circRAB3IP modulates cell proliferation by reorganizing gene expression and mRNA processing in a paracrine manner

Abstract

Circular RNAs are an endogenous long-lived and abundant noncoding species. Despite their prevalence, only a few circRNAs have been dissected mechanistically to date. Here, we cataloged nascent RNA-enriched circRNAs from primary human cells and functionally assigned a role to circRAB3IP in sustaining cellular homeostasis. We combined “omics” and functional experiments to show how circRAB3IP depletion deregulates hundreds of genes, suppresses cell cycle progression, and induces senescence-associated gene expression changes. Conversely, excess circRAB3IP delivered to endothelial cells via extracellular vesicles suffices for accelerating their division. We attribute these effects to an interplay between circRAB3IP and the general splicing factor SF3B1, which can affect transcript variant expression levels of cell cycle–related genes. Together, our findings link the maintenance of cell homeostasis to the presence of a single circRNA.

INTRODUCTION

The question of noncoding RNA authenticity and, in many cases, functionality has generally been settled via advances in the technologies developed to catalog and study transcriptomes. However, due to their structure and unconventional biogenesis (Jeck and Sharpless 2014; Xiao et al. 2020), circular RNAs (circRNAs) remained overlooked in transcriptome studies. Lately, circRNAs have been robustly detected across the evolutionary tree (Jeck and Sharpless 2014; Wu et al. 2020), yet their expression is cell type- (Salzman et al. 2013), tissue- (Maass et al. 2017; Xia et al. 2017), and context-specific (Venø et al. 2015; Knupp and Miura 2018; Lee et al. 2019b; Farooqi et al. 2021). It then follows that circRNA expression is regulated, advocating for their functional relevance in vivo.

It was not until recently that circRNAs emerged as important players in numerous cellular processes, including development and cell differentiation (Lee et al. 2019a; Di Agostino et al. 2020), senescence (Du et al. 2016), immunity (Zhou et al. 2019b), cancer (Visci et al. 2020), Alzheimer's disease (Zhao et al. 2016), and diabetes (Braicu et al. 2019). Mechanistically, most of the functions ascribed to circRNAs to date involve their interaction with microRNAs or RNA binding proteins (RBPs) (Du et al. 2016, 2017; Schneider et al. 2016; Abdelmohsen et al. 2017; Xia et al. 2018). Thus, circRNAs have been shown to act as “sponges” sequestering specific microRNAs or as “decoys” sequestering RBPs. Via such circRNA-microRNA/-RBP interactions the targets of sequestered microRNAs or RBPs are affected as regards expression (Hansen et al. 2013; Li et al. 2015a; Zhao et al. 2016), RNA processing (Zhang et al. 2013; Li et al. 2015b; Chen et al. 2018), or translation (Ashwal-Fluss et al. 2014; Abdelmohsen et al. 2017). Despite compelling data on the mode-of-action of a handful of circRNAs and a growing body of literature on circRNAs in cancer (Liu et al. 2020; Visci et al. 2020), how this large class of endogenous noncoding RNAs contribute to cell homeostasis remains rather poorly understood.

In this study, we investigated circRNAs specifically enriched in the nascent RNA fraction of primary human cells. Focusing on the two most enriched cirRNAs in this fraction, we show that circRAB3IP, recently linked to prostate cancer resistance (Chen et al. 2021a), can regulate the proliferation potency of cells and, thus, their path to senescence. We also show that circCAMSAP1, earlier implicated in diabetes (Haque et al. 2020), cancer cell proliferation (Zhou et al. 2020a; Chen et al. 2021b; Luo et al. 2021; Wang et al. 2022), and chronic inflammation (Liu et al. 2019), also acts as a modulator of senescence-associated inflammatory responses. Notably, circRAB3IP can be packaged into extracellular vesicles and delivered to cells to exert its function. Thus, we provide evidence that the depletion or paracrine action of a single circRNA to a normal human cell can profoundly remodel its gene expression program and modulate its phenotype.

RESULTS

Identification and characterization of circRNAs enriched in nascent RNA

We initially hypothesized that circRNAs with a direct role in gene regulation would be enriched in nascent RNA as a reflection of a required proximity to the sites of transcription (as proposed by Li et al. 2015b). To identify such circRNAs in primary human endothelial cells (HUVEC; pooled from three donors), we used factory-seq (Melnik et al. 2016). In the factory-seq protocol, HUVEC nuclei are isolated in a close-to-physiological buffer, chromatin is digested with DNase I and washed out (this fraction contains chromatin-associated transcripts), before the nucleoplasmic sites of transcription are released from the nuclear substructure into the supernatant by treatment with a mixture of Group-III caspases (this fraction contains nascent RNAs; Fig. 1A). We reasoned that such a biochemical enrichment would facilitate discovery of circRNAs without a need for treatment with the highly processive RNase R. We analysed HUVEC nascent transcriptomes for circRNA enrichment using DCC (Cheng et al. 2016; see Materials and Methods) to identify 244 candidates (Supplemental Fig. S1A,B). Of these, only circCAMSAP1 (hsa_circ_ 0001900) and circRAB3IP (hsa_circ_0000419) showed robust circRNA/mRNA enrichment (DCC-ratio >0.5; Fig. 1B) and were also supported by detection and high DCC-ratios in total RNA-seq data (Supplemental Table S1). Moreover, both could not be detected in chromatin-associated RNA isolated via factory-seq (Supplemental Fig. S1A; Supplemental Table S1).

FIGURE 1.

circRNAs enriched in nascent RNA fractions. (A) Nascent RNA (red) is collected from HUVECs following nuclei purification, and DNase I and Group-III caspase treatment to detach transcription factories (gold) from the substructure. (B) Heatmap showing circRNA enrichment over their linear counterparts (DCC-ratio) in factory-seq data. circRNAs with a ratio >0.5 are indicated (starred). (C) Factory and total RNA-seq coverage along CAMSAP1 and RAB3IP. Zoomed-in view: signal enrichment between circularized exons. (D) Bar graph showing RT-qPCR mean fold enrichment (from two replicates) of RNase R-treated over untreated samples. (*) P < 0.05; unpaired two-tailed Student's t-test. (E) Relative circRNAs and mRNA enrichment in polysome fractions normalized to whole-cell levels (±SD, from three replicates) determined by RT-qPCR. PL: light polysomes; PH: heavy polysomes. (F) Decay plots of the RNAs from panel E following transcriptional inhibition for 0–24 h. RNA levels (±SD, from three replicates) were normalized to YWHAZ and plotted relative to 0-h levels. Half-lives were CAMSAP1_mRNA = 1.9 h, circCAMSAP1 = 32.2 h, RAB3IP_mRNA = 3.8 h, circRAB3IP = 4.6 h. (G) Representative circCAMSAP1 and circRBA3IP FISH signal (arrowheads) in HUVEC counterstained with DAPI. The nonnascent RNA-enriched circHSGP2 signal provides a control. Bar: 10 µm. (H) Bar plots showing percent of cells (from two experiments) without focal circRNA FISH signal (gray), with at least one cytoplasmic (green), or at least one nuclear focus (blue) in images like those in panel G.

