Disruption of the Extracellular Matrix Progressively Impairs Central Nervous System Vascular Maturation Downstream of β-Catenin Signaling

For the mouse data, no significant deviations from normality were detected using the Kolmogorov-Smirnov test with Dallal-Wilkinson-Lillie for approximation at a threshold of α<0.05, and 2-tailed heteroscedastic t tests with Welch correction for unequal variance between groups was applied using Prism 6 (GraphPad Software) if not stated differently. For formulas and details on RNA sequencing data, please see below Sample Processing, RNA Sequencing, and Statistical analysis section. In brief, for differential expression testing, we used a similar test as previously described.²⁹ False discovery rates were calculated by the Benjamini-Hochberg method. Genes with <5% false discovery rate were considered differentially expressed. For the E14.5 Axin1^iEC−^OE (Gene Expression Omnibus entry: GSE66848) and Ctnnb1-KO^iEC samples³⁰, we applied a <5% false discovery rate cutoff to the DESeq P values. These samples were prepared according to endothelial-specific translating ribosome affinity purification (TRAP) sequencing.³¹

For the zebrafish analysis and analysis of Cldn5 (claudin5) expression (Figure IV in the online-only Data Supplement) the data did not pass the normality test and differences between groups were, therefore, evaluated using a nonparametric Mann-Whitney test (please see below zebrafish manipulations, imaging, and statistical analysis).

Mouse strains are summarized in the Major Resources Tables (Table I in the online-only Data Supplement). Briefly, generation of Cdh5CreERT2,³² Flk1Cre,³³ Ctnnb1^{flox(ex2-6)/−)},^6,34 mCherryTRAP,^30,31 the Wnt-reporter mouse strain: TCF/Lef:H2B-GFP³⁵ (Jackson Laboratory, stock no. 013752³⁵) and Adamtsl2^−/− or Adamtsl2^+/−36 mice were described previously. R26Axin1IL (AOE [Axin1 overexpressing] mice) were generated for this study as described in the paragraph below. Vascular endothelium-specific and inducible AOE (Axin1^iEC−^OE) embryos were generated by breeding male Tg(Cdh5-cre/ERT2)1Rha (Cdh5-CreERT2^+/−)³² mice with R26Axin1IL females and were kept on a CD1 background. Ctnnb1-KO^iEC embryos resulted from intercrosses between Chd5-CreERT2^+/−32 male and β-catenin lox-P flanked (floxed) β-catenin^flox/flox (floxed Ctnnb1)³¹ female mice and were maintained on a C57BL/6J background.

Inducible EC-specific TRAP^30,31 was performed by initially crossing male Cdh5CreERT2^+/−32 with female Gt(ROSA)26Sor-mCherry-Rpl10a (mCherryTRAP^+/+)^30,31 mice to activate the mCherry transgene in vascular endothelium: the stop cassette (tpA: 3 SV40 polyAs) is deleted and the ubiquitous enhancer/promoter (CAGGS) drives expression of mCherry-tagged Rpl10a in the presence of Cre recombinase. To generate endothelial-specific AOE or Ctnnb1 knockout TRAP animals, we crossed male Cdh5CreERT2^+/−;mCherryTRAP^+/− with female R26Axin1IL^Homozygous mice or a floxed Ctnnb1 allele was crossed into the Cdh5CreERT2;mCherryTRAP mouse line, and then, translated and transcribed RNAs from Ctnnb1-KO or AOE embryo forebrains were isolated and subjected to RNA-seq. RNA-seq data from Axin1^+/+;Cdh5CreERT2;mCherryTRAP or Ctnnb1^+/+;Cdh5CreERT2;mCherryTRAP embryos were used as control (for details see TRAP-Seq and TRAP-Seq of β-catenin/Ctnnb1-KO^30,31).

Crosses between these strains as indicated in the text were produced by natural breeding according to the schemes presented in the Major Resources Tables (Table II in the online-only Data Supplement), and genotyping was performed using the primers also specified in the Major Resources Tables (Table III in the online-only Data Supplement). Embryos of both sexes were collected and used for the analyses as illustrated in Figure I in the online-only Data Supplement. However, no discrepancies between the genders were detected at E11.5, revealed by dividing the average RPKM expression levels of the Y chromosome expressed genes Ddx3y, Eif2s3y, and Uty, respectively, with the RPKM expression levels in the E11.5 RNA-Seq and E14.5 TRAP-Seq samples (Figure I in the online-only Data Supplement).

Animal care and research protocols were in accordance with institutional guidelines and approved by the appropriate Stockholm Norra Djurförsöks Etiska Nämnd or Cleveland Clinic committees. For staging of embryos, the morning of vaginal plug was designated as E0.5. To induce recombination in CreERT2 mice, tamoxifen (Sigma) was resuspended in corn oil (20 mg/mL) and administered by oral gavage (200 μL per mouse) at E8.5 to E10.5 as indicated.

