Shear-induced Notch-Cx37-p27 axis arrests endothelial cell cycle to enable arterial specification

Establishment of a functional vascular network is rate-limiting in embryonic development, tissue repair and engineering. During blood vessel formation, newly generated endothelial cells rapidly expand into primitive plexi that undergo vascular remodeling into circulatory networks, requiring coordinated growth inhibition and arterial-venous specification. Whether the mechanisms controlling endothelial cell cycle arrest and acquisition of specialized phenotypes are interdependent is unknown. Here we demonstrate that fluid shear stress, at arterial flow magnitudes, maximally activates NOTCH signaling, which upregulates GJA4 (commonly, Cx37) and downstream cell cycle inhibitor CDKN1B (p27). Blockade of any of these steps causes hyperproliferation and loss of arterial specification. Re-expression of GJA4 or CDKN1B, or chemical cell cycle inhibition, restores endothelial growth control and arterial gene expression. Thus, we elucidate a mechanochemical pathway in which arterial shear activates a NOTCH-GJA4-CDKN1B axis that promotes endothelial cell cycle arrest to enable arterial gene expression. These insights will guide vascular regeneration and engineering.

stablishment of a well-organized and perfused circulatory system is essential to oxygenate tissues and remove metabolic waste. When new blood vessels form, during development or in response to tissue injury, newly generated endothelial cells rapidly proliferate and coalesce into disorganized capillary plexi. Coincident with the onset of blood flow through vessel lumens, endothelial cell proliferation is reduced and primitive vessels remodel into arterial-venous networks that acquire mural cell coverage (reviewed in Ribatti et al. 1 ). Although we have made progress in identifying factors that stimulate endothelial cell proliferation and sprouting (reviewed in Marcelo 2013a 2 ), limited understanding of the regulation of endothelial cell growth suppression and phenotypic specialization during vascular remodeling remains a significant roadblock for clinical therapies, tissue engineering and regenerative medicine.
We recently found that endothelial cells require NOTCHinduced cell cycle arrest via regulation of CDKN1B (commonly, p27) for acquisition of a hemogenic phenotype that enables blood-forming potential 16 . Since NOTCH is also implicated in arterial 11 , as well as lymphatic 17 , endothelial cell development, we considered whether NOTCH might play a common role in these processes. That is, perhaps NOTCH-induced cell cycle arrest is required for endothelial cells to acquire all of these specialized phenotypes and functions. Indeed, cell cycle state of undifferentiated embryonic stem cells strongly influences cell fate decisions 18 , but it is unclear whether a similar mechanism applies to endothelial cell specification. We, therefore, investigated whether NOTCH signaling mediates flow-induced endothelial cell growth control, and whether endothelial cell cycle state determines their propensity to acquire an arterial identity.
Examining both post-natal retina neovascularization and cultured endothelial cells, we define a novel signaling pathway whereby FSS, at arterial magnitudes, maximally activates NOTCH signaling, which upregulates GJA4, more commonly known as Connexin37 (Cx37), and downstream CDKN1B to promote   endothelial G1 arrest and to enable expression of arterial genes. This link between endothelial cell cycle and cell fate was not previously known, and is critically important for controlling blood vessel development and remodeling. Insights gained from these studies will facilitate efforts to optimize vascular regeneration of injured and diseased tissues in vivo and blood vessel engineering ex vivo.

Results
Flow-dependent endothelial quiescence is mediated by NOTCH. Preliminary experiments confirmed that physiological FSS (12 dynes/cm 2 ) suppressed the incorporation of EdU, a measure of DNA synthesis and indicator of proliferation, in human umbilical vein endothelial cells (HUVEC) at 12-24 h. To identify mediators of flow-dependent endothelial cell quiescence,   Table 1). Several NOTCH-associated genes, including ligands DLL1, DLL4, JAG1, and JAG2, and canonical NOTCH transcriptional targets HES1, HES2, HEY1, and HEY2, were altered by shear (Fig. 1a). Interestingly, mRNA levels for endothelial-expressed receptors NOTCH1 and NOTCH4 were not affected by FSS. Activation of shear-dependent signaling was confirmed by strong upregulation of KLF2, KLF4, and COX genes.
To verify that flow induces NOTCH signaling, HUVEC were exposed to shear. Cleavage of the NOTCH intracellular domain (NICD), which mediates gene transcription upon receptor activation, was increased with 1 h FSS and blocked by pretreatment with 10 µM DAPT, a pharmacological inhibitor of NOTCH cleavage and signaling ( Fig. 1b and Supplementary  Fig. 1). In addition, NOTCH target genes HEY1 and HEY2 were significantly upregulated by 16 h of FSS (Fig. 1c). Inhibiting NICD cleavage with 10 µM DAPT also significantly alleviated FSS-mediated suppression of endothelial cell EdU incorporation (Fig. 1d). Altogether, these data show that NOTCH signaling mediates shear-induced endothelial cell growth suppression.
