Abstract
Non-coding RNAs are emerging as novel regulators in adipocyte differentiation and function. Circular RNAs (circRNAs) are a new class of non-coding transcripts generated across all eukaryotic tissues, but their function in adipose biology remains unknown. Here we perform deep sequencing of visceral and subcutaneous fat to discover thousands of adipose circRNAs, many of which are species conserved, tissue specific and dynamically regulated during adipogenesis and obesity. We identify circTshz2-1 and circArhgap5-2 as indispensable regulators of adipogenesis in vitro. To characterize the function of circRNAs in vivo, we deliver adenoviral shRNA targeting circArhgap5-2 into mouse inguinal tissue and show that the expression of this circRNA is essential in maintaining the global adipocyte transcriptional programme involved in lipid biosynthesis and metabolism. We also demonstrate that the pro-adipogenic function of circArhgap5-2 is conserved in human adipocytes. Our results provide important evidence that circRNAs serve as important regulators in adipocyte differentiation and metabolism.
Similar content being viewed by others
Main
Obesity poses an escalating health burden worldwide1. Understanding the mechanisms underlying adipose tissue development, energy storage and nutrient homeostasis is necessary for developing new therapies for obesity and its co-morbid diseases.
Non-coding RNAs are emerging regulators in adipogenesis and other metabolic processes. For instance, microRNAs (miRNAs) exhibit altered expression patterns in adipocytes during differentiation and obesity2,3. Advances in next-generation sequencing have also uncovered thousands of long non-coding RNAs (lncRNAs) in adipose tissue, many of which play important roles in adipogenesis and lipid metabolism4,5. Some lncRNAs are preferentially expressed in white or brown fat and are responsive to browning stimuli, acting under master adipogenic regulators such as PPARγ to establish adipocyte identity4,5,6,7,8,9,10,11.
While miRNAs and lncRNAs are extensively studied in adipose biology, the role of other highly expressed non-coding transcripts remains to be explored. CircRNAs are a newly described class of RNAs, generated by a ‘backsplicing’ reaction that covalently links the 3′ end of a downstream exon to the 5′ end of an upstream exon at their canonical splice sites, thus forming a closed, non-polyadenylated loop12. They are generally expressed at lower levels than linear transcripts from the host gene, but can exist in greater abundance than their linear cognates at specific loci13. Once thought to be rare splicing by-products14,15,16, high-throughput sequencing has now confirmed their expression and conservation across thousands of mammalian genes13,17,18,19,20.
Important progress has been made towards deciphering circRNA biogenesis. CircRNAs are derived mostly from protein-coding exons21, but can originate from intronic, antisense, intergenic and 5′/3′ untranslated regions19,22,23. They may form post-transcriptionally24 or co-transcriptionally in direct competition with pre-mRNA splicing25. CircRNA production is largely dependent on flanking internal introns, which typically contain long reverse complementary sequences (for example, Alu elements) that pair-hybridize and promote circularization by the spliceosome20,21,26,27,28. Backsplicing may be regulated by RNA-binding proteins (for example, Mbl and QKI) or RNA-editing enzymes (for example, ADAR1) targeting the intronic sequences21. As circRNA expression changes dynamically across different tissues and during development18,19,29, their biogenesis is probably regulated in a tissue/developmental stage-specific manner.
To date, only a few circRNAs have been functionally and mechanistically characterized14,19,30,31,32. One of the first and most well-studied circRNAs is the neuron-enriched ciRS-7/CDR1as, which harbours multiple miR-7 binding sites and acts as an miRNA sponge19,31,32. In vivo deletion of CDR1as in mice abolishes sequestering of miR-7 and results in synaptic dysregulation and aberrant sensorimotor behaviour32. Some intron-retaining circRNAs interact with upstream promoters, RNA polymerase II and other regulatory proteins to modulate gene transcription22,23,33. Moreover, despite their general lack of coding potential, some circRNAs are polysome-bound and actively translated into novel proteins in a cap-independent manner34,35,36.
The emerging regulatory role of circRNAs in disease pathophysiology has attracted much interest37,38,39. Due to their intrinsic stability, context-specific patterns of expression and presence in human plasma40, they harbour potential as diagnostic biomarkers in cancer39, neurodegenerative disorders29 and cardiovascular disease41. In the field of metabolic disorders, the sponging activity of CDR1as against miR-7 reportedly influences insulin expression and secretion in β cells42, while circHIPK3 plays a role in diabetic retinopathy by blocking miR-30a function, leading to increased endothelial proliferation and vascular dysfunction43. A recent study profiled circRNA expression in pig subcutaneous adipose tissue, implicating circRNAs in lipid metabolism44. Despite this progress, the role of circRNAs in the context of obesity remains to be systematically explored.
To address this gap, we began by analysing our deep RNA-sequencing (RNA-seq) datasets and identified thousands of adipose-expressed circRNAs. We selected a subset of highly abundant, conserved and adipose-enriched candidates and examined their dynamic expression patterns during adipogenesis and obesity. Our study identifies the circular isoform circArhgap5-2 as an indispensable driver of the adipogenic gene programme both in vitro and in vivo.
Results
Global expression analysis of adipose circRNAs
We performed deep RNA-seq on ribosomal-depleted total RNA extracted from human visceral (omental) and subcutaneous fat tissue, as well as mouse epididymal and inguinal fat. Sequencing for each of the four RNA-seq libraries achieved 127 ± 10 million paired-end reads. These were aligned to reference genomes (hg19 for human, mm9 for mouse) using Tophat2, with high mapping efficiency of ≥80%. Gene expression was quantified using Cufflinks2 (v.2.2.1), and reported as fragments per kilobase per million reads (FPKM). Each library showed a consistently high number of expressed genes (FPKM > 1), with over 16,000 genes and 14,000 genes detected in all human and mouse samples, respectively.
Applying an efficient and unbiased algorithm for de novo circRNA identification, CIRI2 (refs. 45,46), we detected 6,925 backsplicing events in human tissue, and 2,380 circRNAs in mouse tissue (Fig. 1a and Supplementary Fig. 1a). CircRNAs, represented by head-to-tail backspliced junctions (BSJ), were called by CIRI2 with stringent filtering criteria. After annotating human circRNAs with GENCODE v.19, we found that the majority (5,651/6,925, 81.6%) were generated from coding exons, and the rest arose from introns (376/6,925, 5.4%), exon–intron boundaries (190/6,925, 2.7%) and other loci (708/6,925, 10.2%) (Fig. 1a). Similar distributions were observed in mice (Supplementary Fig. 1a). Comparison of our adipose list with circRNAs identified in the publicly available circBase database (http://www.circbase.org/) revealed 952 novel isoforms detected by our RNA-seq analysis. (Summary data for all detected circRNAs in Supplementary Table 1a.)
a, Chart of circRNAs detected in human adipose, indicating proportions of exonic, intronic, exon–intron and other types of BSJs. ‘Others’ includes intergenic circRNA (169) and circular RNA spanning two genes (539). b, Most highly expressed circRNA isoforms in human adipose, ranked by mapped backspliced junction read count. The read counts are normalized by the library size, which is represented by SRPBM. c, Human adipose genes producing highest numbers of circRNA isoforms. The isoform numbers are normalized to the number of exons. d, Scatter plot of expression abundance (SRPBM) of circRNA isoforms between human subcutaneous and omental adipose. e, Scatter plot of BSR between human subcutaneous and omental adipose. f, Integrated panel showing relationship between linear host gene expression and corresponding circRNA expression in human adipose. Heat map shows linear genes ranked by expression abundance (yellow: highest FPKM, blue: lowest FPKM). Scatter plot shows abundance of circRNA expression (SRPBM) corresponding to linear genes in the rank position of the heat map. Size of each point adjusted according to number of circRNA isoforms generated by each linear gene. g, Relative expression abundance (SRPBM) of conserved and non-conserved circRNAs in human and mouse adipose. n = 2 biologically independent animals; human conserved (yes) = 133, minimum = 9.46, Q1 = 14.31, median = 33.30, Q3 = 66.33 maximum = 1,822.07; human non-conserved (no) = 6,792, minimum = 4.73, Q1 = 9.46, median = 14.19, Q3 = 28.53 maximum = 11,441.41; mouse conserved (yes) = 133, minimum = 7.64, Q1 = 8.65, median = 16.29, Q3 = 46.05, maximum = 1,621.07; mouse non-conserved (no) = 2,247, minimum = 3.82, Q1 = 7.64, median = 1.46, Q3 = 26.75, maximum = 9,457.24; *P < 0.05 using two-sided Wilcoxon’s sum ranked test, P(human) = 3.59 × 10−15, P(mouse) = 7.10 × 10−5.
We annotated the 100 most highly expressed circRNAs in our adipose libraries (Fig. 1b, Supplementary Fig. 1b and Supplementary Table 1b). The most abundant human circRNA is circHIPK3-1, whose host gene produces another circular isoform that has been found to act as a miRNA sponge30. circZNF609-1, the fourth most highly expressed human candidate, is reported to be actively translated into a truncated peptide34. Other abundant circRNAs ranking within the top 10 are transcribed from the genes RHOBTB3, COL3A1, SMARCA5, ESYT2, ASPH, LPAR1 and CORO1C, whose linear mRNAs encode structural or enzymatic regulatory proteins. Among the host genes generating the highest numbers of circRNA isoforms (for example, AHNAK, ZNF124, NRIP1, ASPH, PDE3B), many are known to undergo extensive alternative splicing to create multiple linear transcripts (Fig. 1c, Supplementary Fig. 1c and Supplementary Table 1c). Overall, the median predicted number of exons for each human exonic circRNA was 3 (range: 1–78), with 10.6% of circRNAs containing only a single exon and 6.2% predicted to cover ≥10 exons. Interestingly, key white adipocyte regulators, such as PPARγ, LEP, ADIPOQ, FABP4 and INSR, all produce detectable exonic circRNAs in human adipose (Supplementary Table 1a).
