eIF3j inhibits translation of a subset of circular RNAs in eukaryotic cells

Abstract

Increasing studies have revealed that a subset of circular RNAs (circRNAs) harbor an open reading frame and can act as protein-coding templates to generate functional proteins that are closely associated with multiple physiological and disease-relevant processes, and thus proper regulation of synthesis of these circRNA-derived proteins is a fundamental cellular process required for homeostasis maintenance. However, how circRNA translation initiation is coordinated by different trans-acting factors remains poorly understood. In particular, the impact of different eukaryotic translation initiation factors (eIFs) on circRNA translation and the physiological relevance of this distinct regulation have not yet been characterized. In this study, we screened all 43 Drosophila eIFs and revealed the conflicting functions of eIF3 subunits in the translational control of the translatable circRNA circSfl: eIF3 is indispensable for circSfl translation, while the eIF3-associated factor eIF3j is the most potent inhibitor. Mechanistically, the binding of eIF3j to circSfl promotes the disassociation of eIF3. The C-terminus of eIF3j and an RNA regulon within the circSfl untranslated region (UTR) are essential for the inhibitory effect of eIF3j. Moreover, we revealed the physiological relevance of eIF3j-mediated circSfl translation repression in response to heat shock. Finally, additional translatable circRNAs were identified to be similarly regulated in an eIF3j-dependent manner. Altogether, our study provides a significant insight into the field of cap-independent translational regulation and undiscovered functions of eIF3.

INTRODUCTION

Circular RNA (circRNA) is a class of covalently closed RNA molecules discovered in diverse species (1–4). A small amount of circRNAs are outputs of non-coding regions (e.g. intronic circRNAs) (5–8), whereas the majority of circRNAs are generated from one or multiple exons of eukaryotic protein-coding genes via back-splicing, by which a splicing donor joins an upstream splicing acceptor (9–12). Due to lacking of canonical features that are usually utilized by linear RNAs, the regulation of circRNAs is distinct from that of their linear counterparts. Take nuclear export as an example. Once generated, linear mRNAs are typically capped at the 5′ end and polyadenylated at the 3′ end. A subset of export adaptors can recognize RNA cargoes as/with cap- or poly (A)-binding proteins and establish a physical bridge for linear mRNAs and their export receptors (13,14). Since circRNAs have no free ends, they must be exported via a different mechanism. In support, our recent study has demonstrated that the evolutionarily conserved receptor Exportin-4 (XPO4) directly binds to a subset of circRNAs, but not their linear counterparts, to facilitate their nuclear export (15).

Perturbations in circRNA expression are closely associated with cellular physiology and many diseases (16–20). A plenty of circRNAs have been demonstrated to directly regulate various physiological or pathological processes through diverse mechanisms, such as sponging microRNAs, forming DNA-RNA hybrids (R-loops), and interacting with RNA binding proteins (RBPs) (21–26). For example, circTLK1, an ischemic stroke-associated circRNA, can sequester miR-335-3p away from its target TIPARP, pathologically aggravating brain infarction and neuronal injury (25). CircSMARCA5 blocks SMARCA5 transcription via an R-loop structure formed at exons 15–16 in breast cancer cells (26). CircNSUN2 recruits the RBP IGF2BP2 to enhance the stability of HMGA2 mRNA, which results in an increased level of HMGA2 protein that promotes colorectal liver metastasis (24).

Although it was long assumed that circRNA is a type of non-coding RNAs without translation ability, emerging studies have demonstrated that a subgroup of endogenous circRNAs are bound by polyribosomes and the translation of circRNAs may be pervasive in eukaryotes (27–30). Moreover, circRNAs have been found to exert physiological or molecular roles through their encoded proteins in recent years (31–33). For example, the insulin-sensitive circRNA circSfl shares the start codon with its linear counterpart and encodes a truncated sulfateless (Sfl) protein which contributes to the lifespan extension of fruit flies (34). CircE-Cad-derived C-E-Cad protein maintains the tumorigenicity of glioma stem cells by activing an array of cancer-relevant pathways, such as STAT3, PI3K-AKT and MAPK-ERK signaling (35). Additionally, a circRNA generated from the long non-coding RNA LINC-PINT is capable of encoding a tumor-suppressive protein which regulates the transcriptional elongation of oncogenes in glioblastoma (7). These studies suggest that circRNA-derived proteins represent essential regulators in normal physiology and multiple diseases, and that the proper regulation of circRNA translation is required for cellular homeostasis maintenance. In fact, some RNA regulons, including internal ribosomal entry site (IRES)-like and m⁶A-modified elements, have been implicated as cis-acting factors to initiate circRNA translation (27–29). However, what trans-acting eukaryotic translation initiation factors (eIFs) function in the translation initiation of circRNAs is still poorly understood.

In eukaryotic cells, eIF3 is a multi-subunit complex containing 12 subunits and consists of two interconnected modules which are assembled by the nucleation core eIF3a and eIF3b (36–38). eIF3 plays essential roles in several steps of translation initiation of linear mRNAs, such as 43S pre-initiation complex (PIC) assembly, mRNA recruitment to the 43S PIC, and start codon recognition/selection (36–38). Beyond its canonical roles, eIF3 has also been implicated as an inhibitor in the translation of certain stress-responsive and proliferation-relevant mRNAs, such as FTL and BTG1 (39,40). Moreover, eIF3 can recognize 5′ end of specific mRNAs and promote initiation complex formation in an eIF4E-independent manner (41). Although eIF3j was first thought to represent the 13th subunit of eIF3, emerging evidence supports that it often functions in an eIF3-independent manner and is not a bona fide eIF3 subunit (36–38).

Using a model translatable circRNA (Drosophila circSfl), we here evaluated the impact of all 43 Drosophila eIFs on circRNA translation by a systematic RNAi screening. The eIF3 complex was found to promote the translation efficiency of circSfl, while eIF3j was identified as the most potent inhibitor. Mechanistically, eIF3j induces translation repression by promoting the disassociation of the eIF3 complex from circSfl. The binding of eIF3j to circRNA templates requires its C-terminus and is essential for the inhibitory activity of eIF3j. Moreover, we demonstrated an RNA regulon within the circSfl untranslated region (UTR) that facilitates eIF3j recruitment and, in turn, translation repression, supporting a combinatorial control of circRNA translation initiation by cis-regulatory RNA elements and trans-regulatory protein factors. In addition, we revealed that eIF3j negatively regulates the heat resistance of circSfl-enriched cells by attenuating circSfl translation in cellular response to heat stress, suggesting a stress-responsive mechanism to ensure clean of damaged cells. Finally, we identified additional translatable circRNAs whose translation is similarly regulated by eIF3j. Focused studies on circPde8 confirmed the general role of eIF3j in circRNA translation. Altogether, our findings provide an insight into the previously undiscovered eIF3j-mediated circRNA translational control and illustrate the physiological relevance of this distinct regulation.

MATERIALS AND METHODS

Cell culture and stable cell line construction

Drosophila Schneider 2 (S2) cells were grown in Schneider's Drosophila medium (Sigma, S9895) supplemented with 10% fetal bovine serum (v/v; HyClone, SH30910.03) and 1% penicillin streptomycin (v/v; Thermo Fisher Scientific, 15140122) at 25°C. To generate a stable cell line, 1.0 × 10⁶ S2 cells in a well of the 12-well plate were transfected with 1 μg of the indicated plasmid (Supplementary Plasmid Information) using Lipo6000 (Beyotime, C0526) for 3 days according to the manufacturer's protocol. The transfected cells were then transferred to fresh medium and maintained by selection with 150 μg/ml hygromycin B (Biofroxx, 1366ML010) for another 3–4 weeks.

Plasmids and cloning

All plasmids used in this study were generated by modifying the Hy_pMT EGFP SV40 pA plasmid (Addgene, #69911), in which the copia transposon LTR promoter drives HygroR transcription and the MtnA promoter (a copper-inducible promoter) drives EGFP transcription (42). All cloning details are provided in Supplementary Plasmid Information.

Double-stranded RNA preparation

The detailed information of each double-stranded RNA (dsRNA) used in this study is provided in Supplementary Table S1. DNA templates of dsRNAs were prepared by PCR reactions with primer pairs containing the T7 promoter sequence (TAATACGACTCACTATAGGG) on the 5′ end. DsRNAs were then generated by in vitro transcription using ScriptMAX^® Thermo T7 Transcription Kit (TOYOBO, TSK-101) according to the manufacturer's protocol.

RNAi and ASO directed knockdown

For dsRNA bathing, a total of 1.5 × 10⁶ S2 cells were suspended in 600 μl of serum-free medium containing 8 μg of the indicated dsRNA for 30 min. 400 μl of medium containing 20% fetal bovine serum (v/v) was then added and cells were maintained for 3 days at 25°C. For small interfering RNA (siRNA) or antisense oligonucleotide (ASO) transfection, a total of 1.5 × 10⁶ S2 cells were transfected with the indicated siRNA (final concentration: 80 nM) or ASO (final concentration: 50 nM) for 2 days using Lipofectamine^® RNAiMAX Reagent (Invitrogen, 13778-100) according to the manufacturer's protocol. The detailed information of each siRNA or ASO used in this study is provided in Supplementary Table S2.

