Circular RNA ZNF827 tunes neuronal differentiation by facilitating transcriptional repression of Neuronal Growth Factor Receptor

Circular RNAs (circRNAs) comprise a large class of conserved non-coding RNAs that regulate a number of biological processes, including transcription, alternative splicing, translation and mRNA decay. circRNAs are enriched in the brain and mammalian cells subjected to neuronal differentiation markedly change their circRNA transcriptomes. Here, we have mapped high-confidence circRNA inventories of mouse embryonic stem cells, neuronal progenitor cells and in differentiated glutamatergic neurons and identify hundreds of differentially expressed circRNAs. Among several candidate circRNAs, knockdown of circZNF827 significantly induces expression of key neuronal markers, suggesting that this molecule negatively regulates neuronal differentiation. Using Nanostring analyses we demonstrate that among 770 tested genes linked to known neuronal pathways and neuropathological states, knockdown of circZNF827 deregulates expression of several genes including a robust upregulation of neuronal growth factor receptor (NGFR). We show that NGFR becomes transcriptionally upregulated and that this functionally enhances NGF signalling. Our results suggest that circZNF827, although being highly enriched in the cell cytoplasm, can elicit transcriptional changes, which in turn balances proliferation and neuronal differentiation signalling.


Introduction
The mammalian non-coding transcriptome, which includes long noncoding RNAs (lncRNAs) and circular RNAs (circRNAs), play pivotal roles in biological decisions during differentiation and normal cell maintenance (reviewed in 1-3 ).
Even though circRNAs were already identified several decades ago 4-7 , they have only recently emerged as a large class of abundant noncoding RNAs that exhibit cell type-and tissue-specific expression patterns 8-14 (reviewed in 3,15,16 ). CircRNAs are generated by the canonical spliceosome in a nonlinear backsplicing fashion 9,10,12,17,18 . During circRNA biogenesis, flanking intronic sequences are thought to bring splice sites within critically close proximity, either by direct basepairing between inverted repeats (e.g. Alurepeats in primates) or facilitated by interactions between flanking intronbound RNA-binding proteins (RBPs) 13,15,19 . circRNAs are largely localized to the cell cytoplasm 9-14 , and recent evidence suggests that nuclear export of circRNAs in human cells is influenced by the size of the given molecules 20 .
At the functional level, several reports have provided evidence that circRNAs play important roles in various fundamental cellular processes. A well described example is the CDR1as/CiRS-7 and SRY circRNAs that function to negatively regulate miR-7 and miR-138 activity, respectively, by sequestration (miRNA sponging), leading to increased mRNA expression of their respective miRNA-targets 11,12 . However, it has been suggested that the majority of circRNAs are likely not bona fide miRNA sponges, simply due to relatively low copy numbers and a low number of miRNA binding sites per molecule, leaving efficient miRNA regulation ambigous in many cases 3,15 . Examples of circRNAs acting as binding scaffolds for RBPs, or RBP sponges, which in turn affect their canonical function in e.g. pre-mRNA splicing and protein translation, have been reported 13,21 . Nuclear variants coined exon-intron circular RNAs (EIciRNAs), have, due to their retention of intronic sequences, been shown to promote transcription by recruitment of U1 snRNP to transcription units by a not fully clarified mechanism 22 . Many abundant circRNAs originate from the 5' end of their precursor transcripts, often giving rise to backsplicing into parts of the 5'UTR of their linear relative 10,12,14 , suggesting that at least some circRNAs might have protein-coding potential via a cap-independent translation mechanism, which is consistent with both early studies of Internal Ribosome Entry Sites (IRES) placed in a circRNA context 23 , as well as more recent studies reporting examples of translationcompentent circRNAs [24][25][26][27] . However, global analyses of hundreds of publicly available ribosome profiling datasets questions the prevalence, frequency and importance of such events 28 . Evidence from RNA-sequencing of RNA isolated from mouse and human tissues along with various cell lines, suggest that circRNAs are most abundantly expressed in the brain, compared to other tissues and that circRNAs are particular enriched in neuronal synaptosomes 14  CDR1as knockout mouse displayed downregulated miR-7 levels, alterations in sensorimotor gating associated with neuropsychiatric disease and abnormal synaptic transmission, suggesting that CDR1as and miR-7 is important for normal brain function in the mouse 29 . Adding to the complexity of this regulatory network, a long noncoding RNA (lncRNA), Cyrano, promotes the destruction of miR-7, which in turn upregulates CDR1as by a still unidentified mechanism 30 . Despite this intricate molecular interplay between a circRNA, miRNA and lncRNA, many important questions regarding neuronal differentiation and function remain unanswered. For example, it is currently unknown whether the tightly controlled expression of circRNAs affect neuronal development. Here, we present the circRNA inventory of mouse embryonic stem cells (mESC), neuronal progenitor cells (NPC) and differentiated glutamatergic neurons, which represents a well-established model for CNStype neuronal differentiation 31 . We report thousands of RNase R-resistant circRNAs of which many are differentially regulated during neuronal development. In a miniscreen for circRNA function using a well-established human model for neuronal differentiation, we identify circZNF827 as a negative regulator of neuronal differentiation, since multiple neuronal markers (TrkB, NEFL, MAP2, TUBB3, RARs and NGFR) become upregulated upon circZNF827 knockdown, while known negative regulators (PTEN, NQO1 and STAT3) become downregulated. Interestingly, although being almost exclusively localized to the cell cytoplasm, circZNF827 knockdown impact several upregulated genes, including NGFR, at the level of transcription.
Taken together, our results suggest that circZNF827 serves to tune the balance between neuronal proliferation and differentiation.