When comparing the normalized abundance (measured by spliced reads per billion mapped, or SRPBM, equation (1)) of circRNAs between subcutaneous and visceral fat in both humans and mice, we observed good correlation between expression patterns in both adipose depots (Fig. 1d and Supplementary Fig. 1d). Nonetheless, a few circRNAs showed depot-specific distribution and were detected either in only visceral (for example, circEPB41L2-2, circNCL-2, circHNRNPA2B1-1, circEFEMP1-1, circCOL1A1-1, circMETRNL-1) or in only subcutaneous fat (for example, circVIM-1, circFBLN2-1, circCOL1A2-1, circCOL4A2-1, circFBN1-4) (Supplementary Table 1a). Additionally, the distribution of backspliced-to-linear ratios (BSR, a measure of the relative proportion of linear and circular junction read counts, equation (2)) were consistent between the two types of white adipose tissue (WAT), indicating that circRNA production in both tissues is regulated similarly, relative to linear transcription in the host genes (Fig. 1e and Supplementary Fig. 1e). Globally, the abundance of circular transcripts was largely outnumbered by their linear cognates (based on raw junction reads) in both visceral and subcutaneous adipose (Supplementary Fig. 1f), but a small subset of circRNAs (320/4,384, 7.2% for visceral; 209/2,974, 7.0% for subcutaneous) were expressed at similar or higher levels relative to the linear transcript. Host genes with higher linear expression tended to yield more abundant circRNAs (Fig. 1f), but this proportional relationship was not perfectly linear (Supplementary Fig. 1f), suggesting that circRNA biogenesis is modulated by additional layers of regulation beyond basal transcription at their loci.
Next, we wanted to identify circRNAs enriched in adipose. We applied the same discovery algorithm to human tissue RNA-seq datasets downloaded from the ENCODE Project Consortium (https://www.encodeproject.org/), and compared circRNA expression in these libraries with our own list of adipose candidates. We identified 1,810 circRNAs (26.1%) that were detected exclusively in our human adipose dataset, indicating a high degree of tissue specificity (Supplementary Fig. 1g and Supplementary Table 1d).
Using a criterion of ≥70% sequence identity between the human and mouse BSJ sequences (extended 25 base pairs (bp) on either end of the splice site) originating from homologous genes, we also identified 120 species-conserved circRNAs (Supplementary Table 1e). These candidates showed similar relative expression between human and mouse adipose samples (Supplementary Fig. 1h). Interestingly, they also exhibited significantly higher abundance compared to non-conserved circRNAs, suggesting that these conserved circRNAs may be more functionally relevant (Fig. 1g).
Experimental validation of adipose circRNAs
Next, we selected circRNAs for further validation and characterization in adipose tissue (Fig. 2a). From the subset of conserved circRNAs (Supplementary Table 1e) we selected the 20 most highly expressed candidates from either human or mouse adipose tissue on the basis of BSJ read count, as well as the 15 most adipose-enriched circRNAs. CircRNAs of length <90 bp were excluded due to PCR detection limits, resulting in a final union set of 41 candidates. We performed PCR with reverse transcription (RT–PCR) with divergent primers that specifically amplify the BSJ but not the canonical linear sequence of the host gene (Fig. 2b), and successfully detected backsplicing in 37 out of 41 (90.2%) circRNA candidates in mouse adipose tissue. For each candidate, we gel-extracted bands of the expected size and confirmed their correct head-to-tail junction sequences by Sanger sequencing. (Summary of selection and validation of candidates in Supplementary Table 2.)
a, Selection criteria for final pool of circRNA candidates. Exonic circRNAs identified in human and mouse WAT were shortlisted on the basis of BSJ sequence conservation, highest SRPBM abundance in human and/or mouse samples and enrichment in human adipose compared to other tissues. Backsplicing in 37 out of 41 candidates successfully validated by RT–PCR and Sanger sequencing. b, Schematic for validation of circRNAs, shown for circArhgap5-2. Divergent primers that amplify only the circularized head-to-tail junction but not the linear transcript were used for RT–PCR of circRNAs in mouse adipose. Sanger sequencing of gel-extracted bands confirmed the predicted BSJ sequence. c, RNase R assay to confirm circularity of candidates via resistance to exonuclease digestion. RT–PCR to detect circRNAs and linear cognate transcripts was performed on mock-treated and RNase R-treated adipose RNA. Repeated twice independently with similar results. d, Heat map showing relative expression of circRNAs and linear cognates across a mouse tissue panel, including brown (BAT), white inguinal (iWAT) and epididymal (eWAT) adipose tissue. RT–qPCR data presented as normalized z scores; n = 3 biologically independent experiments for iWAT, eWAT and BAT, n = 2 for other tissues; red: increased tissue enrichment, blue: less tissue enrichment. e, Subcellular fraction analysis of the relative proportion of circRNAs expressed in the nuclear and cytosolic compartments of mouse inguinal adipose. Values derived from comparison of RT–qPCR Ct values in each subcellular compartment, n = 3 biologically independent experiments, mean % ± s.e.m.
To confirm whether the backspliced sequences indeed reflect a circular structure or were simply due to linear concatemers derived from genomic tandem duplications or exon reshuffling, we performed RT–PCR for ten circRNAs on mouse adipose RNA treated with RNase R, a 3′ to 5′ exoribonuclease that degrades linear transcripts but not lariats or circRNAs. Ten out of ten tested circRNAs were all resistant to RNaase R digestion, indicating a closed-loop structure, while their corresponding linear cognates were susceptible to exonucleolytic cleavage (Fig. 2c). Therefore, the backspliced sequences detected in our analysis are reliable indicators for the circular structure of these transcripts.
Next, we determined whether our in silico prediction of adipose specificity (Supplementary Fig. 1g and Supplementary Table 1d) truly reflected in vivo expression levels. Using real-time quantitative PCR (RT–qPCR) with BSJ-specific primers, we quantified the relative abundance of 12 putative adipose-enriched circRNAs (circAbca1-2, circCrim1-2, circHspa4-3, circMap2k1-1, circMed13l-1, circPhkb-2, circSkap2-1, circTshz2-1, circAsph-9, circHipk3-1, circNfix-2, circOgdh-2) across different mouse organs and fat depots. All 12 circRNAs were preferentially expressed in brown adipose tissue (BAT) and/or WAT (Fig. 2d). In contrast to circRNAs, the distribution of their linear cognates showed greater dispersal among the different tissues and less specificity in adipose. This aligns with previous findings that the tissue-specific expression of circRNAs is modulated by additional layers of regulation beyond the transcriptional activity of their host genes21,24. We also examined the relative abundance of ten circRNAs that are not predicted to be adipose-enriched. As expected, neither these circRNAs nor their linear cognates exhibit clear adipose-specific expression (Supplementary Fig. 2a).
To determine the subcellular localization of our candidates, we separated the nuclear and cytosolic fractions of inguinal adipose and quantified the relative proportion of circRNAs in each compartment by RT–qPCR. Transcript levels of 47S pre-rRNA and maturely spliced Gapdh were quantified as nuclear and cytosolic controls, respectively (Fig. 2e). In line with previous reports on circRNA localization19,20,31, the majority of our candidates were predominantly expressed in the cytosol, suggesting that they probably exert their functions in this compartment. Furthermore, because adipose tissue is composed of multiple cell types, we examined the expression of these circRNAs in mature adipocytes versus the stromal vascular fraction (SVF) to determine their distribution. Most of the circRNAs are preferentially expressed in mature adipocytes (Supplementary Fig. 2b).
CircRNAs are regulated during adipocyte differentiation
To examine transcriptional changes of our validated circRNAs during adipogenesis, total RNA was collected at different time points from primary mouse pre-adipocytes that were chemically induced to differentiation. We quantified the expression levels of circular candidates and their cognate linear transcripts across the time course by RT–qPCR (Fig. 3a). Most circRNAs (31 of 37) showed ≥2 fold change upregulation during adipogenesis, in close correspondence with increasing linear mRNA levels.
a, Heat map of circRNA and linear cognate expression (shown in the same row) during differentiation time course of primary white adipocytes. Transcript levels log2-transformed and normalized to D0. CircRNA expression hierarchically clustered by row (Euclidean distance, average linkage). RT–qPCR data, n = 4 biologically independent samples, red: fold change increase in transcript levels, blue: fold change decrease. b, Cumulative frequency plot of PCC between RT–qPCR expression values of circular versus linear transcripts during adipogenesis time course. Colour-scaled ranking of circRNA PCC values shown, from 0 (blue) to 1 (red). c, log2FC, where FC is fold change, in expression of circRNAs (red) and linear cognates (blue) in eWAT and iWAT adipose of mice fed with HFD compared with control chow diet. n = 10 biologically independent animals for HFD, n = 11 for chow diet, RT–qPCR data presented as mean log2FC with *P < 0.05, two-tailed Student’s t-test. d, Scatter plot of log2 fold change in expression of circRNAs versus linear cognates in differentiated D5 primary white adipocytes treated with soluble TNF-α for 24 h compared to mock treatment. log2FC presented as mean values, n = 4 biologically independent experiments.
