miR-103 promotes endothelial maladaptation by targeting lncWDR59

Blood flow at arterial bifurcations and curvatures is naturally disturbed. Endothelial cells (ECs) fail to adapt to disturbed flow, which transcriptionally direct ECs toward a maladapted phenotype, characterized by chronic regeneration of injured ECs. MicroRNAs (miRNAs) can regulate EC maladaptation through targeting of protein-coding RNAs. However, long noncoding RNAs (lncRNAs), known epigenetic regulators of biological processes, can also be miRNA targets, but their contribution on EC maladaptation is unclear. Here we show that hyperlipidemia- and oxLDL-induced upregulation of miR-103 inhibits EC proliferation and promotes endothelial DNA damage through targeting of novel lncWDR59. MiR-103 impedes lncWDR59 interaction with Notch1-inhibitor Numb, therefore affecting Notch1-induced EC proliferation. Moreover, miR-103 increases the susceptibility of proliferating ECs to oxLDL-induced mitotic aberrations, characterized by an increased micronucleic formation and DNA damage accumulation, by affecting Notch1-related β-catenin co-activation. Collectively, these data indicate that miR-103 programs ECs toward a maladapted phenotype through targeting of lncWDR59, which may promote atherosclerosis.

A rterial endothelial cells (ECs) are exposed to different types of blood flow patterns and shear stresses, which determine the endothelial phenotype. Indeed, laminar flow results in high shear stress and favors a quiescent EC phenotype characterized by a low proliferation rate and a reduced inflammatory activation 1,2 . In contrast, disturbed flow at arterial branching points and curvatures constantly induces low-grade injury to ECs resulting in apoptosis and inflammation [3][4][5] . Moreover, the injury triggers a repair process by EC proliferation, which is essential to maintain the endothelial lining 5,6 ; however, because disturbed flow-induced EC injury is physiologic, endothelial repair remains ineffective 6,7 . Hence, EC maladaptation generates the Achilles heel of the arterial system, vulnerable to additional damages by hyperlipidemia, which inhibits EC proliferation 8,9 and sustains chronic endothelial wound healing 9,10 .
Different types of shear stresses regulate the expression of specific microRNAs (miRNAs), which play key roles in EC injury and proliferation by negatively regulating post-transcriptional gene expression 11 . Indeed, endothelial Dicer affects inflammation and promotes atherosclerosis partly by miR-103-mediated suppression of Krüppel-like factor-4 (Klf4) 12 , which control the phenotypic adaptation of ECs to mechanical stress. In addition, miRNAs also affects EC proliferation. Indeed, miR-126-5p decreases atherosclerosis and improve endothelial proliferation by targeting Notch1-inhibitor Dlk1 6 . MiRNAs can also interact with long non-coding RNAs (lncRNAs), which have a broad spectrum of action as epigenetic regulators 13,14 and scaffold molecules 15 , to promote angiogenesis and cell differentiation. Among these, lincRNA-p21 16 , ANRIL 15 and MALAT1 16 emerged as miRNA decoys during atherosclerosis. However, the role of lncRNAs as targets of miRNA on EC maladaptation is still unclear.
We found that downregulation of miRNAs after Dicer knockout in ECs increases the expression of lncRNA transcripts, including the novel lncRNA WD Repeated Domain 59 (lncWDR59). In particular, targeting of lncWDR59 by miR-103 affects EC proliferation due to reduced Notch1 activation. Moreover, Notch1 activity on EC proliferation is limited by lncWDR59 through β-catenin, which prevents oxidized lowdensity lipoprotein (oxLDL)-induced micronucleic DNA damage accumulation in proliferating cells. Hence, miRNAs might promote EC maladaptation by negatively regulating lncRNA transcripts, which can reprogram ECs toward a more adaptive phenotype.

