Chemical inhibition of exon junction complex assembly impairs mRNA localization and neural stem cells ciliogenesis

Abstract

The exon junction complex (EJC) is formed by the essential eIF4A3, MAGOH, and Y14 core proteins. It is universally deposited during splicing at exon–exon junctions. The EJC is known to impact almost every post-transcriptional regulatory step throughout the life of messenger RNAs (mRNAs) including their modifications, splicing, decay, and trafficking. Its dysregulation leads to neurodevelopmental pathologies. Here, we show that EJC-i, a compound known to block the ATPase activity of eIF4A3, inhibits de novo EJC assembly. EJC-i and targeted knockdown of either eIF4A3 or Y14 core EJC subunits lead to very similar phenotypes by impacting the destiny of mRNAs due to alterations in alternative splicing, nonsense-mediated mRNA decay, genome-wide m6A methylation, and proper localization of specific transcripts, in particular to the centrosome. Both EJC impairment methods disrupt the centrosome function, which might be responsible for mitotic arrest at prometaphase. As a small molecule that readily diffuses into cells, EJC-i is a particularly easy-to-use and versatile tool to investigate EJC functions in live cells or whole organisms that are not prone to genetic manipulation. Indeed, this property was used to disrupt ciliogenesis in primary neural stem cells.

Graphical Abstract

Graphical Abstract.

Introduction

Dysregulation of post-transcriptional control has gained attention due to a large number of pathogenic mutations in ubiquitously expressed RNA binding proteins (RBPs) [1]. These RBPs bind messenger RNAs (mRNAs) from the moment they emerge from RNA polymerase II and all along their life within a cell, and modulate many post-transcriptional regulatory steps leading to functional and mature molecules [2]. Further, the repertoire of RBPs on individual transcripts evolves with each step of an mRNA life, and this dynamic composition of the mRNA ribonucleoparticles (mRNPs) determines their specific pattern of processing, intracellular localization, translation, and decay [2, 3].

Association of proteins with mRNAs is either dictated by defined cis-acting features of the RNA molecules, such as specific sequences or structures, or result from the history of mRNA processing. The exon junction complex (EJC) is a paradigmatic example of this second class of RBPs [4, 5]. The EJC is a multiprotein complex marking mRNAs that have not yet been translated. It is deposited during the splicing reaction, 27 nucleotides upstream of exon–exon junctions as shown by both biochemical and bioinformatic analyses [6–8]. The core of the complex is composed of three proteins: eIF4A3/DDX48, MAGOH, and Y14/RBM8A [9]. It interacts with accessory proteins such as MLN51/CASC3/BARENTSZ or the ASAP and PSAP complexes, which comprise RNPS1, SAP18, and ACIN1 or PNN, respectively. The DEAD-box RNA helicase eIF4A3 clamps the RNA by binding to the sugar-phosphate backbone of mRNA independently of the sequence and is stabilized and locked onto the RNA by the heterodimer MAGOH/Y14 [10, 11]. Recent evidence suggests a universal deposition of EJC [7]. However, a possible specialization of the EJC may occur depending on transcripts and/or exon junctions and accessory protein composition [12–14]. EJCs are disassembled by translation-dependent or translation-independent mechanisms, and their persistence onto mRNA is variable and transcript-specific [15, 16].

Once clamped on mRNA, the EJC core complex acts as a binding platform for several peripheral factors, first in the nucleus and, after mRNA export, in the cytoplasm. Together, these factors are involved in post-transcriptional events, including splicing, RNA methylation, nuclear export, translation initiation, and nonsense mediated decay (NMD) [13, 17–21]. A function of the EJC in the localization of some mRNAs to specific subcellular departments has been solidly established, although limited to two cases. The first example described was the oskar mRNA, which is transported from nurse cells to the posterior pole in Drosophila melanogaster oocytes [22]. More recently, we identified the NIN mRNA, coding for the Ninein core component of centrosomes as a transcript whose localization to centrosomes relies on the EJC in quiescent cells of the Retinitis Pigmentosum Epithelium 1 (RPE1) cell line [23]. Therefore, EJC-dependent RNA localization is present in vertebrates as well. Furthermore, the EJC contributes to excluding the m6A RNA modification around spliced junctions, thereby leading to the typical m6A topologies of enrichment within long internal and last exons [24–27].

The central role of EJC in packaging the evolving mRNPs is reflected by the fact that an altered dosage of its components leads to pathological states such as cancer or developmental defects [28]. In particular, the EJC is essential for neural stem cells (NSCs) division, differentiation, and brain development [28, 29].

In order to understand the mechanisms by which the EJC participates in the wide array of regulatory steps it has been linked to, the ability to modulate its activity becomes a key issue. However, akin to other key cellular regulatory factors, all three core subunits of the EJC are encoded by essential genes [30], therefore precluding their knockout in cells. So far, blocking EJC activity relies on the core EJC proteins knockdown (KD), either using small interfering RNAs (siRNAs) [17, 18, 21, 23, 31–35], or using dTAG-mediated degradation of the EJC core subunit Y14 in CRISPR/Cas9 edited cell lines [25]. However, KD approaches have important limitations, including slow or poorly controlled kinetics and their difficult application to in vivo systems.

Chemical inhibitors are particularly attractive because they diffuse rapidly into cells, they can be used in different cell types, and potentially, in whole organisms. A 1,4-diacylpiperazine derivative (which we will refer to from now on as EJC-i for EJC-inhibitor) has recently been identified as a selective inhibitor of eIF4A3 ATPase activity [36]. It has been shown to alter gene expression, NMD, and the cell cycle [34, 37, 38]. However, its mechanism of action remains poorly understood. In this study, we show that EJC-i blocks de novo EJC assembly both in vitro and in vivo, impairing a wide spectrum of established EJC-dependent functions. We exploit the extreme flexibility and ease of application of the inhibitor to identify novel EJC-dependently localized mRNAs and to strengthen EJC implication in centrosome function and ciliogenesis in human cells.

Materials and methods

Cell culture and treatments

All cells were maintained in Dulbecco’s modified Eagle’s medium high glucose (Dutscher) supplemented with 10% of fetal bovine serum (Sigma–Aldrich) and 1% of penicillin and streptomycin (Sigma–Aldrich) at 37°C. Cell lines used in this study were derivative of HEK293T or HeLa. HEK293T Y14-HA-dTAG cells have been described [25], and HEK293T eIF4A3-HA-dTAG cells have been obtained following the same procedure. OB9 cells co-expressing MAGOH-LgBiT and FLAG-eIF4A3-SmBiT have been described [15]. HeLa cells expressing Centrin1-GFP have been previously described [39]. Cells were cultivated at 70% confluency and treated with DMSO (Dimethyl sulfoxide) 1:500, or the indicated dose (see “Results” section) ranging from 1 to 20 µM EJC-i (MedChemExpress: eIF4A3-IN-2, Cat. No.: HY-101785; CAS No.: 2095677-20-4) in DMSO, or 50 nM dTAG13 or dTAG1 (Sigma–Aldrich) in DMSO, or 100 μg/ml cycloheximide (TOKU-E) in DMSO, and harvested after the indicated time post-treatment.