To quantify the co-expression of circular and linear transcripts during differentiation, we calculated the Pearson correlation coefficient (PCC) for each circRNA–messenger RNA pair across the time course (Fig. 3b). The expression of all detected circRNAs was positively correlated with linear gene transcription, with 32 of 37 (86.5%) exhibiting r > 0.6 and 18 of 37 (48.6%) crossing a strong correlation threshold of r > 0.9. None of the examined circRNAs exhibited a negative correlation with its linear transcript. Thus, our results indicate that the dynamic expression changes of circRNAs during adipogenesis are largely dependent on the transcriptional regulation of their host genes. Interestingly, six circRNA–mRNA pairs exhibited strikingly different expression patterns between the circular and linear transcripts (circAbca1-2, circArih1-1, circArhgap5-2, circNfib-2, circPhkb2-1 and circTshz2-1). These circRNAs accumulated at a much faster rate compared with their relatively static cognate mRNAs (Supplementary Fig. 3a and Figs. 4b and 5c). Thus, apart from the low decay rates of circRNAs, stage-specific factors regulating pre-mRNA backsplicing may drive the production of circular isoforms independent of linear mRNAs.
a, Schematic of circRNA biogenesis in the Tshz2 locus of mouse chromosome 2. Backsplicing occurs in the head-to-tail junction of the first 2,934 bp of exon 2, producing circTshz2-1. This gene structure is derived from NM_001363023.1 sequences. Transcript-specific DsiRNAs ‘si-circ’ (red), ‘si-lin’ (green) and ‘si-circlin’ (blue) designed to target the circRNA, linear mRNA and both transcripts, respectively. b, Expression levels of circTshz2-1 and linear Tshz2 mRNA during differentiation time course of primary white adipocytes. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments. c, Knockdown efficiency and specificity of DsiRNAs (relative to negative control si-NC) in differentiated day 3 primary white adipocytes transfected at the pre-adipocyte stage (measured 96 h post-transfection). RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, **P < 0.01, two-tailed Student’s t-test. d, BODIPY 493/503 fluorescence imaging and ORO staining to examine effect of DsiRNA knockdown on lipid droplet accumulation. Nuclear imaging with Hoechst 33342. Scale bar: 100 µm. Repeated twice independently with similar results. e, Expression levels of adipogenic markers Pparγ, AdipoQ, Cebpα and Fabp4 in differentiated primary adipocytes where circular, linear or both transcripts were silenced. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, **P < 0.01 two-tailed Student’s t-test. f, Effects of circTshz2-1 knockdown on adipogenic markers on the day of differentiation induction and during the first 2 d of differentiation. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test.
a, Schematic of circRNA biogenesis in the Arhgap5 locus of mouse chromosome 12. Backsplicing occurs in the head-to-tail junction of the first two exons, producing circArhgap5-2. The gene structure is derived from NM_009706.2 transcript. Circularization either retains the intervening intron (inclusion) or splices it out (exclusion). Forward primers (paired with reverse primer R) were designed to detect the BSJ (F0), exon-only (F1, F2, F3) and intron-retaining (F4, F5, F6) isoforms. b, RT–PCR of mouse adipose RNA with primers detecting backsplicing (F0), intron exclusion (F1, F2, F3) and intron inclusion (F4, F5, F6). Bands of expected size indicated with green arrows (smaller, non-specific bands correspond to backsplicing of the first Arhgap5 exon only, which is indicated with red arrows). Repeated twice independently with similar results. c, Expression levels of circArhgap5-2 and linear Arhgap5 mRNA during differentiation time course of primary white adipocytes. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments. d, Knockdown efficiency and specificity of DsiRNAs (relative to negative control si-NC) in differentiated D3 primary white adipocytes transfected at the pre-adipocyte stage (measured 96 h post-transfection). RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, **P < 0.01, two-tailed Student’s t-test. e, BODIPY imaging and ORO staining to examine effect of DsiRNA knockdown on lipid droplet accumulation. Nuclear imaging with Hoechst 33342. Scale bar: 100 µM. Repeated twice independently with similar results. f, Expression levels of adipogenic markers Pparγ, AdipoQ, Cebpα and Fabp4 in differentiated primary adipocytes where circular, linear or both transcripts were silenced. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, **P < 0.01, two-tailed Student’s t-test. g, Effect of circArhgap5-2 knockdown on adipogenic markers on the day of differentiation induction and during the first 2 d of differentiation. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test.
CircRNAs are downregulated in diet-induced obesity
Next, we assessed the impact of obesity on in vivo levels of circRNAs by comparing their expression in adipose tissue from C57BL/6 mice fed with high-fat diet (HFD) versus a control group fed with normal chow diet. Expression of the adipokines leptin and adiponectin, which are respectively upregulated and downregulated in classical models of obesity, were measured as controls (Supplementary Fig. 3b). Remarkably, we observed clear downregulation in epididymal WAT for 33 of 37 circRNAs under HFD conditions, with 19 showing statistical significance. A similar decreasing trend was observed in the subcutaneous inguinal fat of HFD mice (Fig. 3c). For visceral fat, the linear mRNAs were generally downregulated in parallel with decreasing circRNA abundance. Interestingly, in inguinal fat most of these mRNAs were upregulated, in an opposite pattern to the decreased circRNAs. Thus, the HFD-induced regulation of circRNAs in subcutaneous adipose cannot be entirely attributed to lower transcriptional activity of the host genes, and additional modulation should exist at the post-transcriptional level.
Given the general downregulation of circRNAs in obesity, it was intriguing to investigate the regulatory stimuli that might trigger this shift. Excess adiposity typically induces the recruitment of monocytes and other immune cells into fat depots, which release cytokines that activate chronic inflammation47. Tumour necrosis factor-α (TNF-α) is a key proinflammatory cytokine whose levels are highly elevated in WAT during obesity. Thus, we tested whether TNF-α-induced inflammation in adipocytes induces perturbed circRNA expression during obesity. On 24-h treatment of differentiated primary mouse subcutaneous adipocytes with soluble TNF-α (sTNF-α), we found reduced circRNA expression for 31 out of 37 circRNAs, with 19 showing a significant decrease (Supplementary Fig. 3c). Interestingly, the fold changes of circRNAs correlated poorly with linear cognates (R2 = 0.034), as only a few mRNAs were significantly downregulated on sTNF-α treatment (Fig. 3d). Thus, activation of inflammatory pathways in mature adipocytes contributes to the suppressive effect of obesity on circRNA expression, suggesting that inflammation may have a post-transcriptional impact on circRNA biogenesis.
circTshz2-1 and circArhgap5-2 are required for adipogenesis
To explore the function of circRNAs during adipogenesis, we used an RNAi-based strategy to knockdown circTshz2-1 and circArhgap5-2, two candidates whose circular transcripts were highly upregulated during differentiation relative to their linear cognates, which did not change significantly (Figs. 4b and 5c). circTshz2-1 was identified as adipose-enriched and highly abundant in mouse tissue within the conserved subset, while circArhgap5-2 is abundant among human conserved circRNAs (Supplementary Table 2).
circTshz2-1 is a sense-oriented circularized transcript originating from the teashirt zinc finger homeobox 2 (Tshz2) locus, a conserved gene from a protein family of transcription factors involved in cancer progression and embryonic development48,49. The circular isoform comprises the first 2,934 bp of the second exon of the linear sequence, with the BSJ adjoining the 5′ end and a 3′ internal splice site located upstream of the next canonical exon–intron boundary (Fig. 4a). circTshz2-1 abundance increased fourfold by day 6 (D6) of differentiation, whereas Tshz2 mRNA levels remained relatively unchanged (Fig. 4b).
We designed three customized Dicer-substrate small interfering RNAs (DsiRNAs) to specifically target different transcripts: ‘si-circ’ binds only to the BSJ of the circular isoform; ‘si-lin’ is complementary to a 3′ UTR sequence not incorporated into the circRNA and targets only the linear cognate; ‘si-circlin’ binds to an exonic region common to the circular and linear transcripts, thereby knocking down both (Fig. 4a). DsiRNAs were transfected into primary mouse white pre-adipocytes and their knockdown efficiencies were assessed before and after differentiation (Fig. 4c and Supplementary Fig. 4a,b). Protein levels of TSHZ2 decreased correspondingly following si-circlin and si-lin transfection, but not with si-circ (Supplementary Fig. 4c).
To evaluate the effect of circRNA depletion on adipogenesis, we visualized the accumulation of lipid droplets (a hallmark of mature adipocytes) by imaging with the lipophilic fluorophore BODIPY 493/503 and Oil Red O (ORO) staining (Fig. 4d). Knockdown of both circular and linear transcripts by si-circlin abrogated lipid accumulation, as did blocking of circTshz2-1 alone by si-circ; whereas knockdown of linear mRNA alone by si-lin had no such inhibitive effect. Consistently, RT–qPCR analysis revealed that si-circ and si-circlin, but not si-lin, reduced the expression of adipogenic markers Pparγ, AdipoQ, C/ebpα and Fabp4 (Fig. 4e). It is notable that a low-abundance circular isoform, circTshz2-2, is derived from the full-length exon 2 of Tshz2 (Fig. 2c and Supplementary Fig. 5a). However, expression of circTshz2-2 was not affected by circTshz2-1 knockdown (Supplementary Fig. 5b,c). Thus, the anti-differentiation phenotype is specific to circTshz2-1 silencing.
We applied the same RNAi-based strategy to characterize the impact of another candidate, circArhgap5-2, on adipocyte differentiation. The Arhgap5 transcript encodes the regulatory Rho GTPase activating protein 5 (Arhgap5), whose entire first and second exons are incorporated into circArhgap5-2 via backsplicing at canonical splice sites. The circRNA may theoretically include or exclude the intervening intron sequence (Fig. 5a). To deduce the structure, we designed two sets of primers to amplify the exon-only and intron-retaining isoforms specifically. To detect the exon-only structure, we performed RT–PCR with divergent primers corresponding to the 5′ (reverse primer R) and 3′ (forward primers F1, F2, F3) distal ends of the upstream exon, which would amplify the entire downstream exon on circularization (Fig. 5a). Similarly, to observe the intron-retaining isoform, we paired reverse primer R with another set of forward primers (F4, F5, F6) complementary to the intronic sequences. While the exon-only primers amplified PCR products of the expected size, the intronic primers resulted in no detectable bands, thus confirming that circArhgap5-2 exists as an exclusively exonic 4,037-bp structure (Fig. 5b).
After induction of differentiation, levels of circArhgap5-2 increase by 33-fold over 6 d, while mRNA levels remain stable (Fig. 5c). Similar to circTshz2-1, we used DsiRNAs to specifically knockdown circArhgap5-2, its linear cognate and both transcripts in primary adipocyte culture before and after differentiation. We achieved high knockdown specificity of all target transcripts for each DsiRNA (Fig. 5d and Supplementary Fig. 6a) and observed consistent knockdown effects on protein levels (Supplementary Fig. 6b). Silencing of circArhgap5-2 by si-circ and si-circlin resulted in overt inhibition of lipid droplet accumulation (Fig. 5e) and significant downregulation of adipogenic markers (Fig. 5f), whereas knockdown of the linear transcript alone by si-lin did not suppress differentiation into mature adipocytes. Taken together, our in vitro results provide evidence for the functional role of our circRNA candidates in the regulation of adipogenesis.