circCAMSAP1 and circRAB3IP arise from the sequences of CAMSAP1 exons 3 and 4, and RAB3IP exons 7 and 8, respectively (Fig. 1C; Supplemental Fig. S1C). circRNA-forming gene segments are often marked by higher than average intronic signal enrichment between the two circRNA-exons in read-coverage plots (Yang et al. 2011). This was apparent for CAMSAP1 and RAB3IP in factory-seq, but not in total RNA profiles despite greater sequencing depth (Fig. 1C; Supplemental Fig. S1D). The prevalent circRAB3IP isoform was biexonic (237 nt-long), whereas the two circCAMSAP1 isoforms we detected differed by the presence of an intron between exons 3 and 4, but were comparably abundant (Supplemental Fig. S1C). Primers designed for all subsequent experiments cannot distinguish between these two circCAMSAP1 isoforms.

Due to their covalently closed structure, circRNAs are resistant to RNase R degradation, display longer half-lives than their mRNA counterparts (Jeck and Sharpless 2014), and are rarely translated (Huang et al. 2021). circCAMSAP1 and circRAB3IP have been reported in HUVECs (http://www.circbase.org/), but we assessed their RNase R resistance (Fig. 1D) and lack of polysome association (Fig. 1E), validating that both are bona fide noncoding circRNAs. As regards their stability in vivo, we found an expected long half-life for circCAMSAP1, but not for circRAB3IP (Fig. 1F), perhaps indicative of tight temporal regulation. We also tested their dependence on ADAR1, an RNA-editing enzyme influencing circRNA biogenesis (Ivanov et al. 2015). We used ADAR1 knockdowns and total RNA-seq to identify ADAR-sensitive circRNAs genome-wide (Supplemental Fig. S1D–F). Following the knockdown, both circCAMSAP1 and circRAB3IP showed decreased DCC-ratios indicative of reduced circularization (Supplemental Fig. S1G; Supplemental Table S1).

circCAMSAP1 and circRAB3IP enrichment in nascent RNA could imply association with the active sites of transcription (Caudron-Herger et al. 2015; Li et al. 2015b). However, circRNAs often localize in the cytoplasm (Salzman et al. 2013), as does circCAMSAP1 in cancer cells (Zhou et al. 2020a). Therefore, we investigated the subcellular distribution of the two circRNAs using RT-qPCR on nuclear or cytoplasmic RNA to find slight (for circCAMSAP1) to high (for circRAB3IP) nuclear enrichment (Supplemental Fig. S1H). We complemented this analysis with circRNA FISH using fluorescently labeled probes targeting each back-splicing junction (Zirkel and Papantonis 2018). circCAMSAP1 and circRAB3IP exhibited both nuclear and cytoplasmic localization in HUVECs. In ∼5% of cells, circRNA signal accumulated focally at the border of the nucleus (often with nascent RNA from the transcribed gene; Fig. 1G,H; Supplemental Fig. S1I,J). Control experiments targeting circHSPG2, an exon–intron circRNA with a high DCC ratio but not enriched in nascent RNA (Fig. 1D; Supplemental Fig. S1C; Supplemental Table S1), showed exclusive nuclear localization (Fig. 1G,H). These data agree with factory-seq enrichments and imply rapid translocation from the nucleus to the cytosol.

circRNA depletion leads to senescence-like transcriptional profiles

We next asked whether depletion of either circCAMSAP1 or circRAB3IP from HUVECs would alter their gene expression program. Knockdown experiments (KD) using siRNAs spanning the circRNA backsplicing junction followed by total RNA sequencing revealed large-magnitude changes with 3134 and 920 differentially expressed genes (DEGs) upon circCAMSAP1- and circRAB3IP-KD, respectively (Fig. 2A–C; Supplemental Fig. S2A). Genes up-regulated upon circCAMSAP1-KD were mostly linked to immune/inflammatory responses, while those down-regulated to the regulation of the cell cycle and ribosome metabolism (Fig. 2D). Genes down-regulated upon circRAB3IP-KD were also predominantly related to cell cycle regulation, as well as to RNA metabolism (Fig. 2E). Unlike circCAMSAP1-KD, circRAB3IP depletion did not trigger proinflammatory gene expression (Fig. 2D,E). One should note, however, that due to the inflexibility in designing siRNAs against circRNA backsplicing junctions, a subset of these DEGs might represent off-target events.

FIGURE 2.

circCAMSAP1- and circRAB3IP-KD trigger major transcriptional changes in HUVECs. (A) PCA plot of RNA-seq replicates from circCAMSAP1- (magenta), circRAB3IP-KD (blue), and control HUVECs (NTC; gray). (B) Bar plot showing fold change (log₂) for genes up-/down-regulated (orange/blue) in circCAMSAP1-KD data with a P_adj ≤ 0.05 and absolute (log₂) fold change ≥ 0.6. (C) As in panel B, but for circrRAB3IP-KD data. (D) Gene set enrichment analysis of DEGs from panel B. Related terms are clustered (dotted circles) and labeled by the group-dominant gene ontology. (E) As in panel D, but for circRAB3IP-KD DEGs. (F) Scatter plot correlating (log₂) fold-change values for DEGs (N) shared by circCAMSAP1- and circRAB3IP-KD; Pearson's correlation coefficient (R) and its associated P-value are indicated. (G) Plots of circCAMSAP1- (pink) and circRAB3IP-KD (blue) mean normalized coverage across genes linked to inflammation and cell cycle control. NTC data coverage (gray) provides a baseline. (H) Bar plots depicting cell cycle profiles determined via FACS of PI-stained HUVECs in circRNA-KD experiments (two replicates). (*) Significantly different to NTC; P < 0.05, Fisher's exact test. (I) As in panel F, but correlating circRNA-KD DEGs with those from senescent HUVECs. (J) Heatmap of SASP genes that are also DEGs in circRNA-KD data. Color coding reflects (log₂) fold-change values with a P_adj ≤ 0.05.

We next analyzed enrichment of transcription factor motifs known to regulate the DEGs in each circRNA-KD. Motifs for E2F factors, involved in cell cycle control (Dimova and Dyson 2005), dominated down-regulated genes in both data sets, while motifs of proinflammatory regulators like NF-κB, STATs, and IRFs (Platanitis and Decker 2018) were overrepresented for circCAMSAP1-KD up-regulated genes (Supplemental Fig. S2B). The two circRNA-KD data sets shared >450 DEGs, but their respective expression changes showed no significant correlation (Fig. 2F). Nevertheless, genes down-regulated in both experiments were part of E2F-controlled and G2/M cell cycle pathways, and this was reflected in changes of the cell cycle profiles of KD cells showing extended G1 phases (Fig. 2G,H). On the other hand, IFNβ/γ and TNFα signaling genes were stimulated by circCAMSAP1-KD only (Fig. 2G). Thus, the partially overlapping roles of these two circRNAs unfold via nonconvergent regulation of different effector genes in the same pathways.