Western blotting

Protein extracts were prepared using RIPA buffer (50 mM/l Tris–HCl pH 7.4, 150 mM/L NaCl, 0.1% sodium dodecyl sulfate (SDS; w/v), 1% sodium deoxycholate (w/v), and 1% Triton X-100 (v/v)) and analyzed by western blotting as previously described (43–45). Briefly, protein samples were denatured at 100°C for 5 min in the presence of protein loading buffer (62.5 mM Tris–HCl pH 6.8, 10% glycerol (v/v), 0.01% bromophenol blue (w/v), 2.15% SDS (w/v), 1.55% dithiothreitol (w/v), and 5% 2-hydroxy-1-ethanethiol (v/v)), separated on 12% SDS-PAGE gels, and transferred to polyvinylidene fluoride (PVDF) membranes (BioRad, 1620177). Membranes were then processed following the standard ECL protocol (Thermo Fisher Scientific, EI9051). Blots were viewed using Bio-Rad ChemiDoc Imaging System and protein levels were quantified from at least three western blots using ImageJ. Antibodies used were anti-FLAG (Beyotime, AF519; 1:1000 dilution), anti-HDAC1 (Abcam, ab1767; 1:1000 dilution), anti-α-Tubulin (Beyotime, AF0001; 1:2000 dilution), and anti-Histone H3 (Abcam, ab1791; 1:1000 dilution).

Northern blotting

Northern blotting was performed as previously described (43,46). Briefly, the same amount of RNA was denatured at 65°C for 15 min in the presence of formaldehyde and Gel Loading Buffer II (Thermo Fisher Scientific, B8546G5), separated on 1.2% denaturing agarose gels, and transferred to hybond-N⁺ membranes (GE healthcare, RPN303B). After UV crosslinking (254 nm, 120 mJ/cm²), membranes were hybridized with DIG-labeled DNA probes (Sangon Biotech) at 42°C overnight followed by anti-DIG incubation for 2 h at room temperature using DIG Northern Starter Kit (Roche, 12039672910). Blots were viewed using Bio-Rad ChemiDoc Imaging System and RNA levels were quantified from at least three northern blots using ImageJ. The sequences of northern blotting probes are provided in Supplementary Table S3.

Estimating the RNase R resistance of circSfl

To confirm the resistance of circSfl to RNase R digestion, 15 μg of RNA from whole cells (the circSfl stable cell line) was treated with 10 units RNase R (Epicentre, RNR07250) for 20 min at 37°C, purified using RNAiso Plus (Takara, 9108), and subjected to northern blotting analyses.

Immunofluorescence staining

To visualize the subcellular localization of circSfl-derived Sfl protein (CdSfl) and circPde8-derived Pde8 protein (CdPde8), the circSfl and circPde8 stable cell line were treated with 500 μM CuSO₄ for 12 h to induce protein expression. A total of 0.5 × 10⁶ cells were seeded on a coverslip coated with concanavalin A (ConA; Solarbio, C8110) in a well of the 6-well plate for the final 2 hr. Coverslips (cell side up) were washed with 1 × PBS buffer (phosphate buffer saline pH 7.4: 137 mM NaCl, 2.7 mM KCl, 10 mM Na₂HPO₄ and 1.8 mM KH₂PO₄) twice and treated with fixative solution (75% methanol (v/v) and 25% glacial acetic acid (v/v)) for 10 min at room temperature. After washing the coverslips with TBST buffer (Tris-Buffered Saline Tween-20 pH 7.6: 0.242% Tris–HCl (w/v), 0.8% NaCl (w/v), and 0.05% Tween-20 (v/v)), cells were permeabilized with 0.1% Triton X-100 in TBST buffer at room temperature for 10 min. Coverslips were then treated with 5% bovine serum albumin (w/v; Beyotime, ST025) in TBST buffer to block non-specific binding of antibodies for 1 h followed by anti-FLAG (Beyotime, AF519) incubation at 4°C overnight. After washing the coverslips with TBST buffer for 3 times, cells were incubated with the Alexa Fluor 647 or 555 dye-labeled secondary antibody (Beyotime, A0473 and A0460) at room temperature for 2 h in the dark. 1 μg/ml DAPI (Beyotime, C1002) was used to stain nuclei for 10 min before imaging. A confocal laser scanning microscopy (Leica TCS, SP8) was used for imaging and ImageJ was used for quantification of nuclear and cytoplasmic fluorescence signals.

Estimating the half-life of circSfl

A total of 6.0 × 10⁶circSfl stable cells were treated with 500 μM CuSO₄ for 12 h to activate circSfl expression. Cells were washed with 1× PBS buffer twice and with the medium containing 500 μM bathocuproine disulphonate (BCS; Sigma, B1125) for 3 times. 50 μM BCS in fresh medium was then added for the indicated amounts of time (3, 6, 9, 12, 15, 18, 21 or 24 h) followed by RNA extraction and northern blotting analyses.

Estimating the half-life of CdSfl

A total of 6.0 × 10⁶circSfl stable cells were treated with 500 μM CuSO₄ for 12 h to activate CdSfl expression. Cells were washed with 1× PBS buffer twice and incubated with 100 μg/ml cycloheximide (CHX; Yeasen, 40325ES03) in fresh medium for the indicated amounts of time (1, 2, 3, 4, 5, 6, 7, 8 or 9 h) followed by protein extraction and western blotting analyses.

Estimating the sensitivity of CdSfl production to different translation inhibitors

A total of 1.2 × 10⁷circSfl stable cells were treated with the indicated inhibitor and 500 μM CuSO₄ simultaneously for 12 h followed by protein extraction and western blotting analyses. Translation inhibitors used in this experiment were anisomycin (ANS; 0.5 μg/ml; Apexbio, B6674), cycloheximide (CHX; 100 μg/ml; Yeasen, 40325ES03), homoharringtonine (HHT; 1 μg/ml; Apexbio, N1504), chloramphenicol (INN; 150 μg/ml; Biofroxx, 1289GR025), and rocaglamide-A (RocA; 0.3 μM; Apexbio, C5148).

Fluorescence in situ hybridization (FISH)

The FISH RNA probe against the back-splicing junction of circSfl was generated by in vitro transcription using ScriptMAX^® Thermo T7 Transcription Kit (TOYOBO, TSK-101), labeled with the Alexa Fluor 546 dye using ULYSIS^® Nucleic Acid Labeling Kit (Thermo Fisher Scientific, U21652), and denatured at 95°C for 5 min before use. The sequence of the FISH probe is provided in Supplementary Table S4. To visualize the subcellular localization of circSfl, the circSfl stable cell line was treated with 500 μM CuSO₄ for 12 h to induce circSfl expression. A total of 0.5 × 10⁶ cells were transferred onto a ConA-coated coverslip for the final 2 h, washed with ice-cold 1× PBS buffer twice, and treated with fixative solution (75% methanol (v/v) and 25% glacial acetic acid (v/v)) for 10 min at room temperature. Cells were then washed with 2× SSC Tween-20 buffer (300 mM NaCl, 30 mM sodium citrate, and 0.5% Tween-20 (v/v)), treated with 2 × SSC Triton X-100 buffer (300 mM NaCl, 30 mM sodium citrate, and 0.1% Triton X-100 (v/v)), denatured at 80°C for 10 min, and incubated with the FISH probe at 42°C in the presence of 20 ng/μl yeast RNA (Beyotime, R0038) overnight in the dark. 1 μg/ml DAPI (Beyotime, C1002) was used to stain nuclei for 10 min before imaging. A confocal laser scanning microscopy (Leica TCS, SP8) was used for imaging and Image J was used for quantification of nuclear and cytoplasmic fluorescence signals.

Cellular fractionation

To reveal the subcellular distribution of circSfl and CdSfl, cellular fractionation was performed as previously described (43,47). Briefly, the circSfl stable cell line was treated with 500 μM CuSO₄ for 12 h to induce circSfl and CdSfl expression. A total of 1.5 × 10⁸ cells were washed with 1 ml of ice-cold 1× PBS buffer twice and resuspended in 1 ml of ice-cold lysis buffer I (10 mM Tris–HCl pH 8, 140 mM NaCl, 1.5 mM MgCl₂, 0.5% IGEPAL CA-630 (v/v), and 1 mM dithiothreitol) at 4°C followed by 100 rounds of pipetting. After spinning the cell lysate at 1000 × g for 3 min at 4°C, the cytoplasmic fraction was present in the supernatant of the centrifugate. The pellet was resuspended in 1.1 ml of ice-cold lysis buffer II (1 ml of lysis buffer I + 100 μl of detergent buffer [3.33% sodium deoxycholate (w/v) and 6.66% Tween-40 (v/v)]), slowly vortexed for 20 sec, and incubated on ice for 5 min. The mix was then spun at 1000 × g for 3 min at 4°C and washed with 1 ml of ice-cold lysis buffer I for 3 times. The final pellet was saved as the nuclear fraction. Efficient cellular fractionation was verified by examining the RNA levels of rp49, U3 and U6 and the protein levels of α-Tubulin and Histone H3 in each fraction.