During RT–PCR analysis for circArhgap5-2 (Fig. 5b), we observed a shorter isoform of circRNA, circArhgap5-1, which was formed by backsplicing of exon 1 (Supplementary Fig. 7a). We questioned whether circArhgap5-2 knockdown might affect the expression of circArhgap5-1. Using isoform-specific primers, we demonstrated that siRNA-mediated knockdown of circArhgap5-2 did not significantly affect circArhgap5-1 (Supplementary Fig. 7b). Conversely, silencing of circArhgap5-1 with an isoform-specific DsiRNA did not influence the expression of circArhgap5-2 (Supplementary Fig. 7c) and resulted in only mild inhibition of adipogenic markers (Supplementary Fig. 7d). Therefore, the blocking of adipogenesis observed in Fig. 5 is mainly due to the loss of circArhgap5-2.
To examine which adipogenesis stage is affected by circArhgap5-2 and circTshz2-1 knockdown, we transfected DsiRNAs at pre-adipocyte stage and then collected different time points during cell differentiation. Knockdown did not affect adipogenic markers at the pre-adipocyte stage and early adipogenesis stage (until D1). The cells began to manifest a clear defect in adipogenic marker expression only at D2 (Figs. 4f and 5g). Therefore, these circRNAs begin to support adipogenesis in the middle stage of differentiation.
circArhgap5-2 maintains the global adipocyte gene programme
Now we wanted to assess whether circArhgap5-2 plays a crucial role in maintaining the adipogenic gene programme in mature adipocytes. We infected differentiated adipocytes (D4) with an adenoviral shRNA targeting the BSJ of circArhgap5-2 (‘sh-circ’) (Fig. 6a), and achieved >70% knockdown efficiency 2 d post-transfection compared with the negative control (‘sh-NC’). No non-specific knockdown of the linear isoform was observed (Fig. 6b). We then performed poly(A) RNA-seq for the adipocyte samples, followed by gene-set enrichment analysis (GSEA) analysis. A rank-ordered list (by log2FC, where FC is fold change) of differentially expressed genes between sh-circ and sh-NC (Supplementary Table 3) was used as input into a pre-ranked GSEA run querying for MSigDB hallmark gene sets. Fatty acid metabolism and adipogenesis emerged as the most downregulated functional pathways (normalized enrichment score (NES) = −3.85 and −3.75, respectively) (Fig. 6c). Thus, depletion of circArhgap5-2 in mature adipocytes results in global suppression of the adipocyte transcriptional programme. Key adipocyte markers (Pparγ, AdipoQ, C/ebpα, Fabp4, Glut4, Lipe) correspondingly decreased at both the transcript and protein levels on sh-circ knockdown (Fig. 6b,d), although a visible impact on lipid accumulation was not detected (Fig. 6e). Therefore, circArhgap5-2 is required for mature adipocytes to maintain their gene expression programme in vitro.
a, A diagram of experimental design. Culture adipocytes were infected by adenoviral shRNA at day 4 and collected for downstream analysis at day 6. b, In vitro knockdown efficiency and specificity of adenoviral circArhgap5-2 sh-circ (relative to negative control sh-NC) in mature primary white adipocytes (left). Expression levels of adipogenic markers Pparγ, AdipoQ, Cebpα, Fabp4, Glut4 and Lipe in mature primary adipocytes where circArhgap5-2 is silenced (right). RT–qPCR data presented as mean ± s.e.m., n = 3 biologically independent experiments, *P < 0.05, **P < 0.01, two-tailed Student’s t-test. c, GSEA performed on RNA-seq data from control and circArhgap5-2 knockdown samples. Enrichment plot for genes involved in fatty acid metabolism and adipogenesis shown. n = 1 for each knockdown experiment, P < 0.01 using Kolmogorov–Smirnov test. d, Western blot showing protein levels of adipogenic genes Pparγ and Fabp4 on circArhgap5-2 knockdown (48 h post-infection). β-Actin as loading control. Repeated twice independently with similar results. e, BODIPY imaging and ORO staining to examine effect of circArhgap5-2 knockdown on lipid droplet accumulation. Nuclear imaging with Hoechst 33342. Scale bar: 100 µM. Repeated twice independently with similar results.
Next, we examined whether circArhgap5-2 maintains the adipocyte gene programme in vivo. We injected our adenoviral shRNAs into the inguinal fat pads of 8-week-old C57BL/6 mice and observed >70% knockdown of the circular isoform in all biological replicates 7 d post-infection (Fig. 7a). Triplicate control and sh-circ adipose tissue samples were subjected to mRNA-seq, followed by differential gene expression analysis and DAVID pathway analysis (Supplementary Table 4a). In line with our previous in vitro results, many of the top downregulated genes were involved in lipogenesis (Fig. 7b) and the most downregulated gene ontology (GO) terms were related to key metabolic and anabolic functions in adipocytes (Fig. 7c and Supplementary Fig. 8 and Supplementary Table 4b). The most upregulated genes and GO terms were related to inflammation pathways (Fig. 7b,c and Supplementary Fig. 8). We further confirmed the decreased expression of adipocyte makers such as Fabp4 and Pparγ and the induction of inflammatory markers such as F4/80, Itgam and Il6 at the transcript level by RT–qPCR (Fig. 7d,e). Fabp4 and Pparγ protein were also reduced, as examined by western blotting (Fig. 7f). Histological staining of sh-circ-infected adipose tissue showed decreased lipid droplet size, reflecting disrupted lipid metabolism (Fig. 7g). Altogether, our findings confirm that circArhgap5-2 is an indispensable regulator for maintaining the key features of mature adipocytes in vivo.
a, In vivo knockdown efficiency of circArhgap5-2 sh-circ (relative to negative control sh-NC) in inguinal adipose tissue of mice injected with shRNAs (measured 7 d post-infection). n = 9 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test, paired. b, Volcano plot showing the log fold change and P value of differentially expressed genes (coloured in red). n = 3 biologically independent experiments. c, Most over-represented GO biological processes in genes regulated on in vivo circArhgap5-2 knockdown, as identified by functional enrichment analysis P < 0.05, using Mann–Whitney U-test (Wilcoxon rank-sum test). Genes were selected according to the criteria FPKM fold change ≤−2, P ≤ 0.05 and FDR ≤ 0.05. d, Expression levels of adipogenic markers Pparγ, AdipoQ, Cebpα and Fabp4 on in vivo circArhgap5-2 knockdown. n = 9 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test, paired. e, Expression levels of inflammation markers, Itgam, Ccl2, Il6 and F4/80, on in vivo circArhgap5-2 knockdown. n = 9 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test, paired. f, Western blot showing protein levels of adipogenic genes Fabp4 and Pparγ on circArhgap5-2 knockdown (7 d post-injection). β-Actin as loading control. Repeated twice independently with similar results. g, Haematoxylin and eosin staining of inguinal adipose tissue sections injected with sh-NC or sh-circ. Relative cell size (cross-sectional area) of adipocytes in each condition presented as arbitrary units (A.U.), as quantified by ImageJ. Mean as measure of centre, n = 7 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test, paired.
circARHGAP5-1 function is conserved in human adipocytes
circArhgap5-2 exists as a homologous circular transcript (annotated ‘circARHGAP5-1’), identified by our pipeline as the 13th most highly expressed conserved isoform in human fat (Supplementary Table 1e). The BSJ of circARHGAP5-1 arises from backsplicing of the second and third exons of the canonical pre-mRNA (Fig. 8a), which align with 93.5% sequence identity to the same two exons that circularize in the mouse homologue. Correspondingly, the BSJ is also highly conserved (94% identity across 50 bp) (Fig. 8b). In contrast to our observations in mice, circARHGAP5-1 does not change significantly in expression during the differentiation of human white adipocytes (Supplementary Fig. 9a). We wanted to determine whether its essential role in adipogenesis remains conserved. Using a lentiviral shRNA targeting the BSJ of circARHGAP5-1 (‘hsa-sh-circ’) we attained >70% specific knockdown of the circRNA in immortalized human white fat progenitor cells, both at the pre-adipocyte stage and 12 d after inducing differentiation (Fig. 8c and Supplementary Fig. 9b). Remarkably, we confirmed that adipogenesis was drastically suppressed by circRNA depletion, as observed in the downregulation of adipose markers (PPARγ, FABP4, GLUT4, CIDEA) (Fig. 8d) and inhibited lipid accumulation (Fig. 8e). Therefore, the pro-adipogenic function of circARHGAP5-1 is conserved between mouse and human adipocytes.
a, Schematic of human circRNA homologue produced from the ARHGAP5 locus of chromosome 14. Backsplicing occurs between exon 2 and 3, producing circARHGAP5-1. b, Alignment of BSJ sequences of mouse circArhgap5-2 and human circARHGAP5-1 (25 bp flanking each side of junction). c, Knockdown efficiency and specificity of lentiviral circARHGAP5-1 hsa-sh-circ (relative to negative control hsa-sh-NC) in differentiated human white adipocytes at D12 (13 d post-infection). RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, two-tailed Student’s t-test. d, Expression levels of adipogenic markers PPARγ, FABP4, GLUT4 and CIDEA in differentiated human adipocytes where circARHGAP5-1 is silenced. RT–qPCR data presented as mean ± s.e.m., n = 4 biologically independent experiments, *P < 0.05, **P < 0.01, two-tailed Student’s t-test. e, BODIPY imaging and ORO staining to examine effect of circARHGAP5-1 knockdown on lipid droplet accumulation. Nuclear imaging with Hoechst 33342. Scale bar: 50 µM. Repeated twice independently with similar results.
circArhgap5-2 is not mechanistically a miRNA sponge
As described extensively in previous studies19,30,31,32, circRNAs may act as ‘sponges’ that competitively bind and sequester miRNAs, thus preventing base-pairing with their mRNA targets and subsequently disrupting translational repression and RNA-induced silencing complex (RISC)-mediated degradation. As the sequence of circArhgap5-2 potentially contains multiple miRNA-binding sites, we hypothesized that it might exert its effects by sponging anti-adipogenic miRNAs.
