Notch3 establishes brain vascular integrity by regulating pericyte number

The vertebrate vascular wall is composed of two major cell types, endothelial cells (ECs) and mural cells, which envelop the ECs. When associated with large arteries and veins, mural cells are multi-layered and referred to as vascular smooth muscle cells (vSMCs). When associated with capillaries, mural cells appear as sparse solitary cells and are referred to as pericytes. Until recently, mural cells were viewed as stationary cells that structurally support the vascular wall. However, they are now recognized as dynamic regulators of vascular development and homeostasis (Armulik et al., 2011; Bell et al., 2010; Lindahl et al., 1997; von Tell et al., 2006). Brain pericytes, in particular, regulate key aspects of cerebral vascular function. Recent studies using pericyte-deficient mouse models resulting from defective platelet-derived growth factor (PDGF) signaling demonstrated that brain pericytes are required for blood-brain barrier (BBB) function during development (Armulik et al., 2010; Daneman et al., 2010). Brain pericytes have been also shown to regulate capillary diameter in response to neural activity (Peppiatt et al., 2006). Furthermore, in adult and aging brain, pericyte loss can lead to brain vascular damage, including diminished capillary perfusion, impaired cerebral blood flow (CBF) response to stimulus and BBB breakdown (Bell et al., 2010).

Brain vSMCs and pericytes have been implicated in various human diseases and pathological conditions (Armulik et al., 2011; Winkler et al., 2011). For example, CADASIL (cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy), the most common form of hereditary stroke disorder, is characterized by arteriopathy associated with the degeneration and loss of vSMCs in the brain (Salloway and Hong, 1998). Another example is intraventricular hemorrhage (IVH), which initiates in the germinal matrix (GM) and is the most common form of neonatal intracranial hemorrhage (ICH). GM is a highly vascularized brain region that has less pericyte coverage than the rest of the brain in human fetuses (Braun et al., 2007), suggesting that a low density of pericytes might account for the fragile vasculature in premature infants (Braun et al., 2007). Pericyte loss and subsequent increase of vessel permeability may also promote neurodegeneration in the aging brain (Bell et al., 2010; Sengillo et al., 2012). Despite the important implications of brain pericytes in human health, the molecular mechanisms that regulate their development, survival and distribution remain poorly understood.

During development, pericytes, derived from both neural crest and mesodermal sources (Asahina et al., 2011; Bergwerff et al., 1998; Etchevers et al., 2001; Korn et al., 2002; Que et al., 2008; Wilm et al., 2005; Yamanishi et al., 2012), expand in number and migrate to cover endothelial tubes. A combination of cell culture and mouse investigations revealed that signaling mediated by PDGF receptor β (PDGFRβ) promotes pericyte attachment to endothelial cells, migration and proliferation (Hellström et al., 1999; Lindahl et al., 1997; Tallquist et al., 2003), but how these distinct biological functions are controlled by one signaling pathway remains unknown. Another signaling mechanism implicated in pericyte development and maintenance is the Notch pathway (Domenga et al., 2004; High et al., 2007; High et al., 2008; Liu et al., 2010; Manderfield et al., 2012; Wang et al., 2012). vSMCs and pericytes express the Notch3 receptor (Domenga et al., 2004; Joutel et al., 2000; Joutel et al., 2010) and the majority of CADASIL patients have NOTCH3 missense mutations (Joutel et al., 1996; Joutel et al., 1997). Postnatal maturation of arterial vSMCs was defective in Notch3 knockout mice (Domenga et al., 2004), raising the possibility that Notch3 promotes mural cell differentiation. However, in vitro studies have produced conflicting data regarding whether Notch signaling promotes or inhibits vSMC differentiation (Doi et al., 2006; Morrow et al., 2005; Noseda et al., 2006; Proweller et al., 2005). Additionally, Notch3 mutant mice had substantially fewer mural cells on retinal blood vessels (Liu et al., 2010), raising the possibility that Notch3 promotes mural cell specification, proliferation or survival. Consistent with this, in vitro studies showed that Notch signaling can promote vSMC proliferation and survival (Sakata et al., 2004; Sweeney et al., 2004; Wang et al., 2003). Notch signaling can drive PDGFRβ expression in vSMCs (Jin et al., 2008), providing a potential mechanistic link between Notch and PDGFRβ functions. Nevertheless, the precise roles that Notch signaling plays in brain pericyte development remain unclear.

In this study, we set out to test the hypothesis that Notch signaling regulates pericyte coverage of brain blood vessels during development. To do so, we used zebrafish because the relatively small and simple brain vasculature facilitates quantification of pericytes and assessment of cell division and survival. We show that zebrafish pericytes express notch3 and that notch3 mutant zebrafish suffer brain hemorrhage and fail to form a tight BBB. Brain pericytes were initially specified in notch3 mutant embryos but then failed to expand in number, raising the possibility that Notch3 promotes pericyte proliferation. We tested this by expressing, in a time-dependent manner, dominant-negative and constitutively active forms of Notch3 and investigating the changes in pericyte division and number. Our data provide strong evidence that Notch3 promotes expansion of brain pericyte number by promoting cell division, raising the possibility that Notch3 signaling helps determine pericyte coverage of brain blood vessels.

