Circular RNA expression Profiles Alter Significantly in Mouse Brain after Transient Focal Ischemia

Abstract

Background and Purpose

Circular RNAs (circRNAs) are a novel class of non-coding RNAs formed from many protein-coding genes by backsplicing. Although their physiologic functions are not yet completely defined, they are thought to control transcription, translation and miRNA levels. We investigated whether stroke changes the circRNAs expression profile in the mouse brain.

Methods

Male C57BL/6J mice were subjected to transient middle cerebral artery occlusion (MCAO), and circRNA expression profile was evaluated in the penumbral cortex at 6h, 12h, and 24h of reperfusion using circRNA microarrays and real-time PCR. Bioinformatics analysis was conducted to identify miRNA binding sites, transcription factor binding and gene ontology of circRNAs altered after ischemia.

Results

1,320 circRNAs were expressed at detectable levels mostly from exonic (1,064) regions of the genes in the cerebral cortex of sham animals. Of those, 283 were altered (>2-fold) at least at one of the reperfusion time points, whereas 16 were altered at all 3-time points of reperfusion after transient MCAO studied compared to sham. Post-ischemic changes in circRNAs identified by microarray analysis were confirmed by real-time PCR. Bioinformatics showed that these 16 circRNAs contain binding sites for many miRNAs. Promoter analysis showed that the circRNAs altered after stroke might be controlled by a set of transcription factors. The major biological and molecular functions controlled by circRNAs altered after transient MCAO are biological regulation, metabolic process, cell communication, and binding to proteins, ions and nucleic acids.

Conclusions

This is a first study that shows that stroke alters the expression of circRNAs with possible functional implication to post-stroke pathophysiology.

INTRODUCTION

Mammalian brain abundantly expresses various classes of noncoding RNAs (ncRNAs) that include long non-coding RNA (lncRNA) and microRNA (miRNA) which control transcription and translation.^1,2 Recent studies show that stroke significantly alters the expression of both lncRNAs and miRNAs, and the post-stroke brain can be protected by normalizing the levels of specific lncRNAs and miRNAs.^3–9 Despite these studies, the significance of other classes of ncRNAs in brain damage and/or repair after ischemia is not yet evaluated. Circular RNAs (circRNAs) are an evolutionarily conserved class of ncRNAs that are abundantly formed by backsplicing from many primary RNA transcripts from which mRNAs are formed.² In vertebrates, most mRNAs are degraded by RNase R within 4 to 5h of transcription. Whereas, circRNAs are extremely stable with a half-life exceeding 2 days as they are inaccessible to RNase R due to the absence of defined 5′ and 3′ ends.¹⁰ CircRNAs contain both exons and introns and most of them are known to contain 1 to 5 preferentially longer exons (exons in the circRNAs are usually 3X longer than an average exon of an mRNA).¹¹

The physiologic functions of circRNAs are still being discovered, but they are shown to control transcription of parent genes, promote rolling circle translation, help to form alternatively spliced mRNAs and sponge miRNAs.¹² Furthermore, altered levels of specific circRNAs were thought to promote cancer, cardiac hypertrophy, atherosclerosis and neurodegenerative diseases.^13–17 To understand if stroke influences the expression of circRNAs, we profiled the levels of 14,236 circRNAs in the cerebral cortex of adult mice as a function of reperfusion time after transient focal ischemia. Using bioinformatics, we evaluated the transcription factors (TFs) that might control circRNA formation after focal ischemia. We also identified miRNA binding sites and the putative biological functions of the stroke-responsive circRNAs.