If a circRNA sponges miRNAs, circRNA depletion would release the sequestered miRNAs and allow them to interact with their cytoplasmic mRNA targets, thereby affecting target expression levels. Sharp’s laboratory demonstrated that overexpression of a constructed sponge can de-repress miRNA targets as strongly as chemically modified antisense oligonucleotides50. Rajewsky’s group also demonstrated that destruction of CDR1as could trigger downregulation of miR-7’s targets32. Similarly, if circArhgap5-2 deficiency cannot significantly influence expression of the mRNA targets of its putatively binding miRNAs, it should not be considered an miRNA sponge. Using the TargetScanMouse database (http://www.targetscan.org/mmu_71/), we identified 43 different conserved miRNAs with 66 matching seed sites along the circArhgap5-2 sequence, and the corresponding mRNA targets for each of these 43 miRNAs. With our mRNA-seq dataset of circArhgap5-2 knockdown in adipose tissue (Supplementary Table 4a), we plotted the cumulative distribution of fold changes in expression of all target mRNAs for each miRNA (Supplementary Table 5a), and measured its shift compared to the reference distribution of all other transcripts. As a positive control, we applied our pipeline to a published dataset51 for HeLa cells transfected with miR-155 (GEO accession GSE22004), and observed the expected leftward shift in the cumulative distribution curve of its mRNA targets (Kolmogorov–Smirnov test, P < 0.001) (Supplementary Fig. 10a), demonstrating the repressive effect of miR-155 on these mRNAs. However, depletion of circArhgap5-2 did not cause any significant shift in distribution of mRNA targets for all putatively binding miRNAs (Supplementary Fig. 10b and Supplementary Table 5b). A similar analysis of our in vitro knockdown dataset from mature adipocytes yielded the same negative results (Supplementary Fig. 10c).
Furthermore, proving that a circRNA functions as a miRNA sponge involves demonstrating the binding interaction between the circRNA and Argonaute-2 (AGO2), a core protein of the RISC19. We conducted RNA immunoprecipitation with AGO2 antibody (Supplementary Fig. 11a), followed by RT–qPCR to examine whether AGO2 could pull down the circArhgap5-2 transcript. Although AGO2 could retrieve several detected miRNAs as expected (Supplementary Fig. 11b), it did not retrieve circArhgap5-2 (Supplementary Fig. 11c), suggesting no interaction between the two. Taken together, we conclude that circArhgap5-2 is unlikely to exert its regulatory effects on adipogenesis by sponging miRNAs.
circArhgap5-2 does not encode peptides
After precluding the miRNA sponging mechanism, we explored whether circArhgap5-2 might encode novel functional peptides as reported in earlier circRNA studies34,35,36. circArhgap5-2 contains a putative open reading frame that can be potentially translated at the same start codon as the canonical protein sequence. We conducted polysome profiling in primary cultured adipocytes to examine whether circArhgap5-2 and circTshz2-1 were associated with polysomes (Supplementary Fig. 12a). While Srebp1 was exclusively found in polysome fractions (Supplementary Fig. 12d), neither circArhgap5-2 nor circTshz2-1 were similarly detected (Supplementary Fig. 12b,c), thus indicating that these circRNAs are not translated.
Discussion
In recent years, circRNA biology has been a hotbed for intensive research. Mounting evidence now points to the potential for circRNAs as novel regulators in a wide range of physiological systems and disease contexts. There have been few studies implicating circRNAs directly in metabolic disorders, and these have been limited to characterizing the pancreatic and endothelial roles of single candidates in models of diabetes42,43. Our present research extends the scope of disease-relevant functions for circRNAs to obesity, which is marked by excess adiposity and disruption of metabolic homeostasis in adipose tissue. We established a comprehensive genome-wide analysis of circular transcripts in visceral and subcutaneous fat, and found circRNAs transcribed from thousands of loci across the human and mouse genomes. Many of these novel candidates have not been previously identified in publicly available RNA-seq datasets.
On the basis of conservation, adipose specificity and abundance, we selected a set of candidates for further validation. These circRNAs were mostly upregulated during adipocyte maturation. This trend concurs with previous studies on myocyte and neuronal differentiation29,34, which suggests that enhanced circRNA biogenesis could be a general feature of cell differentiation. Furthermore, this pattern possibly reflects the low turnover kinetics of circRNAs, which may drive their passive accumulation in differentiating cells.
Circular transcripts also exhibit a distinct signature of downregulation in adipose tissue of HFD mice, possibly due to extensive remodelling of adipocytes under conditions of obesity and chronic inflammation. In this context of metabolic stress, increased expression of TNF-α and other proinflammatory cytokines in WAT exerts a wide range of downstream effects, such as induction of lipolysis and suppression of lipogenesis, insulin signalling, mitochondrial function and adipogenesis47. In addition, inflammation may affect the activity of RNA-binding proteins and RNA-editing enzymes that regulate host gene transcription, circularization and backsplicing of exons. Distinct patterns of modulation for circRNAs have also been observed in other metabolic organs. For instance, it has been reported that Duchenne muscular dystrophy is marked by inhibited myocyte differentiation, which is accompanied by a unique signature of circRNA expression changes34.
As multiple circRNAs can be generated from the same host gene, we wanted to explore whether different isoforms from the same gene are regulated in a similar manner. We examined the expression of circArhgap5-1 and circArhgap5-2 across different organs and observed their distinct tissue distribution, and also found a similar discordance between circTshz2-1 and circTshz2-2 (Supplementary Fig. 13). Therefore, the biogenesis of each circRNA, even derived from the same locus, is independently regulated.
As the majority of circRNAs arise from protein-coding exons, there are limitations in target region selection for specific knockdown without altering the linear isoform. Nonetheless, we successfully utilized RNAi to identify circTshz2-1 and circArhgap5-2 as crucial regulators of adipogenesis in vitro. Remarkably, we demonstrated the in vivo role of circArhgap5-2 in driving the global transcriptional programme that supports core adipogenic pathways such as lipid metabolism and biosynthesis. Critically, this function is conserved in human adipocytes. Therefore, our present study provides strong evidence that circRNAs, such as miRNAs and lncRNAs, are a novel class of regulators in adipocyte biology and are potential therapeutic targets for obesity-related metabolic diseases.
It is of great interest to determine the mechanism by which circArhgap5-2 supports adipogenesis. Our analysis shows that circArhgap5-2 neither sponges miRNAs nor encodes novel peptides to exert its downstream effects. Furthermore, as circArhgap5-2 does not localize to the nucleus, it is unlikely that it directly influences transcription. Previous reports have demonstrated that circRNAs interact with RNA-binding proteins to alter their function (for example, circ-Foxo3 binding to CDK2 and p21)33 or competitively bind them to affect the translation and stability of target mRNA transcripts (for example, circANRIL and PES1, circPABPN1 and HuR)52,53. It is plausible then that circArhgap5-2 modulates adipogenesis in a similar fashion. As the molecular roles of circRNAs are still poorly understood, we cannot preclude that circArhgap5-2 may act through other unknown mechanisms. More detailed mechanistic studies are thus warranted to elucidate how circArhgap5-2 regulates adipogenesis.
Methods
Preparation and analysis of RNA-seq libraries
Total RNA from human visceral (omental) and subcutaneous adipose, as well as mouse epididymal and inguinal adipose, was extracted using Trizol (Life Technologies) and RNeasy kit (Qiagen). Three replicates from the same tissue type were pooled into one tissue type sample (four sample types in total). Total RNAs were purified with Ribo-Zero kit for library preparation, and sequenced on the Illumina HiSeq2000 platform. Paired-end reads were aligned to hg19 and mm9 genomes using Tophat2 (v.2.2.0.12) with default parameters. Gene expression analysis was performed using Cufflinks2 (v.2.2.1), with gene expression levels reported in FPKM.
For in vitro circArhgap5-2 knockdown experiments, total RNA extracted from cultured mature white adipocytes infected with adenoviral shRNAs (control ‘sh-NC’ or ‘sh-circ’ targeting the circRNA) were processed using NEBNext Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs). For in vivo injection experiments, total RNA from inguinal adipose samples injected with sh-NC or sh-circ (triplicate samples for each shRNA) were similarly processed. Reads were mapped using Tophat2 and differential gene expression analysis was performed with edgeR.
Detection, annotation and quantification of circRNAs
RNA-seq data were analysed using CIRI2 (ref. 45), an efficient and unbiased de novo circRNA discovery pipeline. This algorithm analyses the SAM file generated by Burrows–Wheeler Aligner (BWA–MEM) and scans for potential BSJ reads consisting of two segments aligning to the reference genome in chiastic order. Candidate BSJ reads must align in the same chromosome and strand. CIRI2 utilizes paired-end mapping information to exclude false-positive candidates whose paired reads do not map within the region of the putative circRNA range. Candidate junction reads not flanked by canonical GT-AG splicing signals or supported by GENCODE v.19 exon boundaries are also filtered out. CircRNA candidates mapped by at least two unique BSJ reads were included in our analysis and were annotated to hg19 and mm9 GENCODE v.19.
Relative expression abundance of circRNA isoforms is reported in SRPBM (spliced reads per billion mapped reads):
where x is the total number of BSJ reads and n is the total number of mapped reads.
The BSR was calculated to quantify the relative expression of circRNA versus linear cognate mRNA produced from the same exon.
where c denotes the total BSJ read count that spans both ends of the backspliced exon(s), l1 and l2 denote the total linear read counts that span left and right linear-spliced junctions of the same exon(s), respectively.
Adipose specificity and conservation analysis of circRNAs
To identify adipose-specific circRNAs in silico, we applied the CIRI2 pipeline to ribosome-depleted RNA-seq datasets of various human tissues downloaded from the ENCODE Project Consortium, and compared the circRNA output with our own dataset of adipose candidates. ENCODE experimental accessions: ENCSR000AFB (liver), ENCSR000AFF (skeletal muscle tissue), ENCLB223ZZZ (heart), ENCSR000AFC (lung), ENCSR344MQK (gonad), ENCSR586SYA (pancreas), ENCSR900SGE (spleen).