To initiate our investigation of pericyte development in zebrafish, we set out to identify molecular markers for zebrafish vSMCs and pericytes. A panel of mammalian vSMC and pericyte markers has been reported, including smooth muscle α-actin (αSMA), Rgs5, desmin, NG2 and PDGFRβ (Armulik et al., 2011). Among these, PDGFRβ appears to be the most specific and reliable, marking both vSMCs and pericytes (Armulik et al., 2011). Two zebrafish smooth muscle α-actin homologs have been reported, acta2 and transgelin, and shown to be expressed in vSMCs associated with axial vasculature (Santoro et al., 2009). Brain pericytes, however, have not been previously identified in zebrafish. Therefore, we cloned additional zebrafish homologs of mammalian pericyte marker genes and performed whole-mount in situ RNA hybridization (ISH) experiments to examine their expression patterns during embryonic development. We found that pdgfrb, a PDGFRb homolog, and tagln2, another smooth muscle α-actin homolog, appeared to be expressed in the vasculature in developing embryos (Fig. 1A-C; supplementary material Fig. S1A-C). By contrast, we did not detect vascular expression of rgs5a, desmin a or desmin b, or the NG2 homolog cspg4 (supplementary material Fig. S1E-L). Intersegmental vessels (ISVs) and brain vasculature expressed tagln2 in 48 hours post fertilization (hpf) embryos (supplementary material Fig. S1A,C). However, by 72 hpf expression decreased in the trunk (supplementary material Fig. S1B) and became difficult to detect in the brain (supplementary material Fig. S1D), limiting the usefulness of this marker. At 24 hpf, hypochord expressed pdgfrb (data not shown), as previously reported (Wiens et al., 2010). At 48 hpf, pdgfrb expression was associated with the dorsal aorta (DA) and the dorsal longitudinal anastomotic vessel (DLAV) (Fig. 1A). At 72 hpf, pdgfrb⁺ cells were distributed along ISVs, in addition to the DA and DLAV (Fig. 1B). pdgfrb⁺ cells were dispersed throughout the brain, consistent with the possibility that pdgfrb expression marks pericytes associated with blood vessels (Fig. 1C). To determine the cell type that expressed pdgfrb, we performed fluorescence ISH followed by confocal microscopy. To visualize the vasculature, we used the transgenic reporter Tg(fli1a:EGFP) (Lawson and Weinstein, 2002), which labels vascular endothelial cells revealing all embryonic blood vessels. In the trunk, fluorescence ISH revealed that solitary perivascular cells along the ISVs expressed pdgfrb (Fig. 1D). In the brain, solitary pdgfrb⁺ cells were distributed along brain arteries, veins and capillaries and often extended long processes along the vessels (Fig. 1E,F). Some pdgfrb⁺ cells were found at capillary branch points where they extended a process along each branch, forming a Y-shaped cell (Fig. 1E, insets). These morphological features are hallmarks of pericytes (Armulik et al., 2011). Few, if any, other cells in the brain expressed pdgfrb, suggesting that, at this stage, pdgfrb expression specifically labels pericytes. In the retina, pdgfrb⁺ perivascular cells were distributed along hyaloid vessels and hyaloid artery, indicative of retinal pericytes (supplementary material Fig. S1M). These expression patterns closely resemble that of mouse Pdgfrb, suggesting that pdgfrb is a reliable marker for zebrafish pericytes.

To examine pericyte development further in the zebrafish brain, we quantified the number of pdgfrb⁺ cells at various developmental stages. At 48 hpf, an average of ∼20 pdgfrb⁺ pericytes were associated with the large arteries and veins, such as the basilar artery (BA) and primary hindbrain channels (PHBCs) (Fig. 1I,M; supplementary material Fig. S1N). By 56 hpf, the average number had increased to ∼50 (Fig. 1J,M; supplementary material Fig. S1N). At this stage, a few pdgfrb⁺ cells were also associated with the angiogenic sprouting front of capillaries penetrating the brain (Fig. 1J,M). Between 56 hpf and 80 hpf, the number of pericytes steadily increased as the embryos developed (Fig. 1J-M; supplementary material Fig. S1N).

The number of pericytes could increase over time by proliferation of pdgfrb⁺ cells or by specification of new pericytes from pdgfrb^- precursors. To help distinguish between these possibilities, we investigated whether pdgfrb⁺ cells divide using anti-phosphohistone H3 (pH3) labeling and incorporation of the thymidine analog 5-ethynyl-2′-deoxyuridine (EdU) as M and S phase markers, respectively. At 54 hpf, occasional pH3⁺pdgfrb⁺ pericytes could be identified (data not shown), and ∼5% incorporated EdU (average 4.8±1.8%; n=15) (Fig. 1P), indicating that these cells were actively dividing. At 78 hpf, ∼5% of pdgfrb⁺ pericytes were co-labeled with pH3 (average 5.2±1.2%; n=12) (Fig. 1N), and ∼11% incorporated EdU during a 1-hour exposure (average 11.2±2.4%; n=15) (Fig. 1O,P). These observations are consistent with a model in which a small population of pericyte progenitors is specified at ∼48 hpf and then proliferate to expand the population.