Luciferase assays

These assays were performed essentially as described [15]. One hundred nanograms of furimazine (1 mg/ml stock solution in 50:50 ethanol/propylene glycol) in 30 μl of 0.5× Passive Lysis Buffer (Promega #E1941) per assay were added to 50 μl of lysates or cell suspensions. The luciferase activity was detected with a Berthold TriStar LB941 luminometer. For kinetic experiments, OB9 cells were seeded on polylysine-coated 12-well plates, and EJC-i was added at defined times prior to lysis. Incubation was arrested on ice, the medium was rapidly sucked off, and cells were scraped in 400 μl ice-chilled 0.5× Passive Lysis Buffer per well. Each measure is an average for three wells (technical replicates) treated independently. Three wells per plate were exposed to DMSO 1:500, and their average luciferase activity was used to normalize the data for drug-treated cells on the same plate. Standard deviations are determined from biological replicates obtained on different days.

Western blot and co-immunoprecipitation

For coimmunoprecipitation, 10 cm plates of confluent cells were collected and washed in 1 ml phosphate buffered saline (PBS). Cell were lysed for 15 min at 4°C in 300 μl of lysis buffer (20 mM Tris–HCl, pH 7.5, 150 mM NaCl, 1 mM Na2EDTA, 1 mM EGTA (ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid), 1% NP40, 1% sodium deoxycholate, 1% proteaseinhibitor mixture), with 10 U RNase T1 (Fermentas) and 12 U RQ1 DNase (Promega). Cellular debris was removed by centrifugation at 13 000 rpm for 10 min. The lysates were incubated with 20 μl protein A-coupled Dynabeads (Life Technologies) with or without antibody for 2 h at 4°C. The beads were washed extensively with IP buffer 150 (10 mM Tris–HCl, pH 7.5, 150 mM NaCl, 2.5 mM MgCl₂, 1% NP40, 1% proteaseinhibitor mixture), and the bound proteins were eluted with Laemmli buffer. Total proteins, input proteins, or eluted proteins were resolved by 4%–12% sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS–PAGE) and electrotransferred to nitrocellulose membrane (Schleicher & Schuell). Membranes were blocked in PBS containing 5% nonfat dry milk and 0.05% Tween 20. Rabbit polyclonal antiGAPDH, antiY14, antiMAGOH, and antieIF4A3 were used as primary antibodies at 1:1000 dilution. After washing, membranes were incubated with secondary antibodies: either goat anti-rabbitHRP or Clean-Blot™ IP Detection Reagent (HRP) (Thermo Fisher). Protein–antibody complexes were visualized by an enhanced chemiluminescence detection system (SuperSignal West Pico Plus, Thermo Scientific).

EJC reconstitution assay

These assays were performed essentially as described [40]. Proteins (2 µg of each) were mixed in binding buffer (BB-125) containing 20 mM HEPES (pH 7.5), 125 mM NaCl, 1 mM magnesium diacetate, 1 mM DTT (Dithiothreitol), 5% (v/v) glycerol, and 0.1% (w/v) NP-40, complemented or not with 2 mM ADPNP (5'-adenylyl beta,gamma-imidodiphosphate), 1–20 µM EJC-i, or 1.6% DMSO, and 0.5 µM biotinylated ssRNA (single stranded RNA) in a final volume of 60 µl. After 20 min at 30°C, 5 µl of magnetic streptavidin beads (MyOne, Dynal) and 140 µl of BB-250 (250 mM NaCl) were added. The setup was similar for calmodulin pulldown assays with the exception that buffers contained 2 mM CaCl2, ssRNA and ADPNP were excluded, and after incubation mixtures were added 12 µl of calmodulin resin (50% slurry, Agilent Technologies). After rotation for 1 h at 4°C, beads were washed three times with 500 µl BB-250 and then eluted with Laemmli buffer, boiled, and loaded on 4%–12% SDS–PAGE along with inputs and with a protein marker (broad range, New England Biolabs). Proteins were visualized by Coomassie staining.

RNA extraction, reverse transcription, end-point and quantitative RT-PCR

Total RNA was extracted from cells using TRI reagent (Ambion) according to manufacturer’s protocol. The RNA was digested with 5 U RQ1 DNase (Promega) for 1 h at 37°C before phenol extraction and precipitation. Reverse transcription was performed using 1 µg RNA with random hexamers and Superscript IV reverse transcriptase (Invitrogen) according to manufacturer’s protocol. After heat inactivation, samples were treated with 5 U RNase H for 30 min at 37°C. For end-point PCRs, 1% complementary DNA (cDNA) was used as template using GoTaq G2 Flexi (Promega) and 0.2 µM final concentration of sense and antisense primer. After 30 PCR cycles, the samples were resolved by electrophoresis on ethidium bromide-stained, 2% agarose TBE gels and visualized by trans-UV illumination. For qPCRs, 1% cDNA was used as template using PowerUP SYBR Green Master Mix (Applied Biosystems) on CFX384 Real time system (Bio-Rad). Quantification was performed using the ∆∆Ct method with RNA levels normalized by housekeeping gene GAPDH. Controls without reverse transcriptase were systematically run in parallel to estimate the contribution of contaminating DNA. Amplification efficiencies were calculated for every primer pair in each amplification reaction.

m6A-seq library preparation

Whole cell RNA from the samples in the EJC-i time course experiment was poly-A selected using oligo dT-beads (Dynabeads mRNA DIRECT Kit). A total of 200 ng of poly-A selected mRNA per sample was used to prepare m6A-seq2 libraries following the step-by-step protocol, as previously described [41]. The m6A-IP and Input NGS libraries were sequenced paired-end on the NovaSeq 6000 platform.

m6A-seq2 data processing and analysis

Sequencing reads were aligned to the hg19 human reference genome using STAR/2.7.9a. The BAM alignment files were filtered to include only unique alignments, and gene coverage was extracted using the bam2Endreads R script [41] based on the canonical hg19 UCSC gene annotation. m6A gene-level (m6A-GI) estimation was performed as described [41]. Meta-analysis of m6A coverage for long internal exons (excluding the first or last exon and those >600 nucleotides) and last exons (for genes with >1 exon and last exons >400 nucleotides) was conducted. For the analysis, m6A enrichment scores (m6A-IP/Input coverage) were calculated for each gene with sufficient coverage in 20-nt windows. The median m6A enrichment score for each sample was analyzed, and the mean of the replicates was normalized by the mean of the control timepoint (0 h) to calculate the meta m6A enrichment score.

De novo m6A peak calling

De novo m6A peak calling was performed for each sample as described [42]. From all detected peaks, high-confidence m6A sites were defined as de novo m6A peaks in which the summit was within five bases of the nearest adenosine in a DRACH-motif context. m6A site score calculation (m6A-IP/Input coverage) was performed for the detected high-confidence m6A sites across all samples.

Alternative splicing analysis

Alternative splicing was quantified with rMATS-turbo v4.1 [43]. For each analysis batch, rMATS was supplied a comma-separated BAM list via –b1 and the GENCODE v19 GTF (hg19). Runs used paired-end, stranded settings (–libType fr-secondstrand, -t paired, –readLength 50, –variable-read-length, –allow-clipping, –statoff; defaults otherwise). For downstream statistics, shared percent-spliced-in (PSI) values [44] were merged, and ΔPSI was computed relative to the reference time point.

