Vascular defects of DYRK1A knockouts are ameliorated by modulating calcium signaling in zebrafish

The cerebral vasculature plays an essential role in maintaining the homeostasis of the brain by providing oxygen and nutrients and removing waste products. During development, new branches of the cerebral vasculature are formed through angiogenesis by endothelial cells, the primary cell component of the vasculature, via complex cell-cell interactions and signaling pathways from existing vessels (Udan et al., 2013). The cerebral vasculature also contributes to the formation of the neurovascular unit (NVU), which comprises pericytes, astrocytes, microglia and neurons in addition to endothelial cells. Compromise of the normal development or function of the NVU has been implicated in childhood brain development disorders and adult neurological dysfunction (Quaegebeur et al., 2011; Zlokovic, 2008). In the NVU, cerebral endothelial cells, connected mainly by tight junction proteins, build the blood-brain barrier that acts as a primary semipermeable barrier and confers a high selectivity for molecular exchanges between the blood and the brain parenchyma (Dejana et al., 2009). Therefore, the inappropriate development of cerebral endothelial cells may lead to defects in angiogenesis and/or endothelial permeability, which are closely linked to vascular pathologies such as vascular malformations and stroke (Folkman, 1995; Pandya et al., 2006).

The development of the brain vasculature is coordinated by various extra- and intracellular signals. Among them, calcium signaling is one of the major regulators of vascular development and related pathogenesis. Vascular endothelial growth factor signals and various stimuli trigger the change of intracellular Ca²⁺ levels that act as second messengers in endothelial cells, which in turn affect the activity of transcription factors for angiogenesis, such as nuclear factor of activated T-cells, via Ca²⁺-dependent calmodulin/calcineurin activity (Hogan et al., 2003; Loh et al., 1996). In addition, an overload of Ca²⁺ can cause endothelial barrier dysfunction; intracellular Ca²⁺ release via activation of IP₃ receptors (IP3R) or ryanodine receptors (RyRs) of the endoplasmic reticulum into the cytoplasm can increase vascular permeability through disorganization of VE-cadherin (Cdh5) or cytoskeletal rearrangement (Gao et al., 2000; Shen et al., 2010; Tiruppathi et al., 2002). Intriguingly, it has been reported that calcium supplementation for bone health can unexpectedly induce stroke and cardiovascular diseases (Reid and Bolland, 2008), and calcium channel blockers and calcium antagonists have been clinically used as therapeutic agents for strokes and blood vessel dysfunction (Inzitari and Poggesi, 2005). Thus, understanding the detailed underlying molecular mechanisms of Ca²⁺ signaling in angiogenesis and vascular permeability along with the identification of key players may provide an important therapeutic means for treating vascular diseases.

Dual-specificity tyrosine phosphorylation-regulated kinase 1A (DYRK1A) is a serine-threonine kinase that has a dual kinase activity capable of autophosphorylating its own tyrosine residues and phosphorylating other substrates. The DYRK1A gene was first identified in a Drosophila screening as a mutant minibrain, named for the brain morphology defects with reduced brain size (Tejedor et al., 1995). DYRK1A is located in a Down syndrome critical region (DSCR) and is best known as a major causative gene that is implicated in brain function, neurological defects and neurofibrillary tangle formation in Down syndrome (DS) (Liu et al., 2008; Wegiel et al., 2011).

Interestingly, epidemiological studies suggested that DS patients have a reduced incidence of angiogenesis-related solid tumors (Hasle et al., 2000; Nižetić and Groet, 2012) and carry numerous vascular anomalies such as umbilico-portal system anomalies, vertebral and right subclavian artery defects, and pulmonary vein stenosis (Gowda et al., 2014; Pipitone et al., 2003; Rathore and Sreenivasan, 1989; Stewart et al., 1992). Furthermore, they also have an increased incidence of Moyamoya disease and cerebral amyloid angiopathy, which are associated with a cerebrovascular dysfunction and intracerebral hemorrhage (de Borchgrave et al., 2002; Donahue et al., 1998; Jastrzebski et al., 2015; Mito and Becker, 1992; Sabde et al., 2005). Consistent with the potential role of DYRK1A in angiogenesis, the TS65Dn mouse model of DS, trisomic for the Dyrk1a gene, exhibited reduced tumor growth, presumably by suppressing tumor angiogenesis (Baek et al., 2009). Although vascular defects had not been directly associated with human DYRK1A haploinsufficiency syndrome, it has been reported that nearly 75% of children with autism, some of which may have DYRK1A mutations or its reduced activity (Kim et al., 2017), exhibited hypoperfusion in the brain detected by neuroimaging (Bjorklund et al., 2018; Zilbovicius et al., 2000). Also, retinal angiogenesis is disrupted in Dyrk1a heterozygote mice (Rozen et al., 2018), suggesting a role of DYRK1A loss-of-function in regulating angiogenesis in the brain. Taken together, DYRK1A may be implicated in vascular formation and/or function, and this could provide a new perspective to understanding DYRK1A-related pathogenesis; however, in vivo evidence of DYRK1A function in vascular pathology is scarce.

To investigate the role of DYRK1A in vascular formation, we adopted developing zebrafish as a model organism. Zebrafish is a vertebrate animal model used for genetic studies of human diseases exhibiting a high similarity to humans at anatomical and molecular levels, especially in the vascular and nervous system (Isogai et al., 2001; Schmidt et al., 2013). Zebrafish embryos can be readily manipulated for genetic gain-of-function studies with transgenesis or mRNA overexpression, and loss-of-function studies with gene knockouts or morpholino use (Clark et al., 2011; Hogan et al., 2008; Timme-Laragy et al., 2012; Varshney et al., 2015; Zu et al., 2013). Large clutch sizes and various inexpensive and fast experimental techniques allow the use of zebrafish for unique high-throughput in vivo small molecule screening, which enables the identification of hit compounds and gives insights into potential underlying mechanisms (MacRae and Peterson, 2015).