Results
Dicer regulates lncRNA expression in ECs during atherosclerosis. After a 12-week high fat diet (HFD) feeding period, the expression of 172 out of 2926 lncRNAs was differentially regulated in atherosclerotic arteries from EC-Dicer flox mice compared with EC-Dicer WT mice (fold change (FC) ≥ 1.5, P < 0.05, n = 2 mice per group) as determined by genome-wide microarray analysis (Fig. 1a). Among the differentially regulated lncRNAs, the expression of 97 and of 75 lncRNAs was increased and decreased, respectively, in EC-Dicer flox mice (Fig. 1a). Among the 97 upregulated lncRNAs, 6 were downregulated while 5 were also upregulated in the atherosclerotic arteries from Apoe −/− mice after Dicer knockout in myeloid cells (Supplementary Table 2). For 39 upregulated lncRNAs, including Jpx, Dlx1as, and Six3os, which play a role in cell viability 17 and differentiation 18 , the transcript sequence was previously determined (Supplementary Table 2). To determine the sequences of the other 58 upregulated lncRNAs, including the novel most significantly upregulated lncRNA lncWDR59, we performed a strand-specific RNA sequencing (RNA-seq) of the RNA isolated from murine aortic ECs (MAoECs). In addition to 15,199 protein-coding genes (fragments per kilobase million (FPKM) > 1), like the endothelialspecific genes Cdh5, Nos3, and Pecam1 ( Supplementary Figure 1A, B), we found that 34,524 lncRNA genes, including 18,478 annotated lncRNAs such as Malat1, Neat1, and Fendrr, and 16,046 novel lncRNA genes were expressed in MAoECs (Supplementary Figure 1A,B). All 39 annotated lncRNAs upregulated in EC-Dicer flox mice were also expressed in MAoECs (Supplementary Table 2). In addition, read fragments of 5 out of the 58 non-annotated lncRNAs upregulated in EC-Dicer flox mice, including lncWDR59, were identified by de novo transcript assembly combined with chromatin-state maps 19,20 (Supplementary Figure 1C-E and Fig. 1c). Among these five lncRNAs, the sequence of one new lncRNA, here called Leonardo, was fully determined by RNA-seq (Fig. 1c, Supplementary Figure 1D and Supplementary Notes). Leonardo gene is 1.317 kb long, consists of 5 exons and 4 introns, and is located between Gm29666 and nucleoporin 50 (Nup50) genes on the negative strand of chromosome 15. Analysis of the read fragments showed that the Leonardo transcript is spliced into a 792 bp-long mature lncRNA, which is predicted to fold into a thermodynamically stable secondary structure ( Fig. 1c and Supplementary Figure 2). In contrast to Leonardo, the sequence of lncWDR59 was only partially determined by RNA-seq. To discover the lncWDR59 full sequence, we performed a modified RACE-PCR (rapid amplification of cDNA ends-polymerase chain reaction) (Supplementary Figure 1E and Supplementary Notes). The lncWDR59 transcript was 1.61 kb long, derived from one exon located on chromosome 8 between the fatty acid 2-hydroxylase (fa2h) and the wdr59 genes ( Fig. 1c and Supplementary Figure 1E). Similarly as Leonardo, lncWDR59 transcript showed a thermodynamically stable secondary structure with characteristic functional lncRNA loops (Supplementary Figure 2 and Supplementary Table 3). Moreover, Coding Protein Calculator analysis showed a low protein-coding ability for lncWDR59, similar to Malat1 (used as positive control).
Let-7b and miR-103 target lncRNAs. Dicer knockout in ECs reduces the expression of 20 miRNAs, including miR-103 and let-7b, in atherosclerotic arteries 12 . To study whether miRNAs target lncRNAs in ECs during atherosclerosis, the sequences of 44 lncRNAs upregulated in EC-Dicer flox mice were analyzed for putative binding sites (BSs) for the 20 miRNAs downregulated after Dicer knockout. A total of 189 BSs were predicted in 36 lncRNAs ( Fig. 1d and Supplementary Figure 3). The highest number of BSs was found for miR-103 (22 BSs in 21 lncRNAs) and let-7b (22 BSs in 16 lncRNAs) in the sequence of 28 lncRNAs, including 9 lncRNAs that contain BSs for both miR-NAs ( Fig. 1d and Supplementary Figure 3). These data suggest that miR-103 and let-7b play a prominent role in targeting endothelial lncRNAs. The expression of 15 out of 28 lncRNAs with a number of high-quantity BSs for miR-103 and let-7b was studied in EC-Dicer flox mice and EC-Dicer WT by quantitative PCR (qPCR). Upregulation of 10 lncRNAs, including lncWDR59, Leonardo, and lnc039159, in atherosclerotic arteries from EC-Dicer knockout mice was confirmed by qPCR (n = 4-7 mice per group). By contrast, the other 5 lncRNAs and Malat1, which does not contain miR-103 or let-7b BSs, were not differentially regulated (Fig. 1e). To study whether loss of miR-103 and let-7b mediates the upregulation of the 10 lncRNAs in ECs, MAoECs were treated with Locked Nucleic Acid (LNA)-modified inhibitors of miR-103 or let-7b. Treatment with let-7b inhibitors increased Leonardo, Jpx, lnc064083, lnc039159, and lnc002669, whereas lnc003783 and lnc052749 were not affected (Fig. 1f).