Quantification and statistical analysis

All statistical analyses were performed using R 4.1.0. Details of the analyses are provided in the figure legends and result sections. Figures were prepared using basic R packages and ggplot2 [45].

Flow cytometry analysis

Cells were trypsinized and washed in D-PBS. Following the last centrifugation, cells were fixed and permeabilized by resuspension in cold 70% EtOH. They were incubated for 48 h at −20°C. Cells were washed in D-PBS and counted. Equal number of cells per condition were labeled by 10 μg/ml of HOECHST 33342 in the presence of 100 μg/ml RNase A at 4°C for 48 h. After HOECHST 33342 staining, samples were analyzed on a ZE5 cell analyzer (Bio-Rad) using Everest acquisition software (Version 2.5). Data were analyzed using FlowJo software (FlowJo, LLC, version 10.8.1). Briefly, cells were selected based on their morphology (SSC-A versus FSC-A: “Cells” population), then doublets of cells were excluded based on DNA content parameters (Hoechst-A versus Hoechst-W: “Single cells” population). Cell cycle analysis was performed on “Single cells” population using FlowJo module for cell cycle.

smFISH

Probe sets of DNA oligonucleotides targeting the mRNAs of interest were designed based on the Oligostan script[46] and used as previously described [46]. RNA probes were obtained as described previously [39]. For RNA probes, 50 ng of unlabeled RNA probes and 25 ng of each fluorescent secondary probes (TYE-650-labeled LNA oligonucleotides, Qiagen) were prehybridized in 100 µl of a solution containing 7.5 M urea (Sigma–Aldrich), 0.34 mg/ml tRNA (Sigma–Aldrich), and 10% dextran sulfate. The prehybridization was performed in a thermocycler with the following program: 90°C for 3 min, 53°C for 15 min, forming fluorescent duplexes. Cells were grown in a 96-well plate with glass bottom (PhenoPlate, Revvity), fixed with 4% PFA (Paraformaldehyde) at RT (room temperature) and permeabilized with 70% ethanol overnight at 4°C. They were washed with PBS and incubated 15 min at RT in the hybridization buffer (1× SSC, 7.5 M urea) before the addition of the 100 µl of fluorescent duplexes in each well. Hybridization was performed at 48°C overnight. The next day, the plate was washed eight times for 20 min each in 1× SSC and 7.5 M urea at 48°C. Finally, cells were washed three times with PBS at RT for 10 min each, stained with 5 µg/ml DAPI (4′,6-diamidino-2-phenylindole), and mounted in 90% glycerol (VWR), 1 mg/ml p-Phenylenediamine (Sigma–Aldrich), PBS pH 8.

Imaging

smFISH imaging for HeLa Centrin-GFP cells was performed on an Opera Phenic High-Content Screening System (Revvity), with a 63× water-immersion objective (NA 1.15).

The imaging on SunTag cell lines was performed on a Leica thunder with a 63× oil-immersion objective (NA 1.4). Three-dimensional images were acquired with a z-spacing of 0.6 µm.

Image analysis

Quantifications were based on an image analysis pipeline including cell segmentation, threshold-based spot detection, dense region deconvolution, and density-based clustering algorithm (DBSCAN). First, nuclei and cytoplasm segmentation were performed in 2D on image stack mean projection using neural network Cellpose (https://github.com/MouseLand/cellpose) [47]. Second, single molecules are detected as 3D spots using the Big-FISH package [39] (https://github.com/fish-quant/big-fish), which relies on an intensity Laplacian of Gaussian filter followed with a local maxima filter and a user-adjusted thresholding. During analysis, we were careful that same thresholds were used on stacks targeting the same mRNA. As local maxima detection can lead to a miss estimation of single-molecule number in bright region a dense region deconvolution step is added [39]. It consists in fitting the median detected spot with a 3D Gaussian to construct a reference spot. Then an estimation of single-molecule number is done by reconstructing the bright region signal using the reference spot. Third, a DBSCAN clustering algorithm is used to quantify foci. Finally, Big-FISH spot detection was used with the Centrin-GFP channel to detect centrosomes. The brightest candidate spot was considered as the centrosome, and we allowed cells to have a second centrosome if its intensity was lying in a 10% range of the first centrosome. Normalized distances from the detected RNA spot to centrosome were then computed. In each cell, distances from the RNA spots to the closest centrosome were normalized using distances from a random RNA distribution. A distance inferior to 1 indicates centrosomal enrichment, whereas a distance superior to one indicates anti-colocalization. The same calculation was done for the normalized distance to the cell membrane. Quantification of mRNAs or polysomes in SunTag cell lines and evaluation of the proportion of mRNA in the nucleus was performed using the SmallFISH pipeline (https://github.com/2Echoes/small_fish_gui), which provides a ready-to-use graphical interface to combine Python packages for cell analysis. Within the pipeline, 2D cell segmentation was performed using Cellpose, while spot detection was carried out using Big-FISH. Boxplots graphs were constructed on Pyhton using the Seaborn and Statannotations packages.

Results

EJC-i decreases the amounts of EJC in live cells

Previous studies showed that EJC-i inhibits eIF4A3 ATPase and helicase activities in vitro [34, 36] and impairs eIF4A3-mediated NMD in vivo [34, 37]. However, the effect of EJC-i on eIF4A3’s ability to interact with EJC core components and form an EJC could not be assessed [34]. To test the impact of EJC-i on EJC assembly in live cells, we coimmunoprecipitated EJC components from HEK293T cells treated with 20 µM for 4 h, a non-toxic dose of EJC-i. Western blot analysis showed that EJC-i treatment reproducibly resulted in a reduction of the amount of both Y14 and MAGOH EJC subunits coimmunoprecipitated with eIF4A3 (Fig. 1a, compare lanes 8 and 9). Because inhibition of EJC could derive from destabilization of eIF4A3, we monitored the steady-state levels of this core component of the complex after incubation for different times up to 24 h in wild-type HEK293T cells. Twenty micromolar EJC-i did not alter eIF4A3 concentration even after long exposure to the drug (Supplementary Fig. 1a).

Figure 1.

EJC-i decreases the amounts of EJC in live cells.(a) Representative western blot analysis of EJC proteins coimmunoprecipitated (IP) with anti-eIF4A3 from protein extracts of HEK293T cells treated with either DMSO or a 20 µM dose of EJC-i for 4 h. Lanes labeled “-” represent no antibody controls. FT is the flow-through fraction. Amounts of Y14 (52 ± 20%) and MAGOH (65 ± 10%) in EJC-i-treated samples were measured in three independent experiments. (b) Scheme depicting the EJC-NanoBiT system. (c) Response of a stable HEK293T cell line co-expressing MAGOH-LgBiT and FLAG-eIF4A3-SmBiT to a 2-h incubation with the indicated µM doses of EJC-i expressed as relative luciferase units (RLUs). Means and standard deviations from three independent experiments. (d) Response kinetics measured as RLUs of the same cell line as in panel (c) incubated with a 20 µM dose of EJC-i for the indicated times (minutes). Means and standard deviations from three independent experiments.