Previous evidence from birds and rodents suggests that brain vSMCs and pericytes derive from the neural crest (Bergwerff et al., 1998; Etchevers et al., 2001; Korn et al., 2002; Yamanishi et al., 2012). To test whether zebrafish brain pericytes also have a neural crest origin, we injected embryos with previously described antisense morpholino oligonucleotides (MOs) to reduce function of foxd3, which interferes with neural crest formation (Stewart et al., 2006). foxd3 MO-injected zebrafish embryos had significantly fewer pdgfrb⁺ pericytes (average 34.8±5.9, n=18) in the brain than the control group (average 63.9±5.5, n=18) (supplementary material Fig. S1O,P), therefore supporting a neural crest origin for zebrafish brain pericytes.

To identify Notch receptors that potentially regulate mural cell development, we performed ISH to examine expression patterns of known zebrafish Notch receptor genes, including notch1a, notch1b, notch2 and notch3. Of these, notch3 showed distinct vascular expression. At 48 hpf, cells associated with the DA expressed notch3 (Fig. 2A). By 72 hpf, notch3 was evident along the DA and ISVs (Fig. 2B). Similar to pdgfrb, trunk and brain perivascular cells with mural cell-like morphology, but not endothelial cells, expressed notch3 (Fig. 2C,D). Double ISH experiments confirmed that brain and trunk perivascular cells co-express pdgfrb and notch3 transcripts (Fig. 2E-G). By contrast, fluorescence ISH did not reveal perivascular expression of notch1a, notch1b or notch2 (data not shown). These results are consistent with previous reports from mouse studies showing that mural cells, but not endothelial cells, express Notch3 (Joutel, 2011).

To investigate functions of Notch3 signaling in pericyte development, we first examined embryos and larvae homozygous for a previously identified nonsense mutation of notch3 (Alunni et al., 2013). The notch3 ^fh332 allele is predicted to generate a non-functional truncated protein lacking the entire intracellular domain and half of the EGF repeats of the extracellular domain (Alunni et al., 2013). Mutant larvae showed distinct trunk curvature at 3 days post fertilization (dpf), but otherwise appeared morphologically normal (supplementary material Fig. S2A,B). Later, a large percentage of mutant larvae developed heart edema and the majority died before 15 dpf. A small percentage of homozygous mutant animals were able to survive up to 2 months, but subsequently died before they could reach adulthood. To analyze how this mutation affects Notch activity in the animal, we crossed into the mutant background the Notch activity reporter line Tg(Tp1bglob:hmgb1-mCherry), in which a nuclear mCherry fluorescent reporter is expressed under the control of an element containing 12 RBP-Jκ binding sites (Parsons et al., 2009). At 3 dpf, homozygous mutants showed significantly reduced mCherry fluorescent intensity compared with wild-type siblings (supplementary material Fig. S2C-F). We next used ISH to analyze expression of known targets of canonical Notch signaling including her4.1, her2, hey1, her6 and her9. We found that expression of her4.1, her2 and her6 was similar between mutant and wild-type larvae at 3 dpf (supplementary material Fig. S2G,H; data not shown) and 5 dpf (data not shown). By contrast, hey1 and her9 were significantly reduced in 3 dpf mutant larvae (supplementary material Fig. S2I-L). These observations suggest that hey1 and her9 are most likely to be downstream targets of Notch3 signaling.

Strikingly, a portion of 3 dpf mutant larvae had hemorrhage within the brain but in no other tissues (Fig. 3A-D). The penetrance of the hemorrhage phenotype was variable, affecting 6.3-19.5% and 12.5-28.9% of homozygous larvae produced by matings of heterozygous carriers on AB and Tg(fli1a:EGFP) genetic backgrounds, respectively (supplementary material Table S1). Typically, one to four hemorrhage sites were evident in the brain, mostly located within or near the brain ventricles, characteristic of intraventricular hemorrhage (Fig. 3A-D). Expressivity of the phenotype also varied, ranging from small focal hemorrhages (Fig. 3A,B) to a broader hemorrhage across the entire intraventricular region (Fig. 3C,D). To determine whether the hemorrhagic phenotype could be exacerbated by cardiovascular stress, we raised embryos at 32°C, inducing an ∼20% increase in heart rate in embryos as previously described (Barrionuevo and Burggren, 1999). This increased the hemorrhage rate to >50% in 3 dpf notch3^fh332 homozygous larvae on the Tg(fli1:EGFP) background.

One possible explanation for the brain bleeding resulting from loss of Notch3 activity is that Notch3 is required for BBB function. To test this, we injected the tracing dye Alexa Fluor 488-dextran (2000 kD) into the vasculature (Jeong et al., 2008; Xie et al., 2010) of 4 dpf wild-type larvae and similarly staged notch3^fh332 mutant larvae that did not show evidence of brain hemorrhage. In the wild-type larvae, the fluorescent dye was retained by the brain vasculature 90 minutes post injection (Fig. 3E,G), consistent with previous reports that a functional BBB has been established at this stage (Jeong et al., 2008; Xie et al., 2010). By contrast, the dye diffused from the brain vasculature of notch3^fh332 mutant larvae shortly after injection, with little fluorescence retained by the brain vessels 90 minutes after injection (n=10 wild-type and mutant larvae) (Fig. 3F,H). We also injected a smaller size dextran dye (10 kD FITC-dextran) and the tracing dye Evans Blue (950 Da). These dyes were also partially retained by the wild-type but not the mutant larva (supplementary material Fig. S3A-H). These data support the possibility that the brain hemorrhage of notch3^fh332 mutant larvae results from an abnormally formed BBB.