GJA4 mediates endothelial quiescence downstream of NOTCH. To identify genes regulated by NOTCH under flow, RNA-seq was performed on HUVEC exposed to 6 h of FSS with or without 10 µM DAPT. Of the top five genes, only GJA4 (which encodes GJA4, more commonly known as Connexin37, or Cx37) and TMEM100 (which encodes Transmembrane Protein 100) were both upregulated by shear and suppressed in the presence of DAPT (Supplementary Table 2 and Supplementary Fig. 2). Of these, GJA4 was selected for further study because of its endothelial-specific expression during development 19 , and in vivo evidence of its involvement in blood 20,21 and lymphatic 17,22-24 vascular morphogenesis. We found that GJA4 protein was elevated after 6 h FSS in association with sustained upregulation of NICD ( Fig. 2a and Supplementary Fig. 3). GJA4 knockdown (si-GJA4, validated in Supplementary Fig. 4) impaired shear-induced suppression of cell proliferation, as did DAPT (si-Ctrl+DAPT), while si-GJA4 combined with DAPT had no additional effects (Fig. 2b). Altogether, these data suggest that GJA4 is the predominant mediator of NOTCH-and flow-induced endothelial cell growth arrest.
NOTCH signaling regulates GJA4 expression. To determine whether NOTCH regulates GJA4, we used confluent endothelial cell monolayers as a convenient system, in which there is basal activation of NOTCH through ligand-receptor interactions. As expected, endothelial cells exhibited density-dependent NOTCH activation, which was suppressed by DAPT inhibition of NICD cleavage ( Supplementary Fig. 5), resulting in decreased endogenous HES1 mRNA levels ( Supplementary Fig. 6A). DAPT treatment of confluent endothelial cells also reduced mRNA and protein levels for GJA4 (Fig. 2c, d and Supplementary Fig. 3). Basal NOTCH signaling was undetectable in sub-confluent HUVEC ( Supplementary Fig. 5), but induced by seeding subconfluent cells onto immobilized recombinant DLL4 ligand to constitutively activate NOTCH leading to upregulation of HES1 mRNA ( Supplementary Fig. 4B). Seeding of sub-confluent HUVEC onto recombinant DLL4 also upregulated GJA4 mRNA and protein, along with increased NICD (Fig. 2e, f). NOTCH stimulates transcription of target genes through binding of cleaved NICD to a nuclear gene regulatory complex that includes the DNA-binding protein RBP-Jκ. In silico analysis revealed several high-probability RBP-Jκ binding motifs in the regulatory regions of the mouse and human GJA4 genes including several motifs adjacent to the transcriptional start site (Supplementary Table 3). To determine whether GJA4 is a direct target of the NICD-RBP-Jκ complex, we performed chromatinimmunoprecipitation (ChIP) on confluent HUVEC monolayers (in which NOTCH is basally active, Supplementary Fig. 5). Immunoprecipitates of RBP-Jκ from these cells revealed enrichment of the GJA4 promoter region relative to the IgG control, which was blocked by DAPT (Fig. 2g). Thus, GJA4 appears to be a transcriptional target of NICD/RBP-Jκ.
Endothelial cells express two NOTCH receptors, NOTCH1 and NOTCH4, both capable of inducing canonical NOTCH signaling and downstream transcriptional activation. Expression of each NOTCH receptor was selectively reduced by siRNA against NOTCH1 or NOTCH4, respectively ( Supplementary Fig. 6C). Knockdown of either NOTCH1 or NOTCH4 abolished DLL4induced upregulation of HES1 ( Supplementary Fig. 6D); however, only si-NOTCH1, and not si-NOTCH4, eliminated DLL4-induced upregulation of GJA4 mRNA (Fig. 2h), suggesting that GJA4 expression is specifically regulated by NOTCH1.
GJA4 is expressed in remodeling blood vessel endothelium. To examine the role of GJA4 in vascular remodeling in vivo, we used the well-studied neonatal murine retinal vascularization model 12,[25][26][27] . Beginning at post-natal day 1 (P1), rapid, highly stereotyped vascularization initiated from a central vascular plexus can be visualized over subsequent days by labeling endothelial cells using either antibodies targeting PECAM1 (or, CD31) or with fluorescently conjugated isolectin B4 (IB4). ( Supplementary Fig. 7A). Over subsequent days, endothelial cells at the retinal vascular edge proliferate and migrate radially outward, while nascent blood vessels interior from the vascular front remodel, mature and recruit smooth muscle actin (SMA)-positive mural cells to form a functional circulatory network. At P6, the developing retinal vasculature includes an outer region of active vessel remodeling (R) and a more central region with mature (M), specified arteries and veins. For our studies, we designated the R-M boundary (dashed yellow line) as the distance of SMA investment of arteries, (Supplementary Fig. 7B), which was 519 ± 64μm (n = 6, multiple litters) from the vascular front at P6. Intracardiac injection of rhodamine dextran revealed that vessels of the remodeling plexus were lumenized and were accessible to systemic blood flow ( Supplementary Fig. 7C). This finding was consistent with in silico models of retinal blood flow, which show that the remodeling plexi experience near-arterial levels of shear stress 28 . In addition, activated ERK5 (pERK5), which is acutely induced by flow in endothelial cells in culture 29 , was abundant in remodeling retinal vessels ( Supplementary Fig. 7D).