We have recently reported autistic behavioral phenotypes of knockout mutants of dyrk1aa, a mammalian DYRK1A homolog, named dyrk1aa^krb1, generated by transcription activator-like effector nucleases (Kim et al., 2017). In the current study, we investigated a role of DYRK1A in cerebrovascular development during embryogenesis using the dyrk1aa loss-of-function mutants. The dyrk1aa^krb1 mutants exhibited cerebral hemorrhage and angiogenic defects in the developing hindbrain, as analyzed at high resolution by confocal fluorescent microscopy using transgenic animals and transmission electron microscopy. These vascular abnormalities were rescued by expression of wild-type (WT) dyrk1aa mRNA, but not a kinase-dead form, indicating an essential role of its kinase activity. Chemical screening using a US Food and Drug Administration (FDA)-approved chemical library identified the calcium chelator ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid (EGTA) that efficiently rescued cerebral hemorrhage as well as abnormal cerebrovascular defects in dyrk1aa mutants. Another calcium signaling modulator, FK506, rescued the hemorrhagic and cerebrovascular defects of dyrk1aa mutants in a similar manner as EGTA, and transcriptomic analyses identified changes in calcium signaling as the main pathway affected in dyrk1aa mutants. Together, the cerebral hemorrhage and cerebrovascular defects of zebrafish dyrk1aa mutants and the chemical screening revealed an important but less-known role of DYRK1A in in vivo vascular formation, which involves a mechanism that is mediated by calcium signaling, providing a potential therapeutic target for DYRK1A-related vascular disorders.

We recently reported the generation of the dyrk1aa^krb1 mutants that displayed microcephaly and autistic behavioral phenotypes in adults, whereas no distinct morphological defects were observed during embryogenesis (Kim et al., 2017). Detailed inspection of dyrk1aa^krb1 homozygous mutant embryos, however, revealed a cerebral hemorrhage phenotype as early as 52 h postfertilization (hpf) (arrows in Fig. 1A). Of the offspring of dyrk1aa^krb1 mutants, 27.9% of a clutch displayed cerebral hemorrhage, whereas less than 9.9% of WT offspring showed spontaneous hemorrhage at 52 hpf using o-dianisidine staining, which detects hemoglobin activity and allows observation of cerebral hemorrhage more closely (Fig. 1B). Spontaneous hemorrhage of WT offspring at 52 hpf was detected in less than 5% of the clutch under the brightfield microscope (without o-dianisidine staining, data not shown). Cerebral hemorrhage was detected as patches in the forebrain, midbrain, hindbrain and retina, with some embryos exhibiting multiple hemorrhages simultaneously (Fig. 1A). To obtain more pronounced hemorrhagic effects and observe a more explicit phenotype, we challenged the mutant embryos with heat stress by incubation at 35°C for 2.5 h from 48 hpf to 50.5 hpf, followed by keeping them at 28.5°C until 52 hpf. The following experiments for hemorrhagic phenotype were performed in the heat stress condition. It has been previously reported that cardiovascular stress aggravates the hemorrhagic phenotype in the brain, presumably by increasing the heart rate (Barrionuevo and Burggren, 1999; Wang et al., 2014). Consistently, the cerebral hemorrhage of WT and dyrk1aa^krb1 mutants under the heat-stressed condition increased by 16.3% and 40.5%, respectively (Fig. 1B, Heat-stressed).

Because cerebral hemorrhage is sometimes accompanied by defective cerebrovascular formation (Arnold et al., 2014), we examined the formation of central arteries (CtAs) in the hindbrain, a well-characterized stereotypical developing cerebrovascular structure, using Tg(kdrl:EGFP) transgenic animals (Ulrich et al., 2011) in the WT or dyrk1aa^krb1 mutants. CtAs sprouted from the primordial hindbrain channels, which contained the pool of endothelial cells required for CtA formation, and invaded the hindbrain between 32 and 36 hpf with a stereotypical morphology within rhombomeres, formed with over 50% ipsilateral CtA connectivity at 48 hpf (Bussmann et al., 2011; Ulrich et al., 2011). A detailed, high-resolution confocal imaging analysis to compare the CtA formation of dyrk1aa^krb1 mutants and WT at 52 hpf revealed that the stereotypical structure of CtAs was abolished in dyrk1aa^krb1 mutant embryos (Fig. 1C). To quantitate the CtA vascular defects, the length and branching points of CtAs were measured, representing the migration/proliferation and sprouting activities of CtA endothelial cells, respectively (AlMalki et al., 2014; Phng et al., 2009; Tímár et al., 2001) (Fig. 1D). In dyrk1aa^krb1 mutants, the total lengths and branching points of CtAs were reduced to 70.2% and 73.3% compared to the WT control, respectively (Fig. 1D). In the following experiments, we used the length and branching points of CtAs as quantitative measures of defects in vascular development.

CtA angiogenic defects were also confirmed by examining RNA expression of vascular markers kdrl (vegfr2) and dll4 by whole-mount RNA in situ hybridization (WISH) at various developmental stages including 30, 36 and 52 hpf (Fig. S1, kdrl and dll4). Consistent with the angiogenic defects revealed by Tg(kdrl:EGFP) transgenic animals, expression of kdrl and dll4 in the vasculature of the hindbrain was reduced in dyrk1aa^krb1 mutant embryos at all stages examined (red arrows in Fig. S1, kdrl and dll4). These vascular defects in mutants appeared not to be due to gross defects of brain development because the expression of krox20, which marks rhombomere boundaries (Moens and Prince, 2002; Oxtoby and Jowett, 1993), and isl1, which labels primary motoneurons in the hindbrain (Ericson et al., 1992; Inoue et al., 1994; Korzh et al., 1993), were grossly unaffected in mutant embryos (Fig. S1, krox20 and isl1). These vascular defects appeared not to be due to defective heart development or reduced blood flow, based on the normal heartrate of mutants compared to the WT embryos (Fig. S2, Movies 1 and 2).