As western blots are poorly quantitative, we turned to a highly quantitative split-luciferase assay that we have recently established to monitor assembly and disassembly rates of the EJC in live cells [15]. This assay uses a stable cell line co-expressing MAGOH-LgBiT and FLAG-eIF4A3-SmBiT proteins where N-terminal (LgBiT) and C-terminal (SmBiT) moieties of nanoluciferase are fused to MAGOH and eIF4A3, respectively. Assembly of EJCs reconstitutes nanoluciferase (NanoBiT) activity [15] (Fig. 1b). The split-luciferase cell line was incubated for 2 h with different doses of EJC-i or DMSO as control. EJC-i significantly reduced the observed luciferase activity, reflecting a decrease in assembled EJC down to a 20% plateau between 10 and 20 µM concentrations (Fig. 1c). The effect of EJC-i was rapid and reached 50% inhibition within the first hour of incubation for 20 µM concentration (Fig. 1d). Since the split-luciferase assays suggested a significant reduction in the amount of assembled EJC following EJC-i treatment, a Western blot analysis was performed on the cell line co-expressing MAGOH-LgBiT and FLAG-eIF4A3-SmBiT. Following immunoprecipitation with an anti-FLAG antibody and Western blot analysis of EJC components, reduction in the coimmunoprecipitated fraction of Y14 and both endogenous and LgBiT-tagged MAGOH proteins was observed (Supplementary Fig. 1b, compare lanes 8 and 9).

Finally, we used the EJC-NanoBiT system to test whether the effects of EJC-i are reversible. The split-luciferase cells were incubated for 2 h with a 20 µM concentration of EJC-i and allowed to recover for variable times in fresh medium in the presence or absence of cycloheximide. The luciferase activity increased slowly over time, and the increase was two-fold faster in cycloheximide-containing medium (Supplementary Fig. 1c). Thus, EJC-i effects are reversible. Consistent with our previous findings [15], the results in the presence of cycloheximide indicate both the significant contribution of translation-dependent EJC disassembly and that reversibility does not require protein neosynthesis.

Altogether, these results show that an EJC-i treatment decreases the amounts of assembled EJC in live cells.

EJC-i inhibits de novo EJC assembly

In order to test more directly the effect of EJC-i on the EJC, we used an established in vitro EJC assembly assay [40]. Briefly, recombinant EJC proteins (eIF4A3, MAGOH/Y14, and the Selor domain of MLN51) were first mixed with a 3′-end biotinylated ssRNA and the non-hydrolyzable ATP analog ADPNP before RNA precipitation with streptavidin beads and analysis of RNA-bound proteins on SDS–PAGE [40]. In vitro EJC reconstitutions were performed in the presence or absence of EJC-i, or DMSO as control (Fig. 2a). The complex was formed and the proteins efficiently coprecipitated only when ADPNP was included in the incubation (Fig. 2b, compare lanes 7 and 8). Remarkably, incubation in the presence of 20 µM EJC-i resulted in complete loss of protein coprecipitation, indicative of inhibition of complex assembly (Fig. 2b, compare lanes 9 and 10). In vitro, a stable binding of the EJC core to RNA requires its four protein components [40]. Although much less stable, the interaction of eIF4A3 with RNA can be detected alone or in the presence of one of its partners. Thus, we also tested the effect of EJC-i on eIF4A3 individual interactions with the EJC core components separately. Whether with RNA, MLN51-Selor, or the MAGOH/Y14 heterodimer, incubation in the presence of EJC-i always impaired eIF4A3 binding to RNA (Supplementary Fig. 2a–c). Finally, we tested decreasing concentrations of EJC-i. Protein coprecipitation progressively increased with lowering the amount of inhibitor from 20 to 1 µM, indicating that inhibition of EJC assembly was dose-dependent, reaching its maximum at 20 µM (Fig. 2c). To assess whether EJC-i impacts the stability of already formed EJC, EJC-i or DMSO was added to preassembled EJC coprecipitated on streptavidin beads. Notably, in this case, no effect of EJC-i was observed (Fig. 2b, compare lanes 11 and 12).

Figure 2.

EJC-i inhibits de novo EJC assembly. (a) Scheme depicting the EJC reconstitution assay. (b) Protein coprecipitation with biotinylated ssRNA. Recombinant eIF4A3, MAGOH, Y14, and Selor were mixed with 3′-end biotinylated 30-mer ssRNA and incubated with or without ADPNP, in the presence or absence of 20 µM EJC-i or DMSO before or after coprecipitation, as indicated. Proteins from input (16% of total) and precipitate following denaturing elution were separated on a 4%–12% (w/v) acrylamide SDS–PAGE in parallel with a protein marker. Protein molecular weights (kD) are indicated on the left, identity of proteins on the right. Gel was stained with Coomassie blue. (c) Same as (b) but incubation before coprecipitation was with decreasing doses of EJC-i (20 µM lane 4, 10 µM lane 5, 5 µM lane 6, 2 µM lane 7, 1 µM lane 8).

These reconstitution experiments clearly revealed that EJC-i blocks EJC assembly, but not in vitro pre-assembled EJCs. In order to confirm this observation, we tested the effect of the inhibitor on cell lysates that contain EJCs assembled in live cells [15]. Cellular lysates from EJC split-luciferase–expressing cells were incubated for 2 h with different doses of EJC-i or DMSO as control. Even at the highest concentration of 20 µM, we did not observe any effect of EJC-i on luciferase activity (Supplementary Fig. 2d). These results together strongly indicate that EJC-i efficiently inhibits de novo EJC assembly without affecting the stability of pre-assembled complexes.

As eIF4A3 also binds to CWC22 to initiate assembly of the EJC within the spliceosome [31, 48, 49], we tested whether EJC-i affects this interaction. We first used an in vitro assay with recombinant eIF4A3 and two different CWC22 fragments tagged with Nterminal calmodulinbinding peptides (CBP) [31, 40, 48, 49]. Both CBP-CWC22_100-665 and CBP-CWC22_117–406 were previously shown to interact with eIF4A3 [31, 48], with the shorter one, encompassing only the essential eIF4A3-interacting region, displaying the highest affinity. Interaction between eIF4A3 and either CBP-CWC22 fragment was not affected by EJC-i (Supplementary Fig. 3a, compare lanes 3–4 and 8–9), even under stringent saline conditions (Supplementary Fig. 3a, compare lanes 5–6 and 10–11). We also tested the eIF4A3-CWC22 interaction in vivo by reciprocal co-immunoprecipitations from HEK293T cells treated with 20 µM EJC-i for 4 h, or DMSO as control. Anti-CWC22 antibodies precipitated eIF4A3 and anti-eIF4A3 antibodies precipitated CWC22 without any notable effect due to EJC-i treatment, and both expectedly precipitated CWC27, CWC22 major interaction partner [49] (Supplementary Fig. 3b, compare lanes 10–11 and 13–14). Therefore, EJC-i only inhibits EJC core protein interactions without affecting the formation of the eIF4A3-CWC22 complex involved in eIF4A3 recruitment by spliceosomes.