The R26-Axin1-IRES2-LacZ allele was generated by assembly of a construct containing the splice acceptor (SA) from pBigT, followed by the puromycin resistance gene, the bovine growth hormone polyA sequence, the CAGGS enhancer/promoter/intron/SA cassette from pCAGGS, a floxed tpA (3 SV40 polyA sequences from pBigT), a cassette containing IRES2-LacZ preceded by a nuclear localization signal and finally the pCAGGS polyA sequence. Subsequently, mouse Axin1 gene was polymerase chain reaction amplified using the following primers: NheI-KOZAK-cmyc-mAxin1-5′ GGG AAA GCT AGC GCC GCC ACC ATG GAG CAA AAG CTC ATT TCT GAA GAG GAC TTG AAT GAA ATG AAT GTC CAG GAG CAG GG -3′ and SbfI-5′ GGG AAA CCT GCA GGT CAG TCC ACC TTT TCC ACC T -3′ digested with NheI and SbfI and subcloned in between the floxed tpA and IRES2LacZ cassettes into pBSApBpACAGftILn.⁶ The conditional expression construct was released with PacI and AscI and subcloned into the PacI and AscI sites of pRosa26PAS.³⁷ The construct was linearized and electroporated into F1 ES cells.³⁸ Colonies were screened by polymerase chain reaction using the primers for Rosa26-WT sequence and Rosa26-SA sequence specified in Table III in the online-only Data Supplement. Positive colonies were expanded. One targeted clone was injected into host (C57BL/6J, Jackson Laboratories) blastocysts by the Karolinska Center for Transgene Technologies, Karolinska Institutet. Mice and embryos were genotyped using the primers against R26Axin1IL, or R26-wt as indicated in the Major Resources Tables (Table III in the online-only Data Supplement).

Slides were washed in PBS containing 0.02% NP-40 and 2 mmol/L MgCl2. The X-gal signal was detected with a solution consisting of 5 mmol/L K3Fe(CN)6, 5 mmol/L K4Fe(CN)6, 2 mmol/L MgCl2, 0.01% sodium deoxycholate, 0.02% NP-40, and 1 mg/mL X-gal at 37°C. The sections were counterstained with Nuclear Fast Red Solution (Sigma Aldrich, N3020).

Embryos were dissected and fixed in 4% paraformaldehyde for 2 hours followed by incubation in 15% sucrose for 4 hours and finally in 30% sucrose overnight, before embedding and snap freezing in OCT (optical coherence tomography freezing and embedding medium for cryosectioning; Leica, 14020108926). Cryosections (10–20 μm thick) were cut using a cryostat (Leica) and mounted on Superfrost Plus slides (Thermo Scientific, 4951PLUS4). Tissues were stained with the following primary antibodies (also listed in Table IV in the online-only Data Supplement) diluted in blocking solution (PBS, 3% BSA, and 0.25% TritonX-100) overnight at 4°C: Flk-1 (BD Biosciences, 555307) 1:1000, CD31/PECAM1 (R&D systems, AF3628) 1:1000, Glut1 (Santa Cruz Biotechnology, sc1605) 1:1000, Cldn5 (Life Technologies, 34–1600) 1:500. Cy3- or Alexa Fluor-coupled species-specific secondary antibodies (Molecular Probes, Jackson Immuno Research) were used at 1:1000. Antibody concentrations are listed in the Major Resources Tables (Table IV in the online-only Data Supplement). Mounting was done with ProLong Antifading reagent (Life Technologies, P36931) or Immu-Mount (Thermo Scientific, 99-904-02).

For chromogenic secondary detection, species-specific biotin-conjugated secondary antibodies (Jackson Immuno Research) were used together with the VECTASTAIN ABC system (VECTOR laboratories, PK-4000). Development was done with DAB (3′3 diaminobenzidine) peroxidase substrate solution (Sigma Aldrich, D8001) and before mounting with DPX (Sigma Aldrich, 06522) the slides were dehydrated in 70% ethanol for 1 minute, 95% ethanol for 1 minute, 99% ethanol for 1 minute, and xylene for 5 minutes. Negative controls were performed by the same conditions described above but by omitting the primary antibody or running with only IgG followed by Alexa Fluor-coupled secondary antibodies (Figure II in the online-only Data Supplement) as previously described.

Pericyte abundance was evaluated by normalizing PDGFR-β (platelet-derived growth factor receptor beta) immuno-positive signals in a fixed area within the forebrain PVP to the size of the area leading to an estimation of the relative contribution of pericytes (in percent) to the tissue. Vessel-associated pericyte processes were evaluated as the percentage of the total pericyte signal that overlapped with the endothelial PECAM1 (platelet endothelial cell adhesion molecule; CD31 [cluster of differentiation 31]) immune-positive signal within a fixed area of the forebrain PVP.

