当前位置：文档库 › Nature文章-Circular-RNAs-are-large-class-animal-regulatory-potency

Nature文章-Circular-RNAs-are-large-class-animal-regulatory-potency

ARTICLE

doi:10.1038/nature11928

Circular RNAs are a large class of animal RNAs with regulatory potency

Sebastian Memczak 1*,Marvin Jens 1*,Antigoni Elefsinioti 1*,Francesca Torti 1*,Janna Krueger 2,Agnieszka Rybak 1,Luisa Maier 1,Sebastian D.Mackowiak 1,Lea H.Gregersen 3,Mathias Munschauer 3,Alexander Loewer 4,Ulrike Ziebold 1,Markus Landthaler 3,Christine Kocks 1,Ferdinand le Noble 2&Nikolaus Rajewsky 1

Circular RNAs (circRNAs)in animals are an enigmatic class of RNA with unknown function.To explore circRNAs systematically,we sequenced and computationally analysed human,mouse and nematode RNA.We detected thousands of well-expressed,stable circRNAs,often showing tissue/developmental-stage-specific expression.Sequence analysis indicated important regulatory functions for circRNAs.We found that a human circRNA,antisense to the cerebellar degeneration-related protein 1transcript (CDR1as),is densely bound by microRNA (miRNA)effector complexes and harbours 63conserved binding sites for the ancient miRNA miR-7.Further analyses indicated that CDR1as functions to bind miR-7in neuronal tissues.Human CDR1as expression in zebrafish impaired midbrain development,similar to knocking down miR-7,suggesting that CDR1as is a miRNA antagonist with a miRNA-binding capacity ten times higher than any other known transcript.Together,our data provide evidence that circRNAs form a large class of post-transcriptional regulators.Numerous circRNAs form by head-to-tail splicing of exons,suggesting previously unrecognized regulatory potential of coding sequences.Mature messenger RNAs are linear molecules with 59and 39termini that reflect start and stop of the RNA polymerase on the DNA tem-plate.In cells,different RNA molecules are sometimes joined together by splicing reactions (trans-splicing),but covalent linkage of the ends of a single RNA molecule to form a circular RNA (circRNA)is usually considered a rare event.circRNAs were discovered in plants and shown to encode subviral agents 1.In unicellular organisms,circRNAs mostly stem from self-splicing introns of pre-ribosomal RNA 2,but can also arise from protein-coding genes in archaea 3.In the few unambiguously validated circRNAs in animals,the spliceosome seems to link the 59and downstream 39ends of exons within the same transcript 4–10.Perhaps the best known circRNA is antisense to the mRNA transcribed from the SRY (sex-determining region Y)locus and is highly expressed in testes 6.Evidence from computational analyses of expression data in Archaea and Mammalia suggests that circRNAs are more prevalent than previously thought 3,10;however,it is unknown whether animal circRNAs have any biological function.

In comparison to circRNAs,miRNAs are extremely well studied.miRNAs are ,21-nucleotide-long non-coding RNAs that guide the effector protein Argonaute (AGO)to mRNAs of coding genes to repress their protein production 11–14.In humans,miRNAs directly regulate expression of most mRNAs 15–18in a diverse range of bio-logical functions.However,surprisingly little is known about how and if mRNAs can escape regulation by a miRNA.A recently discov-ered mechanism for miRNA removal in a sequence-specific manner is based on target sites acting as decoys or miRNA sponges 19,20.RNA with miRNA binding sites should,if expressed highly enough,sequester away the miRNA from its target sites.However,all reported mam-malian miRNA sponges have only one or two binding sites for the same miRNA and are not highly expressed,limiting their potency 21–24.To identify circRNAs across animal cells systematically,we screened RNA-seq data for https://www.wendangku.net/doc/4119158658.html,pared to previous approaches 10our

computational pipeline can find circRNAs in any genomic region,takes advantage of long (,100nucleotides)reads,and predicts the acceptor and donor splice sites used to link the ends of the RNAs.We do not rely on paired-end sequencing data or known splice https://www.wendangku.net/doc/4119158658.html,ing published 10,25,26and our own sequencing data,our method reported thousands of circRNAs in human and mouse tissues as well as in different developmental stages of Caenorhabditis elegans .Numerous circRNAs appear to be specifically expressed across tissues or developmental stages.We validated these data and showed that most tested circRNAs are well expressed,stable and circularized using the predicted splice sites.circRNA sequences were significantly enriched in conserved nucleotides,indicating that circRNAs compete with other RNAs for binding by RNA binding proteins (RBPs)or miRNAs.We combined biochemical,functional and computational analyses to show that indeed a known human circRNA,CDR1anti-sense (CDR1as)9,can function as a negative regulator of miR-7,a miRNA with perfect sequence conservation from annelids to human.Together,our data provide evidence that circRNAs form an important class of post-transcriptional regulators.

circRNAs have complex expression patterns

To comprehensively identify stably expressed circRNAs in animals we screened RNA sequencing reads for splice junctions formed by an acceptor splice site at the 59end of an exon and a donor site at a downstream 39end (head-to-tail)(Fig.1a).As standard RNA expres-sion profiling enriches for polyadenylated RNAs,we used data gene-rated after ribosomal RNA depletion (ribominus)and random priming.Such data were used before to detect scrambled exons in mammals 10(see Methods for comparison).However,this approach was not specifically designed to detect circRNAs and (1)only used existing exon–intron annotations,thus missing RNAs transcribed from introns or unannotated transcripts;(2)did not explicitly identify

*These authors contributed equally to this work.

Systems Biology of Gene Regulatory Elements,Max-Delbru

¨ck-Center for Molecular Medicine,Robert-Ro ¨ssle-Strasse 10,13125Berlin,Germany.2Angiogenesis and Cardiovascular Pathology,Max-Delbru

¨ck-Center for Molecular Medicine,Robert-Ro ¨ssle-Strasse 10,13125Berlin,Germany.3

RNA Biology and Post-Transcriptional Regulation,Max-Delbru ¨ck-Center for Molecular Medicine,Robert-Ro

¨ssle-Strasse 10,13125Berlin,Germany.4Signaling Dynamics in Single Cells,Max-Delbru ¨ck-Center for Molecular Medicine,Robert-Ro ¨ssle-Strasse 10,13125Berlin,Germany.21M A R C H 2013|V O L 495|N A T U R E |333

the splice sites used for circularization;and (3)assumed that each pair of mates in paired-end sequencing derives from the same RNA mole-cule.To search in a more unbiased way for circRNAs,we designed an algorithm (Methods)that identifies linear and circular splicing events in ribominus data.First,we filtered out reads that aligned con-tiguously to the genome,retaining the spliced reads.Next,we mapped the terminal parts of each candidate read independently to the genome to find unique anchor positions.Finally,we demanded that (1)anchor alignments can be extended such that the original read sequence aligns completely,and (2)the inferred breakpoint is flanked by GU/AG splice signals.Non-unique mappings and ambiguous breakpoints were discarded.We detected circularization splicing from the reversed (head-to-tail)orientation of the anchor alignments (Fig.1a).Our method also recovered tens of thousands of known linear splicing events (Methods and

Supplementary Fig.1a,b).We estimated sen-sitivity (.75%)and false-discovery rate (FDR ,0.2%)using simulated reads and various permutations of real sequencing data (Methods and Supplementary Fig.1c).However,the efficiency of ribominus pro-tocols to extract and sequence circRNAs is limited,reducing overall sensitivity.

