RESEARCH ARTICLE into an intron or beyond transcript bounds,the closest exon was extended to were dissected and tissue samples were collected directly into ice-cold Trizol for match the circRNA boundaries.circRNA start/end coordinates were never RNA preparation.Caenorhabditis elegans RNA was isolated from about 7,000 altered.If no annotated exons overlapped the circRNA we assumed a single-exon mixed stage worms by two rounds of freeze-thaw lysis in Trizol LS reagent circRNA.The resulting annotation of circRNAs is based on the best matching (Invitrogen)according to the manufacturer's protocol.RNA was extracted from transcript and may in some cases not represent the ideal choice.Changing the aqueous phase with phenol:chloroform (Ambion).RNA was precipitated with annotation rules,however,did not substantially change the numbers in Fig.ld.isopropanol and Glycoblue(Ambion)overnight at-2o Cor for30 min at-80'C, Finding circRNAs conserved between human and mouse.We reasoned that respectively.Reverse transcription was performed using M-MLV(Promega)or when comparing two species,the cutoff of two independent reads in each of them Superscript I I with oligo(dT)primer((all Invitrogen)or random primer(Meta- could be dropped,as orthologous circRNAs would automatically be supported by bion).For assaying mRNA expression level,qRT-PCR was performed using SYBR- two independently produced reads via the intersection.We therefore mapped all Green Fluorescein (Thermo Scientific,Fermentas)and a StepOnePlus PCR System mouse circRNA candidates with less stringent filtering to human genome coor-(Applied Biosystems).Expression data in CDRlas knockdown experiments,tran- dinates using the UCSC liftOver tool.The mapped mouse circRNAs were com- scriptional block and RNase R assays were normalized to C.elegans spike-in RNA. pared with independently identified human circRNAs,yielding 229 circRNAs To this end 5-10%C.elegans total RNA was added to the respective Trizol sample with precisely orthologous splice sites between human and mouse.Of these,223 and qPCR primer for ama-I or eif-3.d were used.Mouse expression data were were composed exclusively of coding exons and were subsequently used for our normalized to Actb.miRNA expression levels were assayed using TaqMan conservation analysis (Fig.1f).When intersecting the reported sets of circRNAs microRNA assays(Applied Biosystems)and normalized to sno-234.Expression supported by two independent reads in each species,we found 81 conserved levels of circRNAs described in this study were measured by qPCR using divergent circRNAs (supported by at least 4 reads in total). primers.A list of primer sequences is available in Supplementary Table 8. Conserved element counting.We downloaded genome-wide human (hg19)PCR amplification and Sanger sequencing.DNA templates were PCR amplified phyloP conservation scores tracks derived from genome alignments of placental using BioRad Mastercyclers and ThermoScientific DreamTaq Green PCR Master mammals from UCSC27.We interrogated the genome-wide profile inside Mix according to the manufacturer's protocol.We performed 35 cycles of PCR. circRNAs in two different ways.(1)Intergenic and intronic circRNAs.We read PCR products were visualized after electrophoresis in 2%ethidium bromide- out the conservation scores along the complete circRNA and searched for blocks stained agarose gel.To confirm the PCR results,the PCR products were purified of at least 6-nucleotide length that exceeded a conservation score of 0.3 for through Agencourt AMPure XP PCR purification kit.Direct PCR product Sanger intergenic and 0.5 for intronic circRNAs.The different cutoffs empirically adjust sequencing was performed by LGC Genomics Ready2 Run services.Primer Pl for the different background levels of conservation and were also used on the was provided for sequencing the product for each candidate. respective controls.For each circRNA,we computed the cumulative length of all Primer design.Divergent primers were designed for each candidate(P1,P2)to such blocks and normalized it by the genomic length of the circRNA.Artefacts of anneal at the distal ends of its sequence.As negative controls we used divergent constant positive conservation scores in the phyloP profile,apparently caused by primers for GAPDH