E11.5 embryos were dissected from the yolk sac, genotyped, and fixed in 4% paraformaldehyde overnight, followed by overnight incubation in sterile 1×PBS. Telencephalic vesicles were dissected out from these embryos in sterile 1×PBS and incubated in block/permeabilization buffer (1% BSA, 0.5% TritonX-100 in PBS) overnight. The samples were incubated with the following primary antibodies (also listed in the Major Resources Tables [Table IV in the online-only Data Supplement]) for 48 hours; CD31/PECAM1 (R&D systems, AF3628, goat IgG) 1:1000 and collagen IV (AbDSerotec, 2150-1470, rabbit IgG) 1:100 diluted in 1:1 mixture of block/permeabilisation buffer-PBS with or without 4% normal donkey serum. The tissue was washed with block/permeabilization buffer-PBS for 4×10 minutes and incubated with secondary antibodies (Alexa Fluor-coupled coupled secondary antibodies, Molecular Probes, 1:200) for 2 hours or +4C O/N and finally washed for 2×10 minutes in 1×PBS or with block/permeabilisation buffer-PBS for 4×10 minutes, rinsed in 1×PBS, and flatmounted on microscope slides (Superfrost Plus glass, Thermo Scientific, 4951PLUS4) with ProLong Antifading reagent (Life Technologies P36931). The flat mounts were achieved by 2 cuts through the neocortex exposing the intact medial ganglionic eminence (MGE) and lateral ganglionic eminence (Figure III in the online-only Data Supplement). Negative controls were performed by the same conditions described above but by omitting the primary antibody running with the only IgG followed by Alexa Fluor-coupled secondary antibodies (Figure II in the online-only Data Supplement) as previously described. Imaging was performed using a Zeiss confocal microscope (Zeiss LSM 510).

Four areas within the MGE, (2 per telencephalic vesicle) per embryo were investigated. Photomicrographs of the MGE were taken from the perineural vascular plexus through radial vessels and PVP up to the stretching filopodia. Image analysis was performed with FIJI³⁹ using plug-ins for cell counting, branch identification and length, the number and diameter of radial vessels were calculated. A branch point was defined as a point where branching occurred in at least 3 directions. Diameters of PVP vessels were measured using ImageJ. In total, >1000 individual width measurements were averaged per embryo. Vascular regression was quantified as the number of red pixels in collagen IV (red) and PECAM1 (green) double-stained embryos (where collagen IV-positive vessels appear in yellow), converted to μm² and normalized to an area of interest of 1 mm², calculated using ImageJ. For each embryo, brain snapshots for each of the 2 brain hemispheres were quantified in a semiautomated manner and averaged. For all statistical comparisons, except Apod-stained tissues, 2-tailed heteroscedastic t tests with Welch correction were performed using Prism 6 (GraphPad Software). For Apod-stained tissues, the nonparametric Mann-Whitney test was used.

E12.5 mutant and control embryos underwent cardiac perfusion with Alexa Fluor 555-conjugated 70 kDa dextran (50 μL of 2.5 mg/mL in PBS; Invitrogen) using a pulled glass pipette with mouth connector and tubing (Sigma, P0799) in the course of the 15 to 20 minutes when a visible heart beat could be observed. Fluorescent visualization of the entire vasculature confirmed the perfusion. Tracer was left circulating for 10 minutes at room temperature, embryos were decapitated, their heads fixed in 4% paraformaldehyde overnight at 4°C, cryosectioned and stained as described above. Staining was performed with the following primary antibodies; CD31/PECAM1 (R&D systems, AF3628) 1:1000 and monoclonal TER-119 (R&D Systems, MAB1125) 1:250 followed by a 2 hours incubation at room temperature with DyLight Fluor-coupled secondary antibodies (1:1000, Thermo Scientific Pierce).

To obtain brain ECs, dams were sacrificed and ventral forebrains (MGE/lateral ganglionic eminence) were isolated. Genotyping was performed based on EGFP and LacZ expression. After isolation and dissection in ice-cold PBS (w/o Mg++ or Ca++, Gibco), an equal volume of a mixture of collagenase (2.5 mg/mL Worthington Biochemical Corporation LS004176) and DNase (30 U/μL, SIGMA, to a final of 0,04 mg/mL) in 37°C prewarmed PBS (Mg++ and Ca++, Gibco) was added for 15 to 30 minutes at 37°C on a shaker. Cells were filtered through a cell strainer (40 μm mesh, Falcon) followed by addition of 10% FBS:DMEM to a final volume of 6 mL and centrifugation at 800×g, +4°C for 5 minutes. The supernatant was removed, cells were incubated for 10 minutes in 500 µL blocking solution (2% FBS:PBS), divided in tubes as indicated in Table V in the online-only Data Supplement and stained with antibodies: CD31/PECAM1 PE (BD Biosciences, 553373) 1:1000, CD45 APC-CY7 (BD Biosciences, 557659) 1:100 or control IgG_{2A, K} PE (affymetrix eBioscience, 12432181) and APC-CY7 (BD Biosciences, 552773) and incubated on ice for 10 minutes. Propidium iodide was included to distinguish between dead and living cells. Samples were washed with blocking solution (50 mL/condition) to remove any traces of remaining antibody followed by centrifugation (800×g, +4°C) for 5 minutes. Before flow cytometry cells were recovered in 1.5 mL of blocking solution.