We generated ribominus data for HEK293cells and,combined with human leukocyte data 10,detected 1,950circRNAs with support

from at least two independent junction-spanning reads (Fig.1b).The expression of genes predicted to give rise to circRNAs was only slightly shifted towards higher expression values (Supplementary Fig.1d),indicating that circRNAs are not just rare mistakes of the spliceo-some.We also identified 1,903circRNAs in mouse (brains,fetal head,differentiation-induced embryonic stem cells;Supplementary Fig.1e)25,26;81of these mapped to human circRNAs (Supplemen-tary Fig.1f).To explore whether circRNAs exist in other animal clades,we used sequencing data that we produced from various C.elegans developmental stages (Stoeckius,M.et al.,manuscript in preparation)(Methods)and detected 724circRNAs,with at least two independent reads (Fig.1c).

Numerous circRNAs seem to be specifically expressed in a cell type or developmental stage (Fig.1b,c and Supplementary Fig.1e).For example,hsa-circRNA 2149is supported by 13unique,head-to-tail spanning reads in CD191leukocytes but is not detected in CD341leukocytes (which were sequenced at comparable depth;Supplemen-tary Table 1),neutrophils or HEK293cells.Analogously,a number of nematode circRNAs seem to be expressed in oocytes but absent in 1-or 2-cell embryos.

We annotated human circRNAs using the RefSeq database and a catalogue of non-coding RNAs 27–29.85%of human circRNAs align sense to known genes.Their splice sites typically span one to five exons (Supplementary Fig.1g)and overlap coding exons (84%),but only in 65%of these cases are both splice sites that participate in the circularization known splice sites (Supplementary Table 2),demon-strating the advantage of our strategy.10%of all circRNAs align antisense to known transcripts,smaller fractions align to UTRs,introns,unannotated regions of the genome (Fig.1d).Examples of human circRNAs are shown in Fig.1e.

We analysed sequence conservation within circRNAs.As genomic sequence is subject to different degrees of evolutionary selection,depending on function,we studied three subtypes of circRNAs.Intergenic and a few intronic circRNAs display a mild but significant enrichment of conserved nucleotides (Supplementary Fig.1h,i).To analyse circRNAs composed of coding sequence and thus high overall conservation,we selected 223human circRNAs with circular orthologues in mouse (Methods)and entirely composed of coding sequence.Control (linear)exons were randomly selected to match the level of conservation observed in first and second codon positions (Methods,Fig.1f inset and Supplementary Fig.1k for conservation of the remaining coding sequence (CDS)).circRNAs with conserved circularization were significantly more conserved in the third codon position than controls,indicating evolutionary constraints at the nuc-leotide level,in addition to selection at the protein level (Fig.1f and Supplementary Fig.1j,k).In summary,we have confidently identified a large number of circRNAs with complex expression patterns,which derive often but not always from coding exons.Sequence conservation suggests that at least a subset contains functional sequence elements.

Characterization of 50predicted circRNAs

We experimentally tested our circRNA predictions in HEK293cells.Head-to-tail splicing was assayed by quantitative polymerase chain reaction (qPCR)after reverse transcription,with divergent primers and Sanger sequencing (Fig.2a,b).Predicted head-to-tail junctions of 19out of 23randomly chosen circRNAs (83%)could be validated,demonstrating high accuracy of our predictions (Table 1).In contrast,5out of 7(71%)candidates exclusively predicted in leukocytes could not be detected in HEK293cells,validating cell-type-specific expression.Head-to-tail splicing could be produced by trans-splicing or geno-mic rearrangements.To rule out these possibilities as well as potential PCR artefacts,we successfully validated the insensitivity of human circRNA candidates to digestion with RNase R—an exonuclease that degrades linear RNA molecules 30—by northern blotting with probes which span the head-to-tail junctions (Fig.2c).We quantified RNase R resistance for 21candidates with confirmed head-to-tail splicing by

Circularization

circRNA

Linear splicing

Human (1,950)

+ C. elegans (724) (1,602 inside coding transcripts)

Human circRNAs d

CDR1Conservation score

C u m u l a t i v e f r e q u e n c y

Figure 1|Detection,classification and evolutionary conservation of circRNAs.a ,The termini of junction-spanning reads (anchors)align

sequentially to the genome for linear (top)but in reversed orientation for head-to-tail spliced reads (bottom).Spliced reads must distribute completely to anchors,flanked by AG/GU (Methods).b ,c ,circRNAs in human cell types (b )and nematode stages (c ).d ,Genomic origin of human circRNAs.A total of 96%of circRNAs overlap known transcripts.e ,Examples of human circRNAs.The AFF1intron is spliced out (Supplementary Fig.2e).Sequence conservation:placental mammals phyloP score (Methods),scale bar,200nucleotides.f ,A total of 223human coding sequence circRNAs with mouse orthologues (green)and controls (black)with matched conservation level (inset:mean conservation for each codon position (grey),controls (black);x axis,codon positions;y axis,placental mammals phyloP score;see also Methods and Supplementary Fig.1j,k).Third codon positions are significantly more conserved (P ,4310210,Mann–Whitney U -test,n 5223).

