An intact, non-leaky BBB forms at E15.5²³. We hypothesized that the EC gene cohorts involved in BBB formation are expressed on E13.5, and 3-4 month-old adult CNS EC will express gene cohorts required for BBB maintenance. We defined transcriptomes of primary cortical ECs isolated from E13.5 and 3-4-month-old adults to test this. mRNA-seq analysis revealed strong expression of EC genes, including PECAM1, CDH5, and CLDN5, compared to perivascular cell types such as pericytes, neuronal and glial genes (S.Fig-1A). To define the gene cohorts responsible for BBB formation vs. BBB maintenance, we identified differentially expressed genes (DEGs) expressed in natural log (fold change) in E-13.5 relative to adult mice (Fig-1A,B). DEGs with a P-value < 0.05) were considered significant. Compared to E13.5, 50% of genes were upregulated, 35% were downregulated in adults and 15% were unchanged (Fig-1C).

GO enrichment analysis of the DEGs identified five functional categories: angiogenesis, cell-to-cell junction, transporters, extracellular matrix, and DNA binding transcription factors (Fig-1D). Except for transporters, the other categories were characterized by more downregulated vs. upregulated genes. The DEGs included important BBB genes (Fig-1E). For example, tight junction (TJ) gene CLDN1, CLDN5, BBB transporters MFSD2A, CAV1, and BBB- related transcription factors ZIC3, FOXF2, and SOX17 were differentially expressed (Fig-1E). The complete dataset is available on Geo under accession number GSE214923. Overall, the transcriptomic analysis identified EC gene cohorts that were expressed during the formation or maintenance of the BBB.

We validated the differential expression of important BBB and related genes (e.g., CLDN1, CLDN5, MFSD2A, ZIC3, and SOX17) at E-13.5, E-17.5, P0, P7, and in the adult. This analysis revealed that CLDN1, MFSD2A, and ZIC3 expressions were significantly downregulated by E-17.5 and subsequent developmental stages (Fig. 1F). Thus, the high expression of these genes might be required for the establishment and a baseline expression to maintain the BBB. By contrast, adult expression of CLDN5 was significantly upregulated compared to other developmental stages, and SOX17 was upregulated considerably compared to other developmental stages except for P7 (Fig. 1F). Additionally, we validated the mRNA expression of CLDN11 and FOXF2 (S. Fig. 1B).

Hupe et al ., (2017) demonstrated differentially regulated genes in brain ECs during embryonic development⁵. To further validate our data, we compared our transcriptome with the 1129 genes shown to be regulated during embryonic development. We found 315 of our downregulated transcripts in adults and 490 of our upregulated transcripts in adults matched with their dataset. Key overlapping genes are highlighted (S. Fig. 1C).

Histone deacetylases (HDACs) have critical roles in development and tissue homeostasis, and HDAC inhibitors are instructive experimental tools^24,25. To assess whether the transcriptional downregulation of BBB genes involves HDAC-dependent epigenetic repression, we treated adult ECs with a pan HDAC inhibitor trichostatin A (TSA) for 48 hours. TSA increased CLDN1 (Fig-2A) and ZIC3 (S.Fig-2A) expression relative to the control. Since there are four major HDAC classes, class I (HDAC 1, 2, 3, and 8), class II (HDAC 4, 5, 6, 7, 9, 10), class III (SIRTs 1–7), and class IV (HDAC 11), we tested whether specific HDACs mediate the repression. The class-I HDAC inhibitor MS-275 significantly upregulated CLDN1, MFSD2A and ZIC3 while downregulated CLDN5 (Fig-2A, S.Fig-2A). No significant difference in CLDN1 mRNA expression was observed with class-II HDAC inhibitor (S.Fig-2A). In adult ECs, HDAC2 exhibited greater expression than other class I HDACs (Fig-2B). To analyze the HDAC2 function, we utilized siRNAs to downregulate HDAC1, HDAC2, or HDAC3 in adult ECs. Knockdown (KD) of HDAC2 significantly upregulated the repressed CLDN1, MFSD2A, and ZIC3 genes and reduced CLDN5 expression (Fig-2C). By contrast, HDAC1 and HDAC3 downregulation did not affect these genes (not shown). BBB genes analyzed above were selected based on their expression patterns and to represent important functional attributes of BBBs, such as tight junctions (CLDN1 and CLDN5), transporters (MFSD2A), and transcription factors (ZIC3). We used ChIP-qPCR to test whether HDAC2 directly regulates BBB gene expression in E13.5 and adult cortical EC contexts. Three primers were designed ((-)500, TSS & (+)500) spanning the 1 kb region on each side of the promoter. HDAC2 occupancy was detected in the (-)and(+) 500 regions of CLDN1 in adults with no significant enrichment in E13.5. ZIC3 showed occupancy at all three regions (1kb) in the adult stage, with E13.5 showing enrichment in TSS only (Fig-2D, G &S.Fig2B). Conversely, HDAC2 occupied MFSD2A only at E13.5((-)500 & TSS), and in CLDN5 occupancy was detected at both E13.5 (1kb) and in adults ((-) & (+) 500). These results link HDAC2 to the developmental control of BBB genes (Fig-2D,G & S.Fig2B).