METHODS

Focal ischemia

All surgical procedures were approved by the Research Animal Resources and Care Committee of the University of Wisconsin-Madison, and the animals were cared for in accordance with the Guide for the Care and Use of Laboratory Animals, U.S. Department of Health and Human Services Publication Number. 86–23 (revised). Animals were randomly assigned to study groups (Supplemental Table I). Focal ischemia was induced by intraluminal middle cerebral artery occlusion (MCAO) for 90 min using a 6-0 silicon-coated monofilament (Doccol Corporation USA) in male C57BL/6J mice (12 weeks, 25±3 g, Charles River USA) under isoflurane anesthesia as described earlier.^18,19 Rectal temperature was controlled at 37.0 ± 0.5°C and regional cerebral blood flow and physiological parameters (pH, P_ao₂, P_aco₂, hemoglobin and blood glucose) were monitored. Cohorts of mice were euthanized at 6h, 12h or 24h of reperfusion. Animals that showed symptoms of stroke-induced deficits were included. None of the animals showed any adverse effects or died before they were euthanized. Sham-operated mice underwent similar surgical procedure except MCAO. We used only male animals as this is an exploratory study to show the effects of stroke on circRNA expression.

CircRNA microarrays

Experimental groups were blinded to the person conducting microarray profiling. Briefly, animals were deeply anesthetized, saline perfused and the ischemic penumbral cortex was dissected and used for isolating total RNA, circRNAs were enriched by digesting linear RNAs with RNase R (Epicentre Inc. USA), amplified and transcribed into fluorescent cRNA (Super RNA Labeling Kit, Arraystar USA). The fluorophore-labeled cRNAs were purified (RNeasy Mini Kit; Qiagen USA) and 1 μg sample of each labeled cRNA was fragmented (60°C for 30 min in fragmentation buffer/blocking buffer; Arraystar USA), suspended in hybridization buffer and hybridized to a mouse circRNA microarray (8×15K, Arraystar USA).

CircRNA data analysis and bioinformatics

Arrays were washed and scanned with the Agilent scanner G2505C, and the data from scanned images were imported using Agilent Feature Extraction Software. Using the publicly available R package (https://www.r-project.org/), data was quantile normalized, and low-intensity signals were filtered. A present, absent and marginal calls during raw intensity extraction and probe flagging of circRNAs were given based on the features such as positive and significant signal, saturation, population outlier, above background and uniformity of the background. The quality control flags (present, absent and marginal) are generated by the Agilent Feature Extraction Software during raw intensity extraction while scanning the array for the hybridization signals of the individual RNAs. A marginal call will be generated when the background surrounding the spot is not uniform. However, the software factors-in backgrounds of all other spots considering if a signal is less than a lower threshold or exceeds an upper threshold determined using a multiplier (1.42) for the interquartile range of the population of all other backgrounds to avoid false positives. Based on this the software filters the RNAs with lower thresholds as negative calls and retains those with higher thresholds for further analysis. RNAs with the marginal calls were retained for further analysis only if the expression changes are consistent and across all time points studied. Fold changes (⩾2.0 and p-values ≤ 0.05) were computed between the groups (sham versus each reperfusion time points) by cross-comparison analysis. CircRNA-miRNA interactions were predicted with a proprietary miRNA target prediction software developed based on TargetScan and miRanda (Arraystar USA). Transcription factor analysis was conducted using the promoters of the parent genes of circRNAs altered after focal ischemia to predict if sufficient interaction exists for each gene and each TF using a high stringency TF matrix simulation by Gene-TF Analysis software (Genomatix USA). The functional classification and enrichment analysis of the circRNA parent genes were conducted using the ontology of the terms generated with gene ontology (GO) curation through publicly available WEB-based GEne SeT Analysis Toolkit (WebGestalt; www.webgestalt.org) and DAVID Bioinformatics Resources 6.8 (https://david.ncifcrf.gov).

Real-time PCR validation

Post-stroke changes in the expression of circRNAs observed by microarray analysis were validated by real-time PCR for 3 upregulated (circ_008018, circ_015350, and circ_016128) and 3 downregulated (circ_011137, circ_001729 and circ_006696) circRNAs. Briefly, total RNA was extracted using RNeasy Mini Kit (Qiagen USA) from the ischemic penumbral cortex, and one μg of RNA was reverse transcribed into cDNA with SuperScript® III First-Strand Synthesis System (ThermoFisher Scientific USA). The circRNAs were amplified with circRNA specific outward facing divergent primers that were designed by selecting standard primer design criteria such as primer melting temperature, GC content and product size (Supplemental Table II). The reaction amplification and detection were conducted in a QuantStudio 3 real-time PCR machine (Applied Biosystems USA) using SYBR Green method as described earlier.^7,18 18s rRNA and GAPDH were used as internal controls to normalize the data.

