Induced Endothelial Cell Cycle Arrest Prevents Arteriovenous Malformations in Hereditary Hemorrhagic Telangiectasia

Male and female mice were used to minimize sex-related biased results. All animal protocols and procedures were reviewed and approved by the University of Virginia animal care and use committee (protocol No. 4277) and complied with all ethical regulations. The expanded methods in the Supplement Material contain a list of all mouse strains and protocols.

All continuous variables were represented as mean±SEM. The Mann-Whitney nonparametric test for unpaired samples was used to analyze continuous variables between groups. One-way ANOVA with the Holm-Šidàk multiple comparisons test was used to compare the means of continuous variables among 3 groups. The 2-way ANOVA parametric test was used to compare 2 or more groups to determine whether there was an interactive effect between 2 independent variables (control samples versus treated samples) on a continuous dependent variable (polarity or time) (IBM SPSS Statistics). All graphs and analyses were generated using Prism 8.0 software (GraphPad).

To analyze EC cycle state in AVMs, we first used a preclinical mouse model of HHT that uses BMP9/10 blocking antibodies to create arteriovenous shunts.^12,13 BMP9/10 blocking antibodies were injected intraperitoneally at postnatal day (P) 2 and P4 in mice expressing the Fucci2 cell cycle reporter, which enables dynamic visualization and quantification among ECs residing in the different cell cycle states (Figure 1A and 1B). For example, ECs in S/G2/M phases harbor green fluorescing nuclei, whereas EC nuclei in early G1 state are reporter-negative, and EC nuclei in late G1 state are fluorescing red. Consistent with our previous findings,⁸ we found that in P7 control retinas, ECs forming veins and arteries predominantly reside in the early G1 and late G1 states of the cell cycle, respectively (Figure 1C). Conversely, in anti-BMP9/10–treated pups, we found a lower proportion of retinal arterial ECs in late G1, and significantly more ECs actively cycling in S/G2/M in both arteries and veins, compared with controls. In addition, among ECs forming the capillary plexi, we detected a significantly higher proportion of actively cycling cells in S/G2/M and in early G1 and, concomitantly, significantly less in late G1, compared with control (Figure 1C). In addition, we performed 5-ethynyl-2′-deoxyuridine incorporation studies to identify proliferating ECs that are undergoing DNA synthesis. In anti-BMP9/10–treated pups, we found a significantly higher number of 5-ethynyl-2′-deoxyuridine-positive retinal ECs in arterial, venous, and capillary vessels, and in distal vascular plexi above arterial and venous vessels, compared with controls. Altogether, these results show a dysregulation of cell cycle state and hyperproliferation of all endothelial subtypes in HHT (Figure S1A through S1D).

To further characterize the dysregulation of EC cycle state, we performed bulk RNA sequencing of murine primary retinal ECs isolated from P7 control and anti-BMP9/10–treated pups. Gene ontology analysis revealed that mRNA expression of gene families regulating cell cycle progression and proliferation was highly enriched in the anti-BMP9/10–treated group (Figure S1E). More specifically, analysis of differentially expressed genes in retinal ECs isolated from anti-BMP9/10–treated tissues showed upregulation of several genes that promote G1-to-S cell cycle transition such as CDK4, CDK6, and Mki67, as well as genes that regulate S/G2/M phases (Figure 1D).

Finally, EC cycle status was evaluated in human dermal telangiectases from 3 patients with HHT type 2, and control normal skin biopsies in resection borders from 3 patients with melanoma. Using immunohistochemistry staining for Ki67, a protein expressed during S/G2/M phases, we found a higher number of Ki67-positive ECs in dermal telangiectases from HHT type 2 patients compared with normal skin samples (Figure 1E), which is consistent with previous studies.¹⁴ Collectively, these results demonstrate that EC cycle is dysregulated in both preclinical and clinical HHT conditions leading to EC hyperproliferation. Our data suggest that drugs that induce EC cycle arrest could be clinically relevant for the treatment of patients with HHT to prevent or regress vascular malformations.