Our initial screening showed that the PRC2 inhibitor DZNEP significantly increased CLDN1 expression in adult ECs (S.Fig-2A). To analyze the PRC2 function in this context, we downregulated the PRC2 subunit EZH2 from adult ECs. Compared to the control, EZH2 downregulation significantly increased CLDN1, ZIC3, and MFSD2A expression and decreased CLDN5 expression (Fig 2E). ChiP-qPCR of PRC2 subunit EED revealed EED occupancy at various regions of CLDN1, MFSD2A, and ZIC3 at E13.5 and in the adult (Fig2F,G & S.Fig-2). EED occupied CLDN5 at E13.5, but not in adult ECs. Our data support a model in which HDAC2 and PRC2 are critical determinants of EC BBB transcriptomes during BBB development.

Since HDAC2 and PRC2 regulate multiple BBB genes, we sought to determine whether they share similar post-translational histone modifications. To examine this, we performed ChIP- qPCR of potentially involved histone marks, including the repressive marks H3K27me3, H3K9me3, and the activating marks H3K4me3 and H3K9ac. We have scanned approximately 1kb genomic region surrounding the TSS of CLDN1, CLDN5, MFSD2A, ZIC3, and SOX17 in E13.5 and adults. H3K9me3 ChIP-qPCR on our selected BBB genes didn’t show significant binding in either stage (S.Fig-3A). We detected abundant enrichment of repressive histone mark H3K27me3 on CLDN1 (Fig-3A,C), MFSD2A, and ZIC3 (Fig-3C & S.Fig-3B, C) in adult E.C.s compared to E13.5. CLDN1, MFSD2A, and ZIC3 showed significant enrichment of H3K27me3 compared to IgG in E13.5 (Fig-3A & S.Fig-3B,C). While CLDN5 showed significant enrichment of H3K27me3 in the 1kb region in both E13.5 and adults (Fig-3B). However, E13.5 showed abundant enrichment of H3K27me3 in the -500 region compared to adults (Fig-3B).

Active histone mark, H3K4me3 showed significantly increased enrichment on CLDN1 (TSS), MFSD2A (1 KB), and ZIC3 (-500 and +500) at E13.5 compared to adult (Fig-3A and S.Fig-3B, C). CLDN1 (-)500 region didn’t show any significant binding for H3K4me3 in both stages, while the (+)500 region showed significant binding compared to IgG with no difference between E-13.5 and adult. ZIC3 TSS region showed significant enrichment of H3K4me3 in the adult compared to E-13.5(S.Fig-3). Supporting its abundant expression in adults, CLDN5 (1 KB) showed significant enrichment of H3K4me3 in adults compared to E13.5. Another active histone mark, H3K9ac, showed significant enrichment on CLDN1 (TSS and +500) at E13.5 compared to the adult with no significant binding in the -500 region in both stages (Fig-3A & S.Fig-3). ZIC3 (TSS and +500) showed a significant binding for H3K9ac in E13.5 compared to the adult, while the -500 region didn’t show any enrichment in both stages. MFSD2A showed significant enrichment of H3K9ac in the -500 and +500 regions at E-13.5 compared to adults. At the same time, the TSS region showed significant enrichment in adults compared to E-13.5. Histone modifications on SOX17 are shown in S.Fig-3D. These results indicate that BBB genes acquire a unique epigenetic signature during development.

To examine the role of HDAC2 and PRC2 in BBB maturation, we knock out (KO.) HDAC2 and PRC2 subunit EZH2 from ECs during embryonic development. To conditionally KO HDAC2 or EZH2, we used tamoxifen-inducible Cdh5(PAC)-CreERT2 mice. Tamoxifen was injected into the mother starting at E-12.5 and on alternate days until E-16.5 (Fig-4A & 4E). This allowed the KO of HDAC2 or EZH2 before the maturation of the BBB.