and ACTB linear transcript in HEK293 cells,and elF-3.D in missing alignment data,were removed with an entropy filter (this did not qua- C.elegans.As a further negative control for divergent primers,we used genomic litatively affect the results).circRNAs annotated as intronic by the best-match DNA extracted through Qiagen DNeasy Blood Tissue kit.As positive controls, procedure explained above that had any overlap with exons in alternative tran-we used convergent primers for the corresponding linear transcripts or for house- scripts on either strand (five cases)were removed from the analysis.The resulting keeping genes (elF-3.D for C.elegans). distributions are shown in Supplementary Fig.Ih,i.(2)Coding exon circRNAs.RNase R treatment.HEK293 DNase-treated total RNA (5 ug)was incubated We used the best-match strategy outlined above to construct an estimated 'exon-15 min at 37C with or without 3 U ug of RNase R (Epicentre Bio- chain'for the circRNAs that overlapped exclusively coding sequence.Using this technologies).RNA was subsequently purified by phenol-chloroform extraction, chain we in silico 'spliced'out the corresponding blocks of the conservation score retro-transcribed through Superscript SSIII(Invitrogen)according to the man- profile.We kept track of the frame and sorted the conservation scores into ufacturer's protocol,and used in qPCR. separate bins for each codon position.In addition to this,we also recorded RNA nicking assay.For partial alkaline hydrolysis (nicking)1 ug ul-of conservation scores in the remaining pieces of coding sequence ('outside'the HEK293 total RNA was incubated in 50 mM NaHCO;for 2.5 or 5 min at 90C circRNA)as a control.However,we observed that the level of conservation is or 5 min on ice for controls.After incubation the samples were immediately re- systematically different between internal parts of the coding sequence and the suspended in denaturing RNA sample buffer and analysed on northern blots. amino-or carboxy-terminal parts(not shown).We therefore randomly generated Northern blotting.Total RNA(l0-2oμg)was loaded on a I.2%agarose gel chains of internal exons,mimicking the exon-number distribution of real containing 1%formaldehyde and run for 2-2.5 h in MOPS buffer. circRNAs,as a control.When analysing the circRNAs conserved between human The gel was soaked in 1XTBE for 20 min and transferred to a Hybond-N+ and mouse,it became furthermore apparent that we also needed to adjust for the membrane (GE Healthcare)for 1 h(15 V)using a semi-dry blotting system (Bio- higher level of overall conservation.High expression generally correlates with Rad).Membranes were dried and ultraviolet-crosslinked (at 265 nm)1X at conservation and thus,an expression cutoff was enforced on the transcripts used 200,000 uJ cm2.Pre-hybridization was done at 42C for 1 h and 32p-labelled to generate random controls.This resulted in a good to conservative match with oligonucleotide DNA probes were hybridized overnight.The membranes were the actual circRNAs (Supplementary Fig.1j,k). washed briefly in 2X SSC,0.1%SDS at room temperature and two additional Overlap of identified circRNAs with published circular RNAs.A number of times at 55C for 30 min,followed by two 30-min washes in 0.2X SSC,0.1%SDS studies in human have reported evidence for circRNAs which derive from exons at 50-55C.For data collection,the membrane was exposed to a phosphoimager of DCC',ETSI and a non-coding RNA from the human INK4/ARF locus and screen. the CDRlas locus'.Additionally,circRNAs from exons of the genes CAMSAPI,Genome alignments for detecting miRNA seed complementary sites.Multiple FBXW4,MAN1A2,REXO4,RNF220 and ZKSCANI have been recently experi-species alignments for the genomic intervals,corresponding to circRNAs pre- mentally validated.For the four genes from the latter study,where we had dicted in C.elegans(ce6),human (hg19)and mouse(mm9),were generated via ribominus data from the tissues in which these circRNAs were predicted (leuko- the Galaxy server at UCSCS9-61.In case that a circle was overlapping with an cytes),we recovered validated circRNAs from all of them (ZKSCAN1,CAMSAPI,annotated transcript,the inferred spliced sequence was used for retrieving the FBXW4,MANIA2). alignments. Cell culture and treatments.HEK293(Fig,3f),HEK293TN(for