Cell culturing
L-AN-5 cells were maintained in RPMI and SHSY-5Y and P19 cells were maintained in DMEM medium, both supplemented with 10% fetal calf serum (Gibco) and 1% penicillin/streptomycin (Gibco, 15140122). All cells were cultured at 37°C in 5% (v/v) CO 2 . Differentiation of L-AN-5 and SHSY-5Y cells were stimulatied by addition of 10 µM RA to the cell culture medium. Both cell types were differentiated for four days. inhibitor (CHIR-99021) and 1 µM MEK inhibitor (PD0325901). They were differentiated into neurons as previously described 31 with some modifications. 4 million cells were differentiated into embryoid bodies in suspension in petri dishes for bacterial culture in 15ml medium containing the same as before, but with 10% FBS and without LIF or GSK3 and MEK inhibitors. Every second day, the medium was changed and the embryoid bodies transferred to fresh petri dishes. On days 4 and 6, 5 µM ATRA (sigma, R2625) was added to the medium. On day 8 of differentiation, the embryoid bodies were disgregated with 5% trypsin (Gibco, 15400054) and the cells plated in poly-DL-ornithine (Sigma, P8638) and laminin (Sigma, L2020) coated plates in N2 medium, containing DMEM/F12 and neurobasal 1:1, N2 supplement, sodium pyruvate, glutaMax, 15 nM 2-mercaptoethanol, and 50 µg/ml BSA. The medium was changed after 2 h and after 24 h. 48 h after plating the neuronal precursors, the medium was changed to complete medium, containing B27 supplement, in addition to the N2 medium. Neurons were harvested 2 and 8 days after plating.

Lentiviral production
Third-generation lentiviral vectors were produced in HEK293T cells as previously described 34 . One day before transfection, cells were seeded in 10cm dishes at a density of 4 × 10 6 cells/dish. Transfections were carried out with 3.75 µg pMD.2G, 3 µg pRSV-Rev, 13 µg pMDLg/pRRE and 13 µg lentiviral transfer vector using a standard calcium phosphate or polyethylenimine transfection protocol. Medium was changed to RPMI medium one day after transfection. Two days after transfection viral supernatants were harvested and filtered through 0.45 µm filters (Sartorius).
All lentiviral preparations were made in at least triplicates and pooled before determination of viral titers. To determine viral titers of lentiviral preparations, flow cytometric measurements of EGFP expression were used as previously described 34 . One day prior to transduction, L-AN-5 cells were seeded at a density of 5 x 10 5 cells/well in 12-well plates. For all lentiviral preparations, transductions with 10 2 -and 10 3 -fold dilutions of virus-containing supernatants were carried out. Both viral supernatants and growth medium were supplemented with 4 µg/ml polybrene (Sigma-Aldrich). One day after transduction, medium was changed. Five days after transduction, cells were