Cross-linking immunoprecipitation (CLIP)

CLIP assays were performed as previously described (5,47). Briefly, cells were washed with ice-cold 1× PBS buffer twice, irradiated in a UV cross-linker (Lanyi, LYUV07-11) with 400 mJ/cm² at 254 nm on ice for 1 min, and incubated with RIPA buffer in the presence of RNase inhibitor (80 units/ml; Beyotime, R0102) and 1× Protease Inhibitor Cocktail (Beyotime, P1045) for 30 min on ice followed by 100 rounds of pipetting. After spinning the cell lysate at 12 000 × g for 5 min at 4°C, the supernatant of the centrifugate was saved as protein extracts and precleared with Protein A + G agarose beads (Beyotime, P2055) for 1 h at 4°C to prevent non-specific binding. Precleared protein extracts were incubated with the indicated antibodies or the negative IgG (as a control) for 4 h at 4 °C, and Protein A + G agarose beads (Beyotime, P2055) were then added to the mix for another 6 h at 4 °C. Beads were washed with RIPA buffer in the presence of RNase inhibitor (80 units/ml; Beyotime, R0102) for 4 times at 4°C and treated with 0.5 mg/ml proteinase K (Beyotime, ST535) for 30 min at 55°C to reverse cross-linking followed by RNA extraction and RT-qPCR analyses. Antibodies used in CLIP assays were anti-FLAG (Beyotime, AF519) and anti-V5 (Proteintech, 14440-1-AP).

To analyze the interaction between circSfl and the indicated eIFs, the notag circSfl stable cell line (circSfl from this cell line generates a non-tagged CdSfl) was transfected with the indicated plasmids of FLAG-tagged eIFs for 3 days. 500 μM CuSO₄ was added for the final 12 h to induce circSfl expression and cells were collected for CLIP assays.

To analyze the inhibitory effect of eIF3j on the binding of circSfl to eIF3 or Rps23, the notag circSfl stable cell line was depleted of eIF3j by dsRNA-mediated RNAi or overexpressed with eIF3j (V5-tagged) on day 1, transfected with the expression plasmid of FLAG-tagged eIF3a, eIF3b, or Rps23 on day 2, and collected for CLIP assays on day 4. 500 μM CuSO₄ was added for the final 12 h to induce circSfl expression.

To identify the functional domain of eIF3j which contributes to circSfl binding capacity, the circSfl stable cell line was transfected with the expression plasmid of the wild-type, N-terminus depleted, or C-terminus depleted eIF3j (V5-tagged) for 3 days. 500 μM CuSO₄ was added for the final 12 h to induce circSfl expression and cells were collected for CLIP assays.

To identify the UTR region where eIF3j recognizes, the notag circSfl stable cell line was transfected with the expression plasmid of FLAG-tagged eIF3j for 3 days. 500 μM CuSO₄ was added for the final 12 h to induce circSfl expression and cells were collected for iCLIP assays followed by RT-qPCR with 8 overlapping amplicons tiling through the circSfl UTR. iCLIP was performed as previously described with minor modifications (48).

To analyze the binding of the Δ101–200 circSfl mutant to the indicated eIF3 subunit (eIF3j, eIF3a or eIF3b), the Δ101–200 circSfl stable cell line (circSfl from this cell line is deleted of nucleotides 101–200 of the UTR) was transfected with the expression plasmid of the indicated eIF3 subunit (V5-tagged eIF3j, eIF3a, or eIF3b) for 3 days. 500 μM CuSO₄ was added for the final 12 hr to induce circSfl expression and cells were collected for CLIP assays. The binding of the wild-type circSfl and the Δ301–400 circSfl mutant to eIF3j, eIF3a, or eIF3b was also measured as a control.

To analyze the binding of circSfl to eIF3j in response to heat stress, the circSfl stable cell line was transfected with the expression plasmid of V5-tagged eIF3j for 3 days and maintained at 37°C for the final 4 h. 500 μM CuSO₄ was added for the final 12 h to induce circSfl expression and cells were collected for CLIP assays. The binding of circSfl to eIF3j under unstressed conditions was also measured as a control.

To analyze the binding of endogenous ribo-circRNAs to eIF3j, regular S2 cells were transfected with the expression plasmid of FLAG-tagged eIF3j for 3 days and collected for CLIP assays.

To analyze the inhibitory effect of eIF3j on the binding of endogenous ribo-circRNAs to eIF3, regular S2 cells were depleted of eIF3j by dsRNA-mediated RNAi on day 1, transfected with the expression plasmid of FLAG-tagged eIF3a on day 2, and collected for CLIP assays on day 4.

The detailed information of plasmids used in these experiments is provided in the section ‘plasmids and cloning’ and Supplementary Plasmid Information.

RNA extraction and RT-qPCR

RNAs were extracted using RNAiso Plus (Takara, 9108) from whole cells, nuclear fractions, cytoplasmic fractions, or CLIP samples. Complementary DNAs (cDNAs) were generated by reverse-transcription using PrimeScript RT Master Mix (Takara, RR036A). QPCR assays were then conducted with the CFX connect real-time PCR system (Bio-Rad) using Hieff^® qPCR SYBR Green Master Mix (YEASEN, 11201ES03) according to the manufacturer's protocol. For RNA extracts from whole cells, Ct values were normalized to the level of rp49 mRNA. For RNA extracts from nuclear or cytoplasmic fractions, Ct values were calculated without any normalization (absolute Ct value). For RNA extracts from CLIP samples, signals (relative to input) were normalized to the negative IgG control. For semi-qPCR, the number of reaction cycles was set between 16 and 22 to avoid the saturation phase (49). The detailed information of qPCR primers is provided in Supplementary Table S5.

Estimating cellular sensitivity to stressed conditions

To examine the physiological function of circSfl or CdSfl in response to heat stress, S2 cells stably expressing the wild-type circSfl (WT) or the GCG mutant circSfl (GCG MUT) were cultured at 37°C for the indicated amounts of time (0, 1, 2 or 4 h). Cells were then stained with trypan blue (Biosharp, BS924) and counted using Bright-Line™ Hemacytometer (Sigma, Z359629).

Statistical analyses

Statistical significance for comparisons of means was assessed by Student's t-test (^∗∗P < 0.01; ^∗P < 0.05). All data were generated from at least three independent biological replicates and are shown as means ± SEM in each figure legend.

RESULTS

Generation of a reporter vector for producing a translatable circRNA

For an efficient screening, we developed a translatable circRNA reporter vector by inserting the circularizing exon (including a FLAG sequence after the start codon) of the Drosophila gene Sfl between inverted intronic repeats of the previously described Hy_pMT laccase2 MCS exon vector (Figure 1A), which is able to efficiently express circRNAs under the control of a copper ion inducible promoter (pMT) (10). We chose Sfl-derived circRNA (circSfl) due to its known ability to generate a protein with a characterized physiological function in Drosophila (34). CircSfl uses the same start codon as its cognate mRNA and uses an in-frame stop codon downstream of the back-splicing junction (Figure 1A) (34). CircSfl expression in Drosophila S2 cells stably expressing the circSfl reporter was measured by northern blots and fluorescence in situ hybridization (FISH) with an antisense probe against the back-splicing junction (Figure 1B; Supplementary Figure S1A, B). Note that the vector-derived circSfl contains a multiple cloning site (MCS) sequence at the back-splicing junction such that the junction probe can specifically detect the vector-derived rather than endogenous circSfl (Figure 1A). We found that circSfl accumulated upon transcription induction over time and predominately localized in the cytoplasm (Figure 1B; Supplementary Figure S1A–C).

Figure 1.