In wild-type (WT) P6 mice, GJA4 was expressed in endothelial cells of the remodeling vessel plexus and in associated maturing arteries (Fig. 3a, b), but not at the vascular edge (Fig. 3c) where FSS is limited 28 and where endothelial cells are highly proliferative 13 . In adult tissues, GJA4 is typically observed coexpressed with GJA5 (commonly, Connexin40 or Cx40) in large arteries; 30,31 however, in the developing neonatal mouse retina, GJA5 was restricted to SMA-positive arteries ( Supplementary  Fig. 8), and not detected in remodeling vessels at the R-M boundary where GJA4 was highly expressed (Fig. 3b).
Loss of GJA4 or NOTCH signaling causes remodeling defects. GJA4 is implicated in blood vessel formation, as mice deficient in GJA4 exhibit increased tissue vascularization [20], enhanced neovascularization of injured tissues 20,32 , and dysregulation of venous valve formation 21 . However, developmental blood vessel formation has not been well studied in these mice. We therefore examined vascularization of Gja4 −/− , Gja4 +/− , and Gja4 +/+ (WT) neonatal retinas. Vascular morphology was similar between Gja4 +/− and age-matched Gja4 +/+ controls (Supplementary Fig. 9A (Fig. 3d) compared to agematched Gja4 +/+ controls, and there was a~25% increase in vascular area and vessel branchpoint number within the remodeling zone of Gja4 −/− mice ( Fig. 3f-g). Hyperdensity was observed in nearly 80% of imaged remodeling plexi from Gja4 −/− retinas, indicating high penetrance of this phenotype (Supplementary Fig. 9D). Consistent with the absence of detectible GJA4 in endothelial cells at the vascular edge, GJA4 deletion had no effect on tip cell number ( Supplementary Fig. 10).
The Gja4 −/− phenotype resembles the hyperdense vasculature observed with disrupted NOTCH signaling 12,25 , such as with DAPT treatment of WT mice or with tamoxifen-induced endothelial-specific deletion of NOTCH1 (Notch1 iECKO ) (Fig. 3d). Pharmacological inhibition of endothelial NOTCH signaling via 24 h DAPT was confirmed by observation of a~50% decrease in expression of canonical NOTCH target genes (Hes1, Hey1 and Hey2) in retinal endothelial cells (Fig. 3e). Like Gja4 −/− animals, DAPT-treated WT and Notch1 iECKO -Cre+ animals experienced ã 25% increase in vascular area and branchpoint number in the remodeling zone (Fig. 3f, g) compared to associated controls. However, unlike with GJA4 deletion, NOTCH inhibition (via DAPT) uniquely increased tip cell number ( Supplementary  Fig. 10), suggesting that GJA4 mediates distinct NOTCH effects in the remodeling vasculature. Consistent with this, GJA4 expression was greatly reduced in remodeling vessels of both DAPT-treated WT and Notch1 iECKO retinas (Fig. 3h).
To determine whether Gja4 −/− retinas exhibited vascular morphology defects at later stages of development, we assessed the retinal vasculature at P21, at which time the superficial vascular plexus established in P3-9 ( Supplementary Fig. 7A) has undergone additional remodeling and regression to form a mature network. Meanwhile, smaller vessels have sprouted from the superficial vasculature beginning at P10, and have invaded downward to form a secondary deep vascular plexus below the superficial network. Consistent with our observations at P6, we observed a significant increase in vascular area in the deep vascular plexus of Gja4 −/− mice compared to Gja4 +/− littermates ( Supplementary Fig. 11A). In addition, we observed a significant increase in the number of arterial branches in the superficial plexus of Gja4 −/− compared to littermate controls (Supplementary Fig. 11B) even though overall vascular area was not measurably different at this stage ( Supplementary Fig. 11A). These data suggest that while many of the vessels of the hyperdense Gja4 −/− remodeling plexus ( Fig. 3d) may eventually regress, Gja4 −/− animals retain a vascular hyperbranching phenotype in the mature retinal circulation, which is consistent with our previous observations of excessive collateral connections in the pial circulation and increased hindlimb vascularization in response to ischemic damage 20,32 .