circRNA knockdown and differentiation
One day prior to transduction with lentiviral vectors encoding circRNA-specific dishRNAs, L-AN-5 cells were seeded at a density of 2.2 x 10 6 cells/dish in 6cm dishes or 0.8 x 10 6 cells/well in 6-well plates. Transductions were carried out using equal MOIs calculated based on titers determined by flow cytometry. Both viral supernatants and growth medium were supplemented with 4 µg/ml polybrene. One day after transduction, medium was changed.
Two days after transduction, differentiation was initiation by addition of 10 µM RA to the cell culture medium. The L-AN-5 and SHSY-5Y cells were differentiated for four days.
BrU pulse-chase mRNA decay assay Lentiviral transduction and RA-mediated differentiation of L-AN-5 cells were carried out as described in the section 'circRNA knockdown and differentiation'. The L-AN-5 cells were cultured in 6-cm dishes containing 6 ml cell culture medium supplemented with 10 µM RA. 4 ml cell culture medium was aspirated from each 6-cm dish and pooled from cells transduced with the same dishRNA. For one dish per dishRNA, the residual medium was aspirated and 3.5 ml of the collected medium was added. For the remaining dishes, the residual medium was aspirated and 3.5 ml of the collected medium supplemented with 2 mM BrU (Thermo-Fisher) was added. 1 hour after addition of BrU to the cell culture medium, the cells were washed three times in cell culture medium. 50 min after removal of the BrU-containing cell culture medium the first samples including the samples not treated with BrU were harvested. Subsequently, samples were harvested after 3, 6 and 9 hours. Total RNA was purified using 1 ml TRI Reagent (Sigma-Aldrich) according to manufacturer's protocol. circRNA knockdown and differentiation of L-AN-5 cells were verified by RT-qPCR using total RNA as described in the section 'Quantitative PCR'. BrU-labeled RNA was immunoprecipitated as described elsewhere 35  RNA was purified by phenol/chloroform extraction, ethanol precipitation and the RNA pellets were resuspended in 10 µl nuclease free water. 2 µl of immunoprecipitated RNA was used for quantification of mRNA expression levels by RT-qPCR as described in the section 'Quantitative PCR' except that DNase treatment was omitted and 1 µg yeast RNA (Roche) was added in the cDNA reaction.

BrU-labeling and immunoprecipitation of newly synthesized RNA
The BrU-labeling and immunoprecipitation of newly labeled RNA were carried out as for the mRNA decay assay except that the cells were harvested 45 min after addition of BrU to the cell culture medium. Furthermore, after binding of the RNA to the beads, the beads were washed once in 1x BrU-IP buffer, twice in 1x BrU-IP buffer supplemented with 0.01% Triton X-100 and twice in 1x BrU-IP buffer.

Subcellular fractionation of nuclear and cytoplasmic RNA
Subcellular fractionation of nuclear and cytoplasmic RNA was carried out as previously described in 36 . Briefly, cells were washed in PBS, added 800 µl PBS and scraped off. 100 µl of the cell solution was centrifuged at 12,000 rpm for 10 sec at 4°C. Cell pellets were used for purification of total RNA using 1 ml of TRI Reagent (xx) according to manufacturer's protocol. The remaining 700 µl of the cell solution was used for subcellular fractionation of nuclear and cytoplasmic RNA. After centrifugation at 12,000 rpm for 10 sec at 4°C cell pellets were added 300 µl lysis buffer (20 mM Tris-HCl (pH 7.5), 140 mM NaCl, 1 mM EDTA, 0.5% Igepal-630 (Nonidet P-40)), incubated on ice for 2 min and centrifugated at 1000 g for 4 min at 4°C. Cytoplasmic RNA was purified from the supernatants using 1 ml TRI Reagent (Sigma-Aldrich) according to manufacturer's protocol. Pellets were washed twice in 500 µl lysis buffer, subjected to a single 5 sec pulse of sonication at the lowest settings (Branson Sonifier 250) and nuclear RNA was purified using 1 ml TRI Reagent (Sigma-Aldrich) according to manufacturer's protocol.
The cell lysates were subjected to two 5 sec pulses of sonication at the lowest settings (Branson Sonifier 250) followed by centrifugation at 4000 g for 15 min at 4°C. Glycerol were added to the supernatnats (final concentration: 10%) and protein concentrations were adjusted using Bradford protein assay dye (Bio-Rad). The protein samples were diluted in 6x loading buffer (), heated at 95°C for 3 min and seprated on a Novex WedgeWell 4-12% Tris-Glycine Gel (Invitrogen). Proteins were transferred to an PVDF Transfer Membrane (Thermo Scientific) using standard procedures. The membrans were blocked in 5% skimmed milk powder in PBS for 1 hour at room temperature. The membranes were incubated at 4°C overnight with primary antibodies diluted as indicated in table Sx in 5% skimmed milk powder in PBS. After three times wash, the membranes were incubated with goat polyclonal HRP-conjungated secondary antibodies (Dako) diluted 1:20000 in 5% skimmed milk powder in PBS. After 1 hour of incubation at room temperature, the membranes were washed three times and the bound antibodies were detected using the SuperSignal West Femto maximum sensitivity substrate (Thermo Scientific) according to manufacturer's protocol and the LI-COR Odyssey system (LI-COR Biosciences).