The hemorrhage phenotype and defective BBB could be explained by structural defects in vascular endothelial cells. To investigate this possibility, we examined expression of markers of endothelial cell differentiation. No differences in expression of cdh5, encoding VE-cadherin, and cldn5b, encoding the tight junction protein Claudin 5, transcripts (supplementary material Fig. S3I-L) or Claudin 5 and ZO-1 proteins (Fig. 3I-L) were apparent between mutant and wild-type larvae (n=8 wild-type and mutant larvae). Furthermore, we did not identify defects in endothelial tight junction formation by electron microscopy in either non-hemorrhagic or hemorrhagic mutant larvae (n=7 wild-type and mutant larvae) (supplementary material Fig. S3M-P). Together, these data indicate that Notch3 is required for brain vascular integrity, independently of endothelial cell tight junction formation.

To facilitate analysis of the vascular phenotype, we crossed the Tg(fli1a:EGFP) transgene into the notch3^fh332 mutant background. Transgene expression revealed normal vessel patterning in the trunk of mutant embryos (supplementary material Fig. S4A,B). Brain vessels showed normal patterning at 48 hpf (Fig. 4A,B), but later appeared to have reduced branching, resulting in a less elaborate vascular network at 72 hpf (Fig. 4C,D; supplementary material Fig. S4C). We next investigated pericyte development by examining pdgfrb expression. At 48 hpf, notch3^fh332 mutant embryos had significantly fewer brain pdgfrb⁺ pericytes compared with wild type (Fig. 4A,B,G). At 72 hpf, the brain vasculature was well covered by pdgfrb⁺ pericytes in wild-type larvae, whereas in mutant larvae only an average of ∼20 pdgfrb⁺ pericytes could be detected near the BA and PHBC, and the rest of the brain lacked pdgfrb expression (Fig. 4C,D,G). At 4 dpf, the number of pdgfrb⁺ pericytes remained largely unchanged in mutant larvae, whereas their number had significantly increased in wild type (Fig. 4E-G). ISH using the notch3 probe also revealed a near absence of notch3⁺ pericytes in the mutant brain (supplementary material Fig. S4D,E).

To confirm that the pericyte deficiency phenotype was caused by Notch3 loss of function, we also performed knockdown experiments using a previously described notch3 translation-blocking MO (Leslie et al., 2007). Injection of 7 ng MO resulted in partially penetrant IVH [31.6% to 42.2% on the Tg(fli1a:EGFP) background] and significant reduction of pdgfrb⁺ pericyte number compared with control, resembling the notch3 mutant phenotype (n=26 each) (supplementary material Fig. S5 and Table S1), indicating that the pericyte deficit phenotype of notch3^fh332 mutant larvae is caused by Notch3 loss of function.

Because brain pericytes appear to derive from cranial neural crest, one possible explanation for the brain pericyte deficit in notch3^fh332 mutant larvae is that they are deficient for neural crest. To test this possibility, we examined expression of the pan-neural crest markers crestin and sox10 by whole-mount ISH. At 24 hpf, mutant and wild-type embryos similarly expressed both genes (n=5 each for each gene) (supplementary material Fig. S6A-D), indicating that the brain pericyte deficit in notch3^fh332 mutants was not caused by impaired neural crest specification.

Previous data show that Notch signaling regulates arterial-venous differentiation. Mutation of mind bomb, which encodes an E3 ligase necessary for efficient Notch signaling, reduced expression of arterial markers, including efnb2a and dlc, at the dorsal aorta (Lawson et al., 2001). To determine whether the pericyte deficit might be caused by impaired arterial-venous differentiation as a result of Notch3 loss of function, we examined the expression of ephrin B2a and dlc. Notably, notch3 ^fh332 mutants expressed efnb2a and dlc in a similar pattern to wild type (n=8 wild-type and mutant embryos for each gene) (supplementary material Fig. S6E-H). Expression of the arterial marker flt1 and venous marker dab2 (Bussmann et al., 2011) were also indistinguishable between mutant and wild-type brains (supplementary material Fig S6I-L). These observations suggest that Notch3 function is dispensable for arterial-venous differentiation. Notch signaling also regulates endothelial tip cell selection in the developing vasculature whereby loss of mind bomb or delta-like 4 (dll4) function cause an overabundance of angiogenic cells and ectopic sprouting in ISVs (Leslie et al., 2007; Siekmann and Lawson, 2007; Zygmunt et al., 2011). However, examination of Tg(fli1a:EGFP);notch3^fh332 embryos and larvae between 2 and 3 dpf did not reveal any ectopic angiogenic sprouting (Fig. 4C-F; supplementary material Fig. S4A,B), indicating that Notch3 is not involved in determining endothelial tip cell formation.