EJC-i affects alternative splicing and nonsense-mediated mRNA decay

Recently, we edited cell lines with CRISPR/Cas9 to fuse a dTAG degron [50] to the EJC core subunits Y14 [25] and eIF4A3 (this work). An efficient EJC subunit depletion was obtained within a 24-h time course after addition of 50 nM dTAG [25] (Supplementary Fig. 4). However, both the siRNA and degron knockdown approaches target a single protein and do not necessarily reflect the behavior of the entire complex. Furthermore, several hours to days of treatment are required. To assess the extent of inhibition of EJC functions in vivo, we monitored two established EJC-dependent functions, splicing regulation and NMD, looking at known target genes regulated by the activity of the complex, and compared HEK293T cells treated with a 20 µM of EJC-i for 24 h to HEK293T cells carrying either the Y14 or the eIF4A3 dTAG degron depleted for 16 h.

The EJC impacts some splicing events, notably by contributing to the recognition of neighboring introns and by suppressing cryptic splice site usage [17, 21, 51–53]. Through interaction with peripheral factors, including the ASAP and PSAP complexes, the EJC modulates different splicing choices, and depletion of its core components or of any partners results most often in the exclusion of cassette exons. We analyzed transcripts originating from the MRPL3, SMARCB1, GLRX3, SDHA, and HERC4 genes, which generate mis-spliced forms when individual EJC core proteins are downregulated [17, 21, 52]. For all five genes tested, loss of EJC activity resulted in exon skipping events leading to accumulation of mis-spliced forms, both upon Y14 or eIF4A3 dTAG downregulation and upon EJC-i treatment (Fig. 3a–e, compare lanes 1–3–5 to 2–4–6).

Figure 3.

EJC-i affects alternative splicing and nonsense-mediated mRNA decay. (a) Endpoint RT-PCR analysis of MRPL3 exon 4 skipping with RNA from HEK293T wild-type cells (lanes 1–2) or stably expressing a Y14 (lanes 3–4) or eIF4A3 (lanes 5–6) dTAG degron treated with either DMSO, 20 µM EJC-i for 24 h, or dTAG for 16 h, as indicated. cDNA was analyzed on 2% agarose gels stained with EtBr along with a DNA ladder. Sizes (bp) are indicated on the left. MRPL3 exon architecture and alternatively spliced isoforms are schematized on the right. (b) Same as panel (a) for SMARCB1 exon 4 mis-splicing. (c) Same as panel (a) for GLRX3 exon 3 skipping. (d) Same as panel (a) for SDHA exon 3 and 4 skipping. (e) Same as panel (a) for HERC4 exon 24 skipping. (f) RT-qPCR analysis of the PTB2 and HNRNPL mRNAs and their NMD-sensitive isoforms from HEK293T wild-type cells treated with either DMSO or 20 µM EJC-i for 24 h. Means and standard deviations from three independent experiments. (g) Same as panel (f) with RNA from HEK293T stably expressing a Y14 or eIF4A3 dTAG degron treated with either DMSO or dTAG for 16 h.

We next analyzed the effect of EJC-i on the control of NMD, for which the complex was originally identified [54–56]. To evaluate the impact on EJC-dependent NMD activity, we performed RT-qPCR analysis on two specific genes, PTBP2 and HNRNPL, that generate through splicing both stable mRNAs and unstable isoforms containing premature termination codons and are therefore targeted by degradation through NMD [57]. Following either EJC-i treatment (Fig. 3f) or dTAG depletion (Fig. 3g), the ensuing loss of EJC activity resulted in clear stabilization of NMD-sensitive isoforms. These results together show that chemical inhibition of EJC assembly and degron-mediated depletion of single complex core components have comparable effects on EJC-dependent regulation of alternative splicing and NMD.

To globally assess the effects of EJC inhibition on splicing and/or NMD-specific transcript isoforms, we compared RNA-seq data following either a time course of dTAG depletion of Y14 [25] or an EJC-i treatment (see below). We analyzed changes in gene expression patterns across 10 704 poly(A)-harboring transcripts shared between studies. A principal component analysis (PCA) was performed first to look at changes in the transcriptional landscape. An unbiased principal component analysis revealed that the first principal component (PC1), capturing 26% of the variability in the data, correlated perfectly with time following EJC inhibition, independently of the modality used for inhibiting the EJC (EJC-i versus dTAG), implying that the dominant expression changes in both datasets arise from a common mechanism (Supplementary Fig. 5a). Although the effects are overall similar, the inhibitor appear to act slightly faster, consistent with its direct effect in EJC assembly.

We then sought to assess whether the two forms of inhibition lead to similar impacts on alternative splicing, quantified across both datasets via rMATS-turbo [43]. This analysis yielded a total of 23 757 alternative splicing events across all datasets, for which we obtained PSI [44] values. A principal component analysis of the scaled PSI matrix again placed samples along PC1 in a time-ordered series irrespective of perturbation type (Supplementary Fig. 5b), again suggestive of a common mechanism giving rise to the key differences between the samples. Consistently, a direct comparison of changes in PSI values (∆PSI) in the dTAG versus EJC-i datasets revealed that these were correlated, more so during advanced timepoints, and in particular among intron-retention events (Supplementary Fig. 5c).

To assess global NMD effects we exploited a recently refined set of 147 high-confidence NMD-sensitive transcripts [58]. Comparison of changes in log₂-scaled normalized expression along the time course of Y14 depletion or EJC-i treatment showed that both perturbations resulted in a time-dependent up-regulation of these canonical NMD substrates with an earlier response for EJC-i-treated samples, whereas the bulk transcriptome showed no systematic trend (Supplementary Fig. 5d). Altogether, these results support a global and specific effect of EJC-i.

EJC-i prevents EJC-dependent exclusion of m6A deposition around splice junctions

We and others have recently unveiled a global function of the EJC in preventing m6A RNA modification around splice junctions [24–27]. Using HEK293T cells carrying the Y14 dTAG degron, we have previously shown that Y14 depletion leads to transcriptome-wide increase of m6A within these EJC-mediated m6A exclusion zones [25]. The physical presence of the complex may occlude methylation sites either by steric hindrance or generating an unfavorable mRNP architecture.

To demonstrate that these effects are indeed due to assembled EJCs, we performed an EJC-i 20 µM treatment over a 16-h time-course followed by transcriptome-wide m6A detection and quantification via m6A-seq2 [41], similar to our previous analysis with the Y14 dTAG degron [25]. First, we calculated m6A levels at the gene level, using the previously established m6A gene-index (m6A-GI) [41]. An unbiased principal component analysis of the m6A-GIs over time revealed that the first principle component (PC1) captured a clear, continuous increase over time following inhibition, establishing that the key component underlying changes in m6A levels across samples was time following inhibition of the EJC (Fig. 4a). Inspection of m6A-GI values across genes, binned as a function of whether or not they were single-exon or multi-exon genes, revealed a continuous increase over time in multi-exon genes, but limited changes in single-exon genes (Fig. 4b), consistent with the notion that the loss of the EJC in multi-exon genes is driving the increase in m6A, and consistent also with our previous observations using dTAG mediated EJC depletion [25].

Figure 4.