Next, we assessed whether GJA4 was responsible for the effects of NOTCH on endothelial cell cycle status. We transfected HUVEC with full-length mouse Gja4 (lenti-Gja4 33 ) to constitutively express GJA4 (or with lenti-Ctrl for control cells). As previously described 33 , lenti-Gja4 effectively induced normal levels of GJA4 protein, which properly localized to cell-cell junctions ( Supplementary Fig. 12A). Cells were then treated for 24 h with 10 µM DAPT, which we previously showed effectively blocks NOTCH signaling (Figs. 1, 2 and Supplementary Fig. 6), and cell cycle distribution was assessed. Control (DMSO-treated) cells had the following cell cycle distribution: 27.8 ± 4.9% in G0, 52.5 ± 3.8% in G1 and 19.6 ± 2.1% in S/G2/M. Compared to controls, DAPT-treated HUVEC, in which GJA4 expression is significantly reduced (Fig. 2d), exhibited a decreased proportion of cells in G1, and an increased proportion of actively cells in S/G2/M. However, constitutive re-expression of GJA4 abolished these DAPT-mediated cell cycle effects (Fig. 4d), indicating that NOTCH regulates cell cycle status via GJA4. GJA4 regulates cell cycle status via CDKN1B. Cell cycle progression is tightly regulated by cyclins and associated regulatory proteins. To identify cell cycle regulators targeted by FSSactivated NOTCH and GJA4, we examined our whole RNA-Seq data sets to determine whether the expression of cell cycle effectors were altered by FSS, and whether that expression was further affected by DAPT. We found that several cell cycle regulators were altered by 6 h FSS (vs. 6 h Static) ( Supplementary  Fig. 13A), including the cyclin-dependent kinase inhibitor CDKN1B, which encodes for CDKN1B (commonly, p27), an inhibitor of G1-S transition. CDKN1B expression was upregulated by FSS, and was also significantly downregulated by DAPT under FSS ( Supplementary Fig. 13B), indicating that shearinduced CDKN1B was dependent upon NOTCH signaling.
To determine whether cell cycle effectors identified as both FSS-activated and NOTCH-dependent were also regulated by GJA4, we next assessed expression of candidate cell cycle effectors in sorted Gja4 −/− retinal endothelial cells (which we defined as events positive for PECAM1 (or, CD31) and negative for PTPRC (or, CD45)). Expression of CDKN1B (as well as its regulatory targets CDK4 and CCNE1) was significantly reduced in Gja4 −/− retinal endothelial cells compared to Gja4 +/− littermate controls (Fig. 5a); we detected no differences in the expression of other cycle effectors assessed. In WT (Gja4 +/+ ) P6 mice, CDKN1B was expressed in PECAM1-positive vessels of the remodeling plexus, but not detected in the remodeling plexus of Gja4 −/− P6 mice (Fig. 5b). Furthermore, FSS-induced upregulation of CDKN1B protein expression was abrogated in HUVEC when GJA4 expression was suppressed by si-GJA4 knockdown ( Fig. 5c and Supplementary Fig. 14A). We next investigated whether CDKN1B was necessary for GJA4 effects on endothelial cell cycle status. Control HUVEC treated with empty lentivirus (lenti-Ctrl) and control siRNA (si-Ctrl) had the following cell cycle distribution: 38.9 ± 0.02% in G0, 44.6 ± 0.02% in G1, and 16.9 ± 0.004% in S/G2/M. Constitutive expression of GJA4 (via lenti-Gja4) increased the proportion of cells in G1 and decreased the proportion of cells in S/G2/M (Fig. 5d). In contrast, siRNA knockdown of CDKN1B (si-CDKN1B, validated in Supplementary Fig. 12B, C) increased the proportion of actively cycling cells in S/G2/M and decreased cells in G1, regardless of GJA4 expression (Fig. 5d).
Similarly, HUVEC were transfected with si-GJA4 prior to lentiviral delivery of human CDKN1B (lenti-CDKN1B, validated in Supplementary Fig. 12D, E). The cell cycle distribution of HUVEC transfected with si-Ctrl and infected with lenti-Ctrl was like that of other control groups: 35.5 ± 1.3% in G0, 48.6 ± 1.0% in G1, and 15.6 ± 0.5% in S/G2/M (Fig. 5e). GJA4 knockdown (via si-GJA4) decreased the proportion of cells in G1 and increased the S/G2/M fraction (Fig. 5e). Constitutive expression of CDKN1B, whether in the presence or absence of GJA4, had the   Fig. 7 Endothelial cell cycle arrest, per se, enables arterial gene expression. a Treatment of HUVEC with 10 µM clotrimazole or 2 µM palbociclib reduced RB1, phosphorylated RB1 (pRb1), and E2F1, suggestive of G1 arrest. Clotrimazole tended to upregulate CDKN1B expression and preserve GJA4 expression, and CDK4 expression was lost only with CDK4/6i treatment. Uncropped blots presented in Supplementary Fig. 16. b Using FACS to assess cell cycle distribution, clotrimazole was found to arrest HUVEC in G1, even when CDKN1B was knocked down (via si-CDKN1B) (mean difference in cell cycle % ± . g We hypothesize that in remodeling vessels, arterial shear activates a novel Notch-Cx37-p27 signaling pathway that promotes endothelial cell cycle arrest to enable arterial gene expression opposite effect, increasing the proportion of cells arrested in G0/G1 and decreasing the proportion of actively cycling S/G2/M cells (Fig. 5e). Therefore, GJA4 regulates G1 arrest of endothelial cells via CDKN1B.
