Identification and Characterization of Endothelial circRNAs

Endothelial circRNAs were identified in ribosomal RNA-depleted RNA of human umbilical vein endothelial cells (HUVEC) by using PYTHON scripts provided by circBase.¹² Bowtie2,¹³ a short read aligner which is used to align reads from a deep-sequencing experiment to a reference genome was used to select and analyze splice junction reads followed by filtering out low-quality candidates (Figure 1A). The highest expressed transcripts are shown in Figure 1B. Among the 7388 identified circRNAs of our study, we identified several new circRNAs, which are not deposited in circBase (eg, cSIRT1, cFN1, and cTGFB1; Online Table I). The majority of the circRNAs (95.6%) derive from the gene body, whereas 1.6% are of intergenic origin and 0.5% derive from noncoding sequences (Online Table II). Interestingly, circRNA expression did not correlate with the expression of the host genes suggesting an independent regulation of transcription versus circRNA formation (Figure 1C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Identification of circular RNAs (circRNAs) in endothelial cells. A, Bioinformatics pipeline for the identification of circRNAs using next generation sequencing sequence reads of rRNA-depleted RNA of human umbilical vein endothelial cells (HUVEC). Gene expression was determined from mapped reads using Cufflinks, whereas circRNAs were identified from unmapped reads using Bowtie2 in search for back-splice junction reads, according to the circBase scripts, followed by filtering of low-quality candidates, thereby excluding mitochondrial circRNAs and unique reads <2. Mapping quality of both sequence reads forming the back-splice site had to be >35 bp, whereas distance between them had to be <100 000 bp. B, Highest expressed circRNA transcripts of the 7388 total identified circRNAs in HUVEC. Data are mean+SD. C, Scatter plot of circRNA to host gene expression (selection circRNA>0, host genes<4000 fragments per kilobase of transcript per million mapped reads [FPKM]).

On the basis of basal expression levels estimated using RNA-seq data and the number of reads covering the back-splice sites, we selected 8 circRNAs for further characterization. Reads of 5 selected circRNAs were derived from exon–exon back-splicing events, whereas 2 circRNAs showed back-splicing of exons to intronic sequences or intron–intron head to tail joining (Online Table III). To confirm the expression of these circular transcripts, we performed semiquantitative reverse transcription PCR (RT-PCR) experiments using primers encompassing the circRNA-specific back-splice sites (Online Figure I) and showed that all 8 selected candidates are expressed in HUVEC (Figure 2A) as well as in microvascular and aortic ECs (Online Figure II). Next, we determined whether the selected transcripts are indeed circRNAs by testing their enrichment in poly-A–negative fractions (as circRNAs are expected to have no poly-A tail⁸) and resistance to 3′-5′ exoribonuclease RNase R digestion (as circRNAs are expected to be resistant to exonuclease RNase R¹⁴). We demonstrate that 7 of 8 circRNAs are predominantly detected in the poly-A–negative fraction (Figure 2B and 2C), whereas the protein-coding mRNA for endothelial NO synthase was enriched in the poly-A–positive fraction (Figure 2B and 2C). Only cMED13L was also detected in the poly-A-positive fraction (Figure 2B). Furthermore, we show that all 8 transcripts are resistant to RNase R, whereas linear vascular endothelial growth factor-A mRNA is degraded (Figure 2D and 2E). Notably, all circRNAs except cMED13L were enriched in the cytoplasmic fraction, whereas the nuclear long noncoding RNA MALAT1 is detected in the nuclear fraction (Figure 2F). In summary, 7 of the 8 selected circRNAs fulfilled all criteria of bona fide cytoplasmic circRNAs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Confirmation and characterization of selected circular RNAs (circRNAs) in human umbilical vein endothelial cells (HUVEC). A, Expression of 8 selected circRNAs was validated by semiquantitative reverse transcription polymerase chain reaction (RT-PCR) using primers encompassing the newly formed circRNA-specific back-splice site. cANRIL was used as control. B, RNA was fractionated using oligo-dT beads followed by RT-PCR for circRNAs. Endothelial nitric oxide synthase (eNOS) mRNA was used as polyadenylated positive control. C, Quantification of fractionated poly-A+/−RNA by RT-PCR (n=3). D, Quantification of circRNAs after RNase R digestion by RT-PCR (n=3). E, Total RNA was digested using RNase R followed by RT-PCR for circRNAs. Vascular endothelial growth factor A (VEGF-A) mRNA was used as RNase-sensitive control. F, Expression of circRNAs in cytoplasmic and nuclear fractions. Malat1 was used as a positive control for nuclear localization, RPLP0 as positive control for cytoplasmic localization. All data are n=3; mean+SEM. *P<0.05.