Axin1^iEC−^OE mice were crossed into the Wnt-reporter mouse strain: Tg(TCF/Lef1-hIST1h2BB/EGFP)^61hadj/J.³⁵ Axin1^iEC−^OE/GFP+ cells were analyzed as readout for Wnt signaling. Wnt-reporter activity using flow cytometric analysis (FACScalibur BD) was used (Table V in the online-only Data Supplement). For each sample, cell populations were discriminated from debris in the side scatter-forward scatter (FSC) dot plot. Singlet’s gating was done in the FSC pulse area—FSC pulse width plot. Blood cells were gated out by CD45 labeling with CD45 APC-CY7 (BD Biosciences, rat IgG_2B 557659). The CD45 negative population was presented in the CD31/PECAM1 forward scatter dot plot for EC gating. Finally, the ECs were plotted in the PECAM1/GFP dot plot for determining the endothelial GFP positive/negative events. The percentage of GFP+CD45- versus total CD45- from 5 runs (n=5) were calculated and a 2-tailed heteroscedastic t tests with Welch correction were performed using Prism 6 (GraphPad Software) to gain statistical data of the Wnt positive ECs.

ECs from E11.5 ventral forebrains (MGE/lateral ganglionic eminence) were prepared, and ECs were labeled and flow sorted as described above. CD31/PECAM1 positive cells, obtained by fluorescence-activated cell sorting (FACS) using MoFlo XPD (Beckman Coulter), were further processed for total RNA (RNeasy Micro Kit no. 74004, Qiagen) followed by pre-RNA Sequencing procedures. ECs from E14.5 forebrains were prepared as described in Hupe et al.³⁰ In brief, libraries were prepared using SMARTer Ultra Low Input RNA for Illumina. Sequencing kit (Clontech) and sequenced on an Illumina HiSeq 2000. Reads were aligned using Bowtie 0.12.7⁴⁰ to the genome (assembly mm⁹) and transcriptome (Ensembl annotation, downloaded May 16, 2011). Transcriptome-mapped reads were translated to genome coordinates, merged with genomic alignments, and nonuniquely aligning reads were discarded. We then calculated expression values and read counts using rpkmforgenes (available at http://sandberg.cmb.ki.se/rnaseq/)⁴¹ with settings—full transcript—mRNAnorm and RefSeq gene annotation from July 31, 2011. RPKM values for constructs were calculated from counting bowtie alignments and normalized by lengths and the normalization read sum from rpkmforgenes. Raw data and expression values have been deposited to NCBI Sequence Read Archive with ID SRP056125 and Gene Expression Omnibus with ID GSE66848. For differential expression testing, we used a similar test as previously described.²⁹ False discovery rates were calculated by the Benjamini-Hochberg method. Genes with <5% false discovery rate were considered differentially expressed. For the E14.5 Axin1^iEC−^OE (Gene Expression Omnibus entry: GSE66848) and Ctnnb1KO^iEC samples,³⁰ we applied a <5% false discovery rate cutoff to the DESeq P values. These samples had been prepared according to endothelial-specific TRAP sequencing.³¹

Zebrafish and embryos were maintained, handled, and raised under standard conditions as previously described.²⁵ Transgenic and mutant lines used include Tg(Kdrl:EGFP)^s843, Tg(Kdrl:hsa.HRAS-mCherry)^s896, Tg(Hsp70l:Mmu.Axin1-YFP)^w35, and Adgra2^s984.^25,42–44

Capped-RNA was synthesized from full-length mouse Adamtsl2 template (IMAGE ID 40097418, LifeSciences) using the mMessage mMachine SP6 kit (Ambion, Carlsbad, CA) and injected at the 1-cell stage at 100 pg. Heat-shock was performed on manually dechorionated embryos at 26 or 30 hpf replacing the medium with prewarmed 38°C egg-water and subsequent incubation for 50 minutes. Sixty hours postfertilization embryos and 72 hpf larvae were anesthetized and mounted in 1% Ultrapure LMP Agarose (Invitrogen, Waltham, MA) in glass-bottom dishes (MatTek Corporation, Ashland, MA). Images were acquired using a Zeiss LSM710 confocal microscope (Oberkochen, Germany). Hindbrain cerebral central arteries (CtAs) were quantified and assessed for lumenization. Basilar artery connections were quantified as the total number of CtAs connecting from both hemispheres to the basilar artery. Nonparametric Mann-Whitney test was used to analyze the data by Prism 6 (GraphPad Software).