Statistical Analysis
In biochemical assays (conducted in at least biological triplicates) the significance of difference between samples were calculated by a two-tailed Student's t test to test the null hypothesis of no difference between the two compared groups. The assumption of equal variances was tested by an F test. p < 0.05 was considered statistically significant. Data are presented as mean ± SD. The DEseq2 pipeline 37 , which is based on the negative binomial distribution and estimates the variance-mean dependence in count data from high-throughput sequencing experiments was used to test for differential expression. pAdj <0.05 from biological triplicates were taken as significant differences.

Results
The circRNA profile of mESC changes markedly upon neuronal differentiation.
To determine whether circRNAs influence neuronal differentiation, we initially sought to confidently map the circRNA inventories at different stages of neuronal differentiation and compare this to other available circRNA datasets of neuronal origin from mice and humans 14 . Identification of circRNAs from RNA-seq experiments has often been based on quantification of relatively few reads across the unique circRNA backsplicing junction (circBase 38 ), and current circRNA prediction algorithms inevitably lead to the calling of false positives 39,40 . Hence, to immediately validate the circular nature of to-be called circRNAs, we first performed standard rRNA depletion and subsequently either included or excluded an RNase R treatment step prior to RNAsequencing. Specifically, we used an established differentiation model for CNS-type glutamatergic neurons, based on E14 mouse embryonic stem cells (mESCs) that reportedly yields a purity of glutamatergic neurons of >90% 31 .  Figure 1C and Table S2). We next assessed the circular-to-linear ratio of identified circRNAs (find_circ), by comparing splice site usage in circular vs. linear splicing events. This analysis revealed vast differences in the steady-state levels of these isoforms and demonstrated that many circRNA species are considerably more abundant than their linear precursors ( Figure 1D). Confirming previous studies, the introns flanking the circRNAs are generally longer than average introns and circRNAs often tend to cluster at the 5' end of their respective precursor RNA, yielding an enrichment in canonical 5'UTRs of the precursor RNAs ( Figure S1E-F). Our results suggest that in order to obtain high confidence circRNA inventories from RNA-seq data, it is important to use multiple circRNA prediction algorithms and further beneficial to enrich for bona fide circRNAs, by depletion of linear RNAs using RNase R. We next assessed differential circRNA expression during differentiation, which revealed marked changes in circRNA expression over the 14-day timecourse.
Comparison with previously identified mouse and human homologue circRNAs isolated from mouse brain regions or cell lines of either murine or human origin 14 , revealed significant overlap between circRNAs at differentiated stages (e.g. 80% to differentiated murine p19 cells 45% to human SH-SY5Y and 75% overlap with circRNAs found in the human ENCODE data). We confirmed differential expression of a subset of the most abundant and upregulated circRNAs (circTulp4, circMagi1, circRmst, circEzh2, circHdgfrp3, circZfp827, circMed13L, circZfp609, circSlc8a1, circNfix, circAff3, circAnkib1) using RT-qPCR with amplicons across the backsplicing junction ( Figure 1F-G). 75% of the top-100 expressed mouse circRNAs was also found in human circRNA datasets 14 . We conclude that significant changes in circRNA expression patterns are induced upon neuronal differentiation and that the majority of these circRNAs are conserved between various neuronal cell-types originating from humans and the mouse.