RESULTS

Transient MCAO significantly altered circRNA profiles

The microarrays used in the study contained probes for 14,236 mouse circRNAs. In all the samples (sham, 6h, 12h and 24h reperfusion following transient MCAO; n =3/group), the distribution of circRNA expression patterns was not different (Fig. 1A) indicating that the post-ischemic changes observed in the levels of individual circRNAs are not random. In the cerebral cortex of sham-operated animals, 1,320 circRNAs obtained a present call, and of those, 1,064 are exonic, 42 are intronic, 151 are intragenic, 4 are intergenic, and 59 are antisense to the respective mRNAs formed from the parent genes. A scatterplot (Fig. 1B) and a volcano plot (Fig. 1C) between the sham and a 6h reperfusion time point indicate that the expression of individual circRNAs shows a ~6 fold range between the lowest to the highest expression level. Following transient MCAO, 283 circRNAs were observed to be altered (>2-fold change) at least at one of the reperfusion time points (6h, 12h, and 24h) (Fig. 1D; Supplemental Table III). Of those, 16 were observed to be altered at all 3-time points of reperfusion studied (Table. 1). Of the circRNAs altered after transient MCAO, 75.6% were exonic, and 12.7% were intragenic in origin (Fig. 2A), and were originated from all chromosomes (Fig. 2B).

Fig. 1.

No significant differences were observed in the total expression of circRNAs between groups (sham, 6h, 12h and 24h reperfusion following transient MCAO; n = 3/group) (A). Normalized intensities of all circRNAs expressed in the cerebral cortex between sham and 6h reperfusion groups show a 6 fold range from the lowest to the highest level of expression (B). Volcano plot shows circRNAs altered at 6h reperfusion group compared to sham group (C). 283 circRNAs were altered at one or more time points (6h, 12h and 24h) of reperfusion after transient MCAO compared to sham (D). Sixteen circRNAs showed a persistent change from 6h to 24h of reperfusion compared to sham (D).

Table 1.

CircRNAs altered at all 3 reperfusion time points studied after transient MCAO

circRNA	Δ fold over Sham			Location		Host gene
	6h	12h	24h	Chr	Type	Name	NCBI
circ_016128	2.11	2.82	5.68	chr7+	Exonic	Nars2	NM_153591
circ_007362	2.14	2.22	2.69	chr10−	Antisense	Ccdc6	NM_001111121
circ_006839	2.18	2.65	2.87	chr17+	Intronic	Crim1	NM_015800
circ_000113	2.40	2.69	3.44	chr12−	Antisense	Rian	NR_028261
circ_002664	2.82	2.27	2.02	chr15−	Exonic	Fam49b	NM_144846
circ_008018	3.40	2.82	5.76	chr4+	Intronic	Pum1	NM_001159605
circ_011381	3.58	2.65	3.00	chr11+	Antisense	Grb10	NM_001177629
circ_015350	3.88	2.74	4.66	chr1−	Exonic	Ncoa2	NM_001077695
circ_006696	−3.92	−2.34	−2.88	chr12−	Exonic	Strn3	NM_001172098
circ_001729	−3.38	−2.15	−2.37	chr9−	Exonic	Rasa2	NM_053268
circ_000741	−2.93	−2.48	−2.36	chr10−	Exonic	Akap7	NM_018747
circ_016423	−2.38	−2.11	−2.48	chr17+	Exonic	Plcl2	NM_013880
circ_009396	−2.36	−2.05	−2.56	chr6+	Exonic	Mkln1	NM_013791
circ_017370	−2.35	−2.09	−2.06	chr16+	Exonic	Morc3	NM_001045529
circ_010383	−2.31	−2.52	−2.28	chr10−	Exonic	Akap7	NM_018747
circ_016289	−2.11	2.16	2.00	chr7−	Antisense	Cd81	NM_133655

Values are mean of cross-comparisons between n = 3 samples/group. Chr, chromosome.

Fig. 2.