To study the role of Axin1-mediated repression of β-catenin signaling in ECs during different stages of CNS vascular development, we generated R26-Axin1-IRES2lacZ mice that allowed us to conditionally overexpress Axin1 (AOE) on induction of Cre-mediated recombination (Figure 1A). Our targeting design also allowed visualization of recombination events using β-galactosidase staining. Importantly, crossing these mice to the EC-specific and inducible Cdh5:CreERT2 mouse line allowed induction of AOE in vascular ECs (Axin1^iEC−^OE) by administration of tamoxifen at developmental stages of interest (Figure 1A).^32,33 Tamoxifen treatment at E9.5 in Axin1^iEC−^OE heterozygous carriers led to Cre-mediated recombination and robust expression of LacZ in ≈60% of the forebrain ECs at E11.5 (Figure 1B and 1C). RNA Smart-Seq analysis on FACS ECs revealed that Axin1 expression was elevated by 3.2-fold in the entire EC population under these conditions, implying a ≈5-fold induction of Axin1 in recombined ECs (Figure 1D). Consistent with the role of Axin1 in inhibition of canonical Wnt-signaling, we found a robust downregulation of Wnt-target genes, including Dkk3, Apcdd1, Nkd1, and Tcf7 in the Axin1^iEC−^OE mutants (Figure 1D). To further validate reduced Wnt-signaling activity in our Axin1^iEC−^OE mutants in vivo, we crossed Axin1^iEC−^OE mice into an established Wnt-reporter mouse model expressing EGFP (enhanced green fluorescent protein) under the Tcf/Lef1 promoter in ECs. We detected a strong reduction in the EGFP signal in Flk1 (fetal liver kinase 1, also known as Kdr:kinase insert domain receptor, encoding Vegf-r2 [vascular endothelial growth factor receptor 2]) positive ECs compared with control reporters (Figure 1E). Furthermore, we labeled ECs with an anti-PECAM1 (CD31) antibody in our Wnt-reporter mice and performed flow cytometry counting to quantify the reduction in EGFP-expression (Figure 1F). Both methods independently confirmed that Wnt-signaling was downregulated in ECs in Axin1^iEC−^OE mutant mice to ≈40% of that seen in controls.

EC-specific deletion of β-catenin is known to culminate in severe vascular malformation and hemorrhage in the entire CNS.⁶ To examine the underlying causes of these severe vascular phenotypes and understand the role of Axin1 in this context, we analyzed forebrain vascular phenotypes 24 and 48 hours after tamoxifen-induced recombination in Axin1^iEC−^OE mice. To accurately visualize the 3-dimensional vascular architecture of the embryonic telencephalon, that is, the forebrain area where reduced β-catenin signaling was observed in Axin1^iEC−^OE mice, we first developed a protocol for generating open book flat-mount preparations of this tissue (Figure 2A; Figure III in the online-only Data Supplement). Cre-expression was induced in Axin1^iEC−^OE mice at E9.5 or E10.5 and animals were euthanized for flat-mount preparations at E11.5, (Figure 2B through 2Y). Already after 24 hours of Axin1^iEC−^OE, we found significant retraction of capillaries in the PVP, as judged by the appearance of empty, collagen IV positive basement membrane sleeves of Axin1^iEC−^OE mice (Figure 2B and 2C). While the number of tip cells, filopodia, and PVP branch points were unaffected (Figure 2B, 2E, 2F, and 2H), the number of radial vessels (Figure 2D) were significantly reduced, indicating impaired vascular stability but not sprouting shortly after reducing β-catenin signaling. After 48 hours of Axin1^iEC−^OE, while there was no longer any increase in PVP vessel retraction (Figure 2I and 2J), the number of radial vessels were still reduced compared with control mice (Figure 2I and 2K). Also at 48 hours of Axin1^iEC−^OE, the number of tip cells, filopodia, and branch points did not change (Figure 2I, 2L, 2M, and 2O). However, a significant increase in vessel diameter in the PVP was observed, which had not yet developed at 24 hours of Axin1^iEC−^OE (Figure 2B, 2G, 2I, and 2N). Importantly, EC-specific loss-of-function mutants of β-catenin (Ctnnb1-KO^iEC), at E9.5 phenocopied the vascular changes in the forebrain of Axin1^iEC−^OE at E11.5 (Figure 2P through 2V) thus substantiating that the effects we observed in the latter were caused by overall reduced Wnt-signaling activity.

Vascular remodeling in the CNS is often accompanied or caused by impaired basement membrane formation and coverage with vascular mural cells.¹ To examine, if this was also the case in Axin1^iEC−^OE mice, we stained the basement membrane with anti-Laminin-α4 antibodies and the mural cells with anti-Pdgfrβ antibodies (Figures 2W through 2Y and 3A through 3D). Forebrain vessels in both control and Axin1^iEC−^OE mice exhibited robust coverage overall by Laminin-α4 at E11.5 (Figure 2W through 2Y). Axin1^iEC−^OE mice did, however, exhibit areas with reduced basement membrane coverage (Figure 2W), suggesting that vascular maturation might be delayed in the PVP of Axin1^iEC−^OE compared with control mice. Similarly, vascular coverage by Pdgfrβ-positive mural cells was as expected rather poor but similar in both control and Axin1^iEC−^OE mice overall (Figure 3A through 3D). The same areas exhibiting reduced basement membrane coverage (Figure 2W and 2X), however, also demonstrated a lack of mural cell cells (Figure 3B), suggesting that mural cell recruitment and basement membrane formation may coincide during development and is delayed focally in Axin1^iEC−^OE mice.