EJC-i prevents EJC-dependent exclusion of m6A deposition around splice junctions. (a) Barplot showing PC1 of principal component analysis of the measured m6A gene level (m6A-GIs) based on an m6A-seq2 experiment of an EJC-i time course experiment in HEK293T cells on two independent replicates (r1 and r2). (b) Heatmap showing the row-scaled m6A gene index (m6A-GIs) of all the samples of the EJC-i inhibition dataset for multiple exon genes (annotated with > one annotated exon, genes are ordered from lowest number of exons to highest, top) and single-exon genes (annotated with only a single annotated exon, bottom). (c) Meta fold-change of m6A enrichment (m6A-IP/Input) of timepoints after treatment of HEK293T cells with 20 µM EJC-i (2, 4, 8, and 16 h) compared to timepoint 0h. Analysis was performed for 300 bases of the 5′ beginning of long internal exons (>600 nt, left), for the last 240 bases towards the 3′ end (center), and for 300 nt from the beginning into the last exon (right). Calculated fold changes are based on the median m6A-IP/input score for 20-nt-long bins per time point. (d) Heatmap of row-scaled m6A site scores. High confidence m6A-sites are based on de novo peak calling (see “Materials and methods” section) of all EJC-i treatment and control samples. m6A-sites were ordered by the distance to the closest exon-exon junction.

As EJC-i treatment inhibits de novo EJC assembly, its effects on m6A accumulation are expected to be predominantly observed with newly synthesized short-lived mRNAs. We tested this prediction by binning mRNAs into six equally-sized half-life classes [59] and monitoring the median fold-changes of m6A gene levels for each stability bin over time. Indeed, shorter half-life mRNAs (bins 1–3; Supplementary Fig. 6) displayed a faster increase in m6A, especially during the early time points of treatment.

We next performed a meta-gene analysis, seeking to identify where - within genes - m6A enrichment became apparent following loss of EJC-i. This analysis revealed that the increase in m6A was predominantly present in the ‘exclusion zones’ [25], defined as a region of up to 200 nucleotides around the exon-exon junction at which m6A is typically suppressed (Fig. 4c). As a last step, we performed de novo m6A peak calling based on all EJC-i-treated samples and identified 31 000 high confidence m6A sites (see “Material and methods” section). We calculated m6A-site scores across all samples and found that sites located in the EJC exclusion zones showed a significant increase in m6A methylation enrichment over the time course compared to other sites (Fig. 4d). Overall, EJC-i treatment leads to a transcriptome-wide increase in m6A methylation and recapitulates observations obtained with the Y14 dTAG degron [25]. This reinforces the notion that impairing complex assembly specifically and effectively affects EJC-dependent m6A events on a global scale and provides additional validation to the utility of EJC-i in facilitating functional depletion of the EJC.

EJC-i promotes a mitotic arrest at prometaphase

EJC-i had been shown to arrest cells in G2M and to promote apoptosis [37]. HEK293T cells stably expressing the Y14 dTAG degron were either depleted of Y14 through treatment with dTAG or had EJC assembly blocked by treatment with 10 µM EJC-i over a 24-h time course. Cell cycle profiles were assessed by flow cytometry after treatment and compared to DMSO control. Prolonged inhibition of EJC function by either method resulted in a clear cell cycle arrest at the G2/M boundary to a comparable extent, with 37.3% cells in G2/M following dTAG depletion of Y14 and 26.7% after EJC-i block (Fig. 5a). Microscopic inspection of cells revealed that the arrest due to interfering with EJC stability corresponded to cells arrested in prometaphase (Fig. 5b). Therefore, targeted knockdown of Y14 core EJC subunit and EJC-i similarly result in mitotic arrest of cell cycle at prometaphase.

Figure 5.

EJC-i promotes a mitotic arrest at prometaphase. (a) Cell cycle profiles (counts on vertical axes versus DNA content on horizontal axes) were assessed by flow cytometry after treatment of HEK293T cells stably expressing the Y14 dTAG degron with DMSO, dTAG, or 10 μM EJC-i for 24 h. The percentage of cells assigned to the respective stages of the cell cycle is indicated on the respective charts on the right. (b) Representative DAPI staining of the same cells as in panel (a) with fraction of cells assigned to the respective stages of the cell cycle based on morphological inspection and quantified on the respective charts on the right. Two hundred fifty cells per condition were analyzed in three independent experiments. Dotted lines encircle cells arrested in prometaphase.

While the mechanistic explanation of the observed cell cycle arrest is well beyond the scope of this study, we note however that there is an intimate connection between the EJC and centrosomal function (this study) [23]. In a cascade of regulatory events leading to centrosome biogenesis, the SAS-6 protein is a central scaffolding component of the centrioles ensuring their nine-fold symmetry and is required for centrosome function and their duplication [60]. Strikingly, when HEK293T cells stably expressing the Y14 dTAG degron were either treated with the dTAG over a 24-h time course or with increasing doses of EJC-i for 16 or 24 h, a dramatic loss of SAS-6 were observed at late time points of the Y14 depletion and EJC inhibition with peak effect at a dose of 10 µM EJC-i (Supplementary Fig. 7). This observation constitutes an additional evidence that the EJC is critical for centrosome structure and functions.

EJC-i prevents proper intracellular localization of mRNAs

In human cells, NIN mRNAs localization to centrosomes depends both on translation and the EJC, and siRNA-mediated EJC downregulation impairs pericentriolar material organization [23]. A series of transcripts are specifically transported to centrosomes at different stages of the cell cycle [23, 39, 61–63]. To explore in more detail the role of the EJC in their localization, we tested the effect of 10 μM EJC-i treatment for 16 h. First, we looked at NIN mRNAs, using a HeLa cell line stably expressing a GFP tagged version of Centrin1 to label centrosomes (Fig. 6). As previously reported, NIN mRNA decorated centrosomes during interphase and mitosis, but this localization was dramatically lost upon inhibition of EJC assembly by EJC-i (Fig. 6a). We previously identified ASPM and NUMA1 mRNAs among a set of 8 mRNAs localized at the centrosomes in HeLa cells, with ASPM mRNA localizing to centrosome specifically during mitosis [39, 61–63]. Both mRNAs displayed a clear localization to the centrosomes, which was abolished after treatment with EJC-i (Fig. 6b–d). In addition, the same effect was observed for PCNT mRNAs, which also localize to centrosomes [39] (Supplementary Fig. 8a), and for AKAP9 mRNAs, which accumulate on the Golgi apparatus close to centrosomes [62] (Supplementary Fig. 8b). These results identify a set of five transcripts whose localization in proximity to centrosome depends on EJC assembly, generalizing the effect previously reported for NIN mRNA. As the localization of these mRNAs also require ongoing translation, it points toward a localization mechanism in which the EJC and the nascent peptides work together to localize polysomes [23, 39, 63, 64]. Together with the observation that EJC knockdowns affect centrosome organization [23], these data strengthen the notion that the EJC complex is intricately involved in a post-transcriptional regulation program defining centrosomal functions during the cell cycle.

Figure 6.