virus production)The alignments included C.elegans,C.briggsae and C.remanei in the first case and HEK293 Flp-In T-REx 293 (Life Technologies,all other experiments)were and Homo sapiens,Mus musculus,Rattus norvegicus,Bos taurus and Canis famili- cultured in Dulbecco's modified Eagle medium GlutaMax(Gibco)4.5 gl-'glucose,aris in the latter two. supplemented with 10%FCS,20U ml-'penicillin and streptomycin (Gibco)at C.elegans human and mouse miRNAs.Fasta files with C.elegans,human and 37C,5%CO2.Whereas CDRlas/GAPDH ratios were within the given range,we mouse miRNAs were obtained from miRBase release 16(ref.62).Only mature observed two-to fivefold variation of CDRlas/vinculin ratios between different miRNAs were considered for the seed analysis.According to miRBase 16 a HEK lines.Transcription was blocked by adding 2 ug ml actinomycin D or mature miRNA is the predominant miRNA between the two species arising from DMSO as a control (Sigma-Aldrich)to the cell culture medium.For in vitro wound the two arms of the precursor hairpin (information that is not included in more healing assays,cells were grown to confluency,the cell layer was disrupted using a recent versions).The miRNAs were grouped into families that share a common 300 ul pipette tip and cells were washed once with medium.Bright-field images of seed (nucleotides 2-7).There are 117,751 and 723 miRNA families for C.elegans, cells were taken using a Axio Observer.Z1 (Zeiss)right after setting the scratch and human and mouse,respectively. 24h later.The relative scratch areas were measured using Image)software. Detecting putative miRNA seed matches.The C.elegans,human and mouse Quantitative PCR.Total RNA from cell lines was isolated using Trizol (Invi-multiple species alignments were scanned for putative conserved miRNA target trogen)extraction following the manufacturer's protocol Adult B6129SF1/]mice sites for each of the mature miRNA families.A putative target site of a miRNA is a ©2013 Macmillan Publishers Limited.All rights reservedinto an intron or beyond transcript bounds, the closest exon was extended to match the circRNA boundaries. circRNA start/end coordinates were never altered. If no annotated exons overlapped the circRNA we assumed a single-exon circRNA. The resulting annotation of circRNAs is based on the best matching transcript and may in some cases not represent the ideal choice. Changing the annotation rules, however, did not substantially change the numbers in Fig. 1d. Finding circRNAs conserved between human and mouse. We reasoned that when comparing two species, the cutoff of two independent reads in each of them could be dropped, as orthologous circRNAs would automatically be supported by two independently produced reads via the intersection. We therefore mapped all mouse circRNA candidates with less stringent filtering to human genome coor￾dinates using the UCSC liftOver tool57. The mapped mouse circRNAs were com￾pared with independently identified human circRNAs, yielding 229 circRNAs with precisely orthologous splice sites between human and mouse. Of these, 223 were composed exclusively of coding exons and were subsequently used for our conservation analysis (Fig. 1f). When intersecting the reported sets of circRNAs supported by two independent reads in each species, we found 81 conserved circRNAs (supported by at least 4 reads in total). Conserved element counting. We downloaded genome-wide human (hg19) phyloP conservation score58 tracks derived from genome alignments of placental mammals from UCSC27. We interrogated the genome-wide profile inside circRNAs in two different ways. (1) Intergenic and intronic circRNAs. We read out the conservation scores along the complete circRNA and searched for blocks of at least 6-nucleotide length that exceeded a conservation score of 0.3 for intergenic and 0.5 for intronic circRNAs. The different cutoffs empirically adjust for the different background levels of conservation and were also used on the respective controls. For each circRNA, we computed the cumulative length of all such blocks and normalized it by the genomic length of the circRNA. Artefacts of constant positive conservation scores in the phyloP profile, apparently caused by missing alignment data, were removed with an entropy filter (this