Majority (~80%) of the circRNAs altered after stroke are exonic (A). The circRNAs altered after stroke were transcribed from all 20 chromosomes (B). Real-time PCR confirmed the microarray data for 6 circRNAs (3 upregulated and 3 downregulated) at 6h reperfusion compared to sham (C). Values are mean ± SD (n =3/group). Real-time PCR analysis was conducted in triplicate. *p<0.05 compared to sham (Student’s T-test).

Real-time PCR confirmed post-ischemic changes in circRNA expression

As they are circularized, we used real-time PCR with divergent primers that amplify away from each other spanning a junction to confirm the post-ischemic expression changes for 3 upregulated and 3 downregulated circRNAs. This method eliminated the false negative amplification of the mRNAs originating from the same genes as circRNAs. The circRNAs circ_008018, circ_015350, and circ_016128 were observed to be upregulated and circ_011137, circ_001729 and circ_006696 were observed to be downregulated significantly at 6h reperfusion following transient MCAO compared to sham (Fig. 2C) similar to the microarray data.

Post-ischemic circRNA changes might be mediated by a set of transcription factors

As TFs control the expression of most if not all genes, we analyzed if there is a specific set of TFs that are responsible for the post-ischemic changes in the circRNA expression profiles. We conducted this analysis for the 16 most altered circRNAs after stroke (Fig. 3 and Supplemental Fig. I). TF binding site analysis showed that promoters of all 13 out of 16 highly stroke-responsive circRNAs have binding sites for Fork head domain factors (Fig. 3). In addition, 11 out of 16 stroke-responsive circRNA gene promoters have binding sites for TALE homeodomain class recognizing TG motifs, 9 out of 16 circRNAs have binding sites for pleomorphic adenoma gene, 8 out of 16 circRNAs have binding sites for C2H2 zinc finger transcription factors 2, and 7 out of 16 circRNA gene promoters have binding sites for TGF-β induced apoptosis proteins and RNA Pol II transcription factor II B (Fig. 3). This indicates that circRNAs that contain certain TFs might be preferentially altered after stroke.

Fig. 3.

Transcription factors and number of their binding sites in the promoters of 16 circRNAs altered maximally after stroke.

CircRNAs altered after stroke has many miRNA binding sites

Sponging and thereby controlling the levels of miRNAs is one of the functions attributed to circRNAs. Hence, we conducted a detailed analysis of the miRNA binding sites in the 16 stroke-responsive circRNAs that were altered at all 3 reperfusion time points (6h, 12h, and 24h) after transient MCAO compared to sham. 13 out of 16 circRNAs altered after stroke showed >60 miRNA binding sites each (Table 2). The circ_016423 showed 625 miRNA binding sites (Table 2). Four circRNAs showed >245 (range: 245 to 445), 4 circRNAs showed >145 (range: 145 to 191) and 5 circRNAs showed >63 (range: 63 to 89) miRNA binding sites each (Table 2). These circRNAs were observed to bind to a repertoire of miRNAs. The circ_016423 can bind to 521 different miRNAs, while 6 other circRNAs can bind to >150 different miRNAs each and another 6 can bind to >60 miRNAs each (Table 2). This indicates that circRNAs altered after stroke can significantly change the miRNA function and thus translation of a multitude of proteins those miRNAs control.

Table 2.

miRNAs targeted by circRNAs altered at all 3 reperfusion time points (6h, 12h and 24h) after transient MCAO

circRNA	# of miRNA	# of miRNA binding sites	# of miRNAs with >3 binding sites
circ_016128	86	87	1
circ_007362	1	1	0
circ_006839	60	63	0
circ_000113	11	11	0
circ_002664	174	191	2
circ_008018	353	386	1
circ_011381	0	0	0
circ_015350	80	81	0
circ_006696	378	445	6
circ_001729	72	73	0
circ_000741	216	246	3
circ_016423	521	625	13
circ_009396	155	159	0
circ_017370	79	89	3
circ_010383	158	177	3
circ_016289	137	145	1

