in the cytoplasm and, once EIF4A3 returns to the nucleus, promotes EIF4A3 binding to CWC22.⁹⁵ Thus, EIF4A3 phosphorylation regulates 3′UTR EJC-mediated NMD during the cell cycle, possibly by increasing the efficiency of EJC turnover. Conditional knockout of UPF2 in neural progenitor cells derived from mice results in an abnormally long cell cycle, fewer neural progenitors, and microcephaly, since a subset of cell-cycle regulatory factors are encoded by NMD targets that rely on UPF2.⁹⁶ Thus, peripheral EJC constituents can perform crucial functions during the cell cycle.

The importance of core EJC constituents is illustrated by their essentiality during development. Haploinsufficiency and/or depletion of any of the three core EJC proteins leads to microcephaly in mice, impairs mitosis in cultured human and mouse cells, and increases apoptosis of cortical neuron progenitors and their progeny in primary cell cultures.^97–99 Additionally, EIF4A3 promotes microtubule polymerization in an EJC- and RNA-independent manner during axonal development.¹⁰⁰ Mutations in the EIF4A3 or RBM8A genes that result in abnormally low levels of wild-type protein result, respectively, in Richieri-Costa-Pereira syndrome, which is typified by limb and craniofacial defects,¹⁰¹ or Thrombocytopenia Absent Radius syndrome, which is typified by absent radii and low platelet counts.¹⁰²

Many alternative EJC constituents are also essential, as evidenced by findings that one alternative EJC cannot compensate for the disruption of another, and that alternative EJCs play non-redundant roles. As examples, knockout of the gene encoding UPF2¹⁰³, CASC3¹⁰⁴, RNPS1¹⁰⁵, or UPF3A⁹¹ in mice results in embryonic lethality. Haploinsufficiency of UPF2 results in major neurodevelopmental defects and brain inflammation in mice, and autism spectrum disorders in humans.¹⁰⁶ Additionally, haploinsufficiency of RNPS1 or CASC3 in mice leads to impaired immune cell maturation and survival¹⁰⁵, or to delayed brain development¹⁰⁴, respectively. However, since many cellular proteins can multitask in cells, some of these effects could be due to EJC constituent function independently of the EJC.

Elevated AKT signaling contributes to hyperactivated NMD that characterizes Fragile X Syndrome.⁷ AKT signaling is also hyperactivated in hepatocellular carcinoma (HCC), which was attributed largely to activation of pro-proliferative pathways.¹⁰⁷ As an additional mechanism, HCC may have an abnormally high percentage of EJCs that contain AKT and CASC3, now that we know that EJCs can contain AKT along with CASC3.⁷ Moreover, both CASC3 siRNA and the tumor suppressive microRNA miR-124–1, which also downregulates CASC3 mRNA, reduced HCC-cell proliferation¹⁰⁸, indicating that elevated levels of CASC3 promote HCC-cell proliferation. In fact, alternative EJC dysregulation may contribute to many human diseases in which AKT signaling¹⁰⁹ and/or the transcriptome¹¹⁰ is dysregulated.

Conclusions and future perspectives

While it is now clear that EJCs regulate many aspects of mRNA metabolism to maintain proper gene expression, understanding the regulatory functions of alternative EJCs is an ongoing process. In the nucleus, EJCs regulate mRNA processing such as m⁶A methylation and alternative splicing, the latter of which itself regulates where an EJC is deposited by controlling which introns are removed by splicing. Therefore, EJCs leave lasting records that influence mRNA translation and stability in the cytoplasm, even after EJCs are dissociated from mRNA. Furthermore, EJCs may function as an additional regulatory nexus that links cell signaling pathways with mRNA metabolism. Nevertheless, many aspects of EJC function remain unclear. Not detailed here are reports that EJC constituents can regulate gene transcription. For example, depleting MAGOH in human HeLa cells appears to recapitulate more thorough studies in Drosophila melanogaster S2R+ cells indicating that the EJC core stabilizes RNA polymerase II at promoter-proximal pause sites to prevent premature transcriptional elongation.¹¹¹

The full spectrum of EJC-mediated gene regulation, how cells commit to the formation of one type of EJC over another, how the process of commitment is coupled to EJC formation on the spliceosome, if, how, and when alternative EJC target specificity is dictated, and the functional differences between alternative EJCs are important areas of future study.