To assess whether let-7b and miR-103 target lncRNAs in the RISC (RNA-induced silencing complex) of ECs, GW182 immunoprecipitation was performed in MAoECs treated with let-7b or miR-103 mimics. Treatment with let-7b mimics enriched the transcripts of Leonardo, lnc002669, and lnc039159 in the RISC, but not those of Jpx and lnc064083 (Fig. 1g). MiR-103 mimics enriched lncWDR59, Gm4258, lnc064083, and Jpx transcripts in the RISC (Fig. 1g). These results indicate a prominent role of miR-103 in the targeting of lncRNAs in ECs.
miR-103 targets lncWDR59. Next, the effect of pro-atherogenic conditions on the expression of the miR-103 targets lncWDR59,  Gm4258, Jpx, and lnc064083 was studied in ECs and murine arteries. In contrast to lnc064083, the expression of lncWDR59, Jpx, and Gm4258 was higher in MAoECs than in bone marrowderived macrophages (BMDM) (Fig. 2a). Moreover, lncWDR59 levels were similar between ECs and murine aortic smooth muscle cells (MAoSMCs), whereas miR-103 expression was slightly lower in MAoECs than in ECs and BMDM (Supplementary Figure 4A). Low shear stress, which increases endothelial miR-103 expression [21][22][23][24][25] , reduced lncWDR59 and increased lnc064083 compared to high shear stress, whereas Jpx and Gm4258 expression was similar in MAoECs exposed to low or high shear stress (Fig. 2b). Treatment of MAoECs with oxLDL, (100 µg ml −1 ) upregulated miR-103 and decreased lncWDR59, lnc064083, Gm4258, but not Jpx, and their expression was restored by inhibition of miR-103 in oxLDL-treated ECs (Supplementary Figure 4A, B and Fig. 2c). Notably, oxLDL treatment did not affect lncWDR59 and miR-103 expression in BMDM and MAoSMCs (Supplementary Figure 4A). In mice, the expression of lncWDR59, Jpx, lnc064083, and Gm4258 was substantially higher in athero-protected thoracoabdominal aortas than in athero-prone aortic arches (Fig. 2d), in opposite to miR-103 expression (Supplementary Figure 4C). Moreover, HFD feeding strongly downregulated lncWDR59, lnc064083 and Gm4258 and upregulated miR-103 in the aorta of Apoe -/mice, whereas Jpx was not affected ( Fig. 2e and Supplementary Figure 4D). In addition, lncRNA expression was studied in laser-microdissected ECs and macrophages from atherosclerotic lesions from Apoe -/mice (Supplementary Figure 4E). LncWDR59 and Jpx expression was higher in ECs than in macrophages, whereas the lnc064083 and Gm4258 expression levels were similar in both groups (Fig. 2f). In opposite to the expression of miR-103, in situ hybridization showed that lncWDR59 was more prominently expressed in ECs from EC-Dicer flox mice than in those from EC-Dicer WT mice (Fig. 2g).
Taken together, these data show that, in contrast to Gm4258, Jpx, and lnc064083, atherogenic conditions such as disturbed flow and hyperlipidemia consistently suppress lncWDR59 in ECs, which is in opposite to the regulation of miR-103. miR-103 affects EC proliferation by targeting lncWDR59. To study the functional role of miR-103 and lncWDR59 interaction, an antisense oligonucleotide (target site blocker (TSB)) was designed to specifically block the binding of miR-103 to its predicted lncWDR59 target site (Fig. 2h). Treatment of MAoECs with TSB increased the expression of lncWDR59, but not that of Jpx, Gm4258, lnc064083, or Klf4 (Fig. 2h). This result confirms the miR-103 target site located at nucleotides 1266-1291 of lncWDR59 transcript (Supplementary Figure 2).
MiR-103 promotes pro-inflammatory chemokine expression by targeting KLF4 12 . Although inhibition of miR-103 suppressed Ccl2 and Cxcl1 in MAoECs, silencing lncWDR59 or treatment with TSB did not affect the expression of these chemokines ( Fig. 3a and Supplementary Figure 4F). Moreover, treatment of HFD-fed Apoe -/mice with TSB that block the interaction between miR-103 and Klf4 did not alter lncWDR59 expression in arteries (Supplementary Figure 4G). Hence, lncWDR59 does not mediate the pro-inflammatory effects of miR-103 in ECs.
In addition to endothelial inflammation, inhibition of EC proliferation by hyperlipidemia at athero-prone sites promotes atherosclerosis (Supplementary Figure 4H). Accordingly, atheroprotective EC-Dicer knockout restored endothelial proliferation (Supplementary Figure 4I). In vitro, treatment of MAoECs with miR-103 inhibitors or TSB increased EC proliferation and reduced cyclin-dependent kinase inhibitor 1a (Cdkn1a) expression, whereas lncWDR59 inhibition reduced endothelial replication and upregulated Cdkn1a (Fig. 3b). LncWDR59 inhibition did not alter the expression of Fa2h and Wdr59, indicating that lncWDR59 and not its neighboring genes promotes EC proliferation (Supplementary Figure 4J). Moreover, TSBs promoted cell cycle progression of ECs from the G0 phase to the G1 and S+M phases (Supplementary Figure 4K) and increased proliferation of ECs subjected to low shear stress, but not to high shear stress (Fig. 3c). By contrast, TSB treatment did not affect MAoSMC proliferation (Supplementary Figure 4L). Taken together, these data show that miR-103 inhibits EC proliferation and cell cycle progression by targeting lncWDR59, which may enhance atherosclerosis.