We examined transcripts of dyrk1aa in zebrafish embryos using WISH during embryogenesis. dyrk1aa mRNA had a broad expression in the forebrain (black arrowheads/brackets), midbrain (gray arrowheads/brackets), hindbrain (blue arrowheads/brackets), spinal cord (orange arrowheads), heart (asterisks) and retina (red arrows) at 24, 48 and 72 hpf (Fig. 2Aa-Al). Transverse sections of WISH embryos at the midbrain level showed that dyrk1aa was broadly expressed in the tectum (green arrows), tegmentum (black arrow), the ganglionic cell layer (purple arrows) and the inner nuclear layer of the retina (red arrows) (Fig. 2Am,An). We also checked the expression patterns of dyrk1ab mRNA (ZFIN ID: ZDB-GENE-030131-5677), another zebrafish ortholog of human DYRK1A, using WISH, which appeared to be significantly overlapped with those of dyrk1aa (Fig. S3).

The temporal expression patterns of dyrk1aa and dyrk1ab mRNAs according to developmental stages by reverse transcription-PCR (RT-PCR) were also examined. Transcripts of dyrk1aa and dyrk1ab mRNA were strongly detectable at the one-cell stage but decreased after 6 hpf, presumably because of the maternal effect (Harvey et al., 2013; Mathavan et al., 2005) (Fig. 2B). Zygotic dyrk1aa and dyrk1ab expression appeared to start at ∼15 hpf, increasing at 24 hpf, and strong expression was maintained until 5 days postfertilization (dpf) (Fig. 2B).

Furthermore, whether dyrk1aa and dyrk1ab mRNA was specifically expressed in endothelial cells was confirmed by performing RT-PCR analyses using GFP-positive Tg(kdrl:EGFP) endothelial cells isolated by fluorescence-activated cell sorting (FACS). The dyrk1aa and dyrk1ab mRNA was enriched in GFP-positive endothelial cells but was also expressed in GFP-negative cells, suggesting a role in endothelial as well as non-endothelial cells (Fig. 2C). These expression patterns were consistent with the reported mammalian Dyrk1a expression in the heart primordium and the central nervous system of developing mouse embryos, including the inner (neural) layer of the optic cup (Hämmerle et al., 2008; Rahmani et al., 1998), and endothelial cells isolated from Dscr1 transgenic mice, as assessed using western blotting (Baek et al., 2009), suggesting a functional conservation across species. Recently, the expression of DYRK1A at the protein and RNA level in human endothelial cells and in mouse lung endothelial cells has also been shown by Rozen et al. (2018).

To confirm that loss of function of the dyrk1aa gene was responsible for dyrk1aa^krb1 mutant phenotypes, we tested whether WT dyrk1aa mRNA rescued the cerebral hemorrhage and aberrant vascular phenotype in dyrk1aa^krb1 mutants by globally expressing full-length WT dyrk1aa mRNA. Using o-dianisidine staining, the high incidence of cerebral hemorrhage in dyrk1aa^krb1 mutants under heat stress (41.7% of the offspring) was shown to be significantly reduced down to 25.4% by 0.1 ng dyrk1aa mRNA injection (Fig. 3A,B), whereas the same dose injected into the WT background had little effect (Fig. 3A,C). Similarly, the reduced mean percentages of lengths (66.5%) and branching points (61.2%) of CtAs in dyrk1aa^krb1 mutants relative to WT controls (100%) were rescued (89.9% and 122%, respectively) with the expression of 0.1 ng dyrk1aa mRNA (Fig. 3D,E). The rescue of CtA angiogenesis defects were effective only within specific dose ranges of dyrk1aa mRNA expression in dyrk1aa^krb1 mutants, and overexpression of dyrk1aa mRNA in WT background increased CtA formation in a dose-dependent manner (Fig. 3E), suggesting the importance of spatial and temporal contexts for effects of dyrk1aa expression as well as its dose-sensitive nature.

DYRK1A regulates its target substrate proteins via phosphorylation (Hämmerle et al., 2003). To verify whether the cerebral hemorrhage and CtA defects in dyrk1aa^krb1 mutants were dependent on the kinase activity of Dyrk1aa, we performed rescue experiments with K193R-dyrk1aa mRNA, the predicted kinase-dead form of Dyrk1aa (Himpel et al., 2001). The expression of K193R-dyrk1aa mRNA or control mCherryRed mRNA failed to rescue the defects of hemorrhage and CtA formation with comparable doses of WT-dyrk1aa mRNA (Fig. 3B; Fig. S4A and S4B). These results suggest that the kinase activity of Dyrk1aa is critical for normal CtA development and prevention of hemorrhage.

To examine whether the dyrk1aa mutation caused an ultrastructural change in brain vessels, we analyzed the cytoarchitecture of blood vessels in the WT and dyrk1aa^krb1 embryos at 52 hpf using transmission electron microscopy. In the WT group, blood vessels composed of endothelial cells and lumens were well formed and tightly arranged at this stage (Fig. 4A,A′), although smooth muscle cells, pericytes and astrocytes were in the process of differentiation and not yet clearly identified (Liu et al., 2007). Characteristically, brain tissues in dyrk1aa^krb1 mutants exhibited enlarged interstitial spaces, presumably because of loose connections between layers of vessel walls (arrowheads in Fig. 4B,B′), suggesting that this abnormal formation of vessel walls was one of the causes of the hemorrhagic phenotype in dyrk1aa^krb1 mutant embryos.