MAPK/ERK signaling is required for GJA4 regulation of CDKN1B. Although CDKN1B is transcriptionally regulated, CDKN1B protein intracellular localization and function are primarily influenced by its phosphorylation on serine and threonine residues 34 . A significant site of CDKN1B phosphorylation is serine 10 (S10) 35 , which regulates nuclear export 36 and protein stability 35,37 . In cycling cells, serine 10 phosphorylation is associated with CDKN1B ubiquitination and degradation to promote G1/S transition, whereas in quiescent cells, phosphorylation at serine 10 enhances protein stability to limit cell cycle progression 37,38 . We found that serine 10 phosphorylation (pCDKN1B (S10)) was reduced after GJA4 knockdown (via si-GJA4) in association with reduced total CDKN1B protein levels, which was rescued by re-expression of GJA4 ( Fig. 5f and Supplementary  Fig. 14B). These data suggest that GJA4 regulates CDKN1B stability and function and that this may involve serine 10 phosphorylation of CDKN1B protein.
Previous studies have shown that CDKN1B serine 10 phosphorylation is mediated by members of the MAPK family 35 . The GJA4 C-terminus contains a high-probability MAPK target consensus sequence 39 , and C-terminal truncation eliminates GJA4 suppression of cancer cell proliferation 40 , suggesting involvement of this site. To assess whether MAPK/ERK signaling is required for Cx37-mediated phosphorylation and stabilization of CDKN1B, we treated HUVEC for 1 h with 20 µM U0126, a specific inhibitor of MEK1/2, which activates MAPK/ERK signaling. U0126 blocked the capacity of re-expressed GJA4 to preserve serine 10 phosphorylation and total CDKN1B protein levels ( Fig. 5g and Supplementary Fig. 14C). These results demonstrate that GJA4 requires MAPK/ERK signaling to stabilize CDKN1B protein.
Arterial development requires GJA4 and NOTCH. To determine the effects of remodeling plexus hyperproliferation (Fig. 4) on vascular maturation, we assessed mural cell investment in maturing arteries. In WT tissue, SMA labeling extends from the center of the retina to the R-M boundary (Fig. 6a). By contrast, SMA staining was markedly reduced or absent in DAPT-treated WT retinas, as well as in Gja4 −/− retinas. Quantification of these images showed~20% decrease in the radial distance of SMApositive coverage in NOTCH-inhibited and Gja4 −/− animals (Fig. 6b), suggesting delayed arterial development. We further observed a significant decrease in the radial distance of GJA5 (commonly, Cx40) expression along the arteries of DAPT-treated WT and Gja4 −/− tissues (Fig. 6c), consistent with a defect in acquisition of endothelial arterial identity.
FSS-induced NOTCH-GJA4 signaling regulates arterial identity. In our RNA-seq studies, we found that 6 h of FSS significantly increased the expression of arterial identity genes GJA5 and EFNB2 (commonly, EphrinB2). We next explored whether GJA4 and CDKN1B function downstream of FSS-induced NOTCH signaling to promote arterial identity. We found that activation of NOTCH by DLL4 increased expression of EFNB2 (EphrinB2) (and GJA5) (Fig. 6d). However, knockdown of either GJA4 (via si-GJA4) or CDKN1B (via si-CDKN1B) prevented DLL4-induced upregulation of EFNB2, and significantly reduced GJA5 expression (Fig. 6d), indicating that GJA4 and CDKN1B function downstream of NOTCH signaling to enable arterial gene expression.
However, endothelial cells must be sensitive to the magnitude of flow during vascular morphogenesis in order to optimize arterial-venous specification and attain proper tissue perfusion 3 . To test this, endothelial cell monolayers were exposed to 1 h of shear over a wide physiological range of FSS magnitudes (0-50 dynes/cm 2 ). NOTCH cleavage (Fig. 6e), GJA4 expression, and expression of arterial genes EFNB2 and GJA5 (Fig. 6f) peaked at intermediate FSS magnitudes typical of arteries, and was reduced at magnitudes outside this range, demonstrating that arterial FSS activates NOTCH-GJA4 signaling to enable arterial specification.
Next, we used siRNA to knockdown NOTCH1, GJA4, and CDKN1B under arterial FSS magnitudes (18 dynes/cm 2 ) to determine the contribution of NOTCH-GJA4-CDKN1B signaling to shear-induced upregulation of arterial genes. NOTCH1, GJA4 and CDKN1B knockdown did not affect shear-activated KLF2 expression, indicating that mechanotransduction signaling complexes were intact (Supplementary Fig. 15). Basal and FSSinduced EFNB2 and GJA5 expression was reduced in all groups compared to control cells, and GJA4 knockdown (via si-GJA4) completely abolished FSS-induced upregulation of arterial identity genes (Fig. 6g).