We found that CDK4 and CDK6 are upregulated in ECs isolated from anti-BMP9/10–treated mice (Figure 1D). Thus, to test whether the use of pharmacological cell cycle modulators could prevent the development of AVMs in a murine model of HHT, we used palbociclib, a CDK4 and CDK6 inhibitor (CDK4/6i) that blocks the transition from G1 to S state.⁸ Pups were injected at P2 and P4 with BMP9/10 antibodies, followed by palbociclib treatment starting at P4, then at P5 and P6 (Figure 2A). At P7, we found, as expected, that mice receiving BMP9/10 antibodies exhibited arteriovenous shunts and hyperdense retinal vascular plexi compared with control tissues (Figure 2B through 2D). However, pups receiving both BMP9/10 blocking antibodies and CDK4/6i developed fewer arteriovenous shunts and had less hyperdense retinal vascular plexi, and CDK4/6i treatment normalized retinal vascular progression and venous vessel diameters but not arterial vessel diameters (Figure 2B through 2G). It is important to note that the CDK4/6i treatment by itself did not impair retinal vascular development in palbociclib-treated P7 control pups (Figure S2A and S2B).

We then analyzed the effects of BMP9/10 antibodies and CDK4/6i treatment on lungs, brains, and intestinal tracts, which are commonly affected by vascular malformations in HHT patients. To visualize vascular anomalies in these tissues, we performed intracardiac injections of blue latex dye in P7 control and anti-BMP9/10–treated pups, with or without CDK4/6i treatment following the dosing protocol as shown in Figure 2A. In the lungs of anti-BMP9/10–treated animals, blue latex dye injections revealed vasodilated intrapulmonary arteries of first, second, and third order and perfusion of a dense network of disorganized capillaries compared with controls (Figure S2C). Anti-BMP9/10–treated animals that received CDK4/6i treatment exhibited significantly fewer vascular anomalies and intrapulmonary microvasculature vasodilation (Figure S2D). Similarly, in the intestinal tracts, we found significant vasodilation of the small capillaries perfusing the duodenum in anti-BMP9/10–treated mice that was not prevented by CDK4/6i treatment during the dosing timeframe used herein (Figure S2E and S2F). Conversely, in the brains of anti-BMP9/10–treated animals, we did not detect defects in the cerebral vasculature, notably in the middle cerebral artery and basilar artery (Figure S2G and S2H). Taken together, our data show that treatment with BMP9/10 antibodies induces significant vascular anomalies, particularly in the lungs and intestinal tracts. It is interesting that CDK4/6i treatment shows a preventive effect only in the lungs.

Next, we evaluated the effects of CDK4/6i in mice expressing the Fucci2 cell cycle reporter following the same dosing protocol outlined in Figure 2A. BMP9/10 blocking antibody treatment caused a significantly higher proportion of actively cycling ECs in S/G2/M phases in arteries, veins, and capillaries (Figure 2H and 2I), as well as in the distal plexi (Figure S3A), compared with controls, consistent with data shown in Figure 1C. In pups treated with CDK4/6i, after BMP9/10 blocking antibody treatment, vascular malformations were absent, and ECs in all vessel types were predominantly in late G1 state (Figure 2H and 2I). Because palbociclib is a known cell cycle modulator, we assessed its effects on EC cycle state in the retinal vasculature. First, bulk RNA sequencing analysis revealed that mRNA levels of key genes promoting cell cycle progression, including Ccnd2 (cyclinD2), Cdk6, and E2f5, were highly enriched in retinal ECs isolated from anti-BMP9/10–treated pups. In contrast, mRNA levels of genes that promote cell cycle arrest were highly enriched in ECs from pups receiving BMP9/10 antibodies, followed by CDK4/6i, such as Cdkn2b (P15), Trp53 (P53), and Cdkn1a (P21) (Figure 2J). Furthermore, as previously shown,¹⁵ CDK4/6i treatment of human umbilical vein ECs (HUVECs) induced hypo-phosphorylation of Rb (retinoblastoma protein) (Figure S3B), which is known to induce cell cycle arrest by inhibition of E2F transcription factor activity¹⁵ (Figure S3C). These results suggest that palbociclib prevents ECs hyperproliferation and arteriovenous shunt formation by enabling the expression of genes that collectively promote endothelial late G1 state (as schematized in Figure S3C).