Our data raise the possibility that the IVH and leaky BBB of notch3 mutant larvae result from a deficit of pericytes. If so, an independent method of disrupting pericyte development should produce a similar phenotype. To test this hypothesis, we treated developing zebrafish embryos with the PDGFRβ inhibitor AG1295, a potent blocker of mural cell proliferation in vitro (Banai et al., 1998). At 25 μM, larvae treated with AG1295 from 32 hpf to 72 hpf or from 48 hpf to 72 hpf did not show major developmental delay, yet ISH revealed that drug-treated larvae had a significant decrease in pdgfrb⁺ cell number in the brain (n≥30 for each treatment group) (supplementary material Fig. S7A-C). By 3 dpf, drug-treated larvae on the AB genetic background had a consistent intracranial hemorrhage phenotype, affecting 4.8-10.8% of treated embryos (supplementary material Fig. S7D-I and Table S1). Similar to notch3^fh332 mutants, the hemorrhage rate was enhanced on a Tg(fli1a:EGFP) background to 16.7-31% (supplementary material Table S1). The majority of hemorrhage sites were within the brain ventricles and behind the eyes (supplementary material Fig. S7E-I). Taken together, these observations support the notion that a reduced pericyte population, resulting from loss of Notch3 function, compromises the BBB and brain vascular integrity.

To test further the requirement of Notch3 signaling for promoting expansion of pericyte number, we created the transgenic line Tg(hsp70I:N3ECD-EGFP) to express the extracellular domain of Notch3 (N3ECD), which dominantly interferes with Notch3 function (Haruki et al., 2005; Rebay et al., 1993), fused to EGFP under control of a heat-inducible promoter (supplementary material Fig. S8A). Upon heat shock, transgenic embryos induced uniform EGFP expression (supplementary material Fig. S8B). Heat shock of Tg(hsp70I:N3ECD-EGFP) embryos at 30 hpf resulted in trunk curvature at 3 dpf (supplementary material Fig. S8D,E), reminiscent of notch3^fh332 mutant larvae. To assess how N3ECD overexpression affects Notch activity, we crossed Tg(hsp70I:N3ECD-EGFP) carriers to Tg(Tp1bglob:hmgb1-mCherry) carriers. Heat shock of Tg(hsp70I:N3ECD-EGFP);Tg(Tp1bglob:hmgb1-mCherry) embryos at 30 hpf or 48 hpf resulted in significant reduction of mCherry expression between 3 dpf and 5 dpf (supplementary material Fig. S8F-M), resembling notch3^fh332 mutants. N3ECD-expressing larvae and control larvae also showed reduced levels of hey1 and her9 expression (supplementary material Fig. S8P-S), similar to notch3^fh332 mutant larvae. Therefore, N3ECD overexpression appears to reproduce Notch3 loss of function. ISH revealed that pdgfrb⁺ pericyte number was significantly reduced in Tg(hsp70I:N3ECD-EGFP) larvae at 72 hpf, phenocopying notch3^fh332 mutant larvae (Fig. 5A,B,D). A later heat shock at 48 hpf also resulted in significant reduction in pdgfrb⁺ pericyte number, albeit less severely (Fig. 5C,D). Similarly, a heat shock at 72 hpf resulted in significant reduction of pdgfrb⁺ pericyte number at 96 hpf (average 80±9.9, n=15) compared with the control group (average 106±9.2, n=15). These results are consistent with the hypothesis that Notch3 positively regulates pericyte proliferation. Notably, although 48 hpf heatshock caused significant reduction in pericyte number, vascular patterning in the brain appeared largely normal compared with the control (Fig. 5A,C). Therefore, inactivation of Notch3 signaling after initial specification of pericyte progenitors appeared to abolish the expansion of these cells without causing major defects in vascular morphogenesis.

To directly test if Notch3 regulates pericyte proliferation, we examined EdU incorporation after induction of N3ECD expression. Transgenic and non-transgenic larvae were heat-shocked at 72 hpf and EdU-pulsed for 1 hour at 78 hpf. Tg(hsp70I:N3ECD-EGFP) larvae had significantly fewer pericytes than the control group and the percentage of EdU⁺ brain pericytes (number of pdgfrb⁺EdU⁺ brain pericytes/total number of pdgfrb⁺ brain pericytes) was significantly decreased (Fig. 5E).