Endothelial G1 arrest is needed for arterial specification. The above results prompted us to hypothesize that FSS-activated NOTCH-GJA4-CDKN1B signaling enables arterial specification via cell cycle arrest. However, as CDKN1B may also regulate transcription independent of cell cycle effects 34,41 , we tested whether cell cycle arrest, per se, mediates arterial gene expression. HUVEC were treated with 10 μM clotrimazole, which causes G1 arrest via a CKN1B-independent mechanism 42 (although it modestly upregulates CDKN1B as a secondary effect), or with 2 µM palbociclib 43 , which specifically eliminates expression of CDK4/6 without affecting CDKN1B or GJA4 expression ( Fig. 7a and Supplementary Fig. 16). Both drugs reduced total and phosphorylated levels of RB1, as well as E2F1, consistent with block of G1-to-S phase transition (Fig. 7a). Both drugs also increased the proportion of endothelial cells in G1, decreased endothelial cells in S/G2/M (Fig. 7b, d), and increased expression of arterial markers EFNB2 and GJA5 (Fig. 7c, e). Knockdown of CDKN1B had no effect on clotrimazole-induced G1 arrest (Fig. 7b) or on the upregulation of arterial genes (Fig. 7c). Collectively, these data indicate that G1 arrest enables arterial specification. Consistent with this model, evaluation of Cdt1-mOrange reporter mice revealed that endothelial cells in G1 are more abundant in arterial endothelial cells and surrounding plexi compared to veins (Fig. 7f).
Altogether, these data show that in remodeling endothelial cells, arterial shear activates a novel NOTCH-GJA4-CDKN1B (or, Notch-Cx37-p27) signaling pathway that promotes endothelial cell cycle arrest to enable arterial specification (Fig. 7g).

Discussion
During blood vessel remodeling, endothelial cell growth suppression and arterial-venous specification occur simultaneously, but how these processes are regulated and coordinated was not known. Herein, we demonstrate that endothelial G1 cell cycle arrest is required for expression of arterial-specific genes. In addition, FSS, specifically in the range typical of arterioles and arteries, initiates growth arrest by activating NOTCH signaling, which promotes endothelial cell cycle arrest via GJA4 and downstream CDKN1B in remodeling vessels. These findings are consistent with the observation that NOTCH signaling is low in veins 44 , and that upregulation of venous endothelial gene expression is not observed upon exposure to pulsatile blood flow 4 .
NOTCH signaling was previously thought to promote arterial development primarily via upregulation of fate-specific genes (reviewed in Marcelo et al. 2 ). However, our findings suggest a new paradigm in which the primary role of NOTCH signaling in arterial-venous specification is to modulate endothelial cell cycle state, similar to the role of NOTCH in hemogenic endothelial cell specification 16 . How endothelial cell cycle arrest in G1 leads to arterial specification is unknown, but studies in embryonic stem cells may offer a clue. Stem cells in different cell cycle phases respond differently to inductive signals 18 . For example, early G1 arrest enables responses to factors that promote mesodermal or endodermal vs. ectodermal commitment 45,46 . Arterial inductive signals may therefore be similarly confined to endothelial cells in G1. Chromatin structure, epigenetic regulation, and accumulation of transcription factors, cofactors, and miRNAs may all potentially contribute to cell cycle-dependent sensitivity to extrinsic cues.
The mechanism by which FSS at arterial magnitudes activates NOTCH is unknown. That NOTCH is acutely activated at 1 h following onset of shear makes transcriptional regulation unlikely. Furthermore, although NOTCH signaling is innately mechanosensitive 47 , the kinetics of NOTCH activation argue against direct activation of NOTCH by fluid shear forces. Instead, fluid shear stress-induced signal transduction could result in altered membrane-cytoskeletal rearrangements, as is observed within the CDH5 (VE-Cadherin)/PECAM1 (CD31) complex 48 and mechanically induced NOTCH cleavage 47 . Alternatively, shear signaling could modify stabilization and/or membrane trafficking of NOTCH and its ligands. Finally, shear-induced alterations in membrane lipid composition might alter gammasecretase complex activity to indirectly enhance NOTCH cleavage 49 . Thus, understanding activation of NOTCH by flow will require further investigation.
Also unknown is how GJA4, downstream of NOTCH, regulates CDKN1B phosphorylation and possibly MAPK/ERK signaling. In insulinoma cells, which typically do not express connexins, enforced expression of GJA4 suppresses proliferation 50 in a manner that requires both gap junction channel function and the C-terminal region of the protein, which is dispensible for gap junction channel formation and instead mediates intracellular signaling 33,40 . Interestingly, truncation of the GJA4 C-terminus also eliminates a high-probability MAPK consensus site 39 that may be required, either alone and/or in combination with other kinase target sites, for GJA4 to suppress cell growth. Further studies of GJA4 function and interactions in endothelial cells will be needed to address these questions.