RESEARCH ARTICLE

334|N A T U R E |V O L 495|21M A R C H 2013

qPCR.All of these were at least 10-fold more resistant than GAPDH (Fig.2d and Supplementary Fig.2a).We reasoned that circRNAs should generally turn over more slowly than mRNAs.Indeed,we found that 24h after blocking transcription circRNAs were highly stable,exceeding the stability of the housekeeping gene GAPDH 31(Fig.2e and Supplementary Fig.2b).We also validated 3out of 3tested mouse circRNAs with human orthologues in mouse brains (Supplementary Fig.2c).In C.elegans 15out of 20(75%)of the pre-dictions from gametes and early embryos were validated in a mixed stage sample (Supplementary Fig.2d and Supplementary Table 3).

circRNA CDR1as is

densely bound by AGO

Stable transcripts with many miRNA-binding sites could function as miRNA sponges.We intersected our catalogue of circRNAs with transcript annotations,assuming that introns would not occur in mature circRNAs (as observed for 3out of 3tested circRNAs,Supplementary Fig.2e).We screened for occurrences of conserved miRNA family seed matches (Methods).When counting repetitions of conserved matches to the same miRNA family,circRNAs were

significantly enriched compared to coding sequences (P ,2.96310222,Mann–Whitney U -test,n 53,873)or 39UTR sequences (P ,2.76310221,Mann–Whitney U -test,n 53,182)(Supplementary Fig.3a,b).As an extreme case,we discovered that the known human circRNA CDR1as (ref.9)harboured dozens of conserved miR-7seed matches.To test whether CDR1as is bound by miRNAs,we analysed bio-chemical,transcriptome-wide binding-site data for the miRNA effector AGO proteins.We performed four independent PAR-CLIP (photoactivatable-ribonucleoside-enhanced crosslinking and immu-noprecipitation)experiments for human AGO (Methods)and ana-lysed them together with published,lower-depth data 32.PAR-CLIP 32–34is based on ultraviolet crosslinking of RNA to protein and subsequent sequencing of RNA bound to a RBP of interest.The ,1.5-kilobase (kb)CDR1as locus stood out in density and number of AGO PAR-CLIP reads (Fig.3a),whereas nine combined PAR-CLIP libraries for other RBPs gave virtually no signal.Of note,there is no PAR-CLIP read mapping to the sense coding transcript of the CDR1gene,which was originally identified as a target of autoantibodies from patients with paraneoplastic cerebellar degeneration 35.

Sequence analysis across 32vertebrate species revealed that miR-7is the only animal miRNA with conserved seed matches that can explain the AGO binding along the CDR1as transcript (Methods).Human CDR1as harbours 74miR-7seed matches of which 63are

Table 1|Summary of the validation experiments

Sample

Validation experiment

Validation success

Human (HEK293)

Head-to-tail splicing 19of 23Circularity

21of 21Expression .3%vinculin 12of 21Expression specificity (leukocyte specific)5of 7Mouse (adult brain)

Head-to-tail splicing 3of 3Circularity

3of 3Expression .1%b -actin 2of 3C.elegans

Head-to-tail splicing 15of 20Circularity

13of 13Expression .1%eif-3.d

12of 15

Most experimentally tested circRNAs are validated.

5′

miR-7 (nt) 5′ to 3′

P r o b a b i l i t y

Distance from seed match (nt)

–e n t r o p y (b i t s )

HEK293 expression CDR1as

d e

RNA gel

28S

18S 5S

c i r c +l i n

C i r c u l a r

T o t a l I V T l i n .

Figure 3|The circRNA CDR1as is bound by the miRNA effector protein AGO,and is cytoplasmic.a ,CDR1as is densely bound by AGO (red)but not by unrelated proteins (black).Blue boxes indicate miR-7seed matches.nt,nucleotides.b ,c ,miR-7sites display reduced nucleotide variability across 32vertebrate genomes (b )and high base-pairing probability within seed matches (c ).d ,CDR1as RNA is cytoplasmic and disperse (white spots;single-molecule RNA FISH;maximum intensity merges of Z -stacks).siSCR,positive;siRNA1,negative control.Blue,nuclei (DAPI);scale bar,5m m (see also Supplementary Fig.10for uncropped images).e ,Northern blotting detects circular but not linear CDR1as in HEK293RNA.Total,HEK293RNA;circular,head-to-tail probe;circ 1lin,probe within splice sites;IVT lin.,in vitro transcribed,linear CDR1as RNA.f ,Circular CDR1as is highly expressed (qPCR,error bars indicate standard deviation).g ,CDR1as.Blue,seed matches;dark red,AGO PAR-CLIP reads;bright red,crosslinked nucleotide conversions.

18S

28S RNase R exonuclease

Agarose gel

G A P D H

C D R 1a s

hsa-circRNAs

2 3 16– +cDNA gDNA gDNA gDNA gDNA

100

200300cDNA cDNA cDNA Sanger sequencing

...

A P

D H

p 16R e l. e x p r e s s i o n

C D R 1a s hsa-circRNAs

A P D H

V C L C

D R 1a s

hsa-circRNAs

R e l. e x p r e s s i o n

– +– +– +– +– +

Figure 2|CircRNAs are stable transcripts with robust expression.a ,Human (hsa)ZRANB1circRNA exemplifies the validation strategy.Convergent (divergent)primers detect total (circular)RNAs.Sanger sequencing confirms head-to-tail splicing.b ,Divergent primers amplify circRNAs in cDNA but not genomic DNA (gDNA).GAPDH ,linear control,size marker in base pairs.c ,Northern blots of mock (2)and RNase R (1)treated HEK293total RNA with head-to-tail specific probes for circRNAs.GAPDH ,linear control.d ,e ,circRNAs are at least 10-fold more RNase R resistant than GAPDH mRNA (d )and stable after 24h transcription block (e )(qPCR;error bars indicate standard deviation).

ARTICLE RESEARCH

21M A R C H 2013|V O L 495|N A T U R E |335

conserved in at least one other species(Supplementary Fig.4). Interspaced sequences were less conserved,indicating that miR-7 binding sites are probably functional(Fig.3b).Secondary structure analysis of predicted circRNA–miRNA duplexes(Methods)showed reduced base-pairing of miR-7beyond the seed(Fig.3c).None of the ,1,500miR-7complementary sites across32vertebrate sequences was complementary beyond position12of miR-7(only three could form an11-nucleotide duplex)(Supplementary Table4).Slicing by mammalian Argonaute requires complementarity of positions10and

11and depends on extended complementarity beyond position12 (ref.36).Thus,CDR1as seems optimized to be densely bound but not sliced by miR-7.

Single-molecule imaging(Methods)revealed disperse and mostly cytoplasmic CDR1as expression(HEK293cells),consistent with miRNA sponge function(Fig.3d and Supplementary Table5). CDR1as circularization was assayed by northern blotting(Fig.3e). Nicking experiments confirmed that CDR1as circRNA can be linea-rized and degraded(Supplementary Fig.5a).In RNA from HEK293 cells,circularized but no additional linear CDR1as was detected (Supplementary Fig.5b).Circular expression levels were quantified by qPCR with divergent primers calibrated by standard curves (Supplementary Table6).CDR1as was highly expressed(,15%to ,20%of GAPDH expression,Fig.3f).Estimating GAPDH mRNA copy number from HEK293RNA-seq data(,1,400molecules per cell,data not shown)suggests that CDR1as may bind up to,20,000 miR-7molecules per cell(Fig.3g).