The cell cycle deceleration that followed both circCAMSAP1- and circRAB3IP-KD, complemented by inflammatory stimulation and ECM remodeling, was reminiscent of the changes accompanying replicative senescence entry by HUVECs (Zirkel et al. 2018). Therefore, we compared each KD data set with DEGs from senescent HUVECs to discover positive correlation (R > 0.35; Fig. 2I). In fact, ∼30% of circRAB3IP- and 16% of circCAMSAP1-KD DEGs are regulated in the same manner upon HUVEC senescence entry. Even stronger correlation (R > 0.55) was observed when comparing our KD data sets to a consensus gene expression signature of senescence (Supplemental Fig. S2C; Hernandez-Segura et al. 2017), and the few genes showing divergent regulation did not enrich for any particular pathway or GO term. Senescent cells also display cell type-specific secretory phenotypes (i.e., SASP) comprised of molecules acting in a paracrine fashion to promote inflammation and senescence (Coppé et al. 2010). SASP genes induced in our circRNA-KD cells (Fig. 2J) represent ∼1/3 off all relevant entries in the Reactome database (https://reactome.org/). In line with the proinflammatory signature in circCAMSAP1-KD (Fig. 2D,E), this knockdown induced the largest number of SASP genes. Together, our results indicate that depletion of either circRNA suffices for the extensive remodeling of the HUVEC homeostatic program into a senescence-like one.

circRAB3IP modulates the processing of cell cycle–related transcripts

Various studies have suggested that circRNAs regulate cell proliferation via microRNA “sponging” (Zheng et al. 2016; Li et al. 2018; Zeng et al. 2018). However, this requires multiple cognate binding sites for the sequestered microRNA in the circRNA (e.g., as in Hansen et al. 2013), which does not appear to be the case for circRAB3IP or circCAMSAP1 (assessed via http://www.targetscan.org). Thus, we set out to explore the association of either circRNA with regulatory proteins. To this end, we performed RNA antisense pull-down coupled to mass spectrometry (RAP-MS; Fig. 3A; McHugh and Guttman 2018) in HEK293 cells. This choice of cell type was dictated by the need for large cell numbers (4–5 × 10⁸ per replicate) overexpressing circRAB3IP or circCAMSAP1 (Supplemental Fig. S3A,B). Prior to RAP-MS, circRNA-KDs in HEK293 confirmed that the majority of the gene expression changes likened those seen in HUVECs (Supplemental Fig. S2A,D–G).

FIGURE 3.

circRAB3IP interacts with SF3B1 and affects RNA splicing. (A) For RAP-MS, cells were transfected with circRNA-pcDNA3 vectors, UV crosslinked, lysed, and hybridized with biotinylated probes targeting backsplicing junctions. Following pulldown, circRNA-associated proteins were denatured and analyzed by label-free mass spectrometry. (B) Plot showing circRAB3IP-bound proteins and their interactions inferred using STRING (https://string-db.org/; from three replicates). Color coding reflects GO terms (key below) associated with each protein. (C) Plot showing circRAB3IP enrichment following SF3B1-pulldown in HUVECs normalized to input (±SD, from two replicates). IgG pulldown provides a control. (D) Overlap between DEGs and genes displaying isoform usage changes in circRAB3IP-KD data. Overlap was not more than expected by chance; P = 0.6711, hypergeometric test. (E) GO terms enriched for genes (N) with ≥30% isoform usage switching in circRAB3IP-KD data. (F) Bar plot of isoform usage types in circRAB3IP-KD data. (*) FDR < 0.05. (G) Venn diagram (left) showing the overlap between SF3B1-bound mRNAs and DEGs in circRAB3IP-KD. GO terms (right) linked to overlapping up-/down-regulated genes (orange/blue). (H) Venn diagram showing overlap between SF3B1-bound and differentially spliced mRNAs upon circRAB3IP-KD. (*) More than expected by chance; P < 0.001, hypergeometric test. Genes involved in cell cycle regulation are listed (gray box; https://gsea-msigdb.org). (I) MDM4 isoforms (left) detected upon circRAB3IP-KD. Functional domains in each isoform (color code, top) and coding isoforms (*) are indicated. Bar plot (right) showing usage changes between KD (black) and control RNA-seq data (gray) for each isoform. (*) FDR ≤ 0.001.

circRAB3IP RAP-MS robustly enriched for our circRNA target (Supplemental Fig. S3C) and identified 16 proteins as potential interactors. The majority of these were functionally linked to mRNA splicing and stabilization, chemokine production or cell differentiation (Fig. 3B). RAP-MS for circCAMSAP1 returned a catalog of 27 proteins, most of which could be linked to inflammation, gene expression regulation, and RNA metabolism (Supplemental Fig. S3D). Twenty-five percent of circRAB3IP and 55% of circCAMSAP1 interactors can be classified as RNA-binding proteins (based on Castello et al. 2016). The only interactor shared by circRAB3IP and circCAMSAP1 was U2SURP, a component of the general splicing machinery (Will 2002). However, DEGs from our two circRNA-KD experiments showed negligible correlation to DEGs from published U2SURP-KD data from HEK293 cells (R = −0.035 and −0.25 for circRAB3IP- and circCAMSAP1-KD, respectively; De Maio et al. 2018). Thus, circCAMSAP1 operates via a set of partners that are discrete from those of circRAB3IP.

Of all candidate interactions, we chose to focus on that between SF3B1 and circRAB3IP. This was because SF3B1 is an RNA-binding protein involved in mRNA splicing (Sun 2020), senescence (Yin et al. 2019), and aging (Holly et al. 2014), and because circRNAs have been shown to affect splicing regulation (Ashwal-Fluss et al. 2014; Li et al. 2015b). To study the SF3B1–circRAB3IP interaction, we first orthogonally confirmed its association with circRAB3IP in HUVECs using RNA immunoprecipitation (Fig. 3C). Next, we reanalyzed our circRAB3IP-KD data to find >200 genes with substantial (>30%) change in isoform usage upon knockdown (Fig. 3D). In contrast, circCAMSAP1 depletion induced such changes in just 40 genes. circRAB3IP-related isoform switching is not reflected in differential mRNA expression, as <6% of DEGs also switch isoforms (Fig. 3D). Notably, according to GO term analyses, genes displaying differential isoform usage were also linked to cell cycle progression, cytoskeleton reorganization, and centriole regulation (Fig. 3E). Almost half of these differentially-spliced genes now expressed isoforms containing fewer functional domains and/or shorter open reading frames (Fig. 3F), thus providing a possible interpretation of how cell cycle deceleration manifests. Those alternative splicing events that are mostly responsible for the observed isoform switching were increased skipping of canonical 5′ and 3′ splicing sites, intron retention, and changes in termination site choice (Supplemental Fig. S3E). Thus, on top of its effect on gene expression levels, an additional regulatory layer involving widespread mRNA processing changes follows circRAB3IP depletion from growing HUVECs.