Cerebral angiogenic defects and hemorrhage in dyrk1aa^krb1 mutants were also recapitulated by harmine (7-methoxy-1-methyl-9H-pyrido[3,4-b]-indole), a well-known DYRK1A inhibitor (Bain et al., 2007; Göckler et al., 2009). WT zebrafish embryos exposed to different concentrations of harmine (10, 25 and 50 µM) starting at 24 hpf for 28 h showed the hemorrhage and CtA formation defects at 52 hpf, similar to those of the dyrk1aa^krb1 mutants, and the number of CtA sprouts was reduced down to 15.2% of the control number at a concentration of 50 µM harmine (Fig. 5A,B). The harmine-induced hemorrhagic phenotype increased up to 65.7% that of the offspring, compared to 2.3% of the DMSO-treated controls (Fig. 5C). This chemical suppression of DYRK1A using harmine further suggested that the vascular phenotypes of dyrk1aa^krb1 mutants were owing to the loss of DYRK1A function.

Based on the hemorrhagic phenotype induced by harmine treatment, we developed an embryonic screening assay using a chemical library that consisted of 1280 FDA-approved and pharmacologically active compounds (LOPAC 1280; Sigma-Aldrich), which allowed us to identify small molecules that modulated the harmine-induced hemorrhagic phenotype. Five WT zebrafish embryos at 24 hpf were placed in each well of a 48-well plate, exposed to 30 µM harmine together with the individual chemicals of the chemical library at 10 µM as a final concentration for 28 h, and analyzed for the increased or decreased hemorrhagic phenotype (Fig. 5D). As a result, 171 of 1280 compounds tested were found to be phenotype modifiers, which were categorized according to their common features as ‘class’ based on their known functions (Tables S1-S3). Some chemicals have already been reported to cause hemorrhage. For example, atorvastatin, which increased the hemorrhage in our screening, has been previously reported to induce hemorrhagic stroke as a side effect in zebrafish embryos (Gjini et al., 2011). Of the 171 compounds, EGTA, a specific calcium chelator, was identified as one of the most efficient suppressors of the hemorrhagic phenotype induced by harmine treatment, in our chemical screening.

To determine whether our findings from the chemical screening were applicable to the genetic model, we added EGTA to the dyrk1aa^krb1 mutants and examined the hemorrhagic and CtA development. Under heat-stressed conditions, the EGTA treatment significantly reduced the hemorrhagic phenotype of dyrk1aa^krb1 mutants at a specific concentration of 10 nM (from 37.5% to 23.6%, Fig. 6A,B; P<0.05). In addition, a reduced CtA development of mutants in lengths and branching points (76% and 57.4% reduction compared to WT, respectively) was also rescued up to 86.6% and 85.9% of the normal levels, respectively, by treatment with the same concentrations of EGTA (Fig. 6C,D; P<0.01). EGTA treatment appeared to be effective only within a narrow range, because 1 nM or 100 nM EGTA treatment failed to rescue the vascular defects of dyrk1aa^krb1 mutants, except for the rescue of CtA branching points with 1 nM EGTA (Fig. 6B,D). In contrast, EGTA treatment of the WT embryos did not induce any vascular defects (Fig. S5), suggesting a specific role of EGTA on dyrk1aa^krb1 mutants.

To identify the temporal requirement of EGTA for the suppressive effects, developing embryos were incubated with various doses of EGTA during an early period of 8 h (24∼32 hpf), followed by washing, or during a late period of 16 h (32∼48 hpf) (Fig. 6E). Interestingly, only the early treatment with 10 nM EGTA was effective in significantly preventing the cerebral hemorrhage (from 37.7% to 26.3%), whereas the early treatment with other doses and the late treatment was not (Fig. 6F,G). Because sprouting and elongation of cerebral vessels for angiogenesis occurs actively during day 1 postfertilization (Fujita et al., 2011; Isogai et al., 2001), EGTA may exert its suppressive effects on the vascular defects of dyrk1aa^krb1 mutants by regulating early angiogenic processes.

Identification of EGTA as a suppressor of vascular defects in dyrk1aa^krb1 mutants by chemical screening implies that calcium signaling may be compromised in the dyrk1aa^krb1 mutants. To corroborate this finding, we examined transcriptomic changes of dyrk1aa^krb1 mutants compared to WT embryos at 48 hpf using RNA-seq analysis (Fig. 7). This analysis identified 222 transcripts as differentially regulated genes (DEG), of which 101 were upregulated and 121 were downregulated in the dyrk1aa^krb1 mutants (more than 2-fold and less than 0.5-fold, respectively; P<0.05). When DEGs were analyzed for enriched biological gene ontology (GO) categories by the functional annotation tools in the Database for Annotation, Visualization and Integrated Discovery (DAVID; https://david.ncifcrf.gov), the calcium ion binding category was the most enriched GO molecular function (MF) (Fig. 7B). This GO category includes genes encoding several calcium-dependent adhesion proteins, protocadherin (Pcdh) family members (pcdh1g1, pcdh2ab10, pcdh1gc6, pcdh2g17, pcdh1g18 and pcdh1g30) and calcium-dependent calpains (capn8 and capn2l) (Khorchid and Ikura, 2002) (Fig. 7A). Other genes encoding myosin light chain 4 (myl4), low-density lipoprotein receptor b (ldlrb), and mannan-binding lectin serine protease 2 (masp2) in this GO term are also regulated by calcium signaling, directly or indirectly (Kang et al., 1999; Orr et al., 2016; Zhao and Michaely, 2009) (Fig. 7A), although their roles in regulating vascular formation are not understood at present. In addition, the ‘oxidation-reduction process’, the top-ranked GO biological process (BP), is also well known to affect calcium signaling networks (Görlach et al., 2015; Lounsbury et al., 2000) (Fig. 7C), and the genes belonging to the ‘homophilic cell adhesion via plasma membrane adhesion molecules’ BP category consist of the six Pcdh family genes ‘calcium ion binding’ MF category (Fig. 7C).