It was observed that HDAC2 and EZH2 KO embryos grew normally and were alive on the day of sacrifice E17.5. Significantly decreased mRNA expression of HDAC2 in the whole brain analysis confirmed the loss of HDAC2 (S.Fig-4). Compared to WT, HDAC2 ECKO pial vessels were dilated and showed increased angiogenesis (Fig-4B). BBB permeability was assessed in HDAC2 ECKO mice using a 70KD FITC-conjugated dextran tracer. The tracer remained confined inside the vessels of E17.5 WT embryos, supporting previous findings that the BBB matures by E-15.5 (Fig-4C). Conversely, in HDAC2 ECKO, dextran leaked into the cortical parenchyma (Fig-4C). Green fluorescent intensity measurements confirmed the immature BBB in the HDAC2 ECKO (Fig-4D). A vascular analysis using angiotool determined that HDAC2 ECKO had a significantly higher vascularized brain area than WT, indicating that HDAC2 plays a key role in angiogenesis (Fig-4D). The pharmacological inhibition of class I HDAC using MS-275 in timed pregnant WT mice at E-13.5 showed significant tracer leakage into the brain parenchyma at E-15.5 compared to vehicle-treated control (S.Fig-4A). The MS-275 treated embryos also showed a thin cortex compared to the control, possibly due to the leakage of the drug into the brain which affects brain development (S.Fig-4A). These findings should be considered when considering the clinical use of MS-275 in a developing brain.

Compared to WT, at E-17.5 EZH2 ECKO showed dilated vessels with no evident increase in the pial vessel angiogenesis (Fig-4F). Confirming the loss of EZH2, mRNA analysis on the whole brain showed a significant reduction. BBB permeability assay showed a subtle leakage of 70KD fluorescent tracer into the brain parenchyma (Fig-4G). However, the fluorescence intensity analysis showed no significant difference (Fig-4H). Further, the mRNA analysis on the whole brain showed a significant decrease in EZH2 expression with no difference in CLDN1, MFSD2A, and CLDN5 (S.Fig-4C). Together, HDAC2 and EZH2 ECKO data suggest that HDAC2 is critical for BBB maturation, whereas PRC2 is dispensable or a support mechanism required during later BBB development.

Wnt pathway has been shown to influence BBB gene expression⁵. However, this pathway activity is reported to be minimal in adults^15,18. In our transcriptomic analysis, 67% of Wnt signaling-related genes were downregulated in adults, while 33% were upregulated (Fig-5A). Interestingly, downstream Wnt targets genes including AXIN2, LEF1, and VEGFA have downregulated in adult ECs, while the upstream Wnt pathway components like FZD4/6, LRP5, and CTNNB1 were upregulated (Fig-5a). Since Wnt-regulated BBB genes such as CLDN1, and ZIC3 expression was also minimum in adults, we hypothesize that the Wnt pathway is still active in adult CNS ECs. In contrast, Wnt target genes are epigenetically repressed.

To test this, we activated the Wnt pathway in E13.5 and adult ECs using identical concentrations of Wnt3a ligand or GSK3B inhibitor CHIR99021. mRNA analysis revealed that Wnt target genes AXIN2 and LEF1 can be significantly activated in E13.5 ECs when treated with Wnt3a or CHIR99021, while no significant activation was observed in adults (Fig-5B). Further validating this finding, transcriptome analysis on adult ECs after Wnt3a treatment showed only activation of one Wnt target gene CD44 (Fig-5C). Intriguingly, the Wnt3a treatment downregulated 16 Wnt-related genes (Fig-5C). Next using immunohistochemistry of β-catenin in adult control and CHIR 99021 treated ECs we demonstrated that Wnt pathway activation could translocate or stabilize the Wnt transducer β-catenin into the nucleus (Fig-5D). These results indicate that the Wnt pathway is active in adult CNS ECs.

Next, we investigated whether Wnt target genes are epigenetically repressed. Wnt target genes AXIN2 and LEF1 were significantly upregulated in adult CNS ECs when treated with HDAC2 or EZH2 siRNA, MS-275, and in HDAC2 ECKO mutants (Fig-5E & S.F-5A,B). Furthermore, the AXIN2 promoter in adults showed significantly increased occupancy of HDAC2, repressive histone mark H3K27me3 in adults while active histone marks H3K4me3 and H3K9ac were reduced considerably compared to E-13.5 as analyzed by ChIP-qPCR (Fig 5F). LEF1 also showed a similar repressive histone modification (Fig-5C). Further, MS-275 and LiCl (Wnt agonist) treatment increase the AXIN2 protein expression in adult cortical vessels but not by just LiCl (S.Fig-5D). Thus, our data explain the mechanism behind the low Wnt pathway in adult CNS ECs.