Finally, we analyzed publicly available SF3B1 eCLIP data (accession no.: ENCSR133QEA) against our circRAB3IP-KD to discover that ∼40% of DEGs are also bound by SF3B1 and linked to cell cycle regulation and ECM remodeling (Fig. 3G), much like in SF3B1-KD cells (Fig. 2D,E). We also crossed SF3B1 eCLIP targets with mRNAs that switch isoforms upon circRAB3IP-KD. Again, ∼40% of these were also bound SF3B1 (Fig. 3H) and contained key cell cycle regulators like MDM4, ID2, ALMS1, PLK4, FANCD2, CCNT1, and PLCB1. For all of them, strong changes in favor of noncoding/domainless isoforms could help explain cell cycle effects in cells lacking circRAB3IP (Fig. 3I; Supplemental Fig. S3F). Notably, the negligible overlap between DEGs from circRAB3IP-KD and SF3B1-KD experiments (accession no.: GSE88630 and GSE176778; not shown) implies that the circRAB3IP-SF3B1 interaction is selective for regulating the SF3B1 target subrepertoire linked to cell cycle regulation.

circRAB3IP induces HUVEC proliferation in a paracrine manner

Given that cell cycle arrest was a prominent feature of both our circRNA-KDs, we reasoned that circRNA overexpression would lead to increased proliferation, in line with what has been reported for circCAMSAP1 in cancer cells (Liu et al. 2019; Zhou et al. 2020a; Chen et al. 2021b; Luo et al. 2021). Indeed, MTT assays of HEK293 cells overexpressing circRAB3IP or circCAMSAP1 (used in RAP-MS) showed significantly accelerated proliferation compared to control cells (Supplemental Fig. S4A,B; circCAMSAP1-, circRAB3IP- and empty vector-overexpressing HEK293 doubling times were 17.4, 22.7, and 28.1 h, respectively).

To study this further, we generated a cell line allowing for the controllable circRAB3IP production while also negating any negative effects of long-term stable overexpression. We introduced a piggyback construct (pB) into HEK293 cells that efficiently produced circRAB3IP in a doxycycline-inducible manner (Fig. 4A; Materials and Methods). Motivated by reports on the packaging and secretion of circRNAs (Li et al. 2015c; Dou et al. 2016), we asked whether this ∼2000-fold circRAB3IP excess would end up in extracellular vesicles. Indeed, extracellular vesicles collected from pB-circRAB3IP cultures contained significantly higher titers of the circRNA than those purified from unmodified cells (Fig. 4A). In fact, blocking vesicle formation leads to circRAB3IP FISH signal accumulation in the cytoplasm of pB-modified cells (Supplemental Fig. S4C).

FIGURE 4.

circRAB3IP acts in a cell nonautonomous manner to promote HUVEC proliferation. (A) Relative circRAB3IP levels (±SD, from three replicates) in lysates and extracellular vesicles purified from empty vector (ctrl) or pB-circRAB3IP cells (blue). (B, left) circRNA-pB HEK293 lines were induced for 48 h, before extracellular vesicles were purified and added onto growing HUVECs. (Right) As in panel A, but for HUVECs treated with extracellular vesicles from control (pB-ctrl) and pB-circRAB3IP cells, and expressed as fold-change over nontreated cell levels (NoExo). (C) Plot showing normalized area confluence (±SD) derived from live-cell imaging of HUVECs stimulated with nontreated, control, and pB-circRAB3IP exosomes from three replicates. (*) P_adj < 0.05, multiple t-testing. (D) Overlap between SF3B1 eCLIP targets (red) and genes switching isoforms in circRAB3IP-KD (blue) or in pB-circRAB3IP exosome-treated cells (light blue). (E) GO terms associated with isoform-switching genes in pB-circRAB3IP exosome-treated HUVECs that are also bound by SF3B1; cell cycle–related genes are listed (gray box).

This setup allowed us to test whether HUVEC proliferation could be affected by high circRAB3IP titers, given that endothelial cells are not amenable for overexpression studies. We collected extracellular vesicles from circRAB3IP-overexpressing or empty-vector HEK293 and delivered them onto HUVEC cultures. This resulted in a moderate yet detectable uptake of circRAB3IP (Fig. 4B). Next, HUVEC growth rates were monitored over the course of 30 h via automated live-cell imaging. HUVECs showed constantly increasing proliferation rates, despite moderate circRAB3IP uptake (Fig. 4C). These experiments highlight the ability of circRAB3IP to modulate cellular growth via paracrine signaling, even at low intracellular doses.

Since the removal of circRAB3IP leads to apparent reprogramming of gene expression also through splicing changes, we speculated that excess circRAB3IP would induce cell proliferation in a similar manner. We performed RNA-seq in HUVECs treated with circRAB3IP-exosomes to find negligible changes in gene expression (just seven genes satisfied a P_adj < 0.05 threshold; not shown). However, following splicing analysis, we discovered >200 transcripts displaying alternative isoform usage (by >20% or more). While little overlap to circRAB3IP-KD data was observed, circRAB3IP-treated cells shared 79 transcripts with the list of direct SF3B1 eCLIP targets (Fig. 4D). These transcripts were associated with such GO terms as mitotic cell cycle regulation and anaphase-promoting complex-dependent process, and were derived from genes known to exert cell cycle control (Fig. 4E). For example, ANAPC1 and FZR1 encode core components of the anaphase-promoting complex (Yamano 2019), KLF4 transcription factor isoforms have been shown to modulate cell cycle choices (Yang et al. 2020) and RBL1 is a known tumor suppressor (Liban et al. 2017). Changes in isoform usage identified for each of these genes were predicted as favorable for cell cycle progression (Supplemental Fig. S5). Therefore, circRAB3IP is required for maintaining homeostatic cell cycle regulation and its depletion or (moderate) overexpression can alter this primarily via isoform modulation of SF3B1-associated transcripts.