EJC-i prevents proper intracellular localization of mRNAs. (a) Images are micrographs of HeLa cells treated or not with 10 μM EJC-i for 16 h, and labeled for NIN mRNA by smFISH (left), and centrin1-GFP protein (middle). Merged image (right) display NIN mRNAs in green, centrin1 in red, and DNA in blue. Scale bars: 10 microns. (b) Same as panel (a) for ASPM mRNA. (c) Same as panel (a) for NUMA1 mRNA. (d) Index of normalized median distances of mRNAs to centrosomes, for the two conditions, measured on individual cells. In each cell, distances from spots to closest centrosome were normalized using distances from a random RNA distribution, a distance inferior to one indicates colocalization whereas a distance superior to one indicates anti-colocalization. The dots represent the mean for all cells and the bar the median. P-values were determined using an Alexander Govern and a Games-Howell tests. *** P< .001. (e) Same as panel (a) for DYNC1H1 mRNA (left) and KIF1C mRNA (middle). Right: plot of the quantification of the number of mRNA foci per cell, in cells treated or not with EJC-i for 16 h (up) and plot of the median of the normalized distance of mRNAs to the cell membrane for cells treated or not with EJC-i for 16 h (down). Circles indicate single molecule of mRNA, squares a foci and arrows indicate protrusions.

We next tested the impact of EJC-i on transcripts that localize to other cellular compartments [62]. DYNC1H1 code for the large subunit of dynein and its mRNAs accumulate into discrete cytoplasmic foci acting as translation factories [65], and KIF1C codes for a plus-end kinesin motor whose mRNAs concentrate into cytoplasmic protrusions [62]. Remarkably, in both cases, exposure to EJC-i severely prevented their localization (Fig. 6e). These results identify two additional transcripts whose localization is dependent on the integrity of the EJC.

The localization of the centrosomal mRNAs cited above is translation dependent, as is the formation of the DYNC1H1 translation factories. We thus tested the effects of EJC-i on the translation of these transcripts. We previously generated HeLa cells having a SunTag fused at the N-terminus of the endogenous ASPM and DYNC1H1 proteins, effectively labeling their polysomes and nascent peptides [39, 65]. After treatment with EJC-i, we observed a drastic reduction in the number of ASPM polysomes in interphase cells, already visible after 4h of treatment and complete after 16h (Fig. 7a). We also observed that the export of ASPM mRNAs was also inhibited. Its mRNAs molecules remained in the nucleus after EJC-i addition, accounting for the absence of polysomes in the cytoplasm (Fig. 7a). This effect of the EJC-i on nuclear export was specific for ASPM mRNAs as it was not seen for DYNC1H1 transcripts, which were exported and translated similarly whether EJC-i was present or not (Fig. 7b). Next, we checked ASPM translation in mitotic cells. The SunTag-ASPM protein localized at the centrosome and spindle whether EJC-i was present or not (Fig. 7c). As noted above, the ASPM mRNA did not localize to centrosomes in EJC-i treated cells. Interestingly however, some active SunTag-ASPM polysomes were observed in EJC-i treated cells, suggesting that the SunTag-ASPM mRNA could be translated after its release from the nucleus, following the breakdown of the nuclear envelope (Supplementary Fig. 9). Altogether, these data indicate that EJC-i effects on mRNA localization were not due to an inhibition of translation. These data point to the advantageous use of EJC-i as a tool to efficiently monitor the involvement of the EJC in mRNA localization.

Figure 7.

Selective inhibition by EJC-i reveals the EJC-dependent export of ASPM mRNAs. (a) Images are micrographs of HeLa cells, in interphase, treated or not with EJC-I for 4 or 16 h, expressing a SunTag-ASPM allele (left) SunTag-ASPM protein display in green, ASPM mRNAs in red, and DNA stained with DAPI in blue. Right: same as left with cells expressing a SunTag-DYNC1H1. Scale bars: 10 microns. Arrows indicate polysomes. (b) Same as panel (a) for SunTag-ASPM proteins and mRNA for cells in mitosis.

EJC-i disrupts ciliogenesis in mouse neural stem cells

The integrity of centrosome organization is necessary for the differentiation of radial glial mouse NSCs. When these mono-ciliated cells are starved for serum, they become quiescent and differentiate into multi-ciliated ependymal cells that populate the surface of brain ventricles, ensuring the flow of cerebrospinal fluid through the beating of their cilia [66]. In NSC, the primary cilium forms from the mother centriole (basal body), while during differentiation, amplification of centrioles leads to the generation of multiple motile cilia in ependymal cells [67]. We previously reported that EJCs accumulate at basal bodies of quiescent NSC and RPE1 cells [23]. siRNA downregulation of EJC core components results in abnormal ciliogenesis and decrease in the number of ciliated RPE1 cells [23]. However, this procedure is challenging with NSC primary cultures.

This prompted us to analyze the effect of EJC inhibition on ciliogenesis in primary cultures of NSC isolated from newborn mouse forebrain [67]. Cultures were treated with 10 µM EJC-i or control DMSO at the onset of ependymal cell differentiation and over two days. Treatment with the inhibitor led to a marked reduction in the primary cilium length as well as to delayed differentiation (Fig. 8a and b). Thus, ciliogenesis is severely disrupted upon inhibition of EJC in NSC cells like in RPE1 cells, confirming its crucial role in this central process.

Figure 8.

EJC-i disrupts ciliogenesis in mouse NSCs. (a) Representative images of NSC treated with 10 µM EJC-i or DMSO as control. Primary cilia were stained by poly-glutamylated tubulin (PolyGlu-Tub) antibody (magenta), and centrioles by Centrin2-GFP transgenic mice (green). Scale bars in the upper and lower panels are 5 and 2, 5 µm, respectively.(b) Quantification of the difference in primary cilia length between the two conditions. The bars represent the mean ± SEM; one dot corresponds to one cell (three biological replicates). P-values were determined using a two-sided Student’s t-test. ***P < . 001.

Discussion

Here, we provide evidence that a compound known to inhibit eIF4A3 ATPase activity inhibits de novo EJC assembly, without affecting pre-formed complexes. We show that this drug impairs known EJC-dependent functions such as alternative splicing, NMD, m6A deposition, and mitotic progression. Its effects are similar to degron-mediated downregulation of the core EJC components eIF4A3 and Y14. The use of the drug supports the EJC-dependent localization of several centrosomal mRNAs. Inhibition of EJC assembly disrupts ciliogenesis in mouse NSC primary cultures, further substantiating a central role of the EJC in centrosome biology. Moreover, we propose a broader function of the EJC in the targeting of transcript to other cell compartments.