CircRNAs might control many pathophysiologic processes after stroke

A circRNA and an mRNA are formed from the same primary RNA transcript transcribed from a gene. The circRNAs are known to control their parent genes and hence, to understand their pathophysiologic significance after stroke, we conducted bioinformatics analysis of the functions of the mRNAs formed from the parent genes of stroke-responsive circRNAs. The major biological functions controlled by these are biological regulation, metabolic process, response to stimulus and cell communication (Fig. 4). Gene enriched KEGG pathways analysis conducted for the parent genes of stroke-responsive circRNAs showed that MAPK signaling (9 genes), cell cycle (7 genes), regulation of actin cytoskeleton (6 genes) and focal adhesion (5 genes) are the major signal transduction pathways associated with the circRNAs altered (data not shown). Moreover, protein binding, ion binding, and nucleic acid binding are the major metabolic functions of mRNAs associated with circRNAs altered after stroke (Fig. 4). Within the cell, the majority of the stroke-responsive circRNA associated mRNAs were observed to be in the nucleus, membranes and macromolecular complexes (Fig. 4).

Fig. 4.

Gene ontology classifications of the biological process, molecular function and cellular localization of mRNAs originated from the parent genes of stroke-responsive circRNAs.

DISCUSSION

In brief, our study is the first to show that stroke significantly alters circRNA levels within the first day of reperfusion which might have functional consequences in controlling the secondary brain damage and neurological dysfunction. Hence, circRNAs might be attractive future therapeutic targets to protect the brain after stroke as well as other acute CNS injuries.

More than 98% of the transcriptional output in mammals is one or the other class of ncRNAs. The physiologic functions of many of these ncRNAs are still being discovered. However, some of them like miRNAs and lncRNAs are known to control translation, transcription, epigenetics and RNA/protein scaffolding. Overall, ncRNAs are considered as the master regulators of the genome, and altered ncRNA homeostasis is being recognized as a major promoter of disease, and secondary brain damage following CNS insults.²⁰ Interestingly, post-stroke brain damage mediated by several synergistic pathophysiologic mechanisms including inflammation, oxidative stress, endoplasmic reticulum stress, mitochondrial dysfunction, apoptosis, and autophagy,^21,22 could be modulated by rectifying the level of ncRNAs altered after stroke. In particular, the therapeutic potential of modulating specific miRNAs after stroke is well documented.²⁰ Replenishing the levels of miRNAs down-regulated after stroke like miR-9, miR-99a, miR-207, miR-424, let-7c, miR-29c, miR-210, and miR-122 was shown to prevent one or the other pathologic events listed above to induce significant neuroprotection in rodent models of focal ischemia.^23–30 On the other hand inhibiting miRNAs induced after a stroke like miR-181a, miR-479, let-7f and miR-145 were also shown to protect the brain after focal ischemia in rodents.^3,8,31–33

In addition, recent studies showed that stroke significantly influences the expression of several lncRNAs in the brain and many stroke-responsive lncRNAs were thought to participate in post-ischemic secondary brain damage by modulating inflammation, apoptosis, and autophagy.^4,34,35 We recently showed that one such stroke-induced lncRNA called Fos downstream transcript (FosDT) scaffolds Sin3A and coREST which are chromatin modifying proteins essential for the actions of the RE1-silencing TF (REST) that is known to mediate post-ischemic secondary brain damage.^7,34,36 Knocking-down of FosDT significantly decreases the post-stroke brain damage by curtailing the REST downstream genes glutamate ionotropic receptor AMPA-type subunit 2 (GRIA2) and glutamate receptor, ionotropic, N-methyl d-aspartate 1 (GRIN1).⁷ Furthermore, lncRNAs C2dat1 and NILR were shown to promote neuronal survival in the ischemic brain by modulating NF-κB signaling and inhibiting p53 signaling, respectively,^37,38 whereas lncRNAs H19 and TUG1 were shown to promote secondary ischemic brain damage by activating autophagy, apoptosis and p53, respectively.^39–41 A recent study further demonstrated that lncRNA MALAT1 induced after focal ischemia is neuroprotective by curtailing inflammation and apoptosis.⁹