To identify conserved circRNAs, 25-bp DNA sequences flanking each side of the BSJ for each circRNA were extracted from human and mouse, and the 50-bp BSJ sequences were compared using UCSC BLAT (https://genome.ucsc.edu/cgi-bin/hgBlat). CircRNAs originating from homologous genes and crossing a threshold of ≥70% sequence identity between the human and mouse BSJ sequences were considered conserved.
Pathway analysis
To perform GSEA on RNA-seq datasets from cultured adipocytes, all detected transcripts were rank-ordered according to their differential expression between control and circArhgap5-2 knockdown conditions (log2FC of sh-circ/sh-NC FPKM). The list was used as input in a pre-ranked GSEA run querying for MSigDB hallmark gene sets, under default parameters. Normalized enrichment scores were generated to identify the most enriched pathways (nominal P ≤ 0.05).
To perform GO enrichment analysis on in vivo RNA-seq datasets, all detected genes we filtered for the most highly downregulated candidates on circArhgap5-2 knockdown (log2FC ≤ −1 relative to control, P ≤ 0.05, FDR ≤ 0.05). Over-represented GO biological processes in this shortlist were quantified with the DAVID v.6.8 functional annotation tool (https://david.ncifcrf.gov/), using default parameters.
Computational prediction of miRNA sponging function
Conserved miRNAs with putative seed sequences matching the circArhgap5-2 sequence, along with their predicted mRNA targets, were identified using the TargetScanMouse database (http://www.targetscan.org/mmu_71/). Using mRNA-seq data for in vivo circArhgap5-2 knockdown, the cumulative distribution of log2FC in FPKM of all target mRNAs was plotted for each of the 66 miRNAs identified. The shift of this distribution was compared to the same curve plotted for all other detected transcripts (significance was determined using the Kolmogorov–Smirnov test, FDR < 0.05). As a positive control, the pipeline was applied to a published dataset for HeLa cells transfected with miR-155 (GEO accession GSE22004).
Candidate selection and validation
Exonic circRNAs were filtered for further validation and characterization from the subset of candidates exhibiting BSJ sequence conservation between human and mouse. The following filters were subsequently applied to the conserved subset: (1) top 20 highest abundance (based on raw junction count/SRPBM) in both human omental and subcutaneous fat; and/or (2) top 20 highest abundance in both mouse epididymal and inguinal fat; and/or (3) top 15 highest abundance among circRNAs predicted to be enriched in human adipose tissue on the basis of our comparison with ENCODE datasets. CircRNAs with a BSJ length <90 bp were excluded due to RT–PCR detection limits, resulting in a final shortlist of 41 candidates.
To validate circular candidates, divergent PCR primers were used on cDNA templates from inguinal mouse adipose tissue to amplify the putative BSJ. After electrophoresis in a 1% agarose gel to confirm their expected size, bands were gel-extracted using NucleoSpin PCR clean-up kit (Machery-Nagel) and had their predicted sequences confirmed by Sanger capillary sequencing (1st BASE, Singapore).
RNase R resistance assay
Large ribosomal RNAs were removed from total RNA extracted from mouse inguinal adipose (10 μg for each treatment condition) using RiboMinus Transcriptome Isolation Kit (ThermoFisher). Equal amounts (~1 μg) of ribosome-depleted RNA were subjected to nucleolytic digestion (incubated with 3 U RNase R per μg of RNA for 1 h at 37 °C, as per the manufacturer’s instructions (Epicentre)) or control mock treatment (incubation with no RNase R). Remaining RNA was then recovered immediately by standard phenol/chloroform/isoamylic alcohol extraction. cDNA was synthesized from proportional amounts of RNA relative to the input for each treatment. RT–PCR with the respective primers was performed to compare resistance of circular and linear transcripts to digestion.
Subcellular fractionation
Mouse inguinal fat pads (in three biological replicates of at least 100 mg each) were collected, immersed in 1 ml per 0.1 g of tissue in ice-cold lysis buffer (10 mM HEPES pH 7.4, 1.5 mM MgCl2, 100 mM KCl, 1% NP-40 (Sigma), 10 mM ribonucleoside vanadyl complex (NEB)) and homogenized with a micromincer. Homogenates were passed through a 100-μM filter and centrifuged at 3,000g for 10 min. The cytosolic supernatant and nuclear pellet were collected separately. The pellet was washed with lysis buffer and re-centrifuged to eliminate cytosolic contamination. Total RNA was extracted from each subcellular fraction with Trizol/RNAeasy kit and converted to cDNA. Relative expression of circRNAs in each compartment was analysed by RT–qPCR comparison of nuclear and cytosolic Ct values, where ∆Ct = n is equivalent to a 2n fold change in expression. Expression of 47S pre-rRNA and maturely spliced Gapdh transcripts were measured as nuclear and cytosolic controls, respectively.
Animal protocol and diet-induced obesity model
All mice were bred under specific pathogen-free conditions in the animal vivarium at Duke-NUS Medical School. All animal experimental protocols were approved by the Singapore SingHealth Research Facilities Institutional Animal Care and Use Committee.
For circRNA identification, validation, tissue specificity and cellular localization experiments, epididymal and inguinal adipose and other organs/tissues were collected from adult 12-week-old C57BL/6 male mice. For the diet-induced obesity model, mice were fed either regular chow diet or HFD (Research Diets, D12492) for 18 weeks after weaning at 3 weeks of age. At least n = 10 mice for each diet condition were used for all experiments.
Sample size choice, randomization and blinding
For all the animal experiments, sample sizes were predetermined on the basis of effect size, standard deviation and significance level required to attain statistical significance of P < 0.05 with a 90% probability on the basis of previous experiments using similar methodologies. The selected sample size should be sufficient to account for any biological/technical variability. Mice for each condition were randomly assigned to treatment groups. Samples collected from animals were processed in random order and researchers were blinded to sample groups.
SVF isolation, primary mouse adipocyte culture, differentiation and drug treatment
Inguinal white fat pads from 3-week-old C56BL/6 mice were isolated, pooled, minced and incubated in a digestion solution (0.2% collagenase (Sigma), 2% BSA dissolved in Hank’s balanced salt solution (Gibco)) at 37 °C for 30 min at 500 r.p.m. The suspension was filtered through a 100-μm cell strainer and centrifuged at 2,000 r.p.m. for 5 min. The pelleted SVF containing pre-adipocytes was resuspended and cultured in DMEM (Gibco) supplemented with 5% newborn calf serum, 5% FBS, 100 units ml−1 penicillin, 100 μg ml−1 streptomycin and 10 μg ml−1 gentamicin (Invitrogen). To induce differentiation, pre-adipocytes were grown to confluency and treated for 48 h with a standard differentiation cocktail of 0.5 μM dexamethasone, 850 nM insulin, 0.25 mM 3-isobutyl-1-methyxanthine (IBMX) and 1 μM rosiglitazone in DMEM supplemented with 10% FBS.
For the differentiation time course, cells were collected at time points day 0 (D0), D2, D4 and D6. For the DsiRNA knockdown experiments in differentiated adipocytes, adipogenesis was induced 24 h post-transfection of the DsiRNAs, and cells were collected at D3 (96 h post-transfection). For experiments on TNF-α-induced inflammation, primary white adipocytes were treated for 24 h with 5 nM of soluble TNF-α (Sigma) and compared to mock-treated controls. All the experiments were repeated at least two times with 3–4 replicates each time.
Human cell line maintenance and differentiation
Immortalized clonal cell lines of white fat progenitors (a gift from Y.-H. Tseng, Joslin Diabetes Center) were derived from human subcutaneous adipose tissue as previously described54. Human cells were maintained in DMEM supplemented with 10% FBS. To chemically induce adipogenesis, lentivirus-infected pre-adipocytes were grown to confluency and incubated with a differentiation cocktail (33 μM biotin, 0.5 μM insulin, 17 μM pantothenate, 0.1 μM dexamethasone, 2 nM 3,3′,5-triiodo-l-thyronine (T3), 500 μM IBMX, 30 μM indomethacin, 1 μM rosiglitazone) for 14 d. For differentiation time course, cells were collected at D0, D7, D10 and D14.
DsiRNA-mediated knockdown
Customized DsiRNAs (Integrated DNA Technologies) were designed to specifically target the circular and cognate linear isoforms of circTshz2-1 and circArhgap5-2: (1) ‘si-circ’ binds only to the head-to-tail junction of the circRNA; (2) ‘si-lin’ is complementary to a 3′ UTR sequence not spliced into the circRNA and targets only the linear mRNA; (3) ‘si-circlin’ binds to an exonic region common to the circular and linear transcripts, thus silencing both. DsiRNA specific to circArhgap5-1 was similary designed. Negative control DsiRNA (DS NC1, Integrated DNA Technologies) was also used. Primary white pre-adipocytes were seeded onto 24-well plates and transfected with 100 nM DsiRNAs using 5 μl ml−1 Lipofectamine RNAiMAX (ThermoFisher) mixed in Opti-MEM (Gibco). Knockdown efficiency was assessed 96 h post-transfection by RT–qPCR in both pre-adipocytes and differentiated D3 adipocytes (which were induced to differentiate 24 h post-transfection). All knockdown data are representative of four independent replicates.