Finally, although our studies were conducted in a developmental setting, the FSS-NOTCH-GJA4-CDKN1B (or FSS-Notch-Cx37-p27) pathway we characterize herein is likely also relevant to vascular remodeling in adults. Mice deficient in GJA4 20,32 or CDKN1B 51 display abnormal post-ischemic vessel growth and remodeling. Flow-induced Notch-Cx37-p27 endothelial cell cycle control is also likely to be important for maintenance and/or repair of adult vasculature. Thus, Notch-Cx37-p27 signaling may be an attractive candidate for treatment of vascular diseases, and for promotion of proper vessel formation in tissue engineering and regenerative medicine applications.

Methods
Cell culture. Primary human umbilical vein endothelial cells (HUVEC) were obtained from the Yale core facility and only used up to passage 8. The cells were cultured in M199 with 20% FBS (Gibco, Gemini), 100 µg/mL Heparin (Sigma), 50 µg/mL endothelial cell growth supplement (Sigma), and 1× penicillin/streptomycin (Sigma); some received 10 μM DAPT (Sigma) or DMSO, or were seeded onto surfaces pre-treated with 10 μg/mL recombinant human Dll4 (R&D). For siRNA knockdown, cells pre-conditioned with EGM-2 (Lonza) for >16 h were transfected with 20 nM si-GJA4 (Dharmacon L-011669-00), si-CDKN1B (Dharmacon L-040178-00), si-NOTCH1 (Dharmacon L-007771-00), si-NOTCH4 (Dharmacon L-011883-00) or control siRNA (Ambion AM4611) for >96 h. For lentiviral transduction, HUVEC were pre-infected for 48 h with lentivirus containing mouse Gja4 (lenti-Gja4) 33 , human CDKN1B (lenti-CDKN1B, cloned from Addgene 14049) or empty vector in presence of 10 μg/mL polybrene (Sigma). In MEK1/2 inhibitor studies, cells were washed with fresh EGM-2 media overnight, and then treated for 1 h with 20 µM U0126 (Sigma) or DMSO. In cell cycle studies, cells were treated with 24 h 10 μM clotrimazole (Sigma), 8 h 2 µM palbociclib (Sigma), or DMSO for the equivalent period. Cell cultures were routinely tested for mycoplasma contamination, and no positive test was obtained over the course of these studies. Shear. HUVEC were cultured on 25 × 55mm cell plastic slides pre-treated with 10 µg/mL fibronectin. Slides were then either maintained static in a cell culture incubator or mounted into parallel plate flow chambers and subjected to 3-50 dynes/cm 2 laminar shear for up to 24 h 52 . Flow was applied using a pressuredampened, gravity-driven dual reservoir system with media re-circulating via a peristaltic pump (Masterflex, Cole Palmer). Media was maintained at 37 C, 5% CO2 using a Digi-sense temperature control system (Cole Palmer) and humidified bubbler, respectively. Where indicated, cells were sheared in the presence of 10 µM DAPT. For EdU assays, 10 µM EdU was added for 2 h under shear prior to visualization (ThermoFisher Click-iT). Incorporation was quantified by imaging eight random fields at low magnification and averaging the frequency of EdUpositive nuclei vs. total nuclei. For some experiments, cleaved NOTCH receptor was detected by immunocytochemistry (Cell Signaling) following shear, and fluorescent intensity was quantified for~6500 cells per treatment group using Matlab.
RNA-Seq. HUVEC were exposed to shear (or static) conditions in triplicate, mRNA was isolated (Qiagen, RNeasy Plus Micro Kit), and cDNA libraries were prepared (ABM). Next-generation whole-transcriptome Illumina sequencing was performed with the Yale Center for Genome Analysis. RNA-Seq reads from each sample were aligned to human genome build 38 (GRCh38) using short reads aligner STAR 53 . Gene expression quantification was performed using RSEM 54 with GENCODE annotation (release 24: http://www.gencodegenes.org). Differential analysis was performed using edgeR 55 to identify genes with significant expression changes between groups (6 h Static vs. 6 h Shear; or 6 h Shear, Untreated vs. DAPT). Gene transcripts observed to change significantly (false discovery rate < 0.001) between static and shear groups were used for functional enrichment analysis by running GOseq 56 . A modified version of the nEASE algorithm 57,58 was used to assess functional enrichment of nested GOseq (nGO) terms. Briefly, each enriched upperlevel GOseq (GO) term was then used for nGO analysis to identify statistically enriched nGO terms driving upper-level functional enrichment of non-specific GO terms. GO-nGO paired terms relating to cell proliferation, cell signaling and development were selected for further consideration (Supplementary Data 1). Genes in Supplementary Data 1 were searched against established Kegg signaling modules (http://www.kegg.jp/kegg/tool/map_module.html); modules lacking upper-level block hits were discarded to yield relevant flow-sensitive signaling pathways.