All the above studies opened the exciting new possibility of protecting the brain after a stroke by modulating transcription and translation via ncRNAs. Our current study shows that cerebral ischemia also influences circRNAs, the newly discovered class of ncRNAs. Some classes of ncRNAs like miRNAs are highly conserved while other classes like lncRNAs are poorly conserved between humans and rodents. The circRNAs arise from the same genes that transcribe protein-coding RNAs (mRNAs) by back-splicing. ~13% of the protein-coding genes form both linear RNAs and circular RNAs and they are highly conserved.^2,10,12 The mechanisms of actions of circRNAs under normal physiologic conditions are still being discovered, and their significance in protecting or damaging brain after an injury is not yet studied in detail. However, as circRNAs are known to regulate the expression of their host as well as neighboring genes,^42,43 their disruption can modulate many molecular events essential to protect the brain after stroke. Our study shows that the host genes that form stroke-responsive circRNAs participate in biological regulation, metabolic process, response to stimulus and cell communication in addition to protein binding, ion binding, and nucleic acid binding. All these are essential cellular functions needed for the survival of healthy neurons. At this time, we only have in silico evidence to connect the circRNAs altered after stroke with post-ischemic pathophysiology. Experimental tools and reagents are still being developed to either stop or induce a circRNA without changing the mRNA formed by the parent gene, and those will conclusively show the functional significance of circRNAs in modulating post-stroke brain damage.

The circRNAs arise from the same primary RNA transcripts that form mRNAs by alternative splicing. Hence, their transcription uses the same mechanisms that govern mRNA transcription. With bioinformatics, we show that stroke-responsive circRNAs might be formed by a specific set of TFs that include Fork head domain factors, TALE homeodomain class recognizing TG motifs, Pleomorphic adenoma gene, TGF-β induced apoptosis proteins, C2H2 zinc finger transcription factors 2 and RNA Pol II transcription factor II B. This indicates that the effect of stroke on circRNAs is not a random event and might have implications together with altered TF function which is known to play a significant role in post-ischemic pathophysiology.⁴⁴

CircRNAs are thought to act as sponges to titrate the levels of miRNAs in a cell and/or transport miRNAs from one cell to another cell.⁴⁵ The miRNA miR-7a targets apoptotic and inflammatory genes and hence thought to play a role in preventing myocardial ischemia-induced cell death, cancer progression, and Parkinson’s disease.^46–48 Whereas, induction of circRNA ciRS-7 (Cdr1as) that sponges miR-7a was shown to promote ischemic cell death and tumorigenesis by depleting cell-protective miR-7a.^14,15 The miR-223 is known to cause cardiac hypertrophy and the heart-related circRNA HRCR is thought to protect the heart by sponging miR-223.⁴⁹ The circRNA TTBK2 was shown to promote malignancy of gliomas by modulating miR-217 that controls hepatocyte nuclear factor 1β/Derlin-1 pathway,⁵⁰ and circRNA ITCH is known to suppress lung cancer by sponging miR-7 and miR-214 thereby suppressing the activation of Wnt/β-catenin ⁵¹. In the present study, the circRNAs altered after stroke were identified to contain binding sites for many miRNAs indicating their potential to modulate miRNA-mediated translation and thus ischemic pathophysiology. We observed that >80% of the stroke-responsive circRNAs showed >60 miRNA binding sites each. The circ_016423 showed a maximum of 625 miRNA binding sites that can bind to 521 different miRNAs. Hence, the interaction of circRNAs-miRNAs needs to be factored in to understand the pathophysiologic significance of ncRNAs in the post-stroke brain. Overall, our study for the first time shows that circRNAs are sensitive to cerebral ischemia and their altered function might promote the post-stroke pathophysiology. Post-stroke pathology is known to be modulated by specific cell types including neurons and various glial cells. Currently, it is not known if circRNAs are expressed in a cell-specific manner. In the present study, we used ischemic penumbral cortex to show the changes in circRNA expression profiles. However, changes seen from 6h to 24h of reperfusion might be a mixture of all cellular compartments including glial cells. Future studies will decipher the cell-type specific changes in circRNAs and their mechanisms.