To study the temporal role of β-catenin in CNS angiogenesis and vascular maturation in more detail, we analyzed EC invasion, angiogenesis, vascular stability, and maturation in the CNS of inducible β-catenin loss-of-function zebrafish models. Consistent with previous findings,²⁵ gpr124 mutants or morphants that are unable to induce β-catenin signaling in ECs or embryos conditionally AOE on heat shock at 26 hpf, that is, just before initial EC invasion into the CNS (corresponding to E8.5 in mice), were unable to form a CNS vasculature and completely lacked CtAs at 60 hpf (Figure 4A). Heat-shocking embryos at 30 hpf, a few hours after the initial vascular invasion of the CNS (corresponding to E9.5 in mice), however, did not completely impair CNS vascular development but AOE still resulted in a dramatic reduction of total and lumenized CtAs at 60 hpf, while not affecting the formation of endothelial tip cells in the brain (Figure 4B through 4E). This suggested that β-catenin signaling is important for CNS vascular stability but not ongoing angiogenesis from 30 hpf onwards (ie, after initial invasion), similar to our findings in Axin1^iEC−^OE mice.

To analyze long-term effects of AOE during mouse development, we investigated if and when CNS hemorrhaging occurred in EC-specific constitutively AOE, AOE^Flk1 mice. These mice developed without pronounced vascular phenotypes until E11.5, but at E12.5 robust hemorrhage could be observed specifically localized to the forebrain (Figure 5A). To analyze whether Wnt-signaling was important in a particular window during development, we induced AOE at E8.5, E9.5, or E10.5 in Axin1^iEC−^OE mice and analyzed brain hemorrhage phenotypes from E12.5 onwards. Induction of Axin1^iEC−^OE at E8.5 phenocopied the forebrain hemorrhage observed using constitutive Axin1^Flk1 mice at E12.5 consistent with ECs invading the CNS and requiring β-catenin signaling from E8.5 onwards (Figure 5A). Postponing the induction of Axin1 to E9.5 resulted in a delay in forebrain hemorrhage which was not evident at E12.5 but instead prominent at E13.5 (Figure 5B). Axin1^iEC−^OE in these mice did not result in premature death, as a Mendelian 50% of the recovered embryos were Axin1^iEC−^OE mutants and 50% Cre-negative controls (Figure 5D). The penetrance of forebrain hemorrhage within the Axin1^iEC−^OE population was ≈25% (±6%), and as expected 0% (±0%) in the Cre-negative control population (Figure 5E). As the BBB is thought to be developed at E13.5,⁴⁵ we hypothesized that β-catenin signaling would be particularly important at the developmental window immediately before and including E13.5. Surprisingly, however, induction of AOE at E10.5 delayed both the onset and penetrance of forebrain hemorrhage until E14.5 (Figure 5C), whereas no hemorrhage could be observed at E13.5 in these mice. This suggested that 4 days of Axin1^iEC−^OE are required to develop forebrain hemorrhage regardless of the developmental window (ie, E8.5–E12.5, E9.5–E13.5, or E10.5–E14.5) during which AOE was induced. These findings are supported by previous observations from our and other laboratories that brain hemorrhaging was observed in many different Wnt-pathway mutants, including Wnt7a, Wnt7b, Fz4, Lrp5, Lrp6, and Ctnnb1 single or compound deficient strains,^6,9 suggesting that the observed effects on impaired CNS vascular integrity are not Axin1-specific but rather are because of overall perturbed Wnt-signaling activity in ECs. In these other models of perturbed Wnt/β-catenin signaling, bleeding is, however, also observed in other brain regions as well as the spinal cord, whereas in the Axin1^iEC−^OE embryos, the phenotype was restricted to the ventral forebrain. This suggests that while β-catenin is important for maintaining the structural integrity of all CNS vessels, Axin1 is mainly involved in vascular destabilization because of reduced β-catenin signaling in the forebrain ECs.

Cerebral hemorrhage in adults often results from underlying cardiovascular problems such as hypertension. To investigate if Axin1^iEC−^OE embryos (E13.5) exhibited anatomic indicators of changes in cardiac output, which potentially could lead to general or local forebrain hypertension, we dissected the hearts of Axin1^iEC−^OE and control littermates and compared their forebrain vascular anatomy. The hearts of Axin1^iEC−^OE embryos were morphologically indistinguishable from controls without cardiac indicators of hypertension such as left ventricular hypertrophy or dilation (Figure 5F). Furthermore, the forebrain vasculature was well perfused, and the vessels were of similar or larger caliber in Axin1^iEC−^OE mice compared with controls (Figure 5G) suggesting that the local blood pressure in the forebrain was likely not increased in Axin1^iEC−^OE.