Finally, we assessed the effects of palbociclib treatment on arteriovenous identity in anti-BMP9/10–treated animals following the dosing protocol shown in Figure S4A. We found that JAG1 (Jagged1) protein expression is enriched in arterial ECs in control animals and is decreased in anti-BMP9/10–treated retinal vasculature (Figure S4B and S4C). JAG1 protein expression was increased in arterial ECs in animals that received CDK4/6i, after BMP9/10 antibody treatment, compared with animals that received only BMP9/10 antibodies (Figure S4B and S4C). These results are consistent with our previous studies showing that endothelial late G1 state is required to enable arterial specification¹⁶ and, herein, prevent arterial specification defects. In addition, retinal arteries in anti-BMP9/10–treated animals exhibited reduced smooth muscle cell (SMC) coverage, as assessed by αSMA (α smooth muscle actin)–expressing cells, compared with control tissues. It is interesting that CDK4/6i treatment prevented the loss of SMC coverage on arteries (Figure S4D and S4E). Finally, ENDOMUCIN, which is enriched in veins and capillaries in control retinas, was ectopically expressed in arteries and arteriovenous shunts in the retinal vasculature of anti-BMP9/10–treated animals compared with controls (Figure S4F and S4G). However, CDK4/6 inhibition did not prevent the arterial expression of ENDOMUCIN (Figure S4F and S4G) during the short window of time between the end of treatment and the collection of retinas for analysis. Overall, these results demonstrate that endothelial late G1 cell cycle state, induced by palbociclib treatment, prevents AVM formation and loss of arteriovenous identity in HHT.

Other studies reported that defective EC migration is involved in the genesis of arteriovenous shunts in HHT mouse models.^17,18 Herein, we performed gene ontology analysis of bulk RNA sequencing data from retinal ECs isolated from P7 pups treated with anti-BMP9/10, with or without CDK4/6i, as in Figure 2A. We found enriched expression of several gene families that regulate cell migration in retinal ECs isolated from anti-BMP9/10– and CDK4/6i-treated animals compared with mice that received only BMP9/10 antibody treatment (Figure 3A). Further analysis of differentially expressed genes revealed enrichment of genes, such as Slit2, Kdr, Rac1, Nrp1, and Esm1, which are known to control EC migration and angiogenesis^19–21 (Figure 3B), in retinal ECs isolated from animals treated with anti-BMP9/10 and CDK4/6i.

To assess flow-mediated EC polarized migration in response to anti-BMP9/10, with or without CDK4/6i, we coimmunostained P7 retinas with anti-GM130, a peripheral membrane protein located in the cis-Golgi, and IB4 and anti-ERG (ETS-related gene) 1,2,3 to label ECs. To analyze the orientation of the Golgi toward the direction of the flow in retinal vessels, we measured the angle formed by the vector nucleus/Golgi in each EC and the predicted blood flow vector in the vessel (as depicted in Figure 3C). We also quantified the EC polarization using a polarity index²² ranging from 1 (highly polarized) to 0 (random distribution). As previously described,^17,22 in control retinas, both venous and arterial ECs are oriented in the opposite direction of blood flow, with the Golgi positioned in front of the nuclei (Figure 3D through 3F). In retinas from mice treated with BMP9/10 antibodies, venous ECs exhibited a nonpolarized pattern toward blood flow (Figure 3D through 3F). However, in mice treated with anti-BMP9/10 and CDK4/6i, the polarity of retinal ECs in veins and arteries was oriented similarly to controls (Figure 3D through 3F). Altogether, these results demonstrate that late G1 state induced by CDK4/6 inhibition promotes the expression of genes regulating EC migration and prevents EC migration defects caused by BMP9/10 deficiency.

Previous studies revealed that specific metabolic pathways control EC function and behavior.²³ For example, quiescent or proliferating ECs use different metabolic pathways for energy production. A metabolic rewiring towards glycolysis and mitochondrial pathways occurs in ECs when they are activated during both physiological and pathological angiogenesis. In addition, it has been shown that activation of specific metabolic pathways is associated with cell cycle progression.²⁴ However, such metabolic changes in ECs during the formation of AVMs have not been investigated. RNA sequencing analysis of P7 retinal ECs isolated from control and anti-BMP9/10–treated mice revealed enriched mRNA expression of genes that regulate glycolysis, mitochondrial energy production through the tricarboxylic acid or Kreb cycle, and fatty acid signaling in response to anti-BMP9/10 treatment (Figure 4A), which is consistent with increased ECs energy production/consumption associated with AVM development. It is interesting that we found that late G1 cell cycle state induced by palbociclib treatment led to decreased mRNA expression of genes controlling glycolysis, tricarboxylic acid cycle, and fatty acid pathways in retinal ECs (Figure 4B).