If CDR1as functions as a miR-7sponge,its destruction could trigger downregulation of miR-7targets.We knocked down CDR1as in HEK293cells and monitored expression of published miR-7targets by qPCR with externally spiked-in standards(Methods and Supplemen-tary Fig.5c,d).All eight miR-7targets assayed,but also housekeeping genes,were downregulated.Nanostring technology37additionally indi-cated downregulation of many genes(data not shown).Furthermore, stable loss of CDR1as expression by virally delivered small hairpin RNAs led to significantly reduced migration in an in vitro wound clo-sure assay

(Methods,Supplementary Fig.5e,f and Supplementary Table7).Thus,knockdown of CDR1as affects HEK293cells,but we could not delineate miR-7-specific effects,potentially because of indirect or miR-7-independent CDR1as function(see below).

Co-expression of miR-7and CDR1as in brain

If CDR1as indeed interacts with miR-7,both must be co-expressed. miR-7is highly expressed in neuronal tissues,pancreas and pituitary gland38.Apart from HEK293cells,a cell line probably derived from neuronal precursors in embryonic kidney39,we quantified miR-7and CDR1as expression across mouse tissues and pancreatic-island-derived MIN6cells(Methods and Fig.4a).CDR1as and miR-7were both highly expressed in brain tissues,but CDR1as was expressed at low levels or absent in non-neuronal tissues,including tissues with very high miR-7expression.qPCR suggested that CDR1as is exclu-sively circular in adult and embryonic mouse brain(Supplementary Fig.5g,h).Thus,CDR1as and miR-7seem to interact specifically in neuronal tissues.Indeed,when assaying CDR1as and miR-7in mouse brains by in situ hybridizations(Methods),we observed spe-cific,similar,but not identical,expression patterns in the brain of mid-gestation(embryonic day13.5(E13.5))embryos(Fig.4b).Speci-fically,CDR1as and miR-7were highly co-expressed in areas of the developing midbrain(mesencephalon)40,41.Thus,CDR1as is highly expressed,stable,cytoplasmic,not detectable as a linear RNA and shares expression domains with miR-7.Together with extensive miR-7binding within CDR1as,CDR1as has hallmarks of a potent circular miR-7sponge in neuronal tissues.

Effects of miR-7and CDR1as in zebrafish

It would be informative to knock out CDR1as in an animal model system.However,a knockout would also affect CDR1protein,with unknown consequences.This problem is circumvented when using zeb-rafish(Danio rerio)as an animal model.According to our bioinformatic analyses(not shown)zebrafish has lost the cdr1locus,whereas miR-7is conserved and highly expressed in the embryonic brain42.Thus,we can test whether miR-7has a loss-of-function phenotype and if this pheno-type can be induced by introduction of mammalian CDR1as RNA.We injected morpholinos to knock down mature miR-7expression in zebra-fish embryos(Methods).At a dose of9ng of miR-7morpholino,the embryos did not show overall morphological defects but reproducibly, and in two independent genetic backgrounds(Supplementary Fig.6a–c), developed brain defects(Fig.5a,b).In particular,,70%showed a con-sistent and clear reduction in midbrain size,and an additional,5%of animals had almost completely lost their midbrains.Of note,the tel-encephalon at the anterior tip of the brain was not affected in size.Brain volumes were also measured based on confocal three-dimensional stacks (Fig.5c and Supplementary Fig.7).Reduction of the midbrain size correlated with miR-7inhibition in the respective animals(Supplemen-tary Fig.6d).These data provide evidence that miR-7loss-of-function causes a specific reduction of midbrain size.

To test whether CDR1as can function as a miR-7sponge in vivo,we injected embryos with plasmid DNA that expressed a linear version of the full-length human CDR1as sequence(Supplementary Fig.6e,f)or a plasmid provided by the Kjems laboratory that can produce circular CDR1as in human cells(Fig.5d,e).qPCR analysis detected circular RNA in zebrafish embryos injected with the latter plasmid(Sup-plementary Fig.8),which reproducibly and in independent genetic backgrounds lead to reduced midbrain sizes(Fig.5g,h).Similarly, animals injected with in vitro-transcribed partial mouse CDR1as RNA,but not with RNA from the other strand,showed significant midbrain reduction(Supplementary Fig.6g–i).Thus,the phenotype is probably caused by CDR1as RNA and not by an unspecific effect of RNA or DNA injection.These results provide evidence that human/ mouse CDR1as transcripts are biologically active in vivo and impair brain development similarly to miR-7inhibition.The midbrain reduction could be partially rescued by injecting miR-7precursor (Fig.5f,g),arguing that the biological effect of CDR1as expression is caused at least in part by interaction of CDR1as with miR-7. Discussion

We have shown that animal genomes express thousands of circRNAs from diverse genomic locations(for example,from coding and non-coding exons,intergenic regions or transcripts antisense to59and 39UTRs)in a complex tissue-,cell-type-or developmental-stage-specific manner.We provided evidence that CDR1as can act as a C

miR-7 expression H

M i

d M I

Scrambled

CDR1as

U6 control

miR-7miR-124 control

Figure4|CDR1as and miR-7have overlapping and specific expression in neuronal tissues.a,Among mouse tissues and MIN6cells(qPCR,relative to cerebral cortex expression;error bars indicate standard deviations;see Supplementary Fig.9a for miR-122control)neuronal tissues co-express miR-7 and CDR1as.b,In situ staining of CDR1as and miR-7in mouse embryo brain E13.5(U6and miR-124,positive control;scrambled probe,negative control). Scale bar,1mm.

RESEARCH ARTICLE

336|N A T U R E|V O L495|21M A R C H2013

post-transcriptional regulator by binding miR-7in brain tissues:(1)CDR1as is densely bound by miRNA effector molecules;(2)CDR1as harbours 74miR-7seed matches,often deeply conserved;(3)CDR1as is expressed highly,stably and mostly cytoplasmic;(4)CDR1as and miR-7share specific expression domains in mouse embryonic brain;(5)human/mouse CDR1as is circularized in vivo and is not detectable as a linear molecule;(6)human/mouse CDR1as sequences,when injected into zebrafish,and miR-7knock down have similar pheno-types in brain.While zebrafish circularization of human CDR1as may be incomplete,the midbrain phenotype was stronger compared to expressing linear CDR1as RNA that lacks circularization splice sites.Although the two DNA plasmids used carry identical promoters and were injected in equal concentrations,we cannot rule out the possibility that the difference in midbrain phenotype strength may be explained by other factors.However,because of the observed extreme stability of CDR1as and circRNAs in general,our data argue that circRNAs can be used as potent inhibitors of miRNAs or RBPs.Future studies should elucidate how CDR1as can be converted into a linear mole-cule and targeted for degradation.miR-671can trigger destruction of

CDR1as 9.Thus,CDR1as may function to transport miR-7to subcel-lular locations,where miR-671could trigger release of its cargo.Known functions of miR-7targets such as PAK1and FAK1support these speculations 43,44.