We checked the validity of DEG identification by RNA-seq analysis with RT-PCR using RNAs from the whole embryos or from FACS-isolated endothelial cells at 48 hpf (Fig. S6). As a result, nine out of 15 candidate DEGs (pcdh1g1, pcdh2ab10, pcdh1gc6, myl4, pcdh1g18, f7i, capn8, pcdh1g30 and capn2l) were clearly changed in the whole dyrk1aa^krb1 mutant embryos (Fig. S6A). Those nine DEGs were further tested for expression changes in endothelial cells by performing another RT-PCR using RNAs prepared from FACS-isolated endothelial cells of 48 hpf Tg(kdrl:EGFP) WT and dyrk1aa^krb1 mutants. Seven DEGs (pcdh1g1, pcdh2ab10, myl4, pcdh1g18, f7i, pcdh1g30 and capn2l) were found to be correlative with the similar patterns as in the whole embryos (Fig. S6B), suggesting their potential roles in dyrk1aa-mediated calcium signaling in endothelial cells.

In order to check whether calcium signaling changes occur even earlier than 48 hpf, especially when rescue by EGTA treatment is effective (Fig. 6), we also investigated the transcriptomic changes in dyrk1aa^krb1 mutants at 32 hpf using RNA-seq analysis (Fig. S7). Similar to the findings of 48 hpf transcriptomic analysis, DEGs at 32 hpf were already highly enriched in the GO categories of ‘calcium ion binding’ and ‘oxidoreductase activity’ in MF (Fig. S7A,B). Furthermore, expression of six genes in the calcium ion binding category (pcdh1g1, pcdh1g18, f7i, capn8, pcdh1g30 and capn2l) turned out to be concurrently changed in both 32 hpf and 48 hpf transcriptomes (Fig. S7C). Interestingly, five genes (pcdh1g1, pcdh1g18, f7i, pcdh1g30 and capn2l) out of these six DEGs were also changed in the FACS-isolated endothelial cells (Fig. S6B), potentially highlighting their indispensability in dyrk1aa-mediated calcium signaling.

Collectively, alterations of genes in ‘calcium ion binding’ and other GO categories may reflect dysregulation of calcium homeostasis in dyrk1aa^krb1 mutants, consistent with the rescue effects by EGTA on vascular defects of mutants.

As noted previously, EGTA is a chelating agent that has strong selectivity for calcium ions, primarily by lowering the level of extracellular calcium, eventually affecting the calcium homeostasis inside a cell, which is an essential process for regulating angiogenesis as well as other cellular processes including muscle contraction and neurogenesis (Berridge et al., 2000; Munaron, 2006). To assess whether the modulation of intracellular calcium signaling can recapitulate the effects of EGTA treatment, we attempted to rescue the cerebral hemorrhage and defective CtA formation of dyrk1aa^krb1 mutants by inhibiting calcineurin protein with FK506, a well-known specific calcineurin inhibitor (Li et al., 2011). The signaling pathway mediated by calcineurin, a serine/threonine protein phosphatase, is one of the major signaling pathways that is regulated by calcium (Crabtree, 2001; Hogan et al., 2003). Interestingly, the hemorrhagic phenotype of dyrk1aa^krb1 was rescued by treatment with 50 and 100 ng/ml FK506 in a dose-dependent manner, whereas no significant change was observed in the WT control when treated (Fig. 8A,B). The same concentrations of FK506 also rescued defects in the CtA branching points (100 ng/ml) and length (50 ng/ml) of mutants, although FK506 also increased those of CtAs in the WT control (Fig. 8C,D). These data were consistent with the notion that the dysregulation of calcium homeostasis was responsible for the vascular defects of the dyrk1aa^krb1 mutants, and such defects could be rescued, at least partially, by manipulating a calcium-dependent signaling pathway.

Modulation of calcium signaling by EGTA or FK506, however, did not reverse the transcriptional changes of DEGs found in transcriptomic analysis of dyrk1aa^krb1 mutants (Fig. S8), implying that they may not be appropriate readouts for the phenotypic reversal. In addition, expression of potential zebrafish homologs of RCAN1.4, a known target gene in Dyrk1a/Calcineurin pathway (Minami et al., 2004; Rozen et al., 2018), was not changed in dyrk1aa^krb1 mutants (Fig. S9). These additional data suggest an involvement of more complex mechanisms in vivo with regard to calcium signaling regulated by dyrk1aa.

Proper cerebrovascular development and maintenance are essential processes for normal brain development and function, and are governed by diverse and coordinated signaling pathways (Hogan and Schulte-Merker, 2017). In this report, we established a less-known function of DYRK1A, one of the critical genes contributing to some DS phenotypes, in regulating angiogenesis and preventing hemorrhage in the brain, using zebrafish dyrk1aa knockout mutants. The in vivo chemical screening using zebrafish embryos identified small molecules that were able to modify the cerebral hemorrhage upon DYRK1A inhibition. Among them, EGTA, a known specific calcium chelator, was identified as one of the most effective small molecules that rescued the vascular defects of dyrk1aa mutants. Changes in calcium-related signaling pathways revealed by RNA-seq analyses and the rescuing activity by chemical inhibition of calcineurin, a major component of the calcium-dependent pathway, on the vascular defects corroborated the notion that the vascular defects of dyrk1aa mutants were primarily due to calcium dysregulation, which could be reversed by inhibition of excessive calcium-dependent processes.