Notch1 and β-catenin in lncWDR59-induced EC proliferation. In addition to Klf4, genes related to Wnt and Notch signaling pathways, which can increase cell proliferation 26,27 , were most strongly regulated in atherosclerotic arteries from EC-Dicer flox mice 12 . Accordingly, EC-Dicer knockout increased endothelial Notch1 and β-catenin activation in atherosclerotic arteries (Supplementary Figure 5A). Feeding a normal or HFD increased and decreased endothelial Notch1 and β-catenin activation and EC proliferation at predilection sites compared to non-predilection sites (Supplementary Figure 5B-C). In vitro, treatment with the Notch1-inhibitor DAPT or silencing Ctnnb1 reduced EC proliferation ( Fig. 3d and Supplementary   Fig. 1 LncRNAs upregulated in EC-Dicer flox mice during atherosclerosis and targets of let-7b and miR-103. a Pie graph from genome-wide microarray analysis for lncRNA genes differentially expressed in the aortas of EC-Dicer flox mice compared with EC-Dicer WT mice fed 12 weeks of HFD (n = 2 mice per group). Not-regulated lncRNA genes are in green, and modulated lncRNAs are in gray. Among these, up-and downregulated lncRNAs with a P value cutoff of 0.05 were divided according to their fold change. b Pie graph representing the 97 significantly upregulated lncRNAs, following RNA-seq analysis from MAoECs: 53 unknown lncRNAs (green), 5 lncRNAs for which the sequence was discovered by RNA-seq or RACE-PCR (blue), and 39 lncRNAs with a sequence annotated in NONCODE or ENSEMBL databases (gray). Three out of 36 lncRNAs showed a known function (pink). c Gene locus and full transcript sequence for 2 new lncRNAs, upregulated in EC-Dicer flox mice, i.e., Leonardo and lncWDR59. Detected regions for Leonardo (792 bp) and lncWDR59 (1.61 kb) sequences derive from RNA-seq or RACE-PCR analysis, respectively. Probes from genome-wide microarray are reported in red. d Cytoscape interaction network from RNAHybrid binding sites prediction between downregulated miRNAs and upregulated lncRNAs with a determined sequence in EC-Dicer flox mice. LncRNA targets of let-7b or miR-103 are in light blue and red, respectively. LncRNAs with BS for both let-7b and miR-103 are in orange. In gray are the rest of miRNAs target lncRNAs lacking a BS for miR-103 or let-7b. e, f The in vivo upregulation of lncRNAs with BS for let-7b and miR-103 was confirmed in EC-Dicer flox mice (n = 4-7 mice per group) (e) and in MAoECs after let-7b or miR-103 inhibition (f) by qPCR (n = 6 per group). FC > 2. g Fold enrichment of lncRNA transcripts in the RISC complex after transfection of MAoECs with let-7b or miR-103 mimics together with a mutant form of Ago2, following an immunoprecipitation (GW182-IP), measured by qPCR. Results of three independent experiments are shown. GAPDH and B2M were used as housekeeping genes and relative expression analysis. FC fold change of control (ctrl) group. The data are represented as mean ± SEM of the indicated number (n) of repeats. *P < 0.05, and **P < 0.01 by Student's t-test Figure 5D,E), indicating that Notch1 and Wnt signaling promote athero-protective EC regeneration.
Next, we studied whether Notch1 and β-catenin play a role in lncWDR59-mediated EC proliferation. Inhibition of miR-103 and TSB significantly promoted whereas lncWDR59 knockdown reduced the activation of Notch1 and β-catenin (Fig. 3e, f). TSB-induced EC proliferation was prevented by DAPT, indicating that Notch activation acts downstream of lncWDR59 on endothelial proliferation (Fig. 3g). Surprisingly, silencing Ctnnb1 enhanced the proliferation of TSB-treated ECs (Fig. 3g).