To investigate if the BBB breaks down before hemorrhaging, we next determined the severity/extent of leakage in the embryonic brain at E12.5, that is, 3 days after Axin1^iEC−^OE and one day before hemorrhage. Because the BBB gradually forms and tightens, in a size-selective manner over time, small molecular weight substances (≤10 kDa) leak spontaneously in embryonic forebrains at stages before E14.5⁴⁵ under physiological conditions. However, tracers with a higher molecular weight (ie, ≥70 kDa) are restricted within the forebrain vasculature already from E11.5 onwards.⁴⁶ Therefore, we applied 70 kDa dextran by cardiac perfusion and followed the perfusion and leakage of this tracer at E12.5. Leakage of 70 kDa dextran or extravasation of TER119+ erythroblasts or mature erythrocytes were not observed in Axin1^iEC−^OE mice compared with controls (Figure 5G), indicating that vascular integrity was not severely compromised 3 days after AOE. Furthermore, pericyte recruitment or pericyte-EC interaction, an important feature of the developing and maturing BBB,^47,48 was not impaired in Axin1^iEC−^OE mice, as demonstrated by comparable PDGFRβ-positive mural cell numbers and vascular coverage even in hemorrhaging areas of the forebrain vasculature at E13.5 (Figure 5H and 5I). Collectively, these findings indicate that AOE in our Axin1^iEC−^OE mouse line induces inhibition of Wnt-signaling in ECs, which leads to progressive vascular destabilization, regression, and dilation.

To investigate the mechanisms of Wnt-β-catenin signaling specifically in forebrain ECs during CNS vascular development we isolated mRNA from forebrain ECs of 3 independent β-catenin loss-of-function mouse models: (1) Axin1^iEC−^OE embryos at E11.5 (ie, before BBB formation), (2) Axin1^iEC−^OE embryos at E14.5 (ie, after BBB formation) by TRAP (translating ribosome affinity purification),^30,31 and (3) Ctnnb1-KO^iEC embryos at E14.5 (here data were acquired from our previously published resource³⁰), in all cases 48 hours after induction of Cre-mediated recombination. RNA was used for RNA Smart-Seq²⁹ (Figure 6A). Importantly, our cell preparation was pure and viable as cells were negative for propidium iodide and no significant neural, glial, and pericyte contaminations were detected in the RNA Smart-seq data set (Table VI in the online-only Data Supplement). After correcting for multiple testing, we found 455 genes significantly upregulated and 128 genes downregulated in Axin1^iEC−^OE mice at E11.5 on a genome-wide scale (Figure 6B). EC-specific receptors, such as Kdr (Flk1), Flt1 (FMS-like tyrosine kinase 1, encoding vascular endothelial growth factor receptor 1), Gpr124, and Reck, were not differentially expressed at E11.5 and only Reck was slightly downregulated in Axin1^iEC−^OE embryos at E14.5 (Figure 6C), suggesting that canonical Wnt-signaling is not involved in maintaining endothelial identity per se during CNS vascular development but could as previously reported⁶ be involved in specifying CNS-specific endothelial characteristics. To test this hypothesis, we investigated the expression of EC-specific Wnt-pathway receptors at E11.5 and E14.5. Genes encoding Wnt-receptors Gpr124, Reck, FZD4, FZD6, FZD10, Lrp5, and Lrp6 were not differentially regulated at E11.5 but became either upregulated (FZD4) or downregulated (FZD6, FZD10, Lrp6, and Reck) at E14.5 in either Axin1^iEC−^OE (Lrp6 and Reck) or both Axin1^iEC−^OE and Ctnnb1-KO^iEC mice (FZD6 and FZD10; Figure 6C). The endothelial Wnt-target Tnfrsf19/Troy was downregulated under all 3 conditions (Figure 6C). Other BBB markers, such as Slc2a1 (Glut1 [solute carrier family 2, facilitated glucose transporter member 1]), Abcb1a, Slc7a5, Slco1c1, Slco2b1, and Ocln,⁴⁹ were downregulated at both developmental time points and in both mouse strains, and Slc7a1 became downregulated at E14.5, whereas Jam2 was only downregulated at E11.5 (Figure 6D and 6E), consistent with previous studies.⁶ In contrast, the bonafide tight junction marker Cldn5, which is expressed by all tight junction-bearing ECs,⁵⁰ was upregulated at E11.5, but downregulated at E14.5, suggesting a differential role for β-catenin signaling in regulating tight junction formation at early and later stages of brain vascular development (Figure 6E). To further study whether Cldn5 and the BBB-marker Glut1 were also regulated at the protein level we performed immunohistochemistry of E11.5 Axin1^iEC−^OE and control mice 48 hours after tamoxifen injection (Figure IV in the online-only Data Supplement). In agreement with our RNA-seq data, Glut1 staining was reduced, although with borderline significance (P=0.054) in the Axin1^iEC−^OE mice, but Claudin-5 staining was significantly reduced in Axin1^iEC−^OE mice, suggesting that the elevated mRNA level seen at E11.5 in these mice might be a compensation mechanism. Combined, these results indicate that AOE and concomitant suppression of β-catenin signaling results in transcriptional changes involved in BBB formation and maturation that precede vascular leakiness and rupture. Moreover, our data indicate that inhibition of Wnt-signaling does not affect EC specification per se but impairs the acquisition of CNS EC-specific phenotypes.