The circSfl reporter produces a readily detectable protein. (A) A schematic overview of the construction of the circSfl expression vector which is modified from the previously described Hy_pMT laccase2 MCS exon vector. The circSfl vector was used to generate a stable cell line using Drosophila S2 cells. (B) Northern blots with a probe against the circSfl back-splicing junction were performed to measure the expression of copper-activated circSfl in the stable line. (C) Northern blots of circSfl with RNAs digested with or without RNase R. RNA samples were extracted from the stable cell line treated with or without copper sulfate (CuSO₄). (D, E) Northern blots were performed to measure circSfl expression after the stable cell line had been individually transfected with the back-splicing junction-specific siRNA and ASO. CircSfl expression was quantified from three independent northern blots. ^∗∗P< 0.01; ^∗P < 0.05. (F, G) Northern blots of circSfl after a 12 h transcription pulse of circSfl had been induced by adding of copper into the stable line and shut off by sequestering copper with bathocuproine disulphonate (BCS). CircSfl expression was quantified from three independent northern blots. (H) Related to B, western blots were performed to measure CdSfl accumulation after the stable cell line had been induced with copper for the indicated amounts of time. (I, J) Related to F and G, western blots of CdSfl after the transcription of circSfl stopped. The CdSfl level was quantified from three independent western blots. (K, L) Related to D and E, western blots of CdSfl after circSfl had been knocked down. The CdSfl level was quantified from three independent western blots. ^∗∗P< 0.01; ^∗P < 0.05. (M) A schematic overview of the wild-type or mutant circSfl. (N) S2 cells stably expressing the wild-type or mutant circSfl were induced by copper for 12 hr. Northern and western blots were performed to measure the expression level of circSfl and CdSfl, respectively. All data were generated from at least three independent biological replicates and are shown as means ± SEM.

Further confirming that a true circSfl was produced, (i) the vector-derived transcript, but not ribosomal RNAs, was resistant to 3′-5′ exonuclease RNase R-mediated degradation (Figure 1C), (ii) the back-splicing junction-specific siRNA (circ siRNA) and ASO (circ ASO) significantly reduced the expression level of circSfl (Figure 1D, E), and (iii) Sanger sequencing of PCR products spanning across the junction site of circSfl revealed the ligated back-splicing exon (Supplementary Figure S1D, E). Moreover, transcription inhibition experiment, in which a 12 h transcription pulse was induced by copper sulfate and shut off by addition of the copper chelator bathocuproine disulphonate (BCS), demonstrated that circSfl had a long half-life of ∼15 h (Figure 1F, G). Finally, RT-qPCR revealed that circSfl expression of circSfl stable cells was ∼3000-fold higher than that of regular S2 cells (Supplementary Figure S1F). We thus concluded that the circSfl vector was efficiently back-spliced to produce a bona fide cytoplasmic circRNA.

Next, we detected a ∼25 kDa protein which corresponds to the expected protein size of circSfl-derived Sfl protein (CdSfl), and the level of the ∼25 kDa protein exhibited a similar trend to circSfl expression upon transcription induction (Figure 1B, H) or inhibition (Figure 1F, G, I, J), as verified by western blots with anti-FLAG. To further confirm the translation ability of circSfl, we verified that (i) the level of CdSfl drastically dropped to ∼10–20% when cells were individually transfected with the back-splicing junction-specific siRNA and ASO (Figure 1K, L), and that (ii) mutating the AUG start codon to a GCG codon or deleting the FLAG sequence did not affect circSfl expression, but resulted in CdSfl being no longer detected by western blots using anti-FLAG (Figure 1M, N). In addition, we found that CdSfl localized predominately in the cytoplasm (Supplementary Figure S2A-C) and had a half-life of ∼5 hr (Supplementary Figure S2D, E). Finally, translation inhibition experiment demonstrated that CdSfl was sensitive to the four examined translation inhibitors; note that it was not sensitive to the mitochondrial translation inhibitor chloramphenicol (INN) (Supplementary Figure S2F, G). These findings support that translatable circRNAs use factors similar to those in the canonical mRNA translation pathway.

Collectively, our results confirm identity of the ∼25 kDa protein and strongly support that the vector-derived circSfl is efficiently translated into a readily detectable protein. We therefore envisioned that the circSfl stable cell line should enable an efficient screening and follow-up investigation for the underlying mechanism of circRNA translation.

RNAi screening reveals the inhibitory and promoting effect of different eIF3 subunits on the translational control of circSfl

In Drosophila, protein synthesis encompasses a series of initiation steps that are coordinated by 43 canonical trans-acting eIFs (50). To systematically evaluate the impact of eIFs on the translation initiation program of circRNAs, we took advantage of RNAi screening for all 43 eIFs in the circSfl stable cell line using dsRNAs. All dsRNAs were confirmed to efficiently knock down their targets (Supplementary Figure S3). Copper sulfate was added for the final 12 h to activate pMT and circSfl expression. Western blots and RT-qPCR were then performed to quantify the level of CdSfl and circSfl, respectively (Figure 2A). It is known that circRNAs are covalently-closed and thus lack a 7-methylguanylate (m⁷G) cap structure which is recognized by the cap binding protein eIF4E and utilized by the canonical translation pathway (51–54). As expected, depletion of homologs of human eIF4E (e.g. eIF4E1 and eIF4E3) had a limited effect on circSfl translation (Figure 2B, C), confirming that circRNA translation initiation proceeds by a non-canonical mechanism.

Figure 2.

RNAi screening of eIFs that can regulate circRNA cap-independent translation. (A) A schematic overview of the RNAi screening strategy for identification of the impact of all 43 Drosophila eIFs on circRNA translation. The circSfl stable cell line was treated with each eIF dsRNA for 3 days and copper was added to induce circSfl expression for the final 12 h. β-gal dsRNA served as a negative control. (B, C) Western blots of CdSfl with protein extracts from the circSfl stable cell line treated with eIF dsRNAs. The CdSfl level was quantified from three independent western blots. (D) RT-qPCR was used to quantify circSfl expression using RNA extracts from the circSfl stable cell line treated with eIF dsRNAs. All data were generated from three independent biological replicates and are shown as means ± SEM.

To rule out the possibility that changes in circSfl translation were simply caused by altered circSfl biogenesis, we compared the level of CdSfl and circSfl in each knockdown sample and found that knockdown of some eIFs (e.g. eIF2α, eIF2γ, eIF1A, eIF4A and eIF4G1) resulted in a decreased expression of CdSfl and circSfl to a similar extent to each other (Figure 2B–D; Supplementary Screening Data), suggesting that these eIFs play a very limited role in the translation efficiency of circSfl. In support, the level of CdSfl was increased to a similar extent as that of circSfl when these eIFs were individually overexpressed (Figure 3A–D). Therefore, we concluded that a subset of eIFs, such as eIF2α, eIF2γ, eIF1A, eIF4A and eIF4G1, do not participate in the process of circRNA translation at least in the case of circSfl. We also noticed that several eIFs can function in both the translation and biogenesis of circSfl. A case in point is eIF2β, knockdown of which caused a ∼75% reduction in the CdSfl level and a ∼47% reduction in circSfl expression (Figure 2B–D; Supplementary Screening Data). The eIF2 complex is a very stable heterotrimer formed by eIF2α, eIF2β and eIF2γ in eukaryotic cells (51–54). The discrepancy of eIF2 components in our study could be explained by the assumption that the translation initiation of circSfl may proceed in an eIF2-independent manner. In fact, the final 80S initiation complex on certain linear mRNAs (e.g. c-Src mRNA) is still able to assemble without eIF2 (55–57). On the other hand, there are studies implying that eIF2β and eIF2γ may act in the absence of eIF2α under specific conditions (58). This raises another assumption that eIF2β might bring the initiator tRNA (Met-tRNAi) to the 40S ribosome without eIF2α and eIF2γ in the process of circSfl translation.

Figure 3.

Identification of the conflicting functions of eIF3 subunits in the translational control of circRNAs. (A) RT-qPCR of the indicated eIF mRNAs after the circSfl stable cell line had been transfected with the indicated overexpression vectors. Empty vector served as a negative control. (B, C) Related to A, western blots of CdSfl examining the impact of the indicated eIFs on circSfl translation. The CdSfl level was quantified from three independent western blots. (D) Related to A, RT-qPCR of circSfl examining the impact of the indicated eIFs on circSfl biogenesis. (E, F) Rescue assays of the indicated eIF3 subunits using the circSfl stable cell line. UTR dsRNAs deplete endogenous but not vector-derived eIF3 subunits (V5-tagged). Western and northern blots were performed to detect CdSfl and circSfl, respectively. β-gal dsRNA served as a negative control. The level of CdSfl and circSfl were quantified from three independent blots. The gene locus and rescue vector (RV) of each eIF3 subunit with the location of the UTR dsRNA are also shown on the upper right panel. (G, H) CLIP assays of the indicated eIF3 subunits using the notag circSfl stable cell line (see Figure 1M). The expression vectors of FLAG-tagged eIF3 subunits were individually introduced into the notag circSfl stable cell line for 3 days. RT-qPCR and RT-semi-qPCR were performed with RNA extracts from CLIP samples to measure the binding of each eIF3 subunit to circSfl, circlaccase2, circdati, and U6 snRNA. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from three independent biological replicates and are shown as means ± SEM.