Moreover, DAPT reduced the proliferation of ECs treated with siCtnnb1 and TSB, demonstrating that β-catenin can inhibit EC proliferation by limiting Notch1 activation (Fig. 3h). Accordingly, DAPT decreased β-catenin activation in ECs treated with or without TSBs, whereas silencing Ctnnb1 reduced Notch1 activation (Notch intracellular domain (NICD)) in controltreated ECs (Fig. 3i, j). Moreover, silencing Ctnnb1 increased Notch1 activation in TSB-treated ECs (Fig. 3i, j). These data show that the interaction between miR-103 and lncWDR59 inhibits EC proliferation due to reduced lncWDR59-mediated  Relative expression (log 10 ) Relative expression (log 10 ) Relative expression (log 10 ) Relative expression (log 10 ) Relative expression (log 10  lncWDR59 blocks Numb to promote Notch1 activation. To study the mechanisms by which lncWDR59 regulates Notch1 activation, the subcellular localization of lncWDR59 was analyzed. Treatment of MAoECs with miR-103 inhibitors increased the cytosolic levels of lncWDR59 (Fig. 4a), indicating that lncWDR59 may bind to Notch1 or to cytosolic proteins related to the Notch1 signaling pathway. Therefore, we determined the interaction propensity between lncWDR59 and NICD, deltex4, Numb, anterior pharynx-defective 1 (aph-1), presenilin-1 and -2 (psen-1 and -2), and nicastrin using the catRAPID prediction software. LncWDR59 had no binding sites predictable in the amino acid sequence of NICD. Accordingly, lncWDR59 was not enriched in NICD-immunoprecitiated (IP) samples from MAoECs treated with control LNAs or miR-103 inhibitors ( Fig. 4b and Supplementary Figure 6A), indicating that lncWDR59 does not bind to NICD. Among the Notch1 pathwayrelated proteins, Numb contained sites with the highest propensity to interact with lncWDR59, (Fig. 4c and Supplementary Figure 1F). To study whether lncWDR59 binds to Numb, we performed Numb-IP, which revealed that lncWDR59 was not  Relative expression (log 10 ) Ki67 + cells (%) cdkn1a relative expression (log 10 ) . d MAoECs were treated with DAPT or siRNAs against β-catenin (siCtnnb1) for 24 h and stained with anti-Ki67 antibody to analyze their proliferation as described before. DMSO and siRNA were used as controls of DAPT and siCtnnb1, respectively (n = 4-8 per group). e, f MAoECs were transfected with miR-103 inhibitors, TSBs, or GlncWDR59 as described before and stained with an anti-activated Notch1 (NICD) (e) or β-catenin antibody (f). Data were represented as number of nuclear NICD + , and nuclear (star) and perinuclear (arrowheads) β-cat + localization. Inhibition of β-catenin activation makes β-catenin visible only as membrane adaptor protein for cell-to-cell intercellular adhesions. Data were normalized on total number of cells and expressed in percentage (n = 4-8 per group). g-j MAoECs were treated for 24 h with DAPT or siCtnnb1, alone or in combination with TSB to analyze (g) EC proliferation by Ki67 staining. DMSO or siRNAs in combination with LNAcontrols were used as controls. Data were normalized on total number of cells and expressed in percentage (n = 4 per group). h Analysis of Ki67 + TSB +siCtnnb-transfected MAoECs stimulated with DAPT or DMSO for 24 h (n = 4 per group). Analysis of β-catenin (i) or NICD (j) nuclear localization. The graphs correspond to the analysis of nuclear (stars) and perinuclear (arrowheads) β-cat + and nuclear NICD + localization. Data were normalized on total number of cells and expressed in percentage (n = 4-8 per group). Ccl2 C-C motif chemokine ligand 2, Cxcl1 C-X-C motif chemokine ligand 1, cdkn1a cyclindependent kinase inhibitor 1a, TSB target site blockers, DMSO dimethyl sulfoxide, DAPT γ-secretase inhibitor. Data are represented as mean ± SEM of the indicated number (n) of repeats. *P < 0.05, **P < 0.01, and ***P < 0.001 by one-way ANOVA and t-test. Scale bar: 25 µm enriched (Supplementary Figure 6B). However, NICD was high in the Numb-IP fraction, whereas it was low in the input fraction (Fig. 4d, e and Supplementary Figure 6C, D). To test the existence of a possible competitive binding between lncWDR59 and NICD to Numb, Numb-IP was performed in MAoECs after miR-103 inhibition. LncWDR59 was highly enriched in the Numb-IP compared to the input fraction. In addition, miR-103 inhibition reduced NICD levels in the Numb-IP fraction and increased those in the input (Fig. 4d, e and Supplementary Figure 6B-D). Interestingly, one of the four predicted Numb-interaction sites on lncWDR59 transcript comprises miR-103 BS, and involves the phosphotyrosine-binding domain of Numb, through which Numb interacts and induces Notch1 degradation 28 . Hence, these data suggest that binding of miR-103 on lncWDR59 prevents lncWDR59 reduces oxLDL-induced DNA damage. EC proliferation regenerates low shear stress-induced damage of ECs 6 . However, additional injuries, for example oxLDL-induced DNA damage 29 , may exhaust the EC repair capacity and enhance atherosclerosis 6 . In line, treatment of MAoECs for 24 h with oxLDL dose-dependently increased DNA damage and reduced EC proliferation determined by