Zebrafish dyrk1aa knockout mutants exhibited compromised vessel integrity, with cerebral hemorrhage and angiogenesis simultaneously (Figs 1 and 4), compared to WT controls. Although the relationship between angiogenesis, vessel permeability and cerebral hemorrhage during developmental processes is not clearly defined, the interactions of these processes may be important for development of the functional cerebral vasculature. For example, a targeted deletion of miR-126 in mice showed that the cerebral hemorrhage due to defective vascular integrity accompanied a severe reduction in cranial vessel formation (Wang et al., 2008). In addition, excessive angiogenesis was shown to precede the cerebral hemorrhage in the developing brain of neuroendothelium-specific Itgb8 knockout mice (Arnold et al., 2014). Based on the mutant phenotypes described in our report, it is possible that DYRK1A may be a novel dual-function player that regulates both cerebral angiogenesis and the maintenance of vascular integrity simultaneously to prevent hemorrhage in the developing brain. Consistent with this idea, it has been reported that DS fetuses displayed several developmental vascular defects revealed by ultrasound scanning (Chaoui, 2005), whereas DS adults suffered a significantly higher risk of hemorrhagic stroke, as shown in a large cohort study (Sobey et al., 2015), suggesting potential dual roles of DYRK1A in these vascular disorders.

DYRK1A function in angiogenesis and hemorrhage prevention is likely to be a calcium-dependent process, based on the comparative transcriptome analyses of the WT and dyrk1aa mutants (Fig. 7). The DEG analysis identified ‘calcium ion binding’ as the top-ranked GO category, which may reflect dysregulation of calcium homeostasis in dyrk1aa mutants. Of interest, several genes in the ‘calcium ion binding’ category have been associated with endothelial cell permeability and vascular dysfunction. For example, protocadherin genes such as the Pcdh-gamma cluster, one of the prominent groups of DEGs from our analysis, were recently reported to be highly expressed in endothelial cells of the brain microvasculature and may contribute to junctional stability of the blood-brain barrier (Dilling et al., 2017). Altered DEGs in the ‘oxidoreductase activity’ MF category and ‘oxidation-reduction process’ BP category in dyrk1aa mutants are also consistent with the disruption of Ca²⁺ signaling, because redox homeostasis is one of the major factors in regulating intracellular Ca²⁺ signaling events as well as vascular development (Lounsbury et al., 2000).

The calcium signal-related function of DYRK1A in cerebrovascular formation and maintenance is also supported by results from the in vivo chemical screening and calcineurin inhibition studies. The specific calcium chelator EGTA was identified as one of the most potent small molecules that rescued the cerebral hemorrhage elicited by DYRK1A inhibition (Fig. 6). The Ca²⁺ ion level inside the cytosol is normally maintained at a low concentration (approximately 100 nM), compared to a significantly higher (more than 20,000-fold) concentration in the extracellular environment (Clapham, 2007), illustrating the importance of the tight regulation of calcium homeostasis across the membrane in maintaining normal cellular functions. As EGTA is an extracellular calcium chelator, it may affect the calcium function either extracellularly by directly attenuating the activities of calcium-dependent extracellular vascular ligands and cell adhesion molecules (Deli, 2009; Tomita et al., 1996), or intracellularly by lowering the amount of calcium entry into the cells. Our data suggested that both scenarios would be possible based on the facts that cerebral vessels in the mutants were disrupted at the ultrastructural level, presumably because of the altered calcium-dependent vascular integrity (Fig. 4), and rescue of the mutants' vascular phenotypes were mimicked by inhibiting a calcium-dependent intracellular signaling pathway using FK506, a specific calcineurin inhibitor (Fig. 8). Consistent with these findings, dantrolene, a RyR antagonist that decreases the intracellular calcium level (Fruen et al., 1997), was also identified as a modestly effective small molecule in our chemical screening, although it did not reach statistical significance in our detailed analyses (Table S3 and data not shown).

An underlying mechanism by which DYRK1A regulates the calcium signaling in vascular formation is not yet clear. DYRK1A may affect Ca²⁺ flux directly, as in the case of GluN2A-containing N-methyl-D-aspartate glutamate receptor phosphorylation by DYRK1A, leading to the elevation of their density on the membrane, and eventually to neurological dysfunction (Grau et al., 2014). Alternatively, DYRK1A may indirectly influence Ca²⁺ signaling by phosphorylating mediators that determine the expression or activity of calcium-dependent effectors, similar to myocardial pathology, in which phosphorylation of alternative splicing factor by DYRK1A increases the expression of Ca²⁺/calmodulin-dependent protein kinase II δ (He et al., 2015). Because DYRK1A is likely to be functional both in the nucleus and in the cytoplasm, based on its ubiquitous localization at the cellular level (Marti et al., 2003), it may directly or indirectly regulate membrane-bound ion channels near the cell membrane (similar to GluN2A phosphorylation), intracellular signaling molecules in the cytoplasm or transcription factors in the nucleus, eventually maintaining the Ca²⁺ homeostasis of cells.

Together, our characterization of zebrafish dyrk1aa knockout mutants showed that DYRK1A is implicated in cerebral angiogenic activity and the maintenance of vascular integrity during development. The combination of detailed transcriptomic analyses and chemical screening results strongly suggested that the calcium-dependent signaling regulated by DYRK1A is one of the major signaling pathways responsible for such vascular phenotypes. However, detailed signaling molecules and pathways affected in dyrk1aa mutants remain to be further studied. Our results also illustrate the usefulness of zebrafish dyrk1aa mutants in providing an in vivo animal model to understand the pathophysiology of human vascular diseases related to DYRK1A function and in suggesting potential therapeutic approaches for effective treatments.