Identification of Hypoxia-Regulated circRNAs

Because hypoxia is a key stimulus for angiogenesis, we next exposed HUVECs to hypoxia and determined the expression of circRNAs. Interestingly, hypoxia induced a significant regulation of circRNAs as depicted by the heat map (Figure 3A) and the bar graph of selected circRNAs (Figure 3B). The regulation of selected circRNAs by hypoxia and the hypoxia mimic CoCl₂ was subsequently validated by reverse transcription PCR (Figure 3C and 3D; Online Figure IIIA). In this context, cZNF292, cAFF1, and cDENND4C are upregulated, whereas cTHSD1 is reduced by hypoxia. cSRSF4 is only upregulated by 1.2-fold (Figure 3C and 3D). PCR against cFOXJ3 yielded multiple bands excluding its analysis. With the exception of cAFF1, the regulation of circRNAs by hypoxia was even more pronounced when cells were cultured under dense conditions (Online Figure IIIB). To determine whether the observed regulation of circRNAs is because of transcriptional regulation of host genes, host gene expression was analyzed from RNA-seq data (Figure 3E) and validated by quantitative reverse transcription PCR (Figure 3F). A correlation of host gene expression and circRNA formation under hypoxia was observed for cZNF292, cDENNDA4C, and cTHSD1, whereas only the upregulation of cAFF1 was found to be independent of hypoxia-induced transcriptional changes (Figure 3E and 3F).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Regulation of circular RNAs (circRNAs) under hypoxia in endothelial cells. A, Heatmap of hypoxia-regulated circRNAs determined by next generation sequencing (NGS). B, Hypoxic regulation of selected circRNAs in human umbilical vein endothelial cells (HUVEC) after 12 hours (n=2) or 24 hours (n=2) hypoxia compared with normoxic controls (n=4). *P<0.05 for pooled analysis of hypoxic versus normoxic samples. C, Expression of selected circRNAs in HUVECs after 24-hour hypoxia (0.2% O₂) by reverse transcription polymerase chain reaction (RT-PCR; n=5–7) normalized to RPLP0. D, Representative RT-PCRs of selected circRNAs in HUVECs after 24-hour hypoxia. E, Host gene regulation in NGS data (see A). F, Validation of host gene regulation after 24-hour hypoxia (0.2%) using quantitative RT-PCR (n=5) normalized to RPLP0. All data are mean+SEM/SD. Statistical significance was determined by student test; *P<0.05. VEGF indicates vascular endothelial growth factor.

cZNF292 Regulates Angiogenic Sprouting

With respect to circRNAs, one of the key questions is whether these transcripts have biological functions or are mere byproducts of mRNA processing. To address this question, we focused on cZNF292, which was among the highest expressed and significantly hypoxia-regulated circRNAs, and confirmed its expression by Northern blot (Online Figure IV). Using RNase R-treated RNA, we cloned cZNF292 by using primers directly flanking the predicted back-splice site and confirmed the circularization of exon 4 to intron 1 (Figure 4A; Online Figure V). A detailed analysis uncovered an alternative exon 1A located in intron 1 of which the 5′ end functions as splice acceptor site during back-splicing. Next, we silenced cZNF292 by using 2 siRNAs directed against exon 1A (Figure 4B) since the sequence of the back-splice site was not suitable for the generation of specific siRNAs (because of potential off target effects and high GC [guanine and cytosine base pairs] content). We confirmed that both siRNAs silenced cZNF292 but did not affect pre-mRNA levels or ZNF292 protein expression (Online Figure VI). siRNA-mediated silencing of cZNF292 significantly inhibited spheroid sprouting (Figure 4C), tube formation in matrigel assays (Online Figure VII), and reduced endothelial cell proliferation (Figure 4D). Transfection of a plasmid that was designed to generate the circular transcript of ZNF292 enhanced proliferation, whereas overexpression of linear exon 1A to 4 did not affect cell-cycle progression (Online Figure VIII).

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Function of cZNF292 in endothelial cells. A, Scheme illustrating the exon composition of cZNF292 analyzed by cloning using the indicated primer pair. B, siRNA-mediated knockdown of cZNF292 using 2 siRNAs was confirmed using quantitative polymerase chain reaction (qPCR; n=9). Values normalized to RPLP0 mRNA. Representative reverse transcription PCR (insert). C, Sprouting of spheroids after sicZNF292 treatment (n=5). Representative spheroids are shown at the bottom. D, Cell-cycle analysis by FACS after sicZNF292 treatment. E, Expression of ZNF292 mRNA after sicZNF292 was determined by qPCR (n=7). Values normalized to RPLP0 mRNA (n=7). F, Scheme depicting the analysis of miRNA binding of circular RNAs by using Ago high-throughput sequencing of RNA isolated by cross-linking and immunoprecipitation (HITS-CLIP) data. All data are mean+SEM. Statistical significance was determined with student test; *P<0.05.

Having demonstrated that cZNF292 regulates angiogenic sprouting, we aimed to obtain insights into its mechanism. Although nuclear circRNAs may interfere with the generation of their host gene,⁶ siRNA-mediated silencing of cZNF292 did not affect ZNF292 mRNA expression (Figure 4E). Because circRNAs were shown to act as sponges for microRNAs (miRNAs), we determined whether this mechanism may also account for cZNF292. By merging our identified circRNAs with the published high-throughput sequencing of RNA isolated by cross-linking and immunoprecipitation (HITS-CLIP) data from HUVEC¹⁵ (experimentally determined binding sites of Ago proteins), we bioinformatically characterized putative miRNA-binding sites within endothelial circRNAs (Figure 4F; Online Table IV). However, no HITS-CLIP peaks were detected for cZNF292 indicating that cZNF292 does not associate with Argonaute, making a potential function as miRNA sponge unlikely.