DISCUSSION

Despite a growing body of evidence on the roles of circRNAs in various cancer lines, how they function in primary cells under homeostasis remains largely unexplored. Here, we address this question using human endothelial cells to provide the first functional characterization of circRAB3IP, while also providing a resource for the previously studied circCAMSAP1. While our motivation for studying this subset of circRNAs originated in their enrichment in nascent RNA, our results showed that circCAMSAP1 and circRAB3IP exhibit functional specificity and distinct interactomes in line with their disparate roles.

circCAMSAP1 has been studied in the context of cancer (Zhou et al. 2020a; Chen et al. 2021b; Luo et al. 2021; Wang et al. 2022) and chronic inflammation (Liu et al. 2019). These studies found circCAMSAP1 acting as a “sponge” for various microRNAs, whereas the latter implicates its secondary folding in an interaction with PKR. Here, we describe a different role for circCAMSAP1 in cell homeostasis. Its knockdown had overlapping but not fully convergent effects to circRAB3IP depletion, affirming the view of circRNAs as highly specialized molecules carrying out cell type-specific functions. In addition to the senescence-like cell cycle effects in circCAMSAP1-depleted cells, a prominent proinflammatory signature also emerged. This suggests that circCAMSAP1 plays a role in regulating a subset of inflammation-responsive genes, including a considerable fraction of known SASP effectors. However, our RAP-MS data did not recover the previously documented interaction of circCAMSAP1 with PKR that is thought to implicate this circRNA in inflammatory regulation (Liu et al. 2019). On the other hand, and in line with our analysis of E2F transcription factor motif enrichment in its knockdown-regulated genes, circCAMSAP1 has been shown to influence E2F activity in neurons (Zhou et al. 2020a). This could explain how circCAMSAP1 overexpression in HEK293 accelerates also cell cycle progression. However, how its protein interactors, cataloged here by RAP-MS, mediate the functions of circCAMSAP1 remains to be addressed.

circRAB3IP depletion from HUVECs is accompanied by senescence-like gene expression changes that counter homeostatic cell cycle progression and correlate with the consensus gene expression signature of senescence (Hernandez-Segura et al. 2017), as well as with the induction of factors remodeling ECM, which constitutes another senescence hallmark (Levi et al. 2020). Of all the circRAB3IP interactors, the general splicing factor SF3B1 appeared most relevant due to its implication in cell cycle arrest (Dolatshad et al. 2015) and senescence (Yin et al. 2019). Although the precise nature of the SF3B1–circRAB3IP interplay remains to be determined, there appears to be a good correlation between differential isoform production and changes in cell cycle progression (e.g., via isoform switching of RBL1 transcripts). SF3B1 is a core component of the protein complex that regulates U2-snRNP recruitment to its targets (Zhou et al. 2020b) and was recently reported to be guided onto its target mRNAs via cotranscriptional association with nucleosomes, rather than by specific nucleotide sequences in the target transcripts themselves (Kfir et al. 2015). Given that SF3B1 translocation from the nucleosome onto the nascent transcript is thought to be RNA-dependent, we speculate that circRAB3IP can modulate this association, even at low circRNA titers and for a specific subset of RNA targets. In this scenario, its interaction with U2SURP (which also interacts with circCAMSAP1), a component of U2 snRNP, may be relevant. Such a mode-of-action could help explain the well-documented role of splicing in the processes of cellular senescence and tissue aging (Deschênes and Chabot 2017).

Finally, circRAB3IP can also exert its influence on the cell cycle in a paracrine fashion, despite a very moderate uptake in our system. This opens up the possibility of cell–cell communication via the load carried in extracellular vesicles exchanged within a given tissue niche. Therefore, our data support the notion that even mild modulation of the levels of a single circRNA can inflict profound changes to gene expression and cellular homeostasis in both a cell autonomous and a cell nonautonomous fashion.

MATERIALS AND METHODS

Cell culture and circRNA verification

HUVECs from pooled, apparently healthy, donors (passage 4–9; Lonza) were grown in complete Endopan-3 medium kit (PAN-Biotech) at 37°C under 5% CO₂. HEK293 cells were passaged in DMEM (Sigma) supplemented with 10% FBS (Gibco) and 1% Pen/Strep (Gibco) under the same conditions. For RNase R treatments, HUVECs were grown in complete media, RNA was extracted via the Direct-zol RNA purification kit (Zymo) and treated with 1.7 U RNase R per µg of RNA (Epicenter) for 13 min at 37°C. Fractionation experiments were performed by HUVEC nuclei extraction in a physiological buffer (PB; 100 mM CH₃COOK, 30 mM KCl, 10 mM Na₂HPO₄, 1 mM MgCl₂, 1 mM Na₂ATP, 1 mM DTT, 10 mM β-glycerophosphate, 10 mM NaF, 0.2 mM Na₃VO₄, 25 U/mL RiboLock RNase Inhibitor, and 1:1000 dilution of cOmplete Protease Inhibitor Cocktail; adjusted to pH 7.4 with KH₂PO₄) for 30 min on ice, followed by 5 min centrifugation at 1000g. Supernatants corresponding to the cytoplasmic fraction were collected in TRIzol, while nuclear pellets were washed once more in PB before collection in TRIzol and purification using the Direct-zol RNA purification kit (Zymo). HUVEC polysome profiling was performed as previously described (Bartsch et al. 2018). Finally, for a half-life test, HUVECs were treated with 50 µM DRB (5,6-Dichloro-1-β-D-ribofuranosylbenzimidazole; Sigma) for 2–24 h and collected in TRIzol to isolate total RNA for RT-qPCRs.

Factory-seq and circRNA detection

Nascent RNA from ∼10⁷ HUVECs was purified as previously described (Melnik et al. 2016). Briefly, cells were grown in complete Endopan-2 medium (Lonza) until 90% confluency and scraped in PB. Nuclei isolation was performed for 30 min in ice-cold PB supplemented with 0.4% Igepal (Sigma-Aldrich). Lysis efficiency was assessed microscopically. Nuclei were collected by centrifugation and DNase I-treated for 30 min at 37°C (Worthington, 30U per 5 × 10⁶ cells). After digestion, nuclei were pelleted and lysed in a native lysis buffer (NLB; 40 mM Tris-acetate pH 7.4, 2 M 6-aminocaproic acid and 7% sucrose supplemented with 50 U/mL RiboLock RNase Inhibitor and 1:1000 dilution of cOmplete Protease Inhibitor Cocktail) for 20 min on ice. Lysates were further treated with 2U per sample of a Group-III caspase mix (active human caspases 6, 8, 9, and 10; BioCat) for 30 min under vigorous shaking. Supernatants were collected in TRIzol (Invitrogen) and RNA isolated using the Direct-zol RNA purification kit (Zymo). Resulting cDNA libraries sequenced to ∼35 × 10⁶ read pairs aligned to the human reference genome (hg19) using STAR (Dobin et al. 2013) and assessed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/). For the detection of circRNAs enriched in nascent RNA relative to total RNA, tandem programs DCC and Circ.test were applied (Cheng et al. 2016). Briefly, DCC utilizes the chimeric read mapping to detect and quantify circRNAs and Circ.test subsequently uses binomial distribution for modeling and ANOVA testing to find those circRNAs that differ in expression between tested groups, irrespective of the cognate mRNA from the parental gene. Differentially enriched circRNAs are reported on the basis of the “DCC-ratio” reflecting the circRNA over parental mRNA expression using junction and nonjunction read counts. Raw read coverage plots were generated using Gviz (Hahne and Ivanek 2016).