EJC-i targets a key component of the EJC complex without affecting the expression level of its essential subunits. It had been described as a chemical compound that selectively inhibits the ATPase activity of eIF4A3 [34, 36, 37]. However, the ATPase activity is not required for EJC assembly [40, 68]. While detailed structural insights are currently lacking due to limitations of predictive tools, clues about EJC-i mode of action come from a previous hydrogen/deuterium exchange mass spectrometry (HDX-MS) analysis [34]. Binding of EJC-i appears to occur within an allosteric region of eIF4A3 wherein some of the most responsive stretches either include or are in close proximity with residues involved in interactions with RNA and MAGOH [10, 11, 68]. Noteworthy, we find that EJC-i impairs interaction of eIF4A3 with RNA in the absence of its partners. EJC-i binding might induce a conformational change in eIF4A3 that prevents these interactions and thereby EJC assembly [34]. Importantly, EJC-i does not impede formation of the eIF4A3-CWC22-CWC27 complex, and so, eIF4A3 recruitment to spliceosomes [31, 49]. EJC-i most likely blocks the binding of eIF4A3 to RNA and so EJC core positioning onto the exon during the late phase of the splicing reaction [69]. The effect of EJC-i in vivo can be followed quantitatively using an EJC-NanoBiT assay [15]. Consistent with recent observations that EJCs slowly disassemble in vivo, it plateaus at about 80% inhibition, likely reflecting the presence of complexes not undergoing, or only slowly, recycling. Exposure to EJC-i leads to the same alterations in alternative splicing, m6A deposition, NMD, centrosome structure, mitotic delay, and ciliogenesis defects than targeted degradation of Y14, another EJC subunit. Comparison of global effects between the two treatments suggests an earlier response following EJC-i treatment compared to Y14 depletion, sustaining the specificity of the inhibitor. Even regardless of this trend, perhaps more telling is the observed effect on m6A accumulation that is universally present across the vast majority of genes. This is a much more direct consequence and readout of EJC loss compared to the relatively smaller number of genes that are either affected in their alternative splicing or NMD and that could also reflect indirect effects. Thus, although eIF4A3 may have a life independently of the EJC [70, 71] and other EJC-independent effects cannot be excluded, EJC-i can be considered primarily as an EJC inhibitor. Blocking EJC complex assembly provides a unique opportunity to pinpoint events depending upon functional EJCs.

Many mRNAs display a specific intracellular localization. The EJC was initially found to play a key role in the localization of oskar mRNA in Drosophila oocytes [22]. Then, NIN mRNA had been reported to require the EJC to be localized at centrosomes in mammalian cells [23]. We now extend this requirement to other centrosome-localized transcripts such as ASPM, NUMA1 and PCNT, as well as AKAP9 mRNAs which accumulate on the Golgi apparatus close to centrosomes. Furthermore, localization of DYNC1H1 mRNA in translation factories and of KIF1C mRNA in cytoplasmic protrusions also require the EJC. While the localization of NIN, NUMA1, ASPM, AKAP9, PCNT, and DYNC1H1 mRNAs requires translation, that of KIF1C mRNAs does not [23, 39, 62, 63]. Interestingly, the mRNAs whose localization is affected by EJC-i, could localize in either translation-dependent or independent ways, indicating the involvement of the EJC in multiple RNA transport mechanisms. In the case of DYNC1H1 mRNAs, EJC-i did not induce a shutdown of their translation, pointing toward a direct effect of the EJC in mRNA localization. In the case of ASPM mRNAs, EJC-i blocked export from the nucleus at interphase, resulting in nuclear accumulation and a lack of cytoplasmic mRNAs. How EJC depletion specifically impacts ASPM export remains unknown. ASPM mRNA export might have a particularly strong requirement in the human transcription export complex that associates with the EJC [20, 72]. However, these nuclear ASPM mRNAs are released into the cytoplasm at mitosis when the nuclear envelope breaks down. They become translated and yet they fail to localize to centrosomes in EJC-i-treated cells. This observation points towards a coordinated role for the EJC and local translation in mRNA localization. The EJC and the nascent peptides might work together to localize the corresponding polysomes at the right place. Noteworthy, the EJC is associated with stalled ribosomes [15]. Arrested ribosomes having already synthesized N-terminal fragments of proteins, might direct transport of the corresponding polysomes to their destination [39].

EJC-i leads to a mitotic arrest at prometaphase. This is likely linked to the impaired centrosome organization due to EJC inactivation. Indeed, the centrosome and its earlier-mentioned components play a key role in mitosis. Noteworthy, knockdown of Ninein leads to aberrant spindle formation [73]. ASPM regulates mitotic spindle orientation during metaphase [74]. NUMA is an essential player in mitotic spindle assembly and maintenance [75]. AKAP9 is required for association of the centrosomes with the poles of the bipolar mitotic spindle during metaphase [76]. Therefore, impaired localization of mRNAs coding for these proteins might be responsible for the mitotic arrest. A further mechanistic insight originates from the observation that EJC-i treatment and Y14 depletion alike strongly reduce the levels of SAS-6, a central scaffolding component of centrioles [60], offering a lead for future studies.

Both EJC core and peripheral components have been implicated in various cancers [28]. Furthermore, hypomorphic expression of eIF4A3 and Y14 has been linked to two neurodevelopmental genetic disorders in humans: Richieri-Costa Pereira syndrome and Thrombocytopenia with absent radius syndrome, respectively [29, 77, 78]. Mouse models have established a clear connection between imbalances in EJC core components and defects in neural stem cell (mNSC) division, neurogenesis, and cortical development [79, 80]. We had previously shown that Y14 or eIF4A3 siRNA knockdown in RPE1 cells impairs centrosome organization and ciliogenesis [23]. A centrosome forms the basal body of cilia [81]. The defective ciliogenesis might be due to the defective local synthesis of critical centrosome components. The use of siRNA with mNSC is challenging. Thanks to EJC-i, we now show that EJC is also essential for ciliogenesis in primary mNSCs. It might be hypothesized that EJC-dependent mitotic defects are responsible for mNSC division.

Popular methods for the inhibition of essential proteins rely on conditional gene knockouts using CRISPR/Cas9 genome editing or conditional knockdown methods such as shRNA or siRNA or by dTAG degron modules introduced by genome editing [50]. While effective, conditional loss of function of essential genes has several limitations. Typically, RNA silencing methods necessitate extended time (days) to achieve successful downregulation and are more likely to suffer from potential off-target and side effects. Not all cell types are amenable to CRISPR/Cas9-based knockdown or knockout approaches. This is indeed the case for mouse NSC primary cultures. Small molecules that specifically target essential proteins to inhibit or modulate their function circumvent these limitations. Their action is rapid, generally limited by diffusion into cells; they are easily applicable in diverse settings without the necessity for genetic alterations and carry the potential for therapeutic applications [82, 83]. This and previous studies of EJC-i [37] have made use of doses able to elicit specific responses within reasonably short treatment times and for which the effects on EJC assembly still were reversible. Nevertheless, prolonged exposure to the inhibitor can lead to secondary effects deriving from apoptotic response. Thus, like with all compounds, future studies employing EJC-i will have to determine its optimal effective concentrations and time frame of action. In conclusion, EJC-i is the first available specific inhibitor of the EJC as a whole. It opens up new venues to investigate the roles that the complex plays in the regulation of cell cycle and consequences of its dysfunction associated with pathologies and developmental defects.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work was supported by the Agence Nationale de la Recherche (ANR-17-CE12-0021 and ANR-21-CE120041), Fondation pour la Recherche Médicale (FRMEQU202003010226) and by continuous financial support from the Centre National de Recherche Scientifique, the Ecole Normale Supérieure, and the Institut National de la Santé et de la Recherche Médicale (H.L.H.); by La ligue Contre le Cancer (E.B.); by the Israel Science Foundation (grant no. 913/21), and the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant no. 101000970) (S.S.); and by the Agence Nationale de la Recherche (ANR -20-CE45-0019, ANR-21-CE16-0016, and ANR-22-CE16-0011) and the Fondation pour la Recherche Medicale (FRM EQU202103012767) (N.S.). Funding to pay the Open Access publication charges for this article was provided by the Agence Nationale de la Recherche (ANR-21-CE120041).

Data availability

The high-throughput RNA sequencing data are available in the GEO database under accession number GSE286119.