did not qua￾litatively affect the results). circRNAs annotated as intronic by the best-match procedure explained above that had any overlap with exons in alternative tran￾scripts on either strand (five cases) were removed from the analysis. The resulting distributions are shown in Supplementary Fig. 1h, i. (2) Coding exon circRNAs. We used the best-match strategy outlined above to construct an estimated ‘exon￾chain’ for the circRNAs that overlapped exclusively coding sequence. Using this chain we in silico ‘spliced’ out the corresponding blocks of the conservation score profile. We kept track of the frame and sorted the conservation scores into separate bins for each codon position. In addition to this, we also recorded conservation scores in the remaining pieces of coding sequence (‘outside’ the circRNA) as a control. However, we observed that the level of conservation is systematically different between internal parts of the coding sequence and the amino- or carboxy-terminal parts (not shown).We therefore randomly generated chains of internal exons, mimicking the exon-number distribution of real circRNAs, as a control. When analysing the circRNAs conserved between human and mouse, it became furthermore apparent that we also needed to adjust for the higher level of overall conservation. High expression generally correlates with conservation and thus, an expression cutoff was enforced on the transcripts used to generate random controls. This resulted in a good to conservative match with the actual circRNAs (Supplementary Fig. 1j, k). Overlap of identified circRNAs with published circular RNAs. A number of studies in human have reported evidence for circRNAs which derive from exons of DCC4 , ETS15 and a non-coding RNA from the human INK4/ARF locus8 and the CDR1as locus9 . Additionally, circRNAs from exons of the genes CAMSAP1, FBXW4, MAN1A2, REXO4, RNF220 and ZKSCAN1 have been recently experi￾mentally validated10. For the four genes from the latter study, where we had ribominus data from the tissues in which these circRNAs were predicted (leuko￾cytes), we recovered validated circRNAs from all of them (ZKSCAN1, CAMSAP1, FBXW4, MAN1A2). Cell culture and treatments. HEK293 (Fig. 3f), HEK293TN (for virus production) and HEK293 Flp-In T-REx 293 (Life Technologies, all other experiments) were cultured in Dulbecco’s modified Eagle medium GlutaMax (Gibco) 4.5 g l21 glucose, supplemented with 10% FCS, 20 U ml21 penicillin and streptomycin (Gibco) at 37 uC, 5% CO2. Whereas CDR1as/GAPDH ratios were within the given range, we observed two- to fivefold variation of CDR1as/vinculin ratios between different HEK lines. Transcription was blocked by adding 2 mg ml21 actinomycin D or DMSO as a control (Sigma-Aldrich) to the cell culture medium. Forin vitro wound healing assays, cells were grown to confluency, the cell layer was disrupted using a 300 ml pipette tip and cells were washed once with medium. Bright-field images of cells were taken using a Axio Observer.Z1 (Zeiss) right after setting the scratch and 24 h later. The relative scratch areas were measured using ImageJ software. Quantitative PCR. Total RNA from cell lines was isolated using Trizol (Invi￾trogen) extraction following the manufacturer’s protocol. Adult B6129SF1/J mice were dissected and tissue samples were collected directly into ice-cold Trizol for RNA preparation. Caenorhabditis elegans RNA was isolated from about 7,000 mixed stage worms by two rounds of freeze–thaw lysis in Trizol LS reagent (Invitrogen) according to the manufacturer’s protocol. RNA was extracted from aqueous phase with phenol:chloroform (Ambion). RNA was precipitated with isopropanol and Glycoblue (Ambion) overnight at 220 uC or for 30 min at 280 uC, respectively. Reverse transcription was performed using M-MLV (Promega) or Superscript III with oligo(dT) primer (all Invitrogen) or random primer (Meta￾bion). For assaying mRNA expression level, qRT–PCR was performed using SYBR￾Green Fluorescein (Thermo Scientific, Fermentas) and a StepOnePlus PCR System (Applied Biosystems). Expression data in CDR1as knockdown experiments, tran￾scriptional block and RNase R assays were normalized to C. elegans spike-in RNA. To this end 5–10% C. elegans total RNA was added to the respective Trizol sample and qPCR