Knockdown of circZNF827 stimulates neuronal marker expression
To ascertain whether highly upregulated circRNAs might contribute to the process of neuronal differentiation themselves, we next depleted a number of candidate circRNAs by RNAi. Since mESCs are notoriously hard to transfect, we first tested knockdown efficiency of circZfp827 (circZNF827 in humans) by lentivirally delivered dishRNAs 32 targeting the backsplicing junction in either mESC, p19, SH-SY5Y or L-AN-5 cells, of which the latter three cell lines are well established models of neuronal differentiation following retinoic acid treatment. Knockdown efficiency in mESC and p19 proved relatively poor (30-60% remaining circRNA) compared to the two human cell lines, SH-SY5Y (10% remaining) of which L-AN-5 displayed superior results (<8% remaining) ( Figure S2A and Figure 2A). Moreover, when testing the increase of neuronal differentiation markers TrkB, NEFL, MAP2 and TUBB3 upon retinoic acid treatment, L-AN-5 elicited a more dynamic expression pattern compared to SH-SY5Y cells where only TrkB was upregulated upon differentiation ( Figure   S2B and Figure 2B). We therefore tested knockdown of 14 candidate circRNAs (circTULP4, circSLC8A1, circZNF609, circHDGFRP3, circZNF827, circANKIb, circCDYL, circUNC79, circCAMSAP1, circMAGI1, circRMST, circMED13L, circHIPK3, circNFIX) ( Figure S2C) in L-AN-5 cells and subsequently subjected these to retinoic acid-induced differentiation, followed by neuronal marker quantification to probe for changes in differentiation. Only knockdown of circZNF827 produced a significant increase in neuronal marker expression upon differentiation ( Figure 2B-C). Importantly, the linear ZNF827 mRNAs were not affected by backsplicing junction-specific knockdown ( Figure   S2E). The upregulation of neuronal markers following circZNF827 knockdown, were also evident at the protein level for MAP2 and TUBB3 ( Figure 2C). In addition, proliferation assays demonstrated only a marginally smaller S-phase population (lowered from 31% to 23%) upon circZNF827 knockdown, suggesting slightly lowered replication kinetics ( Figure 2D). This phenomenon was accompanied by a slight stall in G 2 /M phase, while G 2 /G 1 phase was not significantly affected between control and circZNF827 knockdown. Taken together, our results suggest that circZNF827 exert a repressive effect on neuronal differentiation.

circZNF827 controls retinoic acid receptor homeostasis
We next asked whether the Retinoic Acid Receptors (RARs), which represent central nodes in relaying anti-proliferative differentiation cues during neuronal development 43 , and are key targets of retinoic acid, also become upregulated upon knockdown of circZNF827. Indeed, knockdown of circZNF827 leads to a moderate but significant increased expression (1.5-2.5 fold) of RARa and RARg ( Figure 3A). Since most circRNAs have been reported to predominantly localize in the cell cytoplasm, we addressed the localization of circZNF827, circTULP4 and circANKIb by cellular fractionation. All three circRNAs are mainly cytoplasmically localized in L-AN-5 cells (~90% cytoplasmic signal) ( Figure 3B). We therefore speculated that circZNF827 could potentially affect RAR-mRNA stability post-transcriptionally in the cell cytoplasm. However, BrU pulse-chase mRNA decay assays demonstrated no significant change in RAR-mRNA decay rates upon knockdown of circZNF827 ( Figure 3C). Next, we investigated transcription rates, by treating cells with a short pulse of BrU, followed by BrU immunoprecipitation to quantify de novo labeled RNA, which serves as a proxy for transcription rates during control-or knockdown of circZNF827. As expected from the constant mRNA decay rates, transcription was slightly, but significantly, upregulated upon circZNF827 knockdown ( Figure S3A). Our results suggest that circZNF827 may contribute to finetuning of RA-receptor transcription, which in turn will likely keep neuronal differentiation in check.

circZNF827 knockdown affect multiple genes in neuronal signaling
Our results indicate that L-AN-5 cells slow their proliferation and promote RAR-signalling by a slight transcriptional upregulation of these receptor transcription factors when circZNF827 levels are low. To address more global effects of circZNF827 knockdown, we performed NanoString analyses using a neuro-differentiation/pathology panel of ~800 genes on RNA purified from differentiated or non-differentiated cells. Numerous genes become differentially expressed due to circZNF827 knockdown after differentiation ( Figure 4A, Table S3). In line with a potential negative regulatory function of circZNF827 on neuronal differentiation, GO-term analyses show enrichment of terms including axon/dendrite structure, neural cytoskeleton, transmitter synthesis, neural connectivity, growth factor signaling and trophic factors among upregulated genes, during circZNF827 knockdown ( Figure 4B).