Next, to determine whether Notch3 signaling is sufficient for promoting brain pericyte proliferation, we overexpressed the Notch3 intracellular domain (Notch3ICD, or N3ICD) as a constitutively activated form of Notch3. To do this, we generated the transgenic line Tg(hsp70I:N3ICD-EGFP), in which EGFP-tagged N3ICD is expressed under the control of the heat shock promoter (supplementary material Fig. S8A). Transgenic embryos induced uniform EGFP signal after heat shock (supplementary material Fig. S8C). In the majority of embryos, the EGFP fluorescence gradually faded away after ∼10 hours, probably owing to protein degradation mediated by the PEST domain at the C-terminus of N3ICD. This transgenic line therefore allows heat shock-induced, transient activation of Notch3 signaling in an ∼10-hour time window. Heat-shocked Tg(hsp70I:N3ECD-EGFP); Tg(Tp1bglob:hmgb1-mCherry) embryos showed a robust increase in mCherry fluorescence intensity compared with control (supplementary material Fig. S8N,O), confirming that N3ICD overexpression is capable of triggering RBP-Jκ-mediated transcriptional activation. We then performed heat shock at different stages and examined pericyte development at 72 hpf. After an early heat shock at 30 hpf, we did not observe a significant change in pdgfrb expression at 72 hpf, suggesting that Notch3 overexpression alone is not sufficient to induce ectopic pericyte specification (supplementary material Fig. S9A,B). By contrast, heat shock at 48 hpf induced a significant increase in the number of pdgfrb⁺ pericytes (Fig. 6A,B,E), supporting the hypothesis that Notch3 signaling expands the pericyte population once they are specified. To test if Notch3 overexpression promotes pericyte division, we examined EdU incorporation after induction of N3ICD expression. Larvae were heat-shocked at 72 hpf and subjected to an EdU pulse of 1 hour at 78 hpf. We found that the number of pdgfrb⁺ pericytes and the percentage of pdgfrb⁺ pericytes that incorporated EdU were significantly increased in Tg(hsp70I:N3ICD-EGFP) larvae compared with controls (Fig. 6F). Therefore, Notch3 appears to be sufficient for promoting brain pericyte proliferation.

Pdgfb/Pdgfrβ signaling can promote pericyte proliferation (Olson and Soriano, 2011) and in cell culture constitutively active Notch3 can elevate Pdgfrb expression (Jin et al., 2008), raising the possibility that Notch3 regulates pericyte number and vascular integrity, at least in part, by regulating Pdgfrb transcription. We performed two experiments to investigate whether Notch3 and Pdgfrβ function in a common pathway to regulate pericyte development in vivo. First, we reasoned that if Notch3 regulates vascular integrity via Pdgfrβ, then a low dose of the AG1295 inhibitor should enhance the notch3 loss-of-function phenotype. Indeed, notch3^fh332 homozygous mutants showed a more penetrant hemorrhage phenotype when treated with a low dose of AG1295 that, by itself, did not alter pericyte number in wild-type larvae (n=5; supplementary material Fig. S7J). Second, if Notch3 works through Pdgfrβ, then the increase in pericyte number resulting from expression of constitutively active Notch3 should require Pdgfrβ function. To test this, we treated heat-shocked Tg(hsp70I:N3ICD-EGFP) larvae with 25 μM AG1295. This resulted in reduction in brain pericyte number to a similar extent as in AG1295-treated control larvae (Fig. 6G).

Finally, we tested whether Notch3 overexpression could rescue Notch3 loss of function. First, we transiently expressed N3ICD by injecting the hsp70I:Notch3ICD-EGFP construct into notch3^fh332 mutant embryos and heat shocking at 48 hpf. This partially rescued the pericyte deficiency phenotype in about half of the injected mutant embryos (n=11/23) (supplementary material Fig. S9C,D). To confirm this observation, we then crossed the hsp70I:Notch3ICD-EGFP transgene into the notch3^fh332 mutant background. Heat shock at 48 hpf significantly rescued the pericyte deficiency phenotype in all homozygous mutant embryos (n=15) (Fig. 6C,D,H). Notably, despite the significant rescue of the pericyte phenotype, the vascular branching defect was not rescued by N3ICD expression (Fig. 6C,D). Heat shock of notch3^fh332;Tg(hsp70I:N3ICD-EGFP) embryos at 30 hpf did not rescue the pericyte deficiency (n=14; data not shown), consistent with our previous observation that heat shock of Tg(hsp70I:N3ICD-EGFP) embryos at 30 hpf did not expand the pericyte population. Overexpression of Notch3ICD in endothelial cells by injection of the expression construct fli1:myc-Notch3ICD (Lawson et al., 2001) failed to rescue the phenotype (n=15) (supplementary material Fig. S9E,F). Therefore, endothelial cell expression of Notch3ICD is not sufficient to promote pericyte proliferation. Taken together with our data showing pericyte-specific expression of notch3, these results provide evidence that Notch3 function is required within pericytes to promote their proliferation. We conclude that Notch3 signaling within pericytes drives expansion of the population through proliferation, thereby providing an essential role in establishing brain vascular integrity.

Consistent with the recently documented requirement of pericytes in BBB formation and function, notch3^fh332 mutant larvae frequently had brain hemorrhage, and even those without obvious blood pooling had a leaky BBB. We found no evidence of endothelial cell abnormalities by electron microscopy, indicating that, at these early stages of development, hemorrhage probably resulted from impaired BBB function rather than from a failure of endothelial cell differentiation. Therefore, the small number of pericytes that develop in notch3^fh332 mutants is not sufficient to direct BBB formation, indicating that Notch3 function is part of a mechanism to promote adequate pericyte coverage of brain vessels. Notably, a recent study using independently identified notch3 hypomorphic alleles suggested that reduced Notch3 signaling compromised vascular tone and vascular integrity in the fins of adult zebrafish (Zaucker et al., 2013). Mural cells could not be unambiguously identified in the fin vasculature of notch3 mutants, supporting our own results.