primer for ama-1 or eif-3.d were used. Mouse expression data were normalized to Actb. miRNA expression levels were assayed using TaqMan microRNA assays (Applied Biosystems) and normalized to sno-234. Expression levels of circRNAs described in this study were measured by qPCR using divergent primers. A list of primer sequences is available in Supplementary Table 8. PCR amplification and Sanger sequencing. DNA templates were PCR amplified using BioRad Mastercyclers and ThermoScientific DreamTaq Green PCR Master Mix according to the manufacturer’s protocol. We performed 35 cycles of PCR. PCR products were visualized after electrophoresis in 2% ethidium bromide￾stained agarose gel. To confirm the PCR results, the PCR products were purified through Agencourt AMPure XP PCR purification kit. Direct PCR product Sanger sequencing was performed by LGC Genomics Ready2 Run services. Primer P1 was provided for sequencing the product for each candidate. Primer design. Divergent primers were designed for each candidate (P1, P2) to anneal at the distal ends of its sequence. As negative controls we used divergent primers for GAPDH and ACTB linear transcript in HEK293 cells, and eIF-3.D in C. elegans. As a further negative control for divergent primers, we used genomic DNA extracted through Qiagen DNeasy Blood & Tissue kit. As positive controls, we used convergent primers for the corresponding linear transcripts or for house￾keeping genes (eIF-3.D for C. elegans). RNase R treatment. HEK293 DNase-treated total RNA (5 mg) was incubated 15 min at 37 uC with or without 3 U mg21 of RNase R (Epicentre Bio￾technologies). RNA was subsequently purified by phenol-chloroform extraction, retro-transcribed through Superscript SSIII (Invitrogen) according to the man￾ufacturer’s protocol, and used in qPCR. RNA nicking assay. For partial alkaline hydrolysis (nicking) 1 mg ml 21 of HEK293 total RNA was incubated in 50 mM NaHCO3 for 2.5 or 5 min at 90 uC or 5 min on ice for controls. After incubation the samples were immediately re￾suspended in denaturing RNA sample buffer and analysed on northern blots. Northern blotting. Total RNA (10–20 mg) was loaded on a 1.2% agarose gel containing 1% formaldehyde and run for 2–2.5 h in MOPS buffer. The gel was soaked in 13TBE for 20 min and transferred to a Hybond-N1 membrane (GE Healthcare) for 1 h (15 V) using a semi-dry blotting system (Bio￾Rad). Membranes were dried and ultraviolet-crosslinked (at 265 nm) 13 at 200,000 mJ cm22 . Pre-hybridization was done at 42 uC for 1 h and 32P-labelled oligonucleotide DNA probes were hybridized overnight. The membranes were washed briefly in 23 SSC, 0.1% SDS at room temperature and two additional times at 55 uC for 30 min, followed by two 30-min washes in 0.23 SSC, 0.1% SDS at 50–55 uC. For data collection, the membrane was exposed to a phosphoimager screen. Genome alignments for detecting miRNA seed complementary sites. Multiple species alignments for the genomic intervals, corresponding to circRNAs pre￾dicted in C. elegans (ce6), human (hg19) and mouse (mm9), were generated via the Galaxy server at UCSC59–61. In case that a circle was overlapping with an annotated transcript, the inferred spliced sequence was used for retrieving the alignments. The alignments included C. elegans, C. briggsae and C. remanei in the first case and Homo sapiens, Mus musculus, Rattus norvegicus, Bos taurus and Canis famili￾aris in the latter two. C. elegans human and mouse miRNAs. Fasta files with C. elegans, human and mouse miRNAs were obtained from miRBase release 16 (ref. 62). Only mature miRNAs were considered for the seed analysis. According to miRBase 16 a mature miRNA is the predominant miRNA between the two species arising from the two arms of the precursor hairpin (information that is not included in more recent versions).The miRNAs were grouped into families that share a common seed (nucleotides 2–7). There are 117, 751 and 723 miRNA families for C. elegans, human and mouse, respectively. Detecting putative miRNA seed matches. The C. elegans, human and mouse multiple species alignments were scanned for putative conserved miRNA target sites for each of the mature miRNA families. A putative target site of a miRNA is a RESEARCH ARTICLE ©2013 Macmillan Publishers Limited. All rights reserved