Finally, we analyzed the effect of CDK4/6 inhibition on the metabolic activity of HUVECs using the Agilent Seahorse Assay. The mitochondrial stress test was performed measuring basal oxygen consumption rate (basal respiration), and after injection of 1 mM oligomycin, 2 mM BAM15 (respiratory capacity), 10 mM antimycin A, and 1 mM rotenone (nonmitochondrial oxygen consumption, used for normalization; Figure 4C). We found that cells pretreated with palbociclib (3 mg/mL) before analysis exhibited significantly lower mitochondrial basal respiration and respiratory capacity compared with control cells (Figure 4C through 4E). These results correlate with the inhibition of genes regulating mitochondrial tricarboxylic acid cycle induced by palbociclib in ECs isolated from anti-BMP9/10–treated mice (Figure 4B). Altogether, our data suggest that the induced late G1 state prevents the upregulation of genes responsible for the pathological metabolic rewiring associated with EC hyperproliferative state because of BMP9/10 signaling deficiency.

Nearly 50% of patients with HHT harbor mutations in the Acvrl1 gene,¹ which encodes for the ALK1 receptor. To validate the clinical relevance of the use of palbociclib for the treatment of HHT, we crossed Cdh5Cre^ERT2 mice with Alk1^flox/flox mice (6–8 weeks old) to generate an endothelial-specific inducible deletion of Alk1 (referred to as Alk1EC^iKO), which is an established preclinical mouse model of human HHT type 2.^12,25 Neonatal Alk1EC^iKO pups received a single intraperitoneal injection of tamoxifen at P3 to induce Alk1 gene deletion and were then treated with palbociclib by oral gavage at P4, and their retinal vasculature was analyzed at P5 (Figure 5A). Although untreated Alk1EC^iKO animals developed numerous arteriovenous shunts, those that received palbociclib exhibited a significantly lower number of AVMs (Figure 5B and 5C). Palbociclib treatment also prevented the formation of hyperdense and disorganized vascular networks at the front of the retinal vascular plexi in Alk1EC^iKO animals (Figure 5B and 5D) and normalized vascular progression (Figure 5B and 5E). It is interesting that, unlike in the brains of mice treated with BMP9/10 blocking antibodies (Figure S2G and S2H), the brains of Alk1EC^iKO mice (Figure 5F) exhibited significant dilations of the basilar and middle cerebral arteries (Figure 5G and 5H; Figure S5C) compared with control brains. In mice treated with a single dose of palbociclib at P4, we observed a trend toward normalization of the diameter of the basilar and middle cerebral arteries (Figure 5G and 5H). Because of the relatively short survival time after tamoxifen induction in Alk1EC^iKO mice (approximately 48 hours), our study is limited to a single palbociclib administration, which represents a technical challenge in exploring additional treatments. However, the number of latex-filled veins in Alk1EC^iKO brains treated with palbociclib was significantly reduced when compared with untreated animals (Figure 5I). In the lungs of Alk1EC^iKO, we observed strong latex blue dye extravasation across the intrapulmonary bed. In Alk1EC^iKO mice treated with palbociclib, dye extravasation was reduced (Figure S5A). Finally, as previously observed,^12,26 in the intestinal tracts of P7 Alk1EC^iKO animals, we noticed the presence of veins filled with blue latex, demonstrating the presence of arteriovenous shunts in the gastrointestinal tract that were absent in Alk1EC^iKO animals treated with CDK4/6i (Figure S5B). These results demonstrate that palbociclib prevents the development of vascular malformations in major organs affected in patients with HHT, and more robustly affects brain AVMs induced by endothelial loss of Alk1 function.

To gain a deeper understanding of the molecular mechanisms underlying the effects of palbociclib on endothelial identity and cell cycle state, and vascular malformations in Alk1EC^iKO mice, we used an scRNAseq approach. We isolated retinal ECs from P5 Alk1^fl/fl, Alk1^fl/fl treated with CDK4/6i, Alk1EC^iKO, and Alk1EC^iKO treated with CDK4/6i animals (Figure 6A). After retinal tissue dissociation, we used fluorescence-activated cell sorting to isolate ECs (CD31⁺CD45^-) that were used for scRNAseq analysis. After filtering, samples were processed for single-cell barcoding and downstream mRNA library preparation and sequencing (Figure 6B). Relative mRNA expression of genes known to be associated with arterial, venous, capillary, proliferative, and tip cell identities were used to annotate the 5 EC populations (Figure 6C and 6D).