Adenovirus generation, infection and injection
Adenoviruses were produced by the co-transfection of pDC316-ZsGreen-shRNA shuttle plasmids containing either sh-NC (negative control) or sh-circ (targeting the BSJ of circArhgap5-2) shRNA sequences and the adenoviral packaging plasmid pBHG into Q293A cells. Culture media (DMEM + 10% FBS) was changed to fresh media 16 h post-transfection, and collected on virus-induced cell lysis 10 d later. Viruses were amplified in Q293A cells grown in 100-mm dishes, with virus seed collected 48 h post-infection. Seed was used to infect Q293A cells grown in 150-mm dishes, and crude viral lysate was collected 72 h post-infection on cell detachment. Adenoviruses were purified using CsCl density gradient centrifugation. Briefly, cell lysates were subject to three freeze–thaw cycles and 8 ml was loaded on top of a CsCl gradient (2.1 mL of light solution (0.29 g/mL CsCl in 10 mM Tris-HCl pH 7.9), 2.8 mL of heavy solution (0.61 g/mL CsCl in 10 mM Tris-HCl pH 7.9)) in an ultracentrifuge tube (Beckman, 344059). Samples were centrifuged in an SW 41 Ti rotor (Beckman) at 32,000 r.p.m. and 4 °C for 90 min. Viral bands were collected and desalted using a CL-4B Sepharose column (Sigma, CL4B200) and eluted in storage buffer (10 mM Tris-HCl pH 7.9, 2 mM MgCl2, 4% sucrose). Purified adenoviruses were titred with Adeno-X Rapid Titre Kit (Clontech)
For in vitro circArhgap5-2 knockdown, primary D4 white adipocytes grown in 24-well plates were infected with purified adenoviruses (2 × 1013 virus particles (vp) per well) and collected 48 h post-infection (D6). Data are representative of three independent replicates. For in vivo knockdown, adenoviruses were injected directly into adipose tissue of 8-week-old C57BL/6 male mice at 100 μl per subcutaneous inguinal depot (2 × 1012 vp ml−1), with sh-NC injected into the left side and sh-circ into the right. Seven days post-injection, inguinal tissue was excised to assess the knockdown phenotype. Data are representative of nine biological replicates.
Lentivirus generation and infection
Lentiviruses were generated by co-transfection of HEK293T cells with pLKO.1 constructs containing either the hsa-sh-NC (control) or hsa-sh-circ (targeting circARHGAP5-1) shRNA sequences, and the packaging vectors pMD2.G and psPAX2. Culture media was changed 16 h post-transfection, and viruses were collected 48 h post-transfection. Immortalized human white pre-adipocytes were infected with viruses (mixed with 10 μg ml−1 polybrene) at 80% confluency, followed by addition of differentiation cocktail after 24 h. Mature adipocytes were collected at D12 of differentiation. All knockdown data are representative of four independent replicates.
Plasmid construction
All plasmids were constructed using standard methods. Adenoviral shRNA oligonucleotides sh-NC (control) and circArhgap5-2 sh-circ were designed using Invitrogen Block-iT RNAi Designer and cloned between the PstI and BamHI restriction sites of pDC316-ZsGreen-shRNA vector. Similarly, lentiviral shRNA oligonucleotides hsa-sh-NC (control) and circARHGAP5-1 hsa-sh-circ were inserted between the AgeI and EcoRI sites of the pLKO.1 vector.
cDNA synthesis, RT–PCR and RT–qPCR analysis
Total RNA was extracted from all tissue/cell culture samples using Trizol (Life Technologies) and RNAeasy (Qiagen). cDNA was synthesized with random primers using M-MLV (Promega). For circRNA detection by RT–PCR and RT–qPCR, a set of divergent primers that amplify the head-to-tail junction (but not the linearly spliced isoform) only on backsplicing of the exons was designed, along with convergent, linear-specific primers. Standard 30-cycle PCR of mouse tissue or plasma cDNA was performed with Q5 High-Fidelity DNA polymerase (New England Biolabs) and amplified products were visualized on 1% agarose gel stained with SYBR Safe (Life Technologies). For RT–qPCR, a similar set of primers designed to have target-specific amplicons of 100–150 bp was used to detect transcript levels of circular and linear isoforms. RT–qPCR was performed with SYBR Green Master Mix (ThermoFisher) on the Applied Biosystems ViiA 7 system. Statistical analyses for differential gene expression were based on Student’s t-test, with P < 0.05 considered significant. All RT–qPCR data were normalized to expression of the housekeeping gene Rpl23 (for mouse genes) or RPL32 (for human genes).
Fluorescence microscopy and ORO staining
For fluorescent staining of lipid droplet formation, live differentiated adipocytes were incubated with 2 μM (1:1000 dilution) BODIPY 493/500 (ThermoFisher) in DMEM for 15 min at 37 °C, while nuclei were co-stained with Hoechst 33342 (1:500 dilution) (ThermoFisher). Multichannel cell images were captured with a Leica Microsystems DMi8 Microscope and processed with Leica Application Suite X.
For ORO staining, 0.5 g ORO (Sigma) was dissolved in 100 ml of isopropanol, mixed with water at a ratio of 6:4 and filtered with Whatman no. 1 filter paper. Differentiated adipocytes were washed with PBS and fixed with formalin overnight at 4 °C. Cells were stained with freshly prepared ORO solution for 1 h at room temperature, and then washed with water and air-dried before imaging.
Western blot
Adipocytes and adipose tissue were collected and lysed in RIPA buffer (50 mM Tris-HCl pH 7.4, 1% Triton X-100, 150 mM NaCl, 1 mM EDTA, 1 mM PMSF, protease inhibitor cocktail). Protein samples were resolved by SDS–PAGE on a 4–10% acrylamide gel and transferred onto a polyvinylidenedifluoride membrane. Membranes were blocked with 4% BSA in 1× TBST for 1 h at room temperature, incubated with primary antibody overnight at 4 °C and incubated with horseradish peroxidase-conjugated secondary antibody for 1 h at room temperature. Bands were visualized with chemiluminescence substrates and the ChemiDoc Touch Imaging System (Bio-Rad). Primary antibodies used: Ago2 (Proteintech, 10686-1-AP, 95 kDa), Arhgap5 (Proteintech, 55165-1-AP, 170 kDa), β-actin (Sigma, A1978, 43 kDa), Fabp4 (Santa Cruz Biotechnology, C-15 sc-18661, 15 kDa), Gapdh (Abcam, ab9485, 37 kDa), Pparγ (Santa Cruz Biotechnology, H-100 sc-7196, 57 kDa), Tshz2 (Abcam, ab140189, 115 kDa).
Histology
Inguinal adipose tissue samples injected with adenoviral circArhgap5-2 sh-NC or sh-circ were fixed and processed according to the manufacturer’s recommended protocol (National Diagnostics, HistoClear). Haematoxylin and eosin staining was performed as follows: rehydrated sections were stained in haematoxylin solution for 20 min and washed, with excess dye removed with 70% ethanol containing 1% HCl. Sections were then stained in eosin solution for 10 min, washed, dehydrated, cleared and mounted.
Polysome profiling
Approximately 5 million cultured adipocytes at day 5 were lysed in 300 μl polysome lysis buffer (NaCl 150 mM, MgCl2 5 mM, Tris 20 mM pH 7.4, dithiothreitol (DTT) 1 mM, NP-40 1%, SUPERaseIn 0.2 unit μl−1, 100 μg ml−1 cycloheximide). Cells were passed twice through a 26-gauge needle and centrifuged at 20,000g for 10 min at 4 °C. The supernatant was loaded on top of a sucrose density gradient (10–50%) (Tris 20 mM pH 7.4, MgCl2 5 mM, NaCl 150 mM, DTT 1 mM) and then centrifuged at 37,000 r.p.m. in a SW 41 Ti rotor for 2 h at 4 °C. The gradient was then fractionated using a Gradient Station (BioComp) coupled with a UV monitor. The RNA from each fraction was isolated with Trizol.
RNA immunoprecipitation
Pre-adipocytes from SVF were isolated from inguinal white fat and differentiated to day 5. RNA immunoprecipitation was performed using the Magna RIP RNA-Binding Protein Immunoprecipitation Kit (Sigma). Briefly, ~2 × 106 adipocytes differentiated at day 5 were collected and lysed in the lysis buffer. Antibodies for Ago2 (Sigma, SAB4200085) and IgG control were conjugated to the magnetic beads by incubating for 30 min at room temperature. The conjugated beads were incubated with cell lysate for 4 h at 4 °C, followed by precipitation and three washes. Beads were resuspended in 100 μl washing buffer. A 10% volume of the resuspended beads was used for western blot to confirm the success of immunoprecipitation with antibodies against Ago2 (Proteintech, 10686-1-AP, 95 kDa) and Gapdh (Abcam, ab9485, 37 kDa). The rest of the complex was resuspended in Trizol for RNA extraction. The same proportion of RNA from each immunoprecipitation was used for RT–qPCR (Supplementary Fig. 11).
Reporting Summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Data availability
The RNA-seq data generated in this study to build a database of adipose circRNAs in human and mouse adipose tissue, and to assess the global transcriptional impact of circArhgap5-2 knockdown in vivo and in vitro, are available in the NCBI Gene Expression Omnibus (GEO) database (Accession no: GSE118762). All other data generated and/or analysed to support the findings of this study are available from the corresponding author upon reasonable request.
References
Ng, M. et al. Global, regional, and national prevalence of overweight and obesity in children and adults during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013. Lancet 384, 766–781 (2014).
Arner, P. & Kulyte, A. MicroRNA regulatory networks in human adipose tissue and obesity. Nat. Rev. Endocrinol. 11, 276–288 (2015).
Kim, H. J. et al. MicroRNAs are required for the feature maintenance and differentiation of brown adipocytes. Diabetes 63, 4045–4056 (2014).
Sun, L. et al. Long noncoding RNAs regulate adipogenesis. Proc. Natl Acad. Sci. USA 110, 3387–3392 (2013).
Knoll, M., Lodish, H. F. & Sun, L. Long non-coding RNAs as regulators of the endocrine system. Nat. Rev. Endocrinol. 11, 151–160 (2015).
Zhao, X. Y., Li, S., Wang, G. X., Yu, Q. & Lin, J. D. A long noncoding RNA transcriptional regulatory circuit drives thermogenic adipocyte differentiation. Mol. Cell 55, 372–382 (2014).
Alvarez-Dominguez, J. R. et al. De novo reconstruction of adipose tissue transcriptomes reveals long non-coding RNA regulators of brown adipocyte development. Cell Metab. 21, 764–776 (2015).
Yang, L. et al. Integrative transcriptome analyses of metabolic responses in mice define pivotal lncRNA metabolic regulators. Cell Metab. 24, 627–639 (2016).
Ding, C. et al. De novo reconstruction of human adipose transcriptome reveals conserved lncRNAs as regulators of brown adipogenesis. Nat. Commun. 9, 1329 (2018).
Bai, Z. et al. Dynamic transcriptome changes during adipose tissue energy expenditure reveal critical roles for long noncoding RNA regulators. PLoS Biol. 15, e2002176 (2017).