circRNA fluorescence in situ hybridization

FISH was performed using back-splicing junction-targeting fluorescent probes for circCAMSAP1 and circRAB3IP (listed in Supplemental Table S4) as previously described (Zirkel and Papantonis 2018). Briefly, HUVECs treated or not with 35 U RNase A (Qiagen) for 30 min or with 10 µM chloroquine (Sigma) for 2 h were grown on coverslips “etched” with 0.1% hydrofluoric acid until 70% confluent and fixed for 15 min (in 10 mL of fixation buffer containing 1 mL 9% NaCl, 2.5 mL 16% paraformaldehyde, 500 µL glacial acetic acid, and 6 mL RNase-free water). Coverslips were then washed in PBS and cells permeabilized in 0.5% Triton-X/0.5% saponin for 5 min. After washing again in PBS, coverslips were post-fixed in 3.7% formaldehyde for 5 min, washed in PBS and dehydrated in ethanol (70%, 90% and absolute) for 3 min. Hybridization of probes was carried out in hybridization buffer [25% formamide, 2× SSC, 200 ng/µL sheared salmon sperm DNA, 5× Denhardt's solution, 50 mM phosphate buffer (20 mM KH₂PO₄, 30 mM KHPO₄·2H₂O, pH 7.0), and 1 mM EDTA] mixed with fluorescent probe (25 ng/µL) in a 9:1 ratio. For competition experiments, the mix also contained 100× excess of unlabeled probe. The hybridization mix was denatured at 90°C for 10 min, quenched on ice for 2 min and added to the coverslips. Hybridization was performed overnight in the dark at 37°C in a humid chamber. The next day, coverslips were washed three times in 2× SCC, once in RNase-free water, and mounted on glass slides with ProLong Gold Antifade presupplemented with DAPI (Invitrogen). Imaging was performed on a Leica DMi8 platform using a 63× magnification objective (oil), and images exported via the LASX software (Leica).

siRNA-mediated circRNA knockdown and analysis

Approximately 35 × 10⁴ HUVEC cells were seeded per each 60 mm plate 1 d before transfection. Each siRNA (180 pmol) was mixed with Opti-MEM (Gibco) and Lipofectamine RNAiMAX according to manufacturer's instructions (Thermo Fisher). Cells were washed once in PBS and incubated in 2.5 mL of Opti-MEM and 500 µL of transfection mix for 4 h, after which the medium was replaced with complete Endopan medium. Forty-eight hours post-transfection, cells were collected in TRIzol (Invitrogen) and RNA extraction was performed using the Direct-zol RNA purification kit (Zymo). RNA integrity was verified with Bioanalyzer, and samples were sequenced to at least 25 × 10⁶ paired end reads. A sequencing quality check was performed using FastQC and reads were aligned to the human reference genome (hg19) with default STAR aligner settings (Dobin et al. 2013). Aligned reads quantification was performed with featureCounts (Liao et al. 2014), selecting for uniquely aligned and properly matched read pairs (selected options –primary -p -B -C). Between-sample normalization of raw quantified reads was performed using the RUVs option of RUVseq (Risso et al. 2014) that estimates factors of unwanted variation using as reference those genes in the replicates for which the covariates of interest remain constant. Differentially expressed genes (DEGs) were estimated via DESeq2 (Love et al. 2014), wherein genes with FDR < 0.01 and absolute (log₂) fold-change ≥0.6 were considered significantly differentially expressed. The resulting DEGs (deposited in GEO alongside raw data) were used as an a priori defined list for gene set enrichment analysis (GSEA; Subramanian et al. 2005) to assess the overrepresentation of GO terms in the RNA-seq data sets and produce network plots via Cytoscape (Franz et al. 2016). Read counts of mapped RNA-seq were normalized per million mapped reads (RPM) and plotted using ngs.plot (Shen et al. 2014) to visualize signal enrichment along genes of pathways chosen from the GSEA Molecular Signatures Database (MSigDB IDs: TNFα via NFkB signaling, M5890; IFNγ response, M5913; E2F targets, M5925; G2/M checkpoint, M5901). GO term analysis for gene sets without assigned expression data (eCLIP and isoform switching genes) was performed using Metascape (Zhou et al. 2019a). Isoform switch analysis was performed via IsoformSwitchAnalyzeR (v1.12; Vitting-Seerup and Sandelin 2019), whereby raw RNA-seq data is first pseudoaligned to hg19 using Salmon (v1.2; Patro et al. 2017) and the resulting count matrices were used to find genes displaying no significant change in expression levels, but rather in transcript usage. After default filtering steps and statistical testing for significance, genes which display at least 20% (for circRNAs delivered by extracellular vesicles) or 30% change in isoform usage (for circRNA-KD) were considered for downstream analyses.

circRNA overexpression constructs

Expression vectors for circCAMSAP1 and circRAB3IP were generated by amplifying genomic DNA containing the two circRNA-generating exons, the intron in between them and ∼500 and ∼100 bp of the up- and downstream flanking introns. PCR amplified fragments were inserted into pcDNA3 or piggyBAC (pB) vectors via appropriate restriction digests and ligation. To facilitate circRNA production, artificial inverted repeats were generated around the circRNA exons by PCR amplification of ∼360 bp of the upstream flanking intron and its insertion in an inverted orientation downstream from the circRNA exons through a restriction digest. For northern blots and RNase R treatment of cells transfected with pcDNA3 overexpression vectors, 5 µg RNA were digested with 1 U RNase R (Epicentre) per µg RNA in a total reaction volume of 10 µL for 10 min at 37°C. After this, samples were prepared for agarose blotting by addition of 20 µL loading buffer to 5 µg RNase R-treated or untreated purified RNA. Samples were then loaded onto a 1% agarose gel containing 3% formaldehyde and 1× MOPS and run at 75 V in 1× MOPS for ∼3 h after which RNA was transferred onto a Hybond N+ membrane (GE Healthcare) overnight. RNA was then UV crosslinked to the membrane and prehybridized in Church buffer (0.158 M NaH₂PO₄, 0.342 M Na₂HPO₄, 7% SDS, 1 mM EDTA, 0.5% BSA at pH 7.5) for 1 h. The membrane was probed with a 5′ radioactively labeled DNA oligonucleotide (60-mer probe) at 55°C overnight and washed twice in 2× SSC, 0.1% SDS for 10 min at 45°C before exposure on a Phosphoimager screen for data collection. Cell proliferation of pB transfected HEK293 cells was monitored via MTT assays. In brief, triplicates of ∼2000 cells were seeded in one well of a 96-well plate. On the next day, the medium was replaced with 100 µL fresh medium plus 10 µL of a 12 mM MTT stock solution (Invitrogen), and incubated at 37°C for 3 h. Subsequently, all but 25 µL of the medium was removed from the wells, and formazan dissolved in 50 µL DMSO was mixed thoroughly with the cells and incubated at 37°C for 10 min. Samples were then mixed again and absorbance was read at 530 nm. Measurements were taken at 24, 48, 72, and 96 h post-seeding and background levels subtracted.