The phenotype induced by CDR1as expression in zebrafish was only partially rescued by expressing miR-7,indicating that CDR1as could have functions beyond sequestering miR-7.This idea is sup-ported by in situ hybridization in mouse adult hippocampus (Sup-plementary Fig.9b)where areas staining for CDR1as but not miR-7were observed.What could be additional functions of circRNAs beyond acting as sponges?As a single-stranded RNA,CDR1as could,for example,bind in trans 39UTRs of target mRNAs to regulate their expression.It is even possible that miR-7binds CDR1as to silence these trans-acting activities.Alternatively,CDR1as could be involved in the assembly of larger complexes of RNA or protein,perhaps similar to other low-complexity molecules 45.

How many other circRNAs exist?In this study,we identified appro-ximately 2,000human,1,900mouse and 700nematode circRNAs from sequencing data,and our validation experiments confirmed most of the 50tested circRNAs.However,we analysed only a few tissues/developmental stages with stringent cutoffs.Thus,the true number of circRNAs is almost certainly much larger.Although CDR1as is an extreme case,many circRNAs have conserved seed matches.For example,circRNA from the SRY locus 6has seed sites for murine miRNAs.Therefore,circRNAs probably compete with other RNAs for miRNA binding.Sequence analyses indicated that coding exons serve additional,presumably regulatory functions when expressed within circRNAs,whereas intergenic or intronic circRNAs generally showed only weak conservation.Because we detected thou-sands of circRNAs,it is appealing to speculate that occasional circu-larization of exons is easy to evolve and may provide a mechanism for rapid evolution of stably and well expressed regulatory RNAs.Of note,we detected multiple seed matches for viral miRNAs within human circRNAs (not shown).However,there is no reason to think that circRNAs function predominantly to bind miRNAs.As known in bacteria,the decoy mechanism underlying miRNA sponges could be important also for RBPs 46,47.Similarly,circRNAs could function to store,sort,or localize RBPs.In summary,our data suggest that circRNAs form a class of post-transcriptional regulators which com-pete with other RNAs for binding by miRNAs and RBPs and may generally function in modulating the local free concentration of RBPs,RNAs,or their binding sites.

Note added in proof :While this paper was under review,circular RNAs in fibroblasts were described 51.

METHODS SUMMARY

Computational pipeline for predicting circRNAs from ribominus sequencing data.A detailed description of the computational methods is given in the Methods.

Cell culture and treatments.HEK293,HEK293TN and HEK293Flp-In 293T-REx (Life Technologies)were cultured following standard protocols.Tran-scription was blocked by adding 2m g ml 21actinomycin D (Sigma).RNase R (Epicentre Biotechnologies)treatment (3U m g 21)was performed on total RNA (5m g)at 37u C for 15min.qPCR primers are listed in Supplementary Table 8.Single-molecule RNA fluorescence in situ hybridization (smRNA FISH).Stellaris Oligonucleotide probes complementary to CDR1as were designed using the Stellaris Probe Designer (Biosearch Technologies).Probe pools were obtained from BioCat GmbH as conjugates coupled to Quasar 670.Probes were hybridized at 125nM at 37u C.Images were acquired on an inverted Nikon Ti microscope.Mouse strains and in situ hybridization.In situ hybridization (ISH)was per-formed on paraffin tissue sections from B6129SF1/J wild-type mice as described 48using locked nucleic acid (LNA)probes or RNAs obtained by in vitro transcrip-tion on PCR products.

Zebrafish methods.Tg(huC:egfp)and Tg(Xia.Tubb:dsRED)transgenic zebrafish lines were used 49,50.Morpholino antisense oligomers were injected into the yolk of single-cell-stage embryos.Furthermore,two pCS21plasmids coding for full-length linear CDR1as or CDR1as plus upstream and downstream sequence that can express circular CDR1as in human cells (courtesy of the Kjems

laboratory)

Circular CDR1as + miR-7Circular CDR1as

Control MO 15 ng

C o n t r o l M O

m i R -7 M U n i n j e c t e L i n e a C D R 1a C i r c u l a C D R 1a V o l u m e (×106 μm MB

**P h e n o t y p e (%)

E m p t y v e c t o L i n e a C i r c u l a C i r c u l a + m i R -m i R -7 (9 n g C o n t r o l (15 n g U n i n j e c t e **m i R -7 (15 n g miR-7 MO 9 ng

MO (ng)CDR1as

Circular CDR1as + miR-7

TC Figure 5|In zebrafish,knockdown of miR-7or expression of CDR1as causes midbrain defects.a ,b ,Neuronal reporter (Tg(huC:egfp))embryos (top,light microscopy)48h post fertilization (bottom,representative confocal z -stack projections;blue dashed line,telencephalon (TC)(control);yellow dashed line,midbrain (MB)).Embryos after injection of 9ng miR-7

morpholino (MO)(b )display a reduction in midbrain size.Panel a shows a representative embryo injected with 15ng control morpholino.c ,Three-dimensional volumetric reconstructions.d ,Empty vector control.e ,Expression vector encoding human circular CDR1as.f ,Rescue experiment with miR-7precursor.g ,Phenotype penetrance (%of embryos,miR-7MO,n 5135;uninjected,n 583;empty vector,n 591;linear CDR1as,n 5258;circular CDR1as,n 5153;circular CDR1as plus miR-7precursor,n 5217).Phenotype distribution derived from at least three independent experiments.Scale bar,0.1mm.**P ,0.01;***P ,0.001in Students t -test for normal midbrain,reduced midbrain (see also Supplementary Fig.6).h ,Phenotype quantification (Methods).Error bars indicate standard deviation n 53per group.

ARTICLE RESEARCH

21M A R C H 2013|V O L 495|N A T U R E |337

were injected.Confocal imaging was performed using Carl Zeiss MicroImaging. Reduced midbrain development was defined as.50%smaller than the mean size of controls.Each experimental group was evaluated in at least three independent experiments;a minimum of80individual embryos per group was examined. Full Methods and any associated references are available in the online version of the paper.

Received11September2012;accepted24January2013.