Lo, K. A. et al. Adipocyte long-noncoding RNA transcriptome analysis of obese mice identified lnc-leptin, which regulates leptin. Diabetes 67, 1045–1056 (2018).
Starke, S. et al. Exon circularization requires canonical splice signals. Cell Rep. 10, 103–111 (2015).
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 ONE 7, e30733 (2012).
Capel, B. et al. Circular transcripts of the testis-determining gene Sry in adult mouse testis. Cell 73, 1019–1030 (1993).
Cocquerelle, C., Daubersies, P., Majerus, M. A., Kerckaert, J. P. & Bailleul, B. Splicing with inverted order of exons occurs proximal to large introns. EMBO J. 11, 1095–1098 (1992).
Nigro, J. M. et al. Scrambled exons. Cell 64, 607–613 (1991).
Wang, P. L. et al. Circular RNA is expressed across the eukaryotic tree of life. PLoS ONE 9, e90859 (2014).
Salzman, J., Chen, R. E., Olsen, M. N., Wang, P. L. & Brown, P. O. Cell-type specific features of circular RNA expression. PLoS Genet. 9, e1003777 (2013).
Memczak, S. et al. Circular RNAs are a large class of animal RNAs with regulatory potency. Nature 495, 333–338 (2013).
Jeck, W. R. et al. Circular RNAs are abundant, conserved, and associated with ALU repeats. RNA 19, 141–157 (2013).
Chen, L. L. The biogenesis and emerging roles of circular RNAs. Nat. Rev. Mol. Cell Biol. 17, 205–211 (2016).
Li, Z. et al. Exon-intron circular RNAs regulate transcription in the nucleus. Nat. Struct. Mol. Biol. 22, 256–264 (2015).
Zhang, Y. et al. Circular intronic long noncoding RNAs. Mol. Cell 51, 792–806 (2013).
Zhang, Y. et al. The biogenesis of nascent circular RNAs. Cell Rep. 15, 611–624 (2016).
Ashwal-Fluss, R. et al. circRNA biogenesis competes with pre-mRNA splicing. Mol. Cell 56, 55–66 (2014).
Ivanov, A. et al. Analysis of intron sequences reveals hallmarks of circular RNA biogenesis in animals. Cell Rep. 10, 170–177 (2015).
Zhang, X. O. et al. Complementary sequence-mediated exon circularization. Cell 159, 134–147 (2014).
Kramer, M. C. et al. Combinatorial control of Drosophila circular RNA expression by intronic repeats, hnRNPs, and SR proteins. Genes Dev. 29, 2168–2182 (2015).
Rybak-Wolf, A. et al. Circular RNAs in the mammalian brain are highly abundant, conserved, and dynamically expressed. Mol. Cell 58, 870–885 (2015).
Zheng, Q. et al. Circular RNA profiling reveals an abundant circHIPK3 that regulates cell growth by sponging multiple miRNAs. Nat. Commun. 7, 11215 (2016).
Hansen, T. B. et al. Natural RNA circles function as efficient microRNA sponges. Nature 495, 384–388 (2013).
Piwecka, M. et al. Loss of a mammalian circular RNA locus causes miRNA deregulation and affects brain function. Science 357, eaam8526 (2017).
Du, W. W. et al. Foxo3 circular RNA retards cell cycle progression via forming ternary complexes with p21 and CDK2. Nucleic Acids Res. 44, 2846–2858 (2016).
Legnini, I. et al. Circ-ZNF609 is a circular RNA that can be translated and functions in myogenesis. Mol. Cell 66, 22–37.e29 (2017).
Pamudurti, N. R. et al. Translation of circRNAs. Mol. Cell 66, 9–21.e27 (2017).
Yang, Y. et al. Extensive translation of circular RNAs driven by N6-methyladenosine. Cell Res. 27, 626–641 (2017).
Chen, Y., Li, C., Tan, C. & Liu, X. Circular RNAs: a new frontier in the study of human diseases. J. Med. Genet. 53, 359–365 (2016).
Guarnerio, J. et al. Oncogenic role of fusion-circRNAs derived from cancer-associated chromosomal translocations. Cell 166, 1055–1056 (2016).
Li, Y. et al. Circular RNA is enriched and stable in exosomes: a promising biomarker for cancer diagnosis. Cell Res. 25, 981–984 (2015).
Memczak, S., Papavasileiou, P., Peters, O. & Rajewsky, N. Identification and characterization of circular RNAs as a new class of putative biomarkers in human blood. PLoS ONE 10, e0141214 (2015).
Viereck, J. & Thum, T. Circulating noncoding RNAs as biomarkers of cardiovascular disease and injury. Circ. Res. 120, 381–399 (2017).
Xu, H., Guo, S., Li, W. & Yu, P. The circular RNA Cdr1as, via miR-7 and its targets, regulates insulin transcription and secretion in islet cells. Sci. Rep. 5, 12453 (2015).
Shan, K. et al. Circular non-coding RNA HIPK3 mediates retinal vascular dysfunction in diabetes mellitus. Circulation 136, 1629–1642 (2017).
Li, A., Huang, W., Zhang, X., Xie, L. & Miao, X. Identification and characterization of circRNAs of two pig breeds as a new biomarker in metabolism-related diseases. Cell. Physiol. Biochem. 47, 2458–2470 (2018).
Gao, Y., Wang, J. & Zhao, F. CIRI: an efficient and unbiased algorithm for de novo circular RNA identification. Genome Biol. 16, 4 (2015).
Zheng, Y. & Zhao, F. Detection and reconstruction of circular RNAs from transcriptomic data. Methods Mol. Biol. 1724, 1–8 (2018).
Reilly, S. M. & Saltiel, A. R. Adapting to obesity with adipose tissue inflammation. Nat. Rev. Endocrinol. 13, 633–643 (2017).
Yamamoto, M., Cid, E., Bru, S. & Yamamoto, F. Rare and frequent promoter methylation, respectively, of TSHZ2 and 3 genes that are both downregulated in expression in breast and prostate cancers. PLoS ONE 6, e17149 (2011).
Santos, J. S., Fonseca, N. A., Vieira, C. P., Vieira, J. & Casares, F. Phylogeny of the teashirt-related zinc finger (tshz) gene family and analysis of the developmental expression of tshz2 and tshz3b in the zebrafish. Dev. Dyn. 239, 1010–1018 (2010).
Ebert, M. S., Neilson, J. R. & Sharp, P. A. MicroRNA sponges: competitive inhibitors of small RNAs in mammalian cells. Nat. Methods 4, 721–726 (2007).
Guo, H., Ingolia, N. T., Weissman, J. S. & Bartel, D. P. Mammalian microRNAs predominantly act to decrease target mRNA levels. Nature 466, 835–840 (2010).
Holdt, L. M. et al. Circular non-coding RNA ANRIL modulates ribosomal RNA maturation and atherosclerosis in humans. Nat. Commun. 7, 12429 (2016).
Abdelmohsen, K. et al. Identification of HuR target circular RNAs uncovers suppression of PABPN1 translation by CircPABPN1. RNA Biol. 14, 361–369 (2017).
Xue, R. et al. Clonal analyses and gene profiling identify genetic biomarkers of the thermogenic potential of human brown and white preadipocytes. Nat. Med. 21, 760–768 (2015).
Acknowledgements
This work was supported by Tanoto Initiative in Diabetes Research (to L.S.), National Medical Research Council’s Cooperative Basic Research Grant (NMRC/CBRG/0101/2016), Open Fund-Individual Research Grant (NMRC/OFIRG/0062/2017) and Ministry of Education Tier 2 grant (MOE2017-T2-2-015). Support was also provided by RNA Biology Center at CSI Singapore, NUS, from funding by the Singapore Ministry of Education’s Tier 3 grants, grant number MOE2014-T3-1-006. We thank S. Y. Chia, B. Pan, Z. Tiang and H. Hamadee for their technical assistance, Y.-H. Tseng (Joslin Diabetes Center) for providing cell lines and B. Lim and J. Li who offered their insight and expertise for this research.
Author information
Authors and Affiliations
Contributions
L.S. and R.F. conceived and designed the study and acquired funds. C.A., W.T., D.X. and W.F. performed data analysis and developed and implemented the methodology. Data generation, curation and laboratory experiments were performed by C.A., W.T., W.F., T.P.D., D.T.C.S., U.D. and D.X. L.S., R.F., C.A. and W.T. wrote the manuscript and all authors contributed, read and approved the final version.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Peer review information: Primary Handling Editor: Elena Bellafante.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Supplementary information
Supplementary Information
Supplementary Figs. 1–14 and Supplementary Note
Supplementary Table 1
Supplementary Table 1
Supplementary Table 2
Supplementary Table 2
Supplementary Table 3
Supplementary Table 3
Supplementary Table 4
Supplementary Table 4
Supplementary Table 5
Supplementary Table 5
Supplementary Table 6
Supplementary Table 6
Rights and permissions
About this article
Cite this article
Arcinas, C., Tan, W., Fang, W. et al. Adipose circular RNAs exhibit dynamic regulation in obesity and functional role in adipogenesis. Nat Metab 1, 688–703 (2019). https://doi.org/10.1038/s42255-019-0078-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1038/s42255-019-0078-z
This article is cited by
-
Circ-phkb promotes cell apoptosis and inflammation in LPS-induced alveolar macrophages via the TLR4/MyD88/NF-kB/CCL2 axis
Respiratory Research (2024)
-
CircDOCK7 facilitates the proliferation and adipogenic differentiation of chicken abdominal preadipocytes through the gga-miR-301b-3p/ACSL1 axis
Journal of Animal Science and Biotechnology (2023)
-
Epigenetic regulation in metabolic diseases: mechanisms and advances in clinical study
Signal Transduction and Targeted Therapy (2023)
-
Interference of a mammalian circRNA regulates lipid metabolism reprogramming by targeting miR-24-3p/Igf2/PI3K-AKT-mTOR and Igf2bp2/Ucp1 axis
Cellular and Molecular Life Sciences (2023)
-
The landscape of the long non-coding RNAs and circular RNAs of the abdominal fat tissues in the chicken lines divergently selected for fatness
BMC Genomics (2022)