Previous evidence suggested that Notch3 signaling promotes Pdgfrb expression. Pdgfrb expression was reduced by 40% in vSMCs of tail lateral arteries in newborn Notch3^lacZ/lacZ mutant mice and in cell culture constitutively active Notch3 increased Pdgfrb transcript and protein levels (Jin et al., 2008). Our observations seem to corroborate these findings. In notch3^fh332 mutant and heat-shocked Tg(hsp70I:N3ECD-EGFP) larvae, there appeared to be a modest but noticeable reduction of pdgfrb transcript level in individual pericytes. Conversely, N3ICD expression appeared to increase pdgfrb transcript levels in individual pericytes. Furthermore, use of a pharmacological inhibitor showed that reducing Pdgfrβ enhanced the brain hemorrhage phenotype of notch3 mutant larvae and that N3ICD-driven pericyte proliferation requires Pdgfrβ function. Because pdgfrb expression level appears to be an intrinsic determinant of pericyte proliferation rate (Tallquist et al., 2003), these data are consistent with a model in which a Notch3-Pdgfrβ hierarchy regulates pericyte proliferation.

Other important roles of Notch signaling in vascular development have been documented, including regulation of tip cell formation and arterial-venous differentiation (Hellström et al., 2007; Larrivée et al., 2012; Lawson et al., 2001; Lawson et al., 2002; Leslie et al., 2007). notch3^fh332 mutant larvae, however, showed no sign of ectopic vessel sprouting or deficient arterial differentiation as assessed by arterial marker expression. Therefore, Notch3 signaling is most likely to be dispensable for tip cell selection and arterial-venous differentiation. Conversely, notch3^fh332 mutant larvae had reduced brain vascular branching at 3 dpf. It remains to be determined whether this defect reflects an endothelial cell-autonomous role of Notch3 in brain vascular patterning. Nevertheless, several lines of evidence argue that the pericyte deficit of notch3^fh332 mutants is not secondary to reduced vascular morphogenesis. First, the pericyte deficit of notch3^fh332 mutant embryos was evident starting from 48 hpf, preceding the appearance of the patterning defect in the brain. Second, induction of N3ECD expression at 48 hpf resulted in reduced pericyte number at 72 hpf without significantly altering vascular patterning in the brain. Third, N3ICD induction at 48 hpf in notch3^fh332 mutant embryos significantly rescued the pericyte deficit phenotype, but failed to rescue the vascular patterning phenotype. Fourth, endothelial expression of N3ICD failed to rescue the pericyte deficit phenotype in notch3^fh332 mutants. These data are consistent with a cell-autonomous role of Notch3 in pericyte development. Indeed, existing evidence suggests that the reduced vessel branching could actually be caused by the pericyte deficit: there was a mild but measurable decrease of the number of vessel branch points in the brain of Pdgfrb knockout mice (Hellström et al., 2001). Additionally, a hypomorphic allele of Pdgfrb produced significantly fewer vessel branch points in mouse retina (Jadeja et al., 2013).

In mice, Notch signaling has been shown to cooperate with Smad4 to promote N-cadherin expression in brain endothelial cells. Brain endothelial cell-specific inactivation of Notch signaling caused impaired endothelial-pericyte adhesion, causing BBB breakdown and perinatal IVH (Li et al., 2011). This process is likely to involve Notch1 and Notch4, two Notch receptors that are expressed in endothelial cells dependent on Smad4 function (Li et al., 2011). These data, along with the results of our study, indicate that distinct Notch signaling pathways in brain pericytes and endothelial cells (Notch3 and Notch1/Notch4, respectively) work together to ensure the proper establishment of brain vascular integrity by promoting brain pericyte expansion as well as pericyte attachment to the endothelium.

Our work could also provide new insights into CADASIL pathology. The vSMC degeneration associated with CADASIL might result from abnormal signaling conferred by particular dominant missense NOTCH3 mutations (Joutel, 2011). Our data provide strong evidence that Notch3 promotes expansion of pericyte number, perhaps by positively regulating pdgfrb expression. Conceivably, NOTCH3 also maintains pericyte number in the adult brain by promoting PDGFRB expression and replacement of pericytes lost during aging. Indeed, in mice, haploinsufficiency of Pdgfrb resulted in an age-dependent progressive loss of brain pericytes (Bell et al., 2010). Therefore, a reduction in the rate of replenishment of pericytes resulting from abnormal NOTCH3 signaling might contribute to the vascular defects characteristic of CADASIL.

The University of Colorado Anschutz Medical Campus Institutional Animal Care and Use Committee approved all zebrafish studies. Embryos were produced by pair matings and raised at 28.5°C in egg water or embryo medium and staged according to hpf or dpf. Strains in addition to new transgenic lines described below included AB, Tg(fli1a:EGFP) (Lawson and Weinstein, 2002), Tg(Tp1bglob:hmgb1-mCherry) (Parsons et al., 2009) and notch3^fh332 (Alunni et al., 2013).