Notably, we identified that eIF3a, eIF3b, eIF3c and eIF3d1 were among the most potent positive regulators of circSfl translation, but did not affect circSfl expression (Figure 2B–D; Supplementary Screening Data). They were referred to positive eIF3 subunits herein. In eukaryotic cells, eIF3a and eIF3b function as the nucleation core of eIF3 for assembly of other eIF3 subunits into the octamer and the yeast-like core (YLC), respectively (37,59). The octameric head subunit eIF3c binds to eIF3a through its C-terminus and PCI (Proteasome-COP9 signalosome-eIF3) domain and was predicted to interact with ribosomal proteins near the mRNA exit channel (37,60–62). As a peripheral eIF3 subunit and a non-canonical cap-binding protein, eIF3d sits on the opposite site of the mRNA channel near the exit and directly interacts with the octameric right arm (41,63–65). To our surprise, the individual depletion of eIF3j, eIF3k, and eIF3l significantly elevated the extent of CdSfl production from the circSfl template (Figure 2B, C; Supplementary Screening Data). RT-qPCR investigating circSfl expression excluded the possibility that eIF3j, eIF3k, and eIF3l each inhibited circSfl biogenesis to limit its translation (Figure 2D; Supplementary Screening Data). We termed eIF3j, eIF3k and eIF3l as negative eIF3 subunits herein. As right leg subunits of the eIF3 octamer, eIF3k and eIF3l were found to be easily dissociated from the whole eIF3 complex and dispensable for eIF3 formation (59,65–67). eIF3j only loosely contacts with other eIF3 subunits and is usually considered as an eIF3-associated factor rather than a bona fide eIF3 subunit (37,68,69). Informed by the previous studies and the findings from our screening, we concluded that different eIF3 subunits can variously exert diametrically opposed functions regarding the translation of circRNA templates.

To confirm the phenotypes generated from our RNAi screening and rule out potential off-target effects of RNAi-directed knockdown, we took advantage of independent non-overlapping dsRNAs targeting the UTR of representative eIF3 subunits to repeat knockdown experiments and observed phenotypes that mirror our prior results (Figure 3E, F; Supplementary Figure S4). Moreover, we developed a series of expression vectors which only harbor the coding region of each examined eIF and are insensitive to UTR dsRNAs. Reexpression of these eIFs in cells treated with UTR dsRNAs significantly restored CdSfl production to levels similar to the ‘β-gal’ control sample (Figure 3E, F; Supplementary Figure S4). In addition, the involvement of eIF3 subunits in circRNA translation was recapitulated using a cell line stably expressing the circSfl vector modified from the previously described Hy_pMT dati Exons 1–3 vector (43,44,70), in which inverted intronic repeats of the Drosophila gene dati promote circSfl biogenesis (Supplementary Figure S5).

eIF3 interacts with circSfl

To understand the underlying mechanism for eIF3-mediated regulation, we applied cross-linking immunoprecipitation-reverse transcription-quantitative PCR (CLIP-RT-qPCR) to examine the recruitment of eIF3 subunits to circSfl (Figure 3G, H; Supplementary Figure S6A–D). We analyzed the CLIP-RT-qPCR results following the Fold Enrichment method, which is a signal-to-noise ratio comparing the amount of the target sequence measured in the IP isolate (relative to input) to the amount measured in the negative control isolate. Unexpectedly, we observed that both positive and negative eIF3 subunits bound to circSfl, suggesting an interplay between positive and negative eIF3 subunits in the translational control of circRNAs. By contrast, control RNAs (without protein-coding ability, including circlaccase2, circdati, and U6 snRNA) did not interact with the examined eIF3 subunits (Figure 3G, H, Supplementary Figure S6A–D). In addition, control eIFs, including eIF4E3 and eIF5, exhibited no significant binding capacity to circSfl (Supplementary Figure S6E, F). Considering that (i) depletion of eIF3j resulted in the largest increase in CdSfl production of all 43 eIFs screened (Figure 2; Supplementary Figure S5) and that (ii) eIF3j exhibited the strongest binding capacity to circSfl among the three negative eIFs (Figure 3H; Supplementary Figure S6C, D), we next focused on eIF3j-mediated circRNA translation repression and defined how the recruitment of different eIF3 subunits to translatable circRNAs leads to opposing translation phenotypes in the subsequent study.

eIF3j-mediated regulation is specific to the circular version of Sfl RNA

To explore whether eIF3j-mediated regulation is specific to circSfl, we constructed a vector which can exclusively produce a linear Sfl mRNA (Supplementary Figure S7A). Note that the sequence of this vector-derived linear Sfl mRNA is same to circSfl, and that the vector-derived linear Sfl mRNA can encode a protein whose amino acid sequence is same to CdSfl. The individual knockdown of eIF3a and eIF3b significantly reduced the Protein/RNA ratio (the relative linear Sfl encoded CdSfl level divided by the relative linear Sfl mRNA level), suggesting that the eIF3 complex is required for linear Sfl mRNA translation (Supplementary Figure S7B–E). In contrast to the results with circSfl, knockdown of eIF3j had no effect on the translation efficiency of the linear Sfl mRNA (Supplementary Figure S7B–E), ruling out that the linear Sfl mRNA is a subject of eIF3j-mediated regulation.

eIF3j inhibits the binding of eIF3 to circSfl

It is now known that eIF3a and eIF3b are the core subunits for the nucleation of the octamer and the YLC which are two interconnected modules of the eIF3 complex (Figure 4A) (37,59). eIF3 collapses without eIF3a and eIF3b (37,59). Codepletion experiments demonstrated that depletion of eIF3j did not increase CdSfl production in the absence of the nucleation core of eIF3 (Figure 4B–F), indicating that eIF3-mediated translation initiation is a prerequisite for the inhibitory role of eIF3j in the process of circSfl translation. In support of this, codepletion of eIF3c and eIF3j exhibited a similar phenotype (Supplementary Figure S8). In contrast to CdSfl, circSfl expression was only marginally affected upon codepletion of eIF3 subunits (Figure 4B–F; Supplementary Figure S8). Together, these results suggest a potential epistatic relationship in which the eIF3 complex binds to circRNAs and promotes their translation that may be subsequently monitored by eIF3j.

Figure 4.

eIF3j inhibits circRNA translation by inducing the disassociation of the eIF3 complex from translatable circRNAs. (A) A schematic model of the eIF3 complex, adapted from (59). eIF3a and eIF3b serve as the nucleation core to bring other subunits together, while the eIF3-associated factor eIF3j only loosely interacts with other subunits. (B) The approximate locations of the indicated dsRNAs are shown at each gene locus for C–F. (C, D) Codepletion of eIF3a and eIF3j in the circSfl stable cell line. Western and northern blots were performed to detect CdSfl and circSfl, respectively. β-gal dsRNA served as a negative control. The level of CdSfl and circSfl were quantified from three independent blots. ^∗∗P< 0.01; ^∗P < 0.05. (E, F) Codepletion of eIF3b and eIF3j in the circSfl stable cell line. Western and northern blots were performed to detect CdSfl and circSfl, respectively. β-gal dsRNA served as a negative control. The level of CdSfl and circSfl were quantified from three independent blots. ^∗∗P< 0.01; ^∗P < 0.05. (G) The gene locus and overexpression vector of eIF3j with the approximate location of the indicated dsRNA are shown for H-K. RT-qPCR was performed to quantify the knockdown and overexpression efficiency of eIF3j. ^∗∗P< 0.01; ^∗P < 0.05. (H, I) CLIP assays of FLAG-tagged eIF3a (H) or eIF3b (I) using the notag circSfl stable cell line (see Figure 1M) depleted of eIF3j. β-gal dsRNA served as a negative control. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of eIF3a (H) or eIF3b (I) to circSfl, circlaccase2, circdati, U6 snRNA, and the endogenous linear Sfl mRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. (J, K) CLIP assays of FLAG-tagged eIF3a (J) or eIF3b (K) using the notag circSfl stable cell line overexpressing V5-tagged eIF3j. Empty vector served as a negative control. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of eIF3a (J) or eIF3b (K) to circSfl, circlaccase2, circdati, U6 snRNA and the endogenous linear Sfl mRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from three independent biological replicates and are shown as means ± SEM.

To explore whether eIF3j inhibits eIF3-mediated circRNA translation, CLIP-RT-qPCR was used to examine the impact of eIF3j on the recruitment of the nucleation core of eIF3 to circSfl (Figure 4G-K). Intriguingly, both eIF3a and eIF3b had a significantly increased ability to bind to circSfl when cells were depleted of eIF3j (Figure 4H, I), and overexpression of eIF3j resulted in the reduced binding of eIF3a and eIF3b to circSfl (Figure 4J, K). The binding of eIF3a and eIF3b to control RNAs (e.g. circlaccase2 and the endogenous linear Sfl mRNA) was largely unaffected, thereby excluding the possibility of non-specific binding (Figure 4H–K). These results demonstrate that eIF3j represses circRNA translation by blocking the eIF3 complex from binding to translatable circRNAs.

eIF3j functions in circSfl translation through its C-terminus

We next aimed to identify the functional domain of eIF3j which contributes to the specialized translation repression, and a series of eIF3j mutant vectors were constructed (Figure 5A). Overexpression of the mutant eIF3j deleted of the C-terminus did not attenuate the extent of circSfl translation (Figures 5B, line 1 versus 6, C); however, mutants containing other deletions were still sufficient to inhibit CdSfl production, which is analogous to what was observed with the wild-type eIF3j (Figure 5B, line 1–5, C). Moreover, northern blots investigating circSfl expression excluded the possibility that these eIF3j mutants affected circSfl biogenesis/circularization (Figure 5B, C). These results indicate that the C-terminus of eIF3j is essential for eIF3j-mediated circRNA translation repression.