RNA antisense-purification coupled to mass spectrometry (RAP-MS)

Approximately 4 × 10⁸ HEK293 were transfected with either circCAMSAP-pcDNA3 or circRAB3IP-pcDNA3 vectors via CaPO₃ transfection. Briefly, cells were transfected 48 h prior to the experiment with a mix containing the plasmid, 2.5 M CaCl₂, and HEBS (50 mM HEPES, 280 mM NaCl, 1.5 mM Na₂HPO₄ at pH 7.05 adjusted using NaOH). An additional dish of HEK293 was transfected with control GFP-pcDNA3 to monitor transfection efficiency. RAP-MS was performed essentially as previously described (McHugh and Guttman 2018) with 55-mer sense and antisense biotinylated probes designed to target circCAMSAP1 or circRAB3IP. Cells were UV-crosslinked at 254 nm (Stratalinker 1800; 0.8 J/cm²), scraped in ice-cold PBS, collected by centrifugation, and counted to ensure 2 × 10⁸ cells per capture reaction. Cell lysis was performed for 10 min on ice in total cell lysis buffer (10 mM Tris-HCl pH 7.5, 500 mM LiCl, 0.5% dodecyl-maltoside, 0.2% SDS, 0.1% sodium deoxycholate, 1× cOmplete Protease Inhibitor Cocktail, and 1000 U of RiboLock RNase Inhibitor). During cell lysis, samples were passed through 26G needles five times and then sonicated (Bioruptor microtip sonicator, 30% amplitude) for 15 cycles with 5 sec “on”/10 sec “off” settings. Following sonication, lysates were treated with 30 U DNase I (Worthington) for 20 min at 37°C and precleared for 30 min at 52°C with 1.2 mL of streptavidin Roti MagBeads beads (Carl Roth) per capture reaction. Beads were previously washed in 10 mM Tris-HCl pH 7.5 and hybridization buffer (10 mM Tris-HCl pH 7.5, 5 mM EDTA, 500 mM LiCl, 0.5% DDM, 0.2% SDS, 0.1% sodium deoxycholate, 4 M urea, 2.5 mM TCEP). After preclearing, 2 nmol of each probe was denatured at 85°C, quenched on ice, added to the lysates, and incubated at 52°C for 4 h with intermittent mixing (12 sec “on”/9 sec “off,” 900 rpm). For probe capture, a fresh aliquot of streptavidin beads was prepared as above and added to the samples for another 30 min at 52°C. Beads were then washed 6× in hybridization buffer for 5 min at 52°C with intermittent mixing. After the final wash, beads were resuspended in benzonase elution buffer (20 mM Tris-HCl pH 8.0, 0.05% NLS, 2 mM MgCl₂, and 0.5 mM TCEP) and hybridized probes eluted from the beads using 120 U Benzonase (Sigma) for 2 h at 37°C with intermittent mixing. Supernatants were collected and treated with 5 mM DTT for 30 min at 55°C, and subsequently with 40 mM CAA for 30 min at room temperature, centrifuged for 10 min at 20,000g, transferred to a new tube, and stored at −20°C until preparation for mass spectrometry. The final sample preparation and mass spectrometry analysis was carried out at the Proteomics Core Facility in Cologne, Germany. To assess the efficiency of each RNA pull-down, input and eluted RNA samples were collected during the experiment (as described by McHugh and Guttman 2018), resuspended in NLS elution buffer (20 mM Tris-HCl pH 8.0, 10 mM EDTA, 2% NLS, 2.5 mM TCEP) and treated with 1 mg/µL of Proteinase K for 1 h at 55°C. RNA was then isolated via phenol:chloroform extraction, precipitated with ethanol, and reverse transcription was carried out using the SuperScript II Reverse Transcriptase according to manufacturer's instructions (Invitrogen). qPCRs were run using the qPCRBIO SyGreen Mix (NIPPON).

RNA immunoprecipitation coupled to qPCR (RIP-qPCR)

HEK293 cells grown to confluence in complete media were harvested by gentle scaping in PBS, counted to ensure ∼2 × 10⁷ cells per IP, pelleted, and lysed in ice-cold polysome lysis buffer (100 mM KCl, 5 mM MgCl₂, 10 mM HEPES-NaOH pH 7.0, 1 mM DTT, 200 U/mL RNaseIn, 1× PIC, 0.5% NP-40) for 30 min. Cell lysis was facilitated by passing the lysates through a 26G needle 10 times during the incubation on ice, and by 6 cycles of sonication (30 sec “on”/30 sec “off,” low input) on a Bioruptor Pico (Diagenode). Lysates were then treated with 40 U DNase I (Worthington) per IP for 30 min and pelleted by centrifugation to remove incompletely lysed cells. Input (5%) samples were taken from supernatants and stored in TRIzol (Invitrogen) until RNA extraction. The rest of the supernatants was subjected to overnight immunoprecipitation with 10 µg of IgG (Milipore, 12-371B; 1 µg/µL) or SF3B1 antibody (Santa Cruz, sc-514655; 0.2 µg/µL). Next day, 30 µL of protein-G Dynabeads (Invitrogen) were prewashed with 1 mL of NT-2 buffer (250 mM Tris-HCl pH 7.4, 750 mM NaCl, 5 mM MgCl₂, 0.25% NP-40) and incubated for 1 h with 10 µg/IP of mouse bridging antibody (Active Motif; 1 µg/µL) at 4°C with end-to-end rotation. Beads were next washed three times in NT-2 buffer, added to the lysates, and incubated for 4h at 4°C with end-to-end rotation. Lysates were next washed six times in 1 mL of NT-2 buffer for 3 min/wash on end-to-end rotor to ensure removal of unbound material. After the last wash, beads were resuspended in TRIzol (Invitrogen) and RNA extracted using the Direct-zol RNA kit (Zymo). Reverse transcription was carried out using the SuperScript II Reverse Transcriptase (Invitrogen) and qPCRs using the qPCRBIO SyGreen Mix (NIPPON).

Extracellular vesicle purification and transfer experiments

HUVECs and pB-circRNA transfected HEK293 were grown in media supplemented with exosome-depleted (Life Technologies). HEK293 were induced with 3 µg/ml doxycycline (Sigma) and 400 µg/mL of G418 (Sigma). After 48 h, cells were collected in TRIzol for qPCR analysis of overexpression, and media collected on ice to purify extracellular vesicles through a series of centrifugation steps (Livshts et al. 2015). Briefly, the first round of centrifugation was at 750g for 10 min to remove leftover cells from the medium. Supernatants were then collected and spun at 1500g for 10 min, followed by 14,000g for 35 min and 100,000g for 2 h to separate microvesicles and exosomes. Pellets containing extracellular vesicles were then washed in PBS (Sigma) and subjected to another centrifugation at 100,000g for 1 h. Pellets carrying extracellular vesicles were resuspended in 200 µL PBS and stored short-term at −80°C. For assessment of circRNA enrichment in HUVECs, exosomes were used for RNA extraction (Total Exosome RNA & Protein Isolation Kit, Invitrogen), according to the manufacturer's instructions and qPCR analysis as above. For extracellular vesicle transfer onto HUVECs, cells were grown in complete media supplemented with purified extracellular vesicles for 24 h. HUVECs were then seeded and monitored for growth in an IncuCyte S3 platform (Sartorius) for 30 h. For RNA-seq analysis of extracellular vesicle-treated HUVECs, cells were washed 2× in PBS and collected in TRIzol to extract RNA.