Among the most significantly upregulated genes is Neuronal Growth Factor
Receptor (NGFR), which plays a central role in regulating neuronal differentiation, death, survival and neurite outgrowth 44,45 . Furthermore, Phosphatase and tensin homolog (PTEN), STAT3 and NAD(P)H quinone dehydrogenase 1 (NQO1) were all significantly downregulated upon circZNF827 knockdown (2-4 fold), consistent with reported positive roles of these factors in neuronal differentiation 46,47 and the induction susceptibility to energetic and proteotoxic stress 48 . We next validated the NGFR upregulation, using both qRT-PCR and western blotting, which demonstrated a massive upregulation at both the protein and mRNA level ( Figure 4C and S4A). This upregulation was not due to changes in mRNA decay rates, since BrU pulsechase mRNA decay assays yielded very similar mRNA half-lives upon circZNF827 knockdown ( Figure 4D). To address whether the observed changes in gene expression is elicited at the transcriptional or posttranscriptional level, we subjected cells to a short BrU-pulse prior to BrU immunoprecipitation and NanoString hybridization. Interestingly, NGFR and also ATP8A2 proved to be highly upregulated (~4-6 fold) at the level of transcription ( Figure 4E-F), while only NQO1 and not PTEN and STAT3 exhibited significantly reduced transcription activity (ranging from ~1.3 to ~4 fold) ( Figure 4E). Also, the MAP2 gene did not change its de novo RNA output, suggesting that the tuning of the steady-state levels of PTEN, STAT3 and MAP2 mRNAs, initially observed are mainly facilitated by posttranscriptional changes to mRNA stability.
Finally, we sought to functionally validate the impact of NGFR upregulation upon circZNF827 knockdown. To this end, we NGF-treated L-AN-5 cells subjected to either control or circZNF827 knockdown, and quantified downstream signaling output by quantification of c-fos, which is a well-known downstream immediate early target of NGFR signaling. c-fos levels increased significantly, strongly indicating that the higher levels of NGFR protein indeed leads to functional increase in NGFR signaling ( Figure 4G), which can at least in part explain the upregulation of classical neuronal markers. Taken together, we conclude that circZNF827 serves to keep neuronal differentiation 'in check' by limiting both RARs and to a larger extent NGFR. suggesting that the circRNA normally exerts a negative role in neuronal differentiation. Among nearly 800 genes, important to neuronal differentiation and disease, we found that NGFR (p75 NTR ) was most strongly induced, also at the protein level, upon circZNF827 knockdown. NGFR is a member of the TNF superfamily of receptors and relays, along with 3 paralogous receptor tyrosine kinases (TrkA, TrkB and TrkC), signals from the 4 mammalian neurotrophins (Nerve Growth Factor (NGF), brain-derived neurotrophic factor (BDNF), neurotrophin-3 (NT-3) and neurotrophin 4 (NT-4, aka. NT-4/5) 45 . The regulation and functional output from the neurotrophins and their receptors, which are interdependent proteins, is extremely complex and involves a multitude of effector proteins and interaction partners 45 . NGFR can, depending on expression levels of also the other neurotrophin receptors and their ligands, either induce death-or survival signaling to promote neuronal differentiation and control axonal growth or apoptosis 45 . Whether NGFR upregulation is instrumental and causal for the enhanced expression of TrkB, NEFL, TUBB3 and MAP2 that we observe in the L-AN-5 neuroblastoma system, remains to be seen. However, we did observe strongly augmented cfos expression (immediate early gene) upon treatment of L-AN-5 cells with NGF, when circZNF827 was downregulated, which may suggest that TrkAmediated NGF response becomes enhanced by increased NGFR expression.
How does circZNF827 regulate transcription rates of the NGFR gene and does this regulation involve a direct effect? Since circZNF827 is largely cytoplasmic (90%) it is tempting to speculate that positive transcription regulators (i.e. chromatin remodeling factors) with RNA-binding capacity, may become sequestered in this compartment by the circRNA. Supporting such a model is the finding that many classical DNA-binding transcription factors and co-regulators also interact with RNA [49][50][51] . That circRNAs can sequester effectors is reminiscent of the reported sequestration of splicing regulator MBL by a circRNA generated from its own pre-mRNA 13 . Alternatively, the nuclear fraction of the circZNF827 population could potentially regulate transcription in a fashion similar to that of long non-coding RNAs (lncRNAs). Transcriptional regulation has already been reported for intron-containing circRNAs