Zebrafish (Danio rerio) embryos of AB (WT) strain, transgenic zebrafish Tg(kdrl:EGFP) and dyrk1aa knockout mutant zebrafish (dyrk1aa^krb1) were maintained in E3 egg water (5 mM NaCl, 0.17 mM KCl, 0.33 mM CaCl₂, 0.33 mM MgSO₄) in a petri dish at 28.5°C. In order to generate transparent zebrafish embryos for imaging confocal microscopy, performing WISH and examining hemorrhagic phenotype, embryos were incubated in 1× PTU (0.003% 1-phenyl 2-thiourea, Sigma-Aldrich)-E3 egg water after 6 hpf. Zebrafish husbandry and animal care were performed in accordance with guidelines from the Korea Research Institute of Bioscience and Biotechnology (KRIBB) and approved by KRIBB-IACUC (approval number: KRIBB-AEC-17126).

Embryos at 52 hpf were fixed with 4% paraformaldehyde in 1× phosphate-buffered saline (PBS) for 4 h at room temperature (RT) and washed with 1× PBS containing 0.1% Tween 20 (1× PBST). The embryos were placed in o-dianisidine stain solution [0.6 mg/ml o-dianisidine (Sigma-Aldrich), 0.01 M sodium acetate (pH 5.5), 0.65% hydrogen peroxide and 40% ethanol] in the dark for several minutes at RT to detect hemoglobin activity. o-Dianisidine is a peroxidase substrate, and hemoglobin catalyzes the H₂O₂-mediated oxidation of o-dianisidine. Stained embryos were washed several times with 1× PBST and stored in 70% glycerol for imaging using an Olympus SZX16 microscope equipped with TUCSEN Dhyana 400DC camera. The cerebral hemorrhagic phenotype was calculated as the mean percentages by counting the number of embryos with hemorrhagic phenotype in the brain and the retina.

To isolate GFP-positive endothelial cells from Tg(kdrl:EGFP) embryos, we adopted the protocol developed by Manoli and Driever (2012). Briefly, 150 embryos at 48 hpf were dechorionated with 1 mg/ml protease (P6911, Sigma-Aldrich) in E3 egg water for 5 min at RT, and washed in 0.5× Danieau's solution [29 mM NaCl, 0.35 mM KCl, 0.2 mM MgSO₄•7H₂O, 0.3 mM Ca(NO₃)₂, 2.5 mM HEPES (pH 7.6)]. The yolks of those embryos were removed using the deyolking buffer (55 mM NaCl, 1.8 mM KCl, 1.25 mM NaHCO₃), followed by washing in 0.5× Danieau's solution, and embryonic cells were dissociated using FACS max cell dissociation solution (T200100, AMS Biotechnology) and the cell suspension passed through a 40 µm strainer (93040, SPL). FACS was performed at RT under sterile conditions using a FACSAria or FACSAria-Fusion (BD Biosciences).

Zebrafish embryos in each developmental stage or cells isolated by FACS were harvested with TRI reagent solution (Ambion), followed by purifying total RNA with Direct-zol RNA miniprep kit (Zymo Research) and synthesizing cDNA with SuperScript III First-Strand Synthesis System (Invitrogen). The synthesized cDNA was amplified by PCR using: the forward primer 5′-TCAGTGATGCTCACCCACAG-3′ and reverse primer 5′-CGTCATAGCCGTCGTTGTAA-3′ for dyrk1aa; the forward primer 5′-GAAACGACGGCATCAACAGG-3′ and reverse primer 5′-CAGCTCGGTCGTAGGCTTTT-3′ for dyrk1ab; the forward primer 5′-GGCAAGCTGACCCTGAAGTT-3′ and reverse primer 5′-TTCTGCTTGTCGGCCATGAT-3′ for EGFP; the forward primer 5′-CCCTTACCCTGGCTTACACA-3′ and reverse primer 5′-TCTTGTTGGTTCCGTTCTCC-3′ for kdrl; and the forward primer 5′-CTGGTTCAAGGGATGGAAGA-3′ and reverse primer 5′-ATGTGAGCAGTGTGGCAATC-3′ for eef1a1l1. The RT-PCR primers used for supplementary figures are shown in Table S4.

Harmine (Sigma-Aldrich) was dissolved in DMSO and added to 1× PTU-E3 egg water to earn final concentrations of 10-50 µM with 0.1% DMSO, and then applied to ∼40 embryos (dechorionated) on a 90 mm plate from 24 to 52 hpf.

For chemical screening, each of 1280 small molecules in the Library of Pharmacologically Active Compounds (LOPAC1280, Sigma-Aldrich) with 10 µM as a final concentration was individually applied into each well of 48-well plates containing 1× PTU-E3 egg water with 30 µM harmine and five embryos from 24 to 52 hpf. As a negative control, 1× PTU-E3 egg water containing 0.4% DMSO was used. Upon 30 µM harmine treatment, on average three embryos displayed the hemorrhage in the brain regions. Based on this criterion, the increased hemorrhage was defined by four or five embryos showing the hemorrhage in brain regions, whereas the reduced phenotype was defined by 0 to two embryos with such defect.

For EGTA and FK506 treatment, EGTA at final concentrations of 1-100 nM and FK506 at final concentrations of 5-100 ng/ml was applied into each well of 6-well plates containing ∼20 dechorionated embryos of WT and dyrk1aa mutants from 24 to 52 hpf, with 1× PTU-E3 egg water containing 0.1% DMSO alone used as a negative control.

To analyze the brain vasculature phenotypes of Tg(kdrl:EGFP) embryos with high resolution, embryos were grown up to 52 hpf and fixed with 1× staining solution (4% paraformaldehyde, 4% sucrose, 0.15 mM CaCl₂, 1× PBS) overnight at 4°C. Fixed embryos were washed briefly with 1× PBST and embedded on glass-bottomed imaging dishes with 1% low melting point agarose (Promega). The embryos were imaged using Olympus FV1000 confocal microscopy and the CtA development in the hindbrain was quantified using the length and branching points, by measuring total lengths with ImageJ and manually counting the junctional sites of the CtAs.