To investigate the effects of CDK4/6i on arteriovenous identity in HHT, we used the PHATE method (Potential of Heat-Diffusin for Affinity-Based Trajectory Embedding). This method evaluates the relative gene expression levels among clusters within scRNAseq data sets to predict lineage relationships among the populations. As previously published by our group, we used arterial- or venous-enriched gene expression data to generate arterial and venous identity module “scores” for each EC cluster in the scRNAseq data set.⁸ Individual EC scores were then visualized on the PHATE plots (Figure 6E and 6G). As expected, in Alk1^fl/fl control ECs, cells in the arterial or venous clusters displayed a high arterial or venous score, respectively (Figure 6E and 6G). However, in retinal ECs from Alk1EC^iKO, both arterial and venous clusters had low identity scoring, consistent with a loss of arteriovenous identity in HHT conditions (Figure 6E and 6G). More specifically, the mRNA expression of genes associated with arterial identity, including Efnb2, Gja4, Bmx, Jag1, and Unc5b, were downregulated in the arterial cluster (Figure 6F), whereas the mRNA expression of genes associated with venous identity was either downregulated such as Ephb4 and Nr2f2, or upregulated such as Nrp2 and Emcn, in the venous cluster (Figure 6H). In the Alk1EC^iKO animals treated with CDK4/6i, the arterial cluster presented a higher arterial score (Figure 6E) and a higher mRNA expression of all genes associated with arterial identity compared with the Alk1EC^iKO group (Figure 6F). On the venous identity scoring and mRNA expression of venous genes in the venous cluster, we did not observe differences between the Alk1EC^iKO and Alk1EC^iKO treated with CDK4/6i groups (Figure 6G and 6H). In the Alk1^fl/fl animals treated with CDK4/6i, palbociclib did not affect arterial or venous scoring or the expression of arterial and venous genes in their respective clusters compared with the Alk1^fl/fl group (Figure 6E through 6H). These results demonstrate that, in the Alk1EC^iKO mouse model of HHT, at a single cell level, both arterial and venous identity of ECs are profoundly dysregulated. Palbociclib treatment prevents the downregulation of arterial-enriched genes in retinal ECs, whereas the expression of venous identity genes was not different within the timeframe of the experimental dosing.

Because endothelial identities and cell cycle states are closely linked,⁸ we further investigated the effects of palbociclib on EC cycle state using Alk1EC^iKO mice expressing the Fucci2 reporter. Alk1EC^iKO-Fucci2 pups received a single intraperitoneal injection of tamoxifen at P3 to induce Alk1 gene deletion, they were then treated with palbociclib by oral gavage at P4, and their retinal vasculature was analyzed at P5 (Figure 7A). In Alk1EC^iKO-Fucci2 pups, we found a significant increase in ECs actively cycling in the S/G2/M phases in arteries, veins, and capillaries compared with controls (Figure 7B and 7C), consistent with the hyperproliferative state of ECs observed in patients with HHT (Figure 1E). However, in Alk1EC^iKO pups treated with CDK4/6i, the percentage of ECs residing in the S/G2/M phases was significantly lower in arteries and capillaries, concomitant with a higher proportion of ECs in late G1 state; no significant effects on venous ECs were observed (Figure 7B and 7C).

In addition, using previously generated cell cycle state–specific gene expression data,⁸ we generated and applied “early G1” and “late G1” cell cycle state scores to each EC in the scRNAseq data set and visualized the results on PHATE plots (Figure 7D and 7E). In accordance with our previous studies,⁸ in the Alk1^fl/fl group, the arterial cluster displayed a high late G1 score (Figure 7D) and the venous cluster a high early G1 score (Figure 7E). The late G1 score was lower in the arterial cluster of the Alk1EC^iKO group and was higher when Alk1EC^iKO animals were treated with CDK4/6i (Figure 7D). These results are consistent with a loss of arterial identity in ECs from Alk1EC^iKO animals and a prevention of arterial identity loss in ECs from Alk1EC^iKO treated with CDK4/6i. Similarly to the venous scoring in other analyses (Figure 6G), the early G1 score in the venous cluster did not show any obvious differences between Alk1EC^iKO animals and Alk1EC^iKO animals treated with CDK4/6i (Figure 7E). Similar results were also found when assessing venous EC cycle state in Alk1EC^iKO-Fucci2 mice treated with CDKi4/6i (Figure 7C).