immunostaining of phosphorylated histone variant H2AX (γH2AX foci) 30 and Ki67, respectively (Fig. 5a). Moreover, we performed combined γH2AX and CD31 immunostaining of en face prepared aortas from Apoe −/− mice by image stack and three-dimensional (3D) reconstruction microscopy to determine the effect of HFD on endothelial DNA damage. Whereas endothelial DNA damage was not different between predilection and non-predilection sites in NDfed mice, a 12-week HFD increased γH2AX in ECs at predilection sites, but not at non-predilection sites of atherosclerosis (Fig. 5b). This result indicates that low shear stress increases the   susceptibility of ECs to oxLDL-induced DNA damage. Hence, hyperlipidemia may promote atherosclerosis by limiting EC proliferation and by increasing endothelial DNA damage. Notably, endothelial Dicer knockout reduced endothelial γH2AX in hyperlipidemic Apoe −/− mice (Fig. 5c), indicating a role of miRNAs in hyperlipidemia-induced endothelial DNA damage. To study whether miR-103 affects endothelial oxLDLmediated DNA damage by targeting lncWDR59, γH2AX was analyzed in MAoECs treated with TSBs and oxLDL. TSBs reduced γH2AX in MAoECs treated with or without oxLDL (Fig. 5d). Blocking Notch1 by DAPT or silencing Ctnnb1 abolished the TSB-mediated reduction of γH2AX in oxLDL-treated ECs (Fig. 5e). However, in the absence of TSB silencing, Ctnnb1 but not DAPT increased γH2AX in ECs treated with or without oxLDL ( Fig. 5e and Supplementary Figure 7A). Taken together, these data indicate that the protective effect of lncWDR59 on endothelial DNA damage is mediated by Notch1 and β-catenin activation.
During proliferation the chance of mitotic errors is high and can lead to the generation of extranuclear chromatin bodies, called micronuclei (MN), where double-strand breaks accumulate 31,32 . In line, the number of MN was high in proliferating ECs (Supplementary Figure 7B). Whereas damaged DNA is efficiently repaired in the nucleus, persistent DNA damage in the MN may further impair transcriptional activity in MN and compromise cell function 32 . Because in vitro oxidative stress can induce MN in ECs, we analyzed the effect of hyperlipidemia on MN formation in ECs in arterial en face prepared aortas. In contrast to the thoracic aorta, where MN-containing ECs were less than in the aortic arch under ND, HFD feeding increased MN formation in aortic arch ECs (Fig. 5f). Moreover, in ND-fed mice micronuclear γH2AX in aortic arch ECs was more frequent than in thoracic ECs (Fig. 5f). HFD feeding increased the micronuclear γH2AX in aortic arch but not thoracic ECs (Fig. 5f). In vitro, oxLDL increased MN formation and micronuclear γH2AX (Fig. 5g). These data indicate that oxLDL induces MN formation and micronuclear DNA damage in ECs at predilection sites of atherosclerosis. Next, we studied the role of miR-103-mediated suppression of lncWDR59 on oxLDL-induced MN formation in ECs. TSBs abolished oxLDL-induced MN formation and micronuclear γH2AX (Fig. 5h), an effect prevented by blocking Notch1 or Ctnnb1 silencing (Fig. 5i). Moreover, in the absence of TSB, silencing Ctnnb1 but not DAPT increased MN formation and micronuclear γH2AX in EC treated with or without oxLDL ( Fig. 5i and Supplementary Figure 7C, D). Hence, these results suggest that suppression of lncWDR59 by miR-103 in ECs increases MN formation and micronuclear DNA damage in response to oxLDL by inhibiting Notch1 and β-catenin. Moreover, hyperlipidemia may promote atherosusceptibility of predilection sites by DNA damage accumulation in the endothelial MN.
To assess the role of the interaction between miR-103 and lncWDR59 in vivo, Apoe −/− mice were treated with lncWDR59-TSBs or control LNA-modified oligonucleotides (0.5 mg per 20 g per injection) during the last 4 weeks of a 12-week HFD feeding program. Compared with mice treated with control LNAs, treatment with lncWDR59-TSBs reduced atherosclerosis, the necrotic core area, and endothelial DNA damage, and increased EC proliferation (Fig. 7a-c), whereas the fibrous cap thickness and the Cxcl1-expressing ECs (Supplementary Figure 7H) were not affected (Fig. 7a). Taken together, these data indicate that miR-103 enhances atherosclerosis and impairs EC regeneration partly by suppressing lncWDR59.
In human atherosclerotic lesions, in situ hybridization indicated that endothelial expression of miR-103 and lncWDR59 was high and low, respectively, in human atherosclerotic lesions compared with control vessels (Fig. 8d). Analysis of hsa-lncWDR59 expression in human atherosclerotic lesions indicated that hsa-lncWDR59 and SOX17 expression levels were negatively correlated with necrotic core area and positively with endothelial proliferation, whereas the levels of CDKN2A were correlated with increased necrotic core area ( Fig. 8e and Supplementary Figure 7I). Moreover, the levels of hsa-lncWDR59 correlated with increased SOX17 levels (Fig. 8e). Taken together, these data indicate that hsa-lncWDR59 shows a conserved function in human ECs and might also play an athero-protective role in humans.