Published online27February2013.

1.Sanger,H.L.,Klotz,G.,Riesner,D.,Gross,H.J.&Kleinschmidt,A.K.Viroids are

single-stranded covalently closed circular RNA molecules existing as highly base-paired rod-like structures.Proc.Natl https://www.wendangku.net/doc/4119158658.html,A73,3852–3856(1976).

2.Grabowski,P.J.,Zaug,A.J.&Cech,T.R.The intervening sequence of the ribosomal

RNA precursor is converted to a circular RNA in isolated nuclei of Tetrahymena.

Cell23,467–476(1981).

3.Danan,M.,Schwartz,S.,Edelheit,S.&Sorek,R.Transcriptome-wide discovery of

circular RNAs in Archaea.Nucleic Acids Res.40,3131–3142(2012).

4.Nigro,J.M.et al.Scrambled exons.Cell64,607–613(1991).

5.Cocquerelle,C.,Mascrez,B.,Hetuin,D.&Bailleul,B.Mis-splicing yields circular

RNA molecules.FASEB J.7,155–160(1993).

6.Capel,B.et al.Circular transcripts of the testis-determining gene Sry in adult

mouse testis.Cell73,1019–1030(1993).

7.Chao,C.W.,Chan,D.C.,Kuo,A.&Leder,P.The mouse formin(Fmn)gene:

abundant circular RNA transcripts and gene-targeted deletion analysis.Mol.Med.

4,614–628(1998).

8.Burd,C.E.et al.Expression of linear and novel circular forms of an INK4/ARF-

associated non-coding RNA correlates with atherosclerosis risk.PLoS Genet.6, e1001233(2010).

9.Hansen,T.B.et al.miRNA-dependent gene silencing involving Ago2-mediated

cleavage of a circular antisense RNA.EMBO J.30,4414–4422(2011).

10.Salzman,J.,Gawad,C.,Wang,P.L.,Lacayo,N.&Brown,P.O.Circular RNAs are the

predominant transcript isoform from hundreds of human genes in diverse cell types.PLoS ONE7,e30733(2012).

11.Ambros,V.The functions of animal microRNAs.Nature431,350–355(2004).

12.Baek,D.et al.The impact of microRNAs on protein output.Nature455,64–71

(2008).

13.Selbach,M.et al.Widespread changes in protein synthesis induced by microRNAs.

Nature455,58–63(2008).

14.Bartel,D.P.MicroRNAs:target recognition and regulatory functions.Cell136,

215–233(2009).

15.Krek,A.et https://www.wendangku.net/doc/4119158658.html,binatorial microRNA target predictions.Nature Genet.37,

495–500(2005).

16.Lewis,B.P.,Burge,C.B.&Bartel,D.P.Conserved seed pairing,often flanked by

adenosines,indicates that thousands of human genes are microRNA targets.Cell 120,15–20(2005).

17.Xie,X.et al.Systematic discovery of regulatory motifs in human promoters and39

UTRs by comparison of several mammals.Nature434,338–345(2005).

18.Friedman,R.C.,Farh,K.K.,Burge,C.B.&Bartel,D.P.Most mammalian mRNAs are

conserved targets of microRNAs.Genome Res.19,92–105(2009).

19.Ebert,M.S.,Neilson,J.R.&Sharp,P.A.MicroRNA sponges:competitive inhibitors

of small RNAs in mammalian cells.Nature Methods4,721–726(2007).

20.Franco-Zorrilla,J.M.et al.Target mimicry provides a new mechanism for

regulation of microRNA activity.Nature Genet.39,1033–1037(2007).

21.Poliseno,L.et al.A coding-independent function of gene and pseudogene mRNAs

regulates tumour biology.Nature465,1033–1038(2010).

22.Tay,Y.et al.Coding-independent regulation of the tumor suppressor PTEN by

competing endogenous mRNAs.Cell147,344–357(2011).

23.Cesana,M.et al.A long noncoding RNA controls muscle differentiation by

functioning as a competing endogenous RNA.Cell147,358–369(2011).

24.Ebert,M.S.&Sharp,P.A.Emerging roles for natural microRNA sponges.Curr.Biol.

20,R858–R861(2010).

25.Vivancos,A.P.,Guell,M.,Dohm,J.C.,Serrano,L.&Himmelbauer,H.Strand-specific

deep sequencing of the transcriptome.Genome Res.20,989–999(2010).

26.Huang,R.et al.An RNA-Seq strategy to detect the complete coding and non-

coding transcriptome including full-length imprinted macro ncRNAs.PLoS ONE6, e27288(2011).

27.Kent,W.J.et al.The human genome browser at UCSC.Genome Res.12,996–1006

(2002).

28.Pruitt,K.D.,Tatusova,T.&Maglott,D.R.NCBI Reference Sequence(RefSeq):a

curated non-redundant sequence database of genomes,transcripts and proteins.

Nucleic Acids Res.33,D501–D504(2005).

29.Cabili,M.N.et al.Integrative annotation of human large intergenic noncoding

RNAs reveals global properties and specific subclasses.Genes Dev.25,

1915–1927(2011).

30.Suzuki,H.et al.Characterization of RNase R-digested cellular RNA source that

consists of lariat and circular RNAs from pre-mRNA splicing.Nucleic Acids Res.34, e63(2006).31.Iwai,Y.,Akahane,K.,Pluznik,D.H.&Cohen,R.B.Ca21ionophore A23187-

dependent stabilization of granulocyte-macrophage colony-stimulating factor messenger RNA in murine thymoma EL-4cells is mediated through two distinct regions in the39-untranslated region.J.Immunol.150,4386–4394(1993). 32.Hafner,M.et al.Transcriptome-wide identification of RNA-binding protein and

microRNA target sites by PAR-CLIP.Cell141,129–141(2010).

33.Lebedeva,S.et al.Transcriptome-wide analysis of regulatory interactions of the

RNA-binding protein HuR.Mol.Cell43,340–352(2011).

34.Baltz,A.G.et al.The mRNA-bound proteome and its global occupancy profile on

protein-coding transcripts.Mol.Cell46,674–690(2012).

35.Dropcho,E.J.,Chen,Y.T.,Posner,J.B.&Old,L.J.Cloning of a brain protein

identified by autoantibodies from a patient with paraneoplastic cerebellar

degeneration.Proc.Natl https://www.wendangku.net/doc/4119158658.html,A84,4552–4556(1987).

36.Wee,L.M.,Flores-Jasso,C.F.,Salomon,W.E.&Zamore,P.D.Argonaute divides its

RNA guide into domains with distinct functions and RNA-binding properties.Cell 151,1055–1067(2012).