Animals. All animals were on a C57Bl/6 background, and all studies were conducted in compliance with Yale's Institutional Animal Care and Usage Committee guidelines. Male and female Gja4 −/− animals 62 were compared to wild-type Gja4 +/+ or Gja4 +/− littermates or age-matched wild-type C57Bl/6 animals (Jackson #025697), as indicated. Gja4 −/− were also crossed with mice bearing a Cdt1-mOrange fusion protein 54 (gift of S. Guo) to generate Cdt1-mOrange+;Gja4 −/− mice (or Cdt1-mOrange+ ;Gja4 +/+ littermate controls). Some wild-type animals received two subcutaneous injections of 100 mg/kg DAPT (or drug vehicle) at 24 h and 12 h prior to killing; in these studies, littermates were randomized with regard to whether they receive drug or vehicle control. Notch1 iECKO animals were generated by crossing Notch1 flox/flox (Jackson #006951) mice with Cdh5CreERt2 mice, which express a tamoxifen-inducible Cre recombinase in endothelial cells driven by a CDH5 (VE-Cadherin) promoter. NOTCH1 was post-natally deleted in endothelial cells by injection of 100 mg/kg tamoxifen (Sigma) at P1-3, prior to animal killing at P6; Cre+ NOTCH-deleted animals were compared against Cre− littermate controls.
Intracardiac injection. Freshly killed P6 animals received a 500 μL bolus of 10 mg/ mL lysine-fixable fluorescently conjugated 70 kDa rhodamine dextran (Thermo-Fisher) into the left ventricle. Eyes were removed and fixed in 4% formaldeyhyde for 18 min before retinal dissection and immunohistochemistry, as described above.
Retinal vascular analysis. Neonatal mouse retinas (P3-21) were collected and stained as previously described 12,[25][26][27] . In brief, eyes were pre-fixed in 4% paraformaldehyde and dissected retinas were post-fixed for 5 min in 4% paraformaldehyde at room temperature or for 20 min in 100% methanol at −20°C. Retinas were incubated in blocking solution (3% BSA, 0.25% Triton-X in PBS) and incubated overnight in staining solution (1% BSA, 0.25% Triton-X in PBS) containing primary antibodies (listed above). Following PBS wash, retinas were incubated for 1 h at room temperature in staining solution containing secondary antibody (listed above); if staining with fluorescently conjugated IB4, staining solution is supplemented with 100 mg/mL CaCl 2 and MgCl 2 . Retinas were then washed and mounted on glass slides for confocal imaging. The remodeling region was delineated as the distance between the vascular edge and the point where SMA signal is indistinguishable from background (ImageJ). GJA5 and SMA expression was calculated as the radial distance from the retinal center to the point at which fluorescent signal for each marker was undetectable. From a 769μm 2 × 923μm 2 ROI of the remodeling region, total vascular (i.e., PECAM1+ or IB4+) area (ImageJ) and plexus (Retina Analyser v.2, Centre de Morphologie Mathematique, École Nationale Superieure des Mines de Paris) and arterial (ImageJ) branchpoint number were calculated. Linear distance from vascular edge was measured in whole retinal quadrants for mitotic endothelial cells (pH3+/ERG+). Endothelial cells in G1 were identified in Cdt1-mOrange+ reporter mice as CDT1+/ERG+ cells. For all studies involving retinal vascular morphometric analysis, the investigator was blinded to animal genotype until imaging processing was complete.
Statistics. Power analysis was performed on retinal vascular morphometry data collected from WT + DAPT (vs. WT + Vehicle) P6 tissue, since NOTCH inhibition is known to induce a hyperdensity phenotype in P6 retinas. Power analysis was also performed on cell cycle data from sorted retinal endothelial cells from WT + DAPT (vs. WT + Vehicle) to assess the statistical effect size of cell cycle distribution changes associated with vascular hyperdensity. Power analysis was also performed on HUVEC treated with 2 µM palbociclib (vs. DMSO), which has been shown to induce G1 arrest. Finally, power analysis was performed on HES1 mRNA transcript levels in HUVEC treated with 10 µM DAPT (vs. DMSO) or 10 μg/mL DLL4 (vs. PBS), which is known to impact transcription of canonical NOTCH targets. In all cases, a minimum sample size of 3 was shown to be sufficient to detect significant differences at a power of 0.80 and α = 0.05, although pilot data also suggested that a sample size of 5 or greater was preferable. Therefore, for all experiments, we constrained sample size to a minimum of 3-5 with a maximum sampling ratio of 4. Analysis of data obtained through these pilot studies revealed that data were normally distributed and that variance was similar between sample and control groups.
For animal studies, in rare cases, pups were a priori discarded if they appeared to be underweight or lethargic at the time of study onset, but no animals were post hoc excluded from the data set. In rare cases, single data points obtained by qPCR were considered outliers and excluded from subsequent analysis if they fell >2 SD from the sample mean.
Unless otherwise indicated, statistics (α = 0.05) were performed using either a standard two-tail Student's t-test, or a two-way ANOVA test, followed by post hoc t-test with Bonferroni correction.
Code availability. Matlab code and other relevant data associated with these studies are available upon reasonable.
Data availability. The RNA-Seq data sets generated and analyzed in the current study are available at the NCBI Sequence Read Archive (Accession Number: #SRP113256).