To investigate the functional outcomes associated with deregulated gene expression in Axin1^iEC−^OE mice, we performed a pathway analysis including all 583 upregulated or downregulated genes, using DAVID.⁵¹ Using stringent criteria (false discovery rate [FDR]<0.1), 3 significantly enriched functional gene clusters were identified; extracellular proteins, chemokine signaling, and metal-binding proteins, of which the extracellular proteins cluster was the most significant and contained the most deregulated genes (Figure 7A). The ECM is important for coordinating brain vascular development, remodeling, and BBB maturation,²⁷ and we confirmed that nonforebrain CNS vessels, as well as peripheral vasculatures in the paw and dorsal intersegmental region, were not affected by 2 days of AOE in E11.5 Axin1^iEC−^OE mice compared with controls (Figure V in the online-only Data Supplement), suggesting that forebrain vessels were selectively affected in these mice. We, therefore, used our recently released resource³⁰ to filter the cluster of deregulated extracellular proteins for genes specifically expressed in brain ECs but not in ECs from other tissues or in non-EC brain cells, resulting in a list of 20 genes, all of which were downregulated (Table VII in the online-only Data Supplement). Twelve of these genes were transmembrane transporters or coreceptors and 8 (Adamtsl2 [ADAMTS-like protein 2], Apod [apolipoprotein D], Ctsw [Cathepsin W], Htra3 [serine protease HTRA3], Pglyrp1 [peptidoglycan recognition protein 1], Spock2 [Testican-2], Ttyh2 [protein tweety homolog 2], and Wfdc1 [WAP four-disulfide core domain protein 1]) belong to the family of bona fide ECM proteins expressed on average 20-fold higher in brain ECs compared with ECs from other tissues (Figure 7B). These 8 genes were significantly repressed (FDR<0.001) by 50% to 84% in Axin1^iEC−^OE mice at E11.5 (Figure 7C). We validated that the expression of 3 representative genes among these 8, Apod, Htra3, and Adamtsl2 were indeed reduced by investigating the abundance of their protein products apolipoprotein D, serine protease HTRA3, and ADAMTS-like protein 2 using immunostaining (Figure 7D and 7E; Figure VIB in the online-only Data Supplement), although this trend did not reach significance in the apolipoprotein D group (Figure 7F), likely because of the large intergroup variation. Using in situ hybridization, Adamtsl2 mRNA was confirmed to be expressed exclusively in the forebrain vasculature (Figure VIA in the online-only Data Supplement). To investigate the potential role of these genes during BBB formation and maturation, we next investigated their temporal expression pattern in brain ECs during development. The genes clustered in 2 families, one exhibiting increasing expression from E11.5 to E17.5, or plateauing expression after initial BBB-maturation at E15.5, and another exhibiting increasing expression that peaked at BBB-maturation stages (E14.5/E15.5) followed by a decrease in expression at later developmental stages (Figure 7G). These expression patterns coincide with the development and initial maturation of the BBB are consistent with genes important for generating and maintaining BBB characteristics during embryonic development. To further substantiate this hypothesis, we investigated if these genes were also repressed in both Axin1^iEC−^OE and Ctnnb1-KO^iEC mice at E14.5, corresponding to a point in development where robust BBB characteristics are observed in the CNS vasculature in control mice. We found that transcripts of 4 genes, Adamtsl2, Htra3, Spock2, and Wfdc1, were all significantly downregulated at this later developmental time point in both of these 2 other mouse models (Figure 7H). That Apod mRNA levels were only reduced transiently at E11.5 but not E14.5 may explain why we failed to observe a robust downregulation of this factor at the protein level. All but one (Spock2) of the 4 persistently downregulated transcripts have not been previously described as Wnt-β-catenin targets and are, therefore, novel downstream potential regulators of β-catenin functions in CNS vessels. While these factors are downregulated at both mRNA (Figure 7H) and protein levels (Figure 7E and 7F; Figure VIB in the online-only Data Supplement), the regulation of other widely expressed EC-produced ECM proteins, such as Collagen IV (Figure 2I) or Laminin-α4 remain unchanged (Figure 2W). This suggests that Adamtsl2, Htra3, Spock2, and Wfdc1 establish a β-catenin–induced vascular ECM signature that is niche specific to the forebrain and required for vascular stabilization in this region. To analyze whether pathological vascular destabilization and hemorrhage in the forebrain of Axin1^iEC−^OE and Ctnnb1-KO^iEC mice could be ascribed to a single of these genes, we tested this focusing on the gene exhibiting the highest basal expression at E11.5, Adamtsl2, as we hypothesized that high levels of this protein might, therefore, be important for maintaining the vascular stability and barrier characteristics during development. No forebrain hemorrhage was observed in Adamtsl2-KO mice at E14.5 (Figure VIC in the online-only Data Supplement), nor was Adamtsl2 overexpression able to rescue the pathological brain vascular phenotypes observed in Axin1^iEC−^OE zebrafish larvae at 60 hpf (Figure VII in the online-only Data Supplement), suggesting that Adamtsl2 (or likely any other of these factors) does not act alone but rather that β-catenin signaling is important for a coordinated expression of this panel of ECM factors which act in concert to promote BBB formation and maintenance during development (Figure 8).

Taken together, the data presented strongly indicate that AOE-induced suppression of β-catenin signaling leads to impaired production of forebrain-specific, vascular ECM proteins, which drives initial vascular regression and subsequent vascular dilation, barrier disruption and eventually hemorrhage, specifically in the forebrain.