37.Geiss,G.K.et al.Direct multiplexed measurement of gene expression with color-

coded probe pairs.Nature Biotechnol.26,317–325(2008).

https://www.wendangku.net/doc/4119158658.html,ndgraf,P.et al.A mammalian microRNA expression atlas based on small RNA

library sequencing.Cell129,1401–1414(2007).

39.Shaw,G.,Morse,S.,Ararat,M.&Graham,F.L.Preferential transformation of human

neuronal cells by human adenoviruses and the origin of HEK293cells.FASEB J.

16,869–871(2002).

40.Kaufman,M.H.&Bard,J.B.L.The Anatomical Basis of Mouse Development

(Academic,1999).

41.Schambra,U.Prenatal Mouse Brain Atlas(Springer,2008).

42.Kapsimali,M.et al.MicroRNAs show a wide diversity of expression profiles in the

developing and mature central nervous system.Genome Biol.8,R173(2007). 43.Jacobs,T.et al.Localized activation of p21-activated kinase controls neuronal

polarity and morphology.J.Neurosci.27,8604–8615(2007).

44.Chacon,M.R.et al.Focal adhesion kinase regulates actin nucleation and neuronal

filopodia formation during axonal growth.Development139,3200–3210(2012).

45.Kato,M.et al.Cell-free formation of RNA granules:low complexity sequence

domains form dynamic fibers within hydrogels.Cell149,753–767(2012). 46.Romeo,T.Global regulation by the small RNA-binding protein CsrA and the non-

coding RNA molecule CsrB.Mol.Microbiol.29,1321–1330(1998).

47.Gottesman,S.The small RNA regulators of Escherichia coli:roles and mechanisms.

Annu.Rev.Microbiol.58,303–328(2004).

48.Huelsken,J.,Vogel,R.,Erdmann,B.,Cotsarelis,G.&Birchmeier,W.b-Catenin

controls hair follicle morphogenesis and stem cell differentiation in the skin.Cell 105,533–545(2001).

49.Park,H.C.et al.Analysis of upstream elements in the HuC promoter leads to the

establishment of transgenic zebrafish with fluorescent neurons.Dev.Biol.227, 279–293(2000).

50.Peri,F.&Nusslein-Volhard,C.Live imaging of neuronal degradation by microglia

reveals a role for v0-ATPase a1in phagosomal fusion in vivo.Cell133,916–927 (2008).

51.Jeck,W.R.et al.Circular RNAs are abundant,conserved,and associated with ALU

repeats.RNA19,1–17(2013).

Supplementary Information is available in the online version of the paper. Acknowledgements We thank M.Feldkamp and https://www.wendangku.net/doc/4119158658.html,ngnick(laboratory of W.Chen)for Illumina sequencing runs.We thank J.Kjems for sending us a plasmid encoding circular human CDR1as for our zebrafish experiments.We thank K.Meier for technical assistance with zebrafish experiments and A.Sporbert from the confocal imaging facility.We thank A.Ivanov for assisting in bioinformatic analysis.N.R.thanks E.Westhof for useful discussions.We acknowledge the following funding sources:PhD program of the Max-Delbru¨ck-Center(MDC)(S.M.,F.T.,L.H.G.);the MDC-NYU exchange program (M.M.);BMBF project1210182,‘MiRNAs as therapeutic targets’(A.E.);DFG for

KFO218(U.Z.);Helmholtz Association for the‘MDC Systems Biology Network’,MSBN (S.D.M.);BMBF support for the DZHK(F.l.N.and N.R.);Center for Stroke Research Berlin(J.K.,F.l.N.).Funding for the group of M.L.is supported by BMBF-funding for the Berlin Institute for Medical Systems Biology(0315362C).

Author Contributions S.M.,M.J.,A.E.and F.T.contributed equally.S.M.performed many experiments,assisted by L.M.M.J.and A.E.carried out most of the computation, with contributions from N.R.and S.D.M.F.T.performed the circRNA validation experiments.A.R.performed all northern experiments.L.H.G.and M.M.contributed AGO PAR-CLIP experiments and HEK293ribominus data,supervised by M.L.C.K. designed and carried out the single molecule experiments,in part together with A.L. U.Z.performed the mouse experiments.J.K.contributed the zebrafish experiments, supervised by F.l.N.N.R.designed and supervised the project.N.R.and M.J.wrote the paper.

Author Information Sequencing data have been deposited at GEO under accession number GSE43574.Reprints and permissions information is available at

https://www.wendangku.net/doc/4119158658.html,/reprints.The authors declare no competing financial interests. Readers are welcome to comment on the online version of the paper.Correspondence and requests for materials should be addressed to N.R.(rajewsky@mdc-berlin.de).

RESEARCH ARTICLE

338|N A T U R E|V O L495|21M A R C H2013

METHODS

Computational pipeline for predicting circRNAs from ribominus sequencing data.Reference genomes(human hg19(February2009,GRCh37),mouse mm9 (July2007,NB137/mm9),C.elegans ce6(May2008,WormBase v.WS190))were downloaded from the UCSC genome browser(https://www.wendangku.net/doc/4119158658.html,/)27.In a first step,reads that aligned contiguously and full length to the genomes were discarded.From the remaining reads we extracted20mers from both ends and aligned them independently to find unique anchor positions within spliced exons. Anchors that aligned in the reversed orientation(head-to-tail)indicated circRNA splicing(compare main Fig.1a).We extended the anchor alignments such that the complete read aligns and the breakpoints were flanked by GU/AG splice sites. Ambiguous breakpoints were discarded.We used the short-read mapper Bowtie 2(ref.52).Initially,ribominus reads were aligned in end-to-end mode to the genome:

$bowtie2-p16--very-sensitive--phred64--mm-M20--score-min5C,-15,0-q-x ,index.-U reads.qfa2.bowtie2.log j samtools view-hbuS-j samtools sort-sample_vs_genome

The unmapped reads were separated and run through a custom script to split the reads as indicated in Fig.1a to obtain20-nucleotide anchors from both ends of the read:

$samtools view-hf4sample_vs_genome.bam j samtools view-Sb-.unmapped_ sample.bam

$./unmapped2anchors.py unmapped_sample.bam j gzip.sample_anchors.qfa. gz

Here is an example of two anchor pairs in the FASTQ format;the original read was kept as part of the first anchors identifier to simplify downstream analysis:

@s_8_1_0001_qseq_14_A__NCCCGCCTCACCGGGTCAGTGAAAAAACGA TCAGAGTAGTGGTCTTCTTCCGGCGGCCCCGCGCGCGCCGCGCTGC NCCCGCCTCACCGGGTCAGT