Figure 5.

eIF3j-mediated circRNA translation repression is dependent of its C-terminus. (A) Schematics of eIF3j mutant vectors. (B, C) The individual overexpression of the wild-type and mutant eIF3j vectors in the circSfl stable cell line. Western and northern blots were performed to detect CdSfl and circSfl, respectively. Empty vector served as a negative control. The CdSfl level was quantified from six independent western blots and circSfl expression was quantified from three independent northern blots. ^∗∗P< 0.01; ^∗P < 0.05. (D) CLIP assays of the wild-type (WT), N-terminus depleted (ΔN), and C-terminus depleted (ΔC) eIF3j (V5-tagged) using the circSfl stable cell line. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of the wild-type and mutant eIF3j to circSfl, circlaccase2, circdati, U6 snRNA, and the endogenous linear Sfl mRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. (E) CLIP assays of FLAG-tagged Rps23 using the notag circSfl stable cell line (see Figure 1M) depleted of eIF3j. β-gal dsRNA served as a negative control. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of Rps23 to circSfl, circlaccase2, circdati, U6 snRNA, and the endogenous linear Sfl mRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. (F) CLIP assays of FLAG-tagged Rps23 using the notag circSfl stable cell line overexpressing V5-tagged eIF3j. Empty vector served as a negative control. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of Rps23 to circSfl, circlaccase2, circdati, U6 snRNA, and the endogenous linear Sfl mRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from at least three independent biological replicates and are shown as means ± SEM.

The C-terminus of eIF3j exhibits evolutionary conservation across eukaryotes (from Drosophila to human) (Supplementary Figure S9) and was predicted to bind to circSfl with a high discriminative power up to 68% (71). In line with the prediction, CLIP-RT-qPCR revealed that, compared with the wild-type eIF3j, the mutant with C-terminus deletion had a significantly reduced ability to bind to circSfl (Figure 5D). By contrast, no change was observed with the mutant with N-terminus deletion (Figure 5D). These results indicate that eIF3j recruitment to translatable circRNAs requires its C-terminus and is indispensable for its accurate function in translation repression. As a control, the binding of eIF3j to circlaccase2, circdati, U6 snRNA, and the endogenous linear Sfl mRNA was not affected in the absence of the C-terminal region (Figure 5D), further confirming eIF3j-mediated translation repression is specific to circSfl. It is also worth noting that eIF3j can bind to the small ribosomal protein Rps23 occurring near the aminoacyl (A) site and mRNA entry channel of the 40S subunit, which subsequently blocks eIF3 loading to the A site and prevents mRNA recruitment (69,72). Notably, the C-terminus of eIF3j has been shown to be indispensable for the high affinity of eIF3j to the 40S subunit (69,72,73). Therefore, we next investigated the role of eIF3j in the interaction between circSfl and Rps23. As observed, neither knockdown nor overexpression of eIF3j interfered with the binding of Rps23 to circSfl (Figure 5E, F), somehow excluding the possibility that eIF3j blocks the recruitment of translatable circRNAs to ribosomes. Nonetheless, we cannot rule out that eIF3j could block the access of eIF3 to circSfl by contacting Rps23 in the A site.

Taken together, we concluded that eIF3j makes specific interactions with translatable circRNAs, and these contacts, in turn, prevent the eIF3 complex from binding to circRNAs and initiating their translation.

Identification of the UTR region required for eIF3j-mediated circSfl translation repression

It has been previously reported that cis-acting RNA elements located in UTRs of protein-coding RNAs can provide additional layers of translational control to ensure proper protein expression (39,41,74,75). For example, the internal stem-loop structure embedded in the 5′ UTR of c-Jun mRNA blocks eIF4E-dependent translation to ensure eIF3d-specialized cap recognition (39,41). Given that circSfl harbors a 467 nt UTR, we thus tested whether an RNA regulon within the UTR of circSfl is functionally essential to eIF3j-mediated regulation. To this end, a series of circSfl mutant vectors containing truncated UTRs were generated and used for stable cell line construction (Figure 6A). Western and northern blots were performed to examine the level of CdSfl and circSfl, respectively. Of five circSfl UTR mutants, four (Δ1–100, Δ201–300, Δ301–400 and Δ401–467) yielded a reduction in CdSfl production, suggesting that these UTR regions promote circRNA translation. Particularly, the translation of circSfl was barely detectable when the circSfl UTR was individually deleted of nucleotides 1–100, 201–300 and 401–467 (Figure 6B, C). Note that circSfl expression also dropped to ∼43% upon deletion of nucleotides 201–300, indicating that the decreased CdSfl production of the Δ201–300 mutant was attributed to a combination of the reduced biogenesis and translation of circSfl (Figure 6B, C).

Figure 6.

Identification of the UTR region required for the inhibitory effect of eIF3j on the eIF3 complex. (A) Schematics of circSfl mutants containing truncated UTRs (Δ1–100, Δ101–200, Δ201–300, Δ301–400 and Δ401–467). (B, C) Western and northern blots were performed to detect CdSfl and circSfl in S2 cells stably expressing these circSfl mutants. The CdSfl level was quantified from five independent western blots and circSfl expression was quantified from three independent northern blots. (D) The approximate locations of the indicated dsRNAs are shown at each gene locus for E and F. (E, F) Western and northern blots were performed to detect CdSfl and circSfl in the Δ101–200 circSfl stable cell line depleted of eIF3a or eIF3b. β-gal dsRNA served as a negative control. The level of CdSfl and circSfl were quantified from three independent blots. (G) The nucleotide sequence and a possible secondary structure of nucleotides 101–200 of the circSfl UTR. (H) iCLIP assays of FLAG-tagged eIF3j using the notag circSfl stable cell line (see Figure 1M). 8 overlapping amplicons were used to tile through the UTR of circSfl. Data were normalized to the negative IgG sample. (I) The gene locus and overexpression vectors of eIF3j with the approximate locations of the indicated dsRNAs are shown for J–M. (J, K) Western and northern blots were performed to detect CdSfl and circSfl in the Δ101–200 circSfl stable cell line depleted of eIF3j (J, left panel) or overexpressing eIF3j (J, right panel). β-gal dsRNA and empty vector served as a negative control in these depletion and overexpression experiments, respectively. The level of CdSfl and circSfl were quantified from three independent blots. (L, M) Western and northern blots were performed to detect CdSfl and circSfl in the circSfl stable cell line depleted of eIF3j (L, left panel) or overexpressing eIF3j (L, right panel). β-gal dsRNA and empty vector served as a negative control in these depletion and overexpression experiments, respectively. The level of CdSfl and circSfl were quantified from three independent blots. (N) The primer set spanning across the junction site and the UTR of circSfl was used for O–Q. (O–Q) CLIP assays of V5-tagged eIF3j (O), eIF3a (P), and eIF3b (Q) using the circSfl stable cell line and the circSfl mutant stable cell line (Δ101–200 and Δ301–400). RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of the indicated eIF3 subunits to the wild-type and mutant circSfl. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from at least three independent biological replicates and are shown as means ± SEM.

By contrast, the CdSfl level had a ∼69% increase when the circSfl UTR was deleted of nucleotides 101–200, despite a ∼26% reduction in circSfl expression. This indicates that nucleotides 101–200 negatively regulate circSfl translation, which is distinct from the results from other four UTR regions (Figure 6B, C). Next, we individually knocked down eIF3a and eIF3b using the Δ101–200 circSfl stable cell line and observed that, although there was no change in circSfl expression, circSfl failed to produce CdSfl in the absence of the nucleation core of eIF3 (Figure 6D–F). This suggests that the translation of the mutant circSfl lacking nucleotides 101–200 of the UTR is still in an eIF3-dependent manner, which is consistent with the result from the wild-type circSfl. It is known that GC-rich RNA elements in UTRs have the potential to form a stable secondary structure to block ribosome scanning (76). In the case of circSfl, the GC content of nucleotides 101–200 of the UTR is only 40% (Figure 6G). It is thus unlikely that this UTR region intrinsically impedes circRNA translation initiation, which is supported by the RNAfold structure prediction (Figure 6G) (77,78). Instead, this UTR region may rely on other trans-acting factors, such as eIF3j, to exert its inhibitory role.