Statistical tests

Assessment of correlation between RNA-seq data sets was performed using the Pearson's correlation core function in R (https://www.r-project.org/); Student's t-tests, Fisher's exact test, and nonlinear regression fits were performed using Prism (https://www.graphpad.com/scientific-software/prism/). Unless stated otherwise, any data with a P-value <0.05 were deemed as significant.

DATA DEPOSITION

All RNA-seq data produced in this study are available via the NCBI Gene Expression Omnibus repository under accession number GSE199553 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE199553).

SUPPLEMENTAL MATERIAL

Supplemental material is available for this article.

MEET THE FIRST AUTHORS

Natasa Josipovic.

Karoline Ebbesen.

Meet the First Author(s) is a new editorial feature within RNA, in which the first author(s) of research-based papers in each issue have the opportunity to introduce themselves and their work to readers of RNA and the RNA research community. Natasa Josipovic and Karoline Ebbesen are co-first authors of this paper, “circRAB3IP modulates cell proliferation by reorganizing gene expression and mRNA processing in a paracrine manner.” Natasa did her PhD research in the lab of Professor A. Papantonis at the University Medical Center in the International Max Planck Research School for Genome Science in Göttingen, Germany, where she studied regulatory cellular mechanisms mediated by noncoding RNA and proteins. Currently, Natasa is investigating the cellular transcriptome and its regulation through method development and application of single-cell sequencing-based approaches as a member of Single Cell Discoveries B.V. in Utrecht, The Netherlands. Karoline did her PhD research in the labs of Dr. Jørgen Kjems and Dr. Thomas B. Hansen at the Department of Molecular Biology and Genetics, Aarhus University, Denmark, where she examined the biogenesis pathway and molecular function of a specific subset of candidate circular RNAs. Karoline continued her studies of circular RNAs as a postdoc in the same lab for an additional two years, and is currently working in the laboratory of Dr. Søren Lykke Andersen at the same department, Aarhus University, investigating applications of small nucleolar RNAs.

What are the major results described in your paper and how do they impact this branch of the field?

The most important finding, in our view, is the regulatory effect that a noncoding circRNA can have. While it would be expected that a transcription factor affects the expression of hundreds or even thousands of genes, observing the same for a small, noncoding, relatively unabundant circRNA was a (welcome) surprise. With this work we extend the current understanding of how circRNAs may affect the regulation of cellular homeostasis. Specifically, we could show that the levels of a particular circRNA, in our case circRAB3IP, can tune the proliferation potential of the cell via changes in its gene expression and splicing repertoire.

What led you to study RNA or this aspect of RNA science?

NJ: I was fortunate to have a chance to explore genomic, transcriptomic, and proteomic-oriented molecular biology through disparate projects during my PhD studies. What shifted my focus and interest toward transcriptomics is the versatility of RNA molecules—some are coding, some regulatory; they come in different shapes and sizes with different roles within cells, while also being exported and serving as paracrine signaling cues. It is a growing field marked by cool technical advances, such as single-cell sequencing and RNA localization assays, with many of its intricacies remaining to be explored.

KE: I was always fascinated by the diversity and multifaceted form and function of RNAs in general and became intrigued by circular RNAs in particular, because the presence of this covalently closed RNA provided so many new potential avenues to explore that could contribute to and potentially even close gaps in our understanding of RNA biogenesis and function in general.

During the course of these experiments, were there any surprising results or particular difficulties that altered your thinking and subsequent focus?

We know that human cells have about 10,000 different circRNA molecules, often originating from coding genes, yet we observe that their roles can be quite distinct from the proteins coded by their parental genes. Moreover, we show that circRNAs that do regulate the cell's transcriptome, distinctly alter the gene expression program. In our example of circRAB3IP, we observe regulation of a stunning ∼1000 genes that relate specifically to the cellular division program, adding this circRNA to the already complex regulatory cascade, largely relying on protein regulation. This specificity of the circRNA's master regulator-like role and profound effect on cells homeostasis was quite a surprising outcome of our research. The difficulty of studying circRNAs stems from their low abundance and the lack of orthogonal approaches to the conventional ones in investigation of protein–RNA interactions. This limits the number of experiments that can be executed close to physiological conditions, making it difficult to show mechanistically how these interactions facilitate circRNAs regulatory roles.

What are some of the landmark moments that provoked your interest in science or your development as a scientist?

KE: As a bachelor student back in 2011, I was fortunate enough to stumble upon the topic of circular RNAs at a poster session held by the Department of Molecular Biology and Genetics. I was immediately intrigued, simply because this was an RNA type that I had never heard of or considered the possibility of, but, at the same time it seemed infinitely logical that such an RNA could be produced once I considered it. As a young student diving into a topic such as circular RNAs that, at the time, was only just emerging, this was an incredible way to realize that what we are able to discover is always defined by the lens we use and to remember that one of the constraints on every experiment is our imagination as scientists. This is something that I always try to keep in mind to this day and that allows me to stay curious and find different ways to approach scientific questions.

If you were able to give one piece of advice to your younger self, what would that be?

NJ: To divide and conquer. The scientific process is often a lonely journey, where one bears the sole responsibility for their work. Science, however, is a joint effort and it is now easier than ever to reach a person specializing in any topic or experimental method. I would advise my younger self to limit my time and effort trying to solve issues on my own and seek collaboration or advice on those things that I am not yet an expert in.

What are your subsequent near- or long-term career plans?

NJ: While I thoroughly enjoyed my time during my PhD and learned the principles of academic work, after its completion I wanted to do research with a more tangible outcome. I believe the future of therapeutic approaches lies in transcriptomic regulation; therefore, I recently joined a company which provides me with a chance to stay close to experimental work, take part in various scientific projects, and work on the development/improvement of single-cell sequencing-based methods. In the short term, this position allows me to stay at the cutting edge of science, all the while learning the ropes of start-up companies. In the longer term, I hope to grow enough as a scientist and entrepreneur to be able to start my own biotech company.

How did you decide to work together as co-first authors?

It was a lucky coincidence that brought a wonderful collaboration between the Papantonis and Hansen labs located in Germany and Denmark, respectively. This project stemmed from a curious finding involving circRNAs in a project with a rather different scope in the Papantonis lab. As the Papantonis lab mainly focuses on genomics, we sought partners with circRNA expertise. Karoline traveled to Germany and spent a few months with us for this purpose, where we jointly performed relevant experiments and where Natasa had the chance to learn the intricacies of circRNA methodology from her. This collaboration was invaluable for the project, but also brought a fresh scientific perspective: We both learned how effortless and fun it can be to work together with someone and get more out of the time put into the project.