Discussion
Chronic, low-grade endothelial injury and regeneration is a characteristic feature of endothelial maladaptation to disturbed flow at arterial bifurcations. We found that miR-103 inhibits endothelial proliferation by targeting lncWDR59, which resulted in reduced activation of Notch1 signaling. Moreover, suppression of lncWDR59 by miR-103 increased the susceptibility of proliferating ECs to oxLDL-mediated DNA damage and mitotic errors by reducing Notch1-mediated β-catenin activation. Hence, miR-103-mediated suppression of lncWDR59 promotes endothelial maladaptation by impairing EC regeneration and increasing mitotic aberrations during hyperlipidemia (Supplementary Figure 8).
Endothelial Dicer promotes arterial susceptibility to atherosclerosis by generating miRNAs 9,10,12 . In addition to mRNAs, miRNAs can interact with other RNA species, like lncRNA transcripts. This interaction can inhibit miRNA function, because lncRNAs can act as molecular decoys or sponges of miRNAs 34,35 or mediate the silencing of lncRNAs 36,37 . Our data show that Dicer suppressed the expression of multiple lncRNAs in ECs, indicating that targeting of lncRNAs by miRNAs plays a role in EC maladaptation. The miRNAs miR-103 and let-7b silenced lncRNAs, such as the novel intergenic lncRNAs lncWDR59 and Leonardo, respectively, which contain mainly non-canonical miRNA binding sites. We confirmed that miR-103 targets lncWDR59, which was downregulated at predilection sites and by hyperlipidemia, via a 6-mer binding site in ECs. Whereas miR-103 promotes endothelial inflammation by silencing KLF4 12 , the interaction between miR-103 and lncWDR59 in ECs inhibited cell proliferation and, in response to oxLDL, increased micronucleic DNA damage. Hence, miR-103 might promote an EC maladaptive phenotype by the synergistic silencing of Klf4 and lncWDR59. In line, our data indicate that lncWDR59 limits the increased susceptibility of proliferating ECs to oxLDL-induced DNA damage and micronuclei formation. Moreover, miR-103induced silencing of lncWDR59 may mediate the effect of endothelial Dicer on disturbed flow-induced EC proliferation and    38,39 , promotes proliferation of adult ECs, and prevents athero-progression 6,40,41 . Notably, endothelial Dicer suppresses EC proliferation and Notch1 signaling pathway at predilection sites of atherosclerosis 12 , and miR-103 targeting of lncWDR59 reduced EC proliferation by inhibiting Notch1 activation. Therefore, these findings support the hypothesis that miRNA silencing of lncRNAs, e.g., lncWDR59 silencing by miR-103, may promote endothelial maladaptation and atherosclerosis. Notably, lncWDR59 promoted Notch1 activation by binding to Numb, which mediates NICD degradation 28 , therefore impeding Numb interaction with Notch1. Therefore, disturbed flow promotes Notch1 activation not only via miR-126-5p-mediated suppression of Notch1-inhibitor Dlk1 6 , but also through lncWDR59 targeting of Notch1-inhibitor Numb. Taken together, downregulation of miR-126-5p and upregulation of miR-103 may both contribute to EC maladaptation by inhibiting Notch1 activity.
Proliferating cells are vulnerable to mitotic aberrations and DNA damage, e.g., by oxidative stress 29 , due to the accumulation of damaged DNA in MN, which can lead to mitotic catastrophe 31,32,42,43 . Although MN frequency was increased in proliferating ECs, lncWDR59 protected proliferating ECs from micronucleic DNA damage accumulation via β-catenin signaling, activated through Notch1-induced Sox17 expression. Accordingly, β-catenin can promote cell cycle checks and DNA damage repair processes in response to DNA damage in proliferating cells 44 . Moreover, the activation of β-catenin generated a negative feedback loop that limited Notch1 and protected ECs from aberrant proliferation. Hence, lncWDR59 plays a dual role in EC maladaptation by promoting EC proliferation through Notch1 activation and protecting ECs during proliferation from micronucleic DNA damage accumulation through β-catenin activation. Decreasing the effect of lncWDR59 by hyperlipidemia-and oxLDL-mediated upregulation of miR-103 increased the vulnerability of proliferating ECs to mitotic aberrations, which may limit EC regeneration and promote atherosclerosis.
Targeting of lncRNAs by miRNAs may represent a new mechanism by which ECs are reprogrammed toward a maladapted phenotype under disturbed flow.