Therefore, we tested the functional relevance of nucleotides 101–200 to eIF3j. To identify the UTR region where eIF3j recognizes, iCLIP-RT-qPCR experiments were performed with 8 overlapping amplicons tiling through the UTR of circSfl. We found that eIF3j exhibited a higher binding capacity to nucleotides 101–200 compared to other UTR regions (Figure 6H). In addition, neither depletion nor overexpression of eIF3j affected CdSfl production of the Δ101–200 mutant (Figure 6I-K), in contrast to the phenotypes obtained from the wild-type circSfl (Figure 6L, M) and the Δ301–400 mutant (Supplementary Figure S10A, B). Note that we also examined the effect of eIF3j knockdown on circSfl translation using the Δ1–100 and Δ401–467 circSfl stable cell line. No change was observed with the CdSfl level, indicating that Δ1–100 and Δ401–467 are simply dead mutants (Supplementary Figure S10C–F). These results thus support that the fate of undergoing eIF3j-mediated translational regulation is encrypted in the context of nucleotides 101–200 of the circSfl UTR. To further gain a mechanistic insight into how this cis-acting RNA regulon coordinates eIF3j, CLIP-RT-qPCR was used to examine eIF3j recruitment to the Δ101–200 mutant. Compared with the wild-type circSfl, this mutant almost completely lost the ability to interact with eIF3j (Figure 6N, O). Correspondingly, the binding of the nucleation core of eIF3 to the Δ101–200 mutant exhibited a significant increase (Figure 6N, P, Q). As a control, the binding of the Δ301–400 mutant to eIF3a, eIF3b, and eIF3j was not altered (Figure 6N–Q). Taken together, these findings demonstrate that nucleotides 101–200 of the circSfl UTR facilitate the binding of eIF3j to circSfl, and, in turn, ensure the inhibitory effect of eIF3j on the eIF3 complex.

eIF3j-mediated translational regulation in response to stressed conditions

The expression of circRNAs is perturbed in response to various stresses and a great many circRNAs serve as regulatory factors under different physiological or pathological conditions to maintain cellular homeostasis (16–20). Notably, emerging studies have demonstrated that some circRNAs exert functions through their encoded proteins instead of themselves (31–33). However, the physiological relevance of eIF3j-mediated circRNA translation program remains an unknown question. To fill this gap, S2 cells stably expressing the wild-type circSfl (WT) or the GCG mutant circSfl (GCG MUT) were used for the subsequent study (Figure 1M, N). Considering that the endogenous circSfl is weakly expressed in the regular S2 cell line (Supplementary Figure S1F) (46,79), these circSfl stable cell lines served as ideal models of circSfl highly expressed cells or tissues (e.g. neuronal cells (34)) with no or marginal influence from the endogenous circSfl. In addition, it was easy to distinguish the physiological role of CdSfl from that of circSfl, since GCG MUT cells only generate the mutant circSfl without translation ability (Figure 1M, N).

Given that circSfl was found to be a type of insulin-sensitive circRNAs (34) and that heat shock response is involved in the regulation of insulin sensitivity (80), we thus investigated the potential physiological role of circSfl and CdSfl in response to heat stress. After heat shock treatment, the number of stressed cells was counted. Compared with regular S2 cells and GCG MUT cells, WT cells exhibited a significant resistance to heat stress (Figure 7A). In addition, neither circSfl nor CdSfl affected cell proliferation under unstressed conditions (Figure 7B). These findings support that circSfl only functions through its encoded protein CdSfl during heat stress.

Figure 7.

The physiological relevance of eIF3j-mediated translation repression in cellular response to stressed conditions. (A) Examination of cellular resistance to heat shock. S2 cells stably expressing the wild-type circSfl (WT) or the GCG mutant circSfl (GCG MUT; see Figure 1M) were cultured at 37°C for the indicated amounts of time and the number of stressed cells was counted. The resistance of regular S2 cells served as a negative control. ^∗∗P< 0.01; ^∗P < 0.05. (B) Examination of cell proliferation of the indicated cell line under unstressed conditions. The number of cells was counted at different time points. ^∗∗P< 0.01; ^∗P < 0.05. (C–E) Examination of cellular resistance of the indicated cell line to heat shock after cells had been depleted of eIF3j. The number of heat-shocked cells was counted at different time points. β-gal dsRNA served as a negative control. ^∗∗P< 0.01; ^∗P < 0.05. (F–H) Examination of cellular resistance of the indicated cell line to heat shock after cells had overexpressed V5-tagged eIF3j. The number of heat-shocked cells was counted at different time points. Empty vector served as a negative control. ^∗∗P< 0.01; ^∗P < 0.05. (I, J) Western blots of CdSfl with protein extracts from the heat-shocked circSfl stable cell line. The CdSfl level was quantified from four independent western blots. ^∗∗P< 0.01; ^∗P < 0.05. (K) CLIP assays of V5-tagged eIF3j using the unstressed and heat-shocked circSfl stable cell line. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of eIF3j to circSfl, circlaccase2, circdati, and U6 snRNA. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. (L) RT-qPCR quantification of the endogenous eIF3j mRNA in the unstressed and heat-shocked circSfl stable cell line. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from at least three independent biological replicates and are shown as means ± SEM.

Based on the above observation, we next asked whether eIF3j affects heat resistance through modulating circSfl translation. Depletion of eIF3j significantly increased the heat resistance of WT cells but not regular S2 cells or GCG MUT cells (Figure 7C–E). Moreover, overexpression of eIF3j significantly reduced the heat resistance, which is also specific to WT cells (Figure 7F-H). These results imply that eIF3j physiologically reduces heat resistance through downregulation of CdSfl production. In support, the CdSfl level exhibited a remarkable decrease during heat stress over time (Figure 7I, J). Particularly, CdSfl was almost completely eliminated after 4 hr heat shock (Figure 7I, J). Mechanistically, eIF3j recruitment to circSfl was significantly elevated in response to heat shock (Figure 7K), despite no change in eIF3j expression (Figure 7L). Taken together, these findings indicate that eIF3j is able to physiologically regulate heat resistance through modulating circRNA translation, thereby ensuring clean of damaged cells.

eIF3j regulates translation of a subset of circRNAs

To explore whether eIF3j-mediated regulation represents a widespread mechanism for circRNA translation, we examined the binding of eIF3j to 15 previously annotated endogenous ribo-circRNAs, which were identified by ribosome footprinting from S2 cells (28). As observed, four were found to significantly interact with eIF3j (Figure 8A). Among these four eIF3j-associated ribo-circRNAs, three (e.g. circPde8) exhibited a significantly increased binding capacity to the nucleation core of eIF3 upon eIF3j knockdown (Figure 8B), which is similar to what was observed with circSfl (Figure 4H, I). To further validate the role of eIF3j in translation repression of other circRNAs, we chose circPde8 for subsequent experiments and constructed a cell line stably expressing circPde8 (Figure 8C). The circPde8 stable cell line was confirmed to successfully generate a cytoplasmic protein (Figure 8D, E), which was referred to circPde8-derived Pde8 (CdPde8) herein. Knockdown of eIF3a and eIF3b reduced the translation efficiency of circPde8, while knockdown of eIF3j elevated the level of CdPde8 (Figure 8F, G). The RNA level of circPde8 was almost not affected in above knockdown experiments (Figure 8H), excluding the possibility of altered circPde8 biogenesis. Taken together, these results support that eIF3j inhibits translation of at least a subset of circRNAs in Drosophila S2 cells.

Figure 8.

eIF3j regulates translation of additional circRNAs. (A) CLIP assays of FLAG-tagged eIF3j using regular S2 cells. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of eIF3j to the indicated endogenous ribo-circRNAs. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P< 0.05. (B) CLIP assays of FLAG-tagged eIF3a using regular S2 cells depleted of eIF3j. β-gal dsRNA served as a negative control. RT-qPCR was performed with RNA extracts from CLIP samples to measure the binding of eIF3a to the eIF3j-associated ribo-circRNAs. Data were normalized to the negative IgG sample. ^∗∗P< 0.01; ^∗P < 0.05. (C) A schematic overview of the construction of the circPde8 expression vector which is modified from the previously described Hy_pMT laccase2 MCS exon vector. The circPde8 vector was used to generate a stable cell line using S2 cells for D–H. (D, E) Immunofluorescence assays were performed to examine the subcellular localization of CdPde8 in the circPde8 stable line. Scale bar = 5 μm. The nuclear (Nuc) and cytoplasmic (Cyto) signals of CdPde8 were quantified from 102 cells. (F–H) The individual depletion of eIF3a, eIF3b, and eIF3j in the circPde8 stable cell line. Western blots and RT-qPCR were performed to detect CdPde8 (F, G) and circPde8 (H), respectively. β-gal dsRNA served as a negative control. The CdPde8 level was quantified from three independent blots. ^∗∗P< 0.01; ^∗P < 0.05. All data were generated from three independent biological replicates and are shown as means ± SEM.