#BB@?@AB@;5@B;B@@58(

@s_8_1_0001_qseq_14_B

CCCCGCGCGCGCCGCGCTGC

;.;((.).0;.8########

Next the anchors were aligned individually to the reference,keeping their paired ordering.The resulting alignments were read by another custom script that jointly evaluates consecutive anchor alignments belonging to the same original read,performs extensions of the anchor alignments,and collects statistics on splice sites.After the run completes,the script outputs all detected splice junctions (linear and circular)in a UCSC BED-like format with extra columns holding quality statistics,read counts etc.The original full-length reads that support each junction are written to stderr:

$bowtie2-p16–reorder--mm-M20--score-min5C,-15,0-q-x genome-U sample_anchors.qfa.gz j./find_circ.py-S hg19-p sample_-s sample/sites.log. sample/sites.bed2.sample/sites.reads

The resulting BED-like file is readily filtered for minimal quality cutoffs to pro-duce the reported circRNA candidates.In particular,we demanded the following: (1)GU/AG flanking the splice sites(built in);(2)unambiguous breakpoint detec-tion;(3)a maximum of two mismatches in the extension procedure;(4)the breakpoint cannot reside more than2nucleotides inside an anchor;(5)at least two independent reads(each distinct sequence only counted once per sample) support the junction;(6)unique anchor alignments with a safety margin to the next-best alignment of at least one anchor above35points(,more than two extra mismatches in high-quality bases);and(7)a genomic distance between the two splice sites of no more than100kb(only a small percentage of the data).As the ribosomal DNA cluster is part of the C.elegans genome assembly(ce6)and ribo-somal pre-RNAs could give rise to circular RNAs by mechanisms independent of the spliceosome,we discarded130candidates that mapped to the rDNA cluster on chrI:15,060,286-15,071,020.

Permutation testing.To test the robustness of the circRNA detection pipeline we altered the sequence of real sequencing reads in different ways at the step of anchor generation.We(1)reversed either anchor;(2)reversed the complete read;

(3)randomly reassigned anchors between reads;or(4)reverse complemented the read(as a positive control).Although the reverse complement recovered the same output as expected,the various permutations led to only very few candidate predictions,well below0.2%of the output with unpermuted reads and in excel-lent agreement with the results from simulated reads(Supplementary Fig.1c). HEK293RNA-seq after rRNA depletion(RibominusSeq).Total HEK293RNA was isolated using Trizol as recommended by the manufacturer.Ribosomal RNA was depleted from total RNA using the Ribominus kit(Invitrogen).A cDNA library was generated from rRNA-depleted RNA according to the Illumina RNA-seq protocol.The cDNA library was sequenced on an Illumina GAIIx by a2376bp run.

C.elegans oocyte isolation.Oocytes were isolated from worms carrying a tem-perature-sensitive(TS)allele for fem-1(unovulated oocytes BA17[fem-1(hc17ts)] strain)and spe-9(partially ovulated oocytes BA671[spe-9(hc88ts)])as described previously53.Oocytes were washed at least four times in PBS containing protease inhibitors(Sigma-Aldrich)to separate from worm debris.Oocyte purity was observed under the dissection scope(Zeiss).Oocytes were extracted from young adults to enrich for non-endomitotic oocytes,which was also checked by fluo-rescence microscopy(Zeiss)with a nuclear dye.Oocytes isolated from fem-1or spe-9mutant background worms are hereafter referred to as fem-1oocytes and spe-9oocytes,respectively.

C.elegans s perm isolation.Sperm was isolated in principle as described prev-iously54from male worms obtained from a fog-2(q71)mutant background.Males were cut in cold PBS containing protease inhibitors(Sigma-Aldrich).Sperm was subsequently purified by filtration(3340m m nylon mesh,2310m m nylon mesh)and a series of differential centrifugations(30min300g,10min450g) and washed twice in cold PBS.Sperm was subsequently activated by incubation in PBS containing200m g ml21Pronase(Sigma-Aldrich)for30min at25u C. Sperm purity is around70%spermatids and spermatozoa contaminated with around30%primary and secondary spermatocytes,as observed under oil immer-sion microscope.

C.elegans isolation of1-cell-and2-cell-stage embryos.1-cell and2-cell-stage embryos were obtained by fluorescence-activated cell sorting as described previ-ously55.Microscopic examination of the sorted embryos indicated that the1-cell-stage sample was virtually pure(.98%one-cell stage embryos),whereas the 2-cell-stage embryo sample was a mixture of1-cell-stage(40%),2-cell-stage (55%)and older(,5%)embryos.Moreover,purity of the stages was further validated by checking for marker gene expression.

Ribominus RNA preparation from C.elegans samples.We used a kit that was developed for human and mouse samples,but still performs sufficiently to enrich mRNAs up to30%in C.elegans.Most of the remaining reads mapped to ribo-somal RNAs.1m g of total RNA per sample was depleted from rRNAs with the Ribominus Transcriptome kit(Invitrogen)according to the manufacturer’s instructions with the modification that annealing of LNA probes to total RNA was performed in a thermocycler(Eppendorf)with a temperature decrease from 70to37u C at a rate of1u C per min.Depletion of rRNAs was validated by capillary gel electrophoresis on a Bioanalyzer(Agilent).The ribominus RNA was then processed for sequencing library preparation according to the Illumina protocol. Cluster generation and sequencing of C.elegans libraries.Cluster generation as well as sequencing of the prepared libraries was performed on the Illumina cluster station(Illumina)and sequenced on the HiSeq2000according to the manufac-turer’s protocols(Illumina).

Human gene models.We obtained gene models for RefSeq transcripts(12 December2011),non-coding RNAs29,56and the rnaGene and tRNA tracks from the UCSC table browser(23April2012)27.

Intersection of circRNAs with known transcripts.Our computational screen identifies only the splice sites that lead to circularization but not the internal exon/ intron structure of circular RNAs.To perform analyses of the sequence content of circRNAs we therefore inferred as much as possible from annotated transcripts. The conservative assumption was that as little as possible should be spliced out. On the other hand,coincidence of circRNA splice sites with exonic boundaries inside a transcript were considered as an indicator for relevant agreement and internal introns appear to be spliced out(Supplementary Fig.2e).We therefore sorted all overlapping transcripts hierarchically by(1)splice-site coincidence (2,1,or0);(2)total amount of exonic sequence between the splice sites;(3)total amount of coding sequence.The latter was used to break ties only and helped the annotation process.If one or both splice sites fell into an exon of the best match-ing transcript,the corresponding exon boundary was trimmed.Likewise,if it fell

ARTICLE RESEARCH