Sequences encoding N3ECD (amino acids 1-1665) and N3ICD (amino acids 1659-2468) were amplified from wild-type cDNA using the following primers: N3ECD forward, 5′-ATGGGGAATTACAGCCTTTGGAT-3′; N3ECD reverse, 5′-GGCAATCAACATGCCAACCATC-3′; N3ICD forward, 5′-ATGGTTGGCATGTTGATTGC-3′; N3ICD reverse, 5′-AGCAAACACCTGCATCTTCC-3′. Products were cloned followed by Gateway recombination to generate hsp70I:N3ECD-EGFP and hsp70I:N3ICD-EGFP using the Tol2 kit (Kwan et al., 2007). Microinjected embryos were screened and those with EGFP⁺ hearts were raised. These were mated with AB fish and progeny were screened for heart EGFP expression. EGFP⁺ embryos were raised to establish transgenic lines. Multiple alleles were identified for each transgenic line, and those with stronger transgene expression were designated Tg(hsp70I:N3ECD-EGFP)^co15 and Tg(hsp70I:N3ICD-EGFP)^co17.

Whole-mount ISH was performed as described (Thisse and Thisse, 2008). pdgfrb, tagln2, cspg4, rgs5a, desmin a, desmin b and notch3 sequences (GenBank accession numbers HM439112, NM_201676, XM_001923422, NM_199962, NM_130963, NM_001077452 and AF152001, respectively) were amplified from cDNA and cloned into pJC-53.2 (Collins et al., 2010). Anti-sense probes were synthesized from PCR products amplified from corresponding cDNA constructs with T7 primers. Fluorescence ISH was performed as described (Lauter et al., 2011). Fluorescent ISH samples were imaged using a microscope equipped with a PerkinElmer UltraVIEW VoX confocal system and Volocity software. Image brightness and contrast were adjusted in Volocity or ImageJ. For comparison purposes, the same adjustment was applied to all images. For quantification of brain pericytes, pdgfrb⁺ cells were counted in confocal z-stacks of whole embryonic or larval brains. For quantification of brain vessel branching, vessel branch points were counted in confocal z-stacks of whole larval brains.

Whole-mount immunohistochemistry was performed after ISH using rabbit anti-phospho-Histone-H3 (1:1000, 06570, Upstate Biotechnology) and rabbit anti-GFP (1:200, A11122, Invitrogen) antibodies. ZO-1 and Claudin 5 antibody staining were performed on 20 μm frozen sections using mouse anti-Claudin 5 (1:25, 18-7364, Invitrogen) and rabbit anti-ZO-1 (1:25, 33-9100, Invitrogen). For EdU labeling, embryos were manually dechorionated and incubated with 10 mM EdU in 10% DMSO in E3 medium for 1 hour. The embryos were then fixed using 4% paraformaldehyde in PBST (0.1% Tween) and processed for ISH. EdU detection was performed using the Click-iT EdU Imaging Kit (Invitrogen) by incubating in EdU detection cocktail for 1.5 hours.

Embryos were fixed at 4 dpf in 2% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) overnight at 4°C. Embryos were washed, postfixed with 1% osmium tetraoxide, dehydrated and embedded. Sections (50-70 nm) were cut and stained with 3% uranyl acetate and Reynolds’ lead citrate.

MOs targeting notch3 and foxd3 were obtained from Gene Tools. For notch3, we injected 7 ng of ATG-targeting morpholino: 5′-ATATCCAAAGGCTGTAATTCCCCAT-3′ (Leslie et al., 2007). For foxd3 knockdown, we co-injected 3.5 ng each of foxd3-MO^5UTR: 5′-CACCGCGCACTTTGCTGCTGGAGCA-3′, and foxd3-MO^AUG: 5′-CACTGGTGCCTCCAGACAGGGTCAT-3′ as described previously (Montero-Balaguer et al., 2006).

For the BBB assay, 2000 kD Alexa-488 dextran (2.5 mg/ml in PBS, FD2000S, Sigma), 10 kD FITC-dextran (2.5 mg/ml in PBS, FD10S, Sigma) and Evans Blue (2% in PBS, 15110805, MP Biomedicals) were injected into the hearts of 4 dpf larvae as described (Jeong et al., 2008).

For imaging brain hemorrhage, embryos were treated with 0.001% 1-phenyl-2-thiourea (PTU) at 22 hpf to prevent pigmentation. For PDGF receptor inhibitor assays, PDGF receptor tyrosine kinase inhibitor AG1295 (Millipore) was reconstituted in DMSO to make a 1 mM stock. Embryos were treated with 1% DMSO alone or with 25 μM or 10 μM AG1295 with 1% DMSO in E3 media.

Quantifications were performed by collecting confocal z-stacks of entire brains and counting pericytes from maximally projected images. z images were examined to distinguish between individual pericytes. Data were evaluated using Student’s two-tailed t-test.

We thank Julie Siegenthaler for valuable discussions and comments on the manuscript; Nathan Lawson for the fli1:myc-Notch3ICD construct; James Collins for the pJC-53.2 vector; and Dorothy Dill for electron microscopy help.