WISH in zebrafish embryos was performed as previously reported (Thisse and Thisse, 2008). The DNA templates for zebrafish dyrk1aa, dyrk1ab and dll4 (GeneBank accession numbers BC129212.1, NM_001347831.1, and NM_001079835.1, respectively) were amplified from cDNA of WT embryos at 52 hpf. Prof. Cheol-Hee Kim (Chungnam National University, Republic of Korea) donated kdrl, krox20 and isl1 DNA. Dig-labeled anti-sense probes were in vitro transcribed using SP6 or T7 RNA polymerase kits (Roche) and purified with NucAway spin columns (Invitrogen). Embryos for WISH were prepared by fixing with 4% paraformaldehyde in 1× PBS, dehydrating using methanol, stored at −20°C for 30 min, and serially rehydrated using 1× PBST. The rehydrated embryos were treated with proteinase K in 1× PBS and post-fixed with 4% paraformaldehyde. The antisense probes were hybridized with the fixed embryos at each developmental stage in hybridizing solution (5 mg/ml torula yeast RNA type VI, 50 µg/ml heparin, 50% formamide, 5× SSC, 0.1% Tween-20, 1 M citric acid used to adjust to pH 6.0) at 70°C overnight. The probes were washed serially using 2× SSCT-F (2× SSCT, 50% formamide, 0.1% Tween-20), 2× SSCT (2× SSCT, 0.1% Tween-20), 0.2× SSCT (0.2× SSCT, 0.1% Tween-20) at 70°C and 1× PBST at RT. The embryos were blocked with blocking solution (5% horse serum, 1× PBST) at RT, and the alkaline phosphatase-conjugated anti-digoxigenin antibody (11 093 274 910, Roche) was added into the blocking solution at 4°C overnight. To detect the expression signal of transcripts, NBT/BCIP solution (11 681 451 001, Roche) was used as alkaline phosphatase substrate. The expression patterns of transcripts were observed using an Olympus SZX16 microscope and imaged with a TUCSEN Dhyana 400DC or Olympus XC10 camera. To observe detailed expression patterns, whole-mount RNA in situ hybridized embryos were prepared for cryosectioning by embedding in an agar-sucrose solution (1.5% agar, 5% sucrose). After the agar blocks containing the embryos were kept in 30% sucrose solution, they were processed for transverse cryosectioning using a LEICA CM1860 cryostat at a thickness of 25-35 µm.

In order to prepare mRNAs of dyrk1aa, dyrk1aa-K193R and mCherryRed for rescue experiments, the pCS2+ vectors inserted with each DNA were linearized, in vitro transcribed using a mMESSAGE mMACHINE kit (Invitrogen) and purified with NucAway spin columns (Invitrogen). One-cell-stage eggs were collected and microinjected with each mRNA construct containing 0.05% Phenol Red solution as a visible indicator using a PV380 Pneumatic picopump (World Precision Instruments).

Tissue samples from embryos of WT and dyrk1aa^krb1 at 48 hpf were fixed immediately with 2% glutaraldehyde and 2% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) for 2 h at 4°C. Following three washes in the phosphate buffer, tissues were post-fixed with 1% osmium tetroxide on ice for 2 h and washed three times in the phosphate buffer. The tissues were then embedded in pure Epon 812 mixture after dehydration in ethanol series and followed by infiltration in propylene oxide:epon mixture series. Polymerization was conducted with pure resin at 70°C for 24 h. Ultrathin sections (∼70 nm) were obtained with a model MT-X ultramicrotome (RMC Boeckeler) and then collected on 100 mesh copper grids. After staining with 2% uranyl acetate (7 min) and lead citrate (2 min), the sections were visualized using the Bio-HVEM system (JEM-1400Plus at 120 kV and JEM-1000BEF at 1000 kV, JEOL).

Total RNA was isolated using Trizol reagent (Invitrogen). RNA quality was assessed by an Agilent 2100 bioanalyzer using the RNA 6000 Nano Chip (Agilent Technologies), and RNA quantification was performed using an ND-2000 Spectrophotometer (Thermo Fisher Scientific). For control and test RNAs, the construction of the library was performed using the SENSE mRNA-Seq Library Prep Kit (Lexogen) according to the manufacturer's instructions. Briefly, 2 μg total RNA are prepared and incubated with magnetic beads decorated with oligo-dT, and other RNAs except mRNA were removed by washing, and the mRNA was isolated from the oligo-dT bead of the poly(A) RNA selection kit (Lexogen). Library production was initiated by the random hybridization of starter/stopper heterodimers containing Illumina-compatible linker sequences to the poly (A) RNA bound to the magnetic beads. A single-tube reverse transcription and ligation reaction extends the starter to the next hybridized heterodimer, where the newly synthesized cDNA insert is ligated to the stopper. Second strand synthesis was performed to release the library from the beads, and the library was then amplified. Barcodes were introduced when the library was amplified. High-throughput sequencing was performed as paired-end 100 sequencing using HiSeq 2500 (Illumina). The sequenced reads were mapped to the University of California Santa Cruz zebrafish genome (danRer10) using STAR (v.2.5.1) (Dobin et al., 2013), and the gene expression levels were quantified using the count module in STAR. The edgeR (v.3.12.1) (Robinson et al., 2010) package was used to select differentially expressed genes from the RNA-seq count data. Meanwhile, the trimmed mean of M-values-normalized counts per million value of each gene was set to a baseline of 1 and log2-transformed for volcano plot drawing (Figs 7A and S7A).

Statistical analyses of the data were performed using a Mann–Whitney U test or one-way ANOVA with Dunnett's multiple comparisons test using Prism software (Ver.7). Data are mean±s.e.m. with *P<0.05, **P<0.01 and ***P<0.005.

The authors thank members of the Dr Lee and Dr Yu laboratories for helpful discussion on the manuscript.