Last, because we previously showed that the early G1 state was essential for BMP-induced venous genes and late G1 state was essential for TGF-β1–induced arterial gene expression,⁸ we evaluated the differential expression of key genes of the BMP and TGF-β pathways in the scRNAseq data sets. Correlating with a loss of arteriovenous identity in HHT, in Alk1EC^iKO ECs, we found lower expression of genes of the TGF-β pathway, such as Tgfbr1, Tgfbr2, small mothers against decapentaplegic (Smad) 6, and Tgfb2 (Figure S6A), as well as genes of the BMP pathway, including Eng (Figure S6B), compared with the control group. In ECs isolated from Alk1EC^iKO mice that received palbociclib treatment, the expression of those genes was higher (Figure S6A and S6B).

To decipher the molecular mechanism(s) underlying the effects of palbociclib-induced EC cycle arrest on the prevention of AVMs, we tested the effects of CDK4/6 inhibition on VEGF-A and BMP9-mediated signaling, which are known to be dysregulated in HHT. As expected, Western blot analysis revealed that a 15-minute VEGF-A stimulation of HUVECs leads to increased pAKT and pERK1/2 (Figure 8A and 8B). It is interesting that pretreatment with palbociclib for 24 hours before VEGF-A stimulation significantly decreased pAKT and pERK1/2 activation. In addition, palbociclib pretreatment of HUVECs enhanced SMAD1/5/8 phosphorylation in response to BMP9 (Figure 8B and 8C). Next, we treated HUVECs with Alk1 small interfering RNA to silence Alk1 and test the VEGF-A–induced response, with or without palbociclib treatment, using pAKT activation as a readout. We did not detect any differences in the pAKT activation induced by VEGF-A between Alk1-silenced HUVECs treated, or not, with palbociclib (Figure S7A and S7B). Consistent with this observation, the scRNAseq data did not show any obvious differences in the expression of genes associated with the AKT pathway between retinal ECs isolated from Alk1^iECKO mice treated, or not, with palbociclib (Figure S7C).

With bulk RNA sequencing analysis, we further investigated the effects of palbociclib on these pathways in vivo on P7 retinal ECs from control pups or those treated with anti-BMP9/10, with or without CDK4/6i, as outlined in Figure 2A. We found enhanced mRNA expression of several genes in the BMP9/10 signaling pathway, such as Eng, but also the antiangiogenic gene FMS-like tyrosine kinase 1 (Flt1), in pups treated with CDK4/6i, after anti-BMP9/10 treatment (Figure 8D). In Alk1EC^iKO mice, we confirmed that palbociclib promoted the expression of ENDOGLIN in venous ECs at both mRNA (Figure S6B) and protein (Figure 8E and 8F) levels. We also investigated the potential role of FLT1, a decoy receptor for VEFG-A, in the mechanism of action of palbociclib, by injecting intraperitoneal FLT1 blocking antibodies in Alk1EC^iKO mice simultaneously with CDK4/6i treatment (Figure 8G). We found that, as expected, Alk1EC^iKO mice that were treated with palbociclib developed fewer arteriovenous shunts than untreated Alk1EC^iKO animals; however, palbociclib-treated Alk1EC^iKO mice that also received FLT1 antibodies exhibited numerous arteriovenous shunts, comparable with untreated Alk1EC^iKO mice (Figure 8G and 8H). These results demonstrate that blocking FLT1 counteracts the effects of palbociclib and emphasizes a mechanism of action inhibiting VEGF-A signaling by the decoy activity of VEGFR1/FLT1. Collectively, these results suggest that the cell cycle arrest, induced by palbociclib, enables the expression of genes and proteins in the BMP9/10 signaling pathway that help to restore this defective signaling axis in HHT and prevent overactivation of proangiogenic signaling.

Expanded Methods

Figures S1–S7

References 43–51