DISCUSSION

eIFs in circRNA translation

Due to the sequence homology between circRNAs and their linear counterparts, circRNA-derived proteins usually share the same amino acid sequence (or even the same start codon) with their cognate full-length proteins (34,81), implying that the functions of these novel truncated proteins are somehow auxiliary to their full-length versions. Indeed, the circSMO-derived protein SMO193aa, identical to amino acids 230–421 of the full-length SMO, can promote SMO-mediated Hedgehog signaling activation and tumorigenicity in glioblastoma patients (81). However, some circRNA-derived proteins have distinct subcellular localizations and biological functions compared to their full-length versions (82). A case in point is circARHGAP35 whose encoded protein promotes cancer progression by forming a complex with the transcription factor TFII-I in the nucleus, whereas the full-length ARHGAP35 inhibits tumor growth by switching off RhoA activity in the cytoplasm (82). These studies indicate that (i) circRNA-derived proteins are not a simple supply to the full-length cognates, (ii) the accurate control of circRNA translation is fundamental to pattern biological processes, and (iii) cells require a mechanism to specifically coordinate the translation of these novel proteins. In this study, we evaluated the impact of all Drosophila eIFs on circRNA translation and the focused investigation regarding circSfl proposes the molecular mechanism of eIF3j-mediated translational control: eIF3j interacts with translatable circRNAs and inhibits translation by preventing the eIF3 complex from binding to circRNA templates (or displacing eIF3 from templates) (Figure 9), providing a pathway that assists the general translational machinery to specifically recognize circRNA templates. The inhibitory activity of eIF3j requires its C-terminus and relies on an RNA regulon located in the circRNA UTR. Moreover, we revealed that the binding of eIF3j to circRNAs varies in response to stressed conditions, thereby influencing the translation ability of stress-responsive circRNAs and, in turn, cellular physiology. In summary, our study provides a significant insight into the field of cap-independent transcript-specific translation.

Figure 9.

A working model for eIF3j-mediated circRNA translation repression. eIF3j inhibits circRNA translation initiation by blocking the eIF3 complex from binding to translatable circRNAs (possibly in the A site).

It is important to emphasize that eIF3j was also shown to directly interact with Rps23 in the ribosomal decoding center (69,72) and the C-terminus of eIF3j is required for its binding to the 40S ribosomal subunit (69,73), reminiscent of the phenotype that the C-terminus of eIF3j facilitates its recruitment to translatable circRNAs (Figure 5A-D). In this regard, we observed that eIF3j was not involved in the association between Rps23 and circRNAs (Figure 5E, F). Therefore, it is not likely that eIF3j prevents the recruitment of translatable circRNAs to ribosomes. But this hypothesis needs to be further validated. Moreover, the binding capacity of eIF3j to the 40S subunit decreases in the presence of linear mRNAs (83,84). We thus speculate a possible role of eIF3j in discrimination of circular and linear translation templates. On the other hand, an interplay between eIF3j and eIF1A has also been revealed (69,72,85,86). They can bind anticooperatively to the 40S subunit surface (69,86) or closely cooperate to orchestrate the process of AUG recognition (72). However, no robust impact of eIF1A on circSfl translation was observed in our screening. eIF1A seems to regulate circSfl biogenesis to affect the expression level of CdSfl (Figure 2B-D). Taken together, the previous findings as well as our study strongly support that eIF3j represents a regulatory/accessory factor for eIF3 and is not a bona fide eIF3 subunit.

In addition to eIF3j, two tightly interacting octameric partners eIF3k and eIF3l were also found to robustly inhibit circSfl translation in our screening (Figure 2B-D), suggesting that they can exert a function different from the canonical role of eIF3. In support, only eIF3k and eIF3l are dispensable for normal growth and viability of Caenorhabditis elegans among all 12 eIF3 subunits (87). The rate of bulk translation initiation is not attenuated in worm mutants lacking eIF3k or eIF3l (87). Moreover, a recent study has demonstrated that eIF3k and eIF3l are non-essential eIF3 subunits with no effect on the integrity of the whole eIF3 complex and are able to antagonize mRNA recruitment to the 43S PIC in human cells (88). As exemplified by RPL41 mRNA, the individual knockdown of eIF3k and eIF3l significantly promoted the recruitment level of RPL41 mRNA up to 124–132% (88). Based on these findings, we thus speculate that eIF3k and eIF3l may not be specific for translational control of circRNAs.

eIF2 is a heterotrimeric complex (containing eIF2α, eIF2β, and eIF2γ) used to transfer Met-tRNAi to the 40S ribosomal subunit. The action of eIF2 is generally considered as a rate-limiting step in mRNA translation (51–54). However, we observed that knockdown of eIF2γ (the core of eIF2) and eIF2α had only a limited effect on the translation efficiency of circSfl (Figure 2B–D). This indicates that circRNA translation (at least for circSfl) may be controlled via an eIF2-independent mechanism. In fact, the eukaryotic translation initiation machinery is able to operate without eIF2 under stressed conditions (55–57). For example, in human cells, the hepatitis C virus (HCV) IRES can make use of a bacterial-like pathway to direct translation and 80S complex formation with the assistance of eIF3 and eIF5B when eIF2 is inactivated by phosphorylation (56). In addition to eIF5B, eIF2A and eIF2D could be other candidates for Met-tRNAi delivery in circRNA translation, since there are studies showing these eIFs can deliver the initiator tRNA into the ribosome under specific contexts (89–91). For example, eIF2A facilitates Met-tRNAi delivery through a direct interaction with a stem-loop structure located in the IRES of c-Src mRNA, which is required for cell growth under stress conditions (89).

eIFs in circRNA biogenesis

Besides circRNA translation, we also identified a subset of eIFs as potent regulators required for circSfl biogenesis. For example, the circSfl level dropped to ∼16% upon depletion of eIF4A (Figure 2D), an essential subunit of the eIF4F complex. eIF4F comprises the scaffold protein eIF4G, the cap-binding protein eIF4E, and the DEAD-box RNA helicase eIF4A (51–54). Different from eIF4A, knockdown of eIF4E1 and its paralogs only had a marginal effect on circSfl biogenesis (Figure 2D). This somehow rules out the possibility that eIF4A indirectly affects circSfl biogenesis through controlling the translation of circRNA biogenesis factors. Instead, eIF4A may function via an eIF4F-independent mechanism. Consistent with our hypothesis, an array of studies have demonstrated that eIF4A3 can promote the back-splicing reaction by directly interacting with the flanking sequence of circularizing exons in mammalian cells (92–94). Since several eIFs have been reported as nucleocytoplasmic shuttling proteins (95–97), the nuclear functions of eIFs should be taken into consideration in future.

RNA regulons in circRNA translation

CircRNA lacks the 5′ cap structure utilized by the canonical translation pathway. To data, different RNA elements have been reported to function in the cap-independent translation of protein-coding circRNAs. For example, the IRES-like regulon, a complex scaffold typically present in the circRNA UTR, is capable of directly recruiting the translation initiation machinery in the absence of the 5′ cap structure and cap-binding protein eIF4E (27,28). In addition, a subset of circRNAs consist of the consensus m⁶A motif which is specifically recognized by the m⁶A reader YTHDF3. YTHDF3 then recruits eIF4G2 to m⁶A-modified circRNAs to initiate their translation (29). Our study is unique in that we revealed that eIF3j negatively regulates circSfl translation through a direct interaction with an RNA regulon present in the UTR, further supporting a combination of cis- and trans-acting regulators in the translational regulation of circRNAs.

Theoretically, exogenous circRNA is an ideal translation tool to generate functional proteins due to its long half-life. However, one fundamental limitation to its broad application is its relatively low translation initiation efficiency in eukaryotic system (27–29). To overcome this obstacle, an array of RNA elements, including IRES and 18S rRNA complementary sequences, are used to engineer circRNAs (98–100). But these inserts also generate some side effects, such as forming unexpected structures with proximal sequences or even distal sequences through long-distance contacts (98,99). Differently, the RNA regulon (nucleotides 101–200 of the circSfl UTR) identified in our study has a low potential to form a complex structure (Figure 6G). Use of this RNA regulon may be an alternative strategy to engineer circRNAs for controllable translation without influences on the overall circRNA structure, which is one of our aims currently.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online.

FUNDING

National Natural Science Foundation of China [32270601, 32070633]; Chongqing Talents Plan for Young Talents [cstc2022ycjh-bgzxm0140]; Fundamental Research Funds for the Central Universities of China [2022CDJXY-004, 2021CDJZYJH-002]; Innovation Support Program for Overseas Returned Scholars of Chongqing, China [cx2019142]. Funding for open access charge: National Natural Science Foundation of China [32270601].

Conflict of interest statement. None declared.