Methods
In vivo TSB treatment and staining. Following 8 weeks of HFD feeding, Apoe −/− mice (The Jackson Laboratory, Bar Harbor, ME, USA) were randomized to the different experimental groups and tail vein injected once weekly for four consecutive weeks with lncWDR59 LNA-TSB or control LNA oligonucleotides (0.5 mg per 20 g; miRCURY LNA Target Site Blocker, in vivo use; Exiqon). During the injection period, mice were fed with a HFD comprising 21% crude fat, 0.15% cholesterol, and 19.5% protein. Tissues were harvested 1 week after the last injection. Paxgene-fixed/paraffin-embedded aortic roots were stained with Elastic van Gieson stain. A bright-field microscope (Leica DM6000B; Leica Microsystems) connected to a CCD camera (Leica DFC365) was used to obtain the images. The lesion area was quantified with ImageJ. Immunostaining of Ki67, γH2AX, C-X-C motif chemokine ligand 1 (Cxcl1), and von Willebrand factor (vWF) was also performed. The positive area or the number of positive cells was normalized to the lesion area or to the total number of vWF + cells, respectively, using ImageJ. The analysis of the staining was performed in a blinded manner.
All animal experiments were reviewed and approved by the local authorities (State Agency for Nature, Environment and Consumer Protection of North Rhine-Westphalia and District Government of Upper Bavaria) in accordance with the German animal protection laws.
RNA fluorescence in situ hybridization (FISH) on human and murine tissues. RNA FISH assay was performed using the Affymetrix protocol (ViewRNA Cell Plus Assay, Affymetrix) with some modifications. Briefly, after DNase digestion with RNase-free DNase (Roche Diagnostics, 10776785001), tissues were incubated with custom probe oligonucleotides specific for miR-103 and lncWDR59 designed and synthesized by Affymetrix as Type 1 and 6, respectively, following the manufacturer instructions. At the end of the hybridization, sections were blocked for 1 h with 1% bovine serum albumin blocking solution, then incubated with anti-vWF antibody.
En face immunostaining of murine tissues for micronuclei analysis. To study MN formation and DNA damage accumulation at predilection and nonpredilection sites, aortic arches and thoracic aortas from 12-week HFD-or chow diet-fed Apoe −/− mice were en face prepared and stained for CD31 and γH2AX. Cell nuclei were counterstained with 4′,6-diamidino-2-phenylindole (DAPI). The stained tissues were visualized using a Leica DM6000B fluorescent microscope. To detect the fluorescent signal in en face prepared tissues, z stacks of two-dimensional images were recorded, and deconvolution was performed using a mathematical algorithm to remove out-of-focus information (AF6000 3D deconvolution software module, Leica Microsystem). The 3D rendering was performed using the LAS X 3D algorithm module (Leica) to consider the staining and MN within CD31 + cells. For quantification, en face were entirely quantified. Data regarding MN formation were expressed as number of CD31 + cells containing MN on total number of CD31 + cells, in percentage. Data regarding γH2AX immunostaining in the MN were expressed as number of CD31 + cells showing γH2AX + MN normalized on total number of micronucleated CD31 + cells (previously normalized on total number of cells), expressed in percentage. The analysis of the staining was performed in a blinded manner.
Laser-capture microdissection. Root sections were collected on ultravioletsterilized and RNase-free polyester-membrane 0.9 µm FrameSlides (Leica). ECs and plaques were collected using a laser microdissection system (LMD7000, Leica) in RNase-free tubes. RNA was isolated with the PAXgene RNA MinElute kit (Qiagen) and followed by pre-amplification and reverse transcription with Ovation PicoSL WTA System V2 (NuGEN) following manufacturer's instructions.
Human carotid lesion samples. Human atherosclerotic lesions were collected during carotid endarterectomy. One part was fixed in 4% paraformaldehyde while another part was immediately stored in RNAlater for RNA isolation and qPCR. The study protocol for the collection of human atherosclerotic lesions was first approved from Ethics Committee of the Medical Faculty of RWTH Aachen University. The written informed consent was obtained from all participating patients. Background information regarding the patients were reported in the Supplementary Table 4. Classification of the type of the lesion was made according to the necrotic core area using a threshold 30% 46 . Compute nonparametric Spearman's rho correlation analysis was used to investigate the effect of athero-progression on hsa-lncWDR59, SOX17, and CDKN2A expression. Statistical analysis. All analyses were performed in a blinded way. Power calculation with StatMate (GraphPad Software) was performed to calculate the number of replicates needed for each experiment. The data represent the mean ± SEM and were compared using unpaired t-test, one-way analysis of variance (ANOVA), or two-way ANOVA (Prism, GraphPad Software), two-sided, unless stated otherwise. A P value of <0.05 was considered significant.
Additional methods are available in the Supplementary Information File (see Supplementary Methods section).
Data availability. All relevant data are included in the manuscript and available from the authors upon reasonable request. MiRNA expression profile from microarray is available online (https://doi.org/10.1038/ncomms10521) and in the Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/) under accession numbers GSE53433, GSE53435, and GSE114805.