We investigated the relevance of the low Wnt pathway to BBB development. To this, we treated primary E13.5 cortical ECs with LF3 (inhibits the interaction of β-catenin and TCF4) for 48 hours. LF3 activity was confirmed by reduced mRNA expression of the Wnt target gene AXIN2 (Fig-6A). LF3 treatment induces adult/BBB maintenance-type gene expression patterns in E13.5 with a significant decrease in CLDN1, ZIC3, and MFSD2A expression and an increase in CLDN5 and SOX17 expression (Fig-6A). Another Wnt inhibitor IWR-1-endo also showed similar results (not shown), while β-catenin siRNA KD showed a similar gene expression pattern for AXIN2, CLDN1, ZIC3, and MFSD2A and no difference in CLDN5 (S.Fig-6A).

To determine whether Wnt pathway inhibition induces epigenetic modifications on its target genes and BBB genes, we performed Chip-qPCR analysis of HDAC2, EED, and histone mark H3K27me3, H3K4me3, and H3K9ac on the promoter of Wnt target genes AXIN2, and LEF1, BBB genes CLDN1, CLDN5, MFSD2A, and ZIC3. An increased HDAC2 occupancy was observed on AXIN2, LEF1, CLDN1, and MFSD2A promoter when E13.5 ECs were treated with LF3 (Fig-6B & S.Fig-6B). ZIC3 and CLDN5 showed no significant difference (not shown). EED and histone mark H3K27me3 showed an increased enrichment on the AXIN2 while LEF1 showed increased enrichment for H3K27me3 in the promoter after LF3 treatment (Fig-6B & S.Fig-6B) with no difference in other genes analyzed (not shown). CLDN1 and MFSD2A showed a significant decrease in the enrichment of active histone mark H3K4me3 after treatment with LF3, whereas CLDN5 showed a significant increase (Fig-6B). Another active histone mark H3K9ac showed significantly increased enrichment on CLDN5 (Fig-6B) with no difference in the promoter of other genes analyzed (not shown).

We then investigate the significance of the physiological reduction of Wnt signaling on BBB maturation. For this we use Ctnnb1lox(ex3)^+/+; Cdh5-CreERT2 mice which are widely used to attain the inducible EC specific β-catenin gain of function (GOF) (Fig-6C). Upon tamoxifen treatment exon 3 of CTNNB1 (encoding β-catenin) will be deleted, leading to the expression of a stabilized form of β-catenin protein and thereby constitutive activation of canonical Wnt signaling (Fig6C). Tamoxifen was injected into the pregnant mother, as explained in section 2.4. At E-17.5, β-catenin GOF embryos showed significantly increased pial angiogenesis (Fig-6D) compared to WT BBB permeability assay using 70KD FITC dextran revealed immature BBB in β-catenin GOF (Fig-6E & F). Confirming this result, the pharmacological activation of Wnt signaling by the Wnt agonist LiCl also showed a significant BBB leakage compared to the control (S.Fig- 6C). Our data suggest that a low Wnt pathway supports BBB maturations by epigenetically modifying BBB genes.

Embryonic deletion of EC HDAC2 and Class-I HDAC inhibitor treatment of adult CNS ECs activate gene expression related to angiogenesis, BBB formation, and the Wnt pathway.

Since the E-17.5 HDAC2 ECKO showed a significant increase in vessel density and BBB leakage, we hypothesize this is due to the inefficient epigenetic repression of angiogenesis and BBB formation genes during development. To investigate this, we utilized E-17.5 HDAC2 ECKO and FACS-sorted CD31+ ECs for ultra-low mRNA sequencing (Fig. 7A). Indicating the critical role of HDAC2 in regulating the EC gene expression 2257 genes show significant upregulation and 1723 genes showed downregulation.

We then performed functional classification of these genes based on key endothelial functions such as angiogenesis, cell-cell junctions, extracellular matrix, transporters, DNA- binding transcription factors, and the Wnt signaling pathway. Except for extracellular matrix function, all other categories were characterized by a significant number of upregulated genes (Fig. 7B). Key differentially regulated genes are shown in Fig. 7C. Critical angiogenesis genes, including VEGFA and ENG; tight junction proteins, including CLDN5, TJP1 (ZO-1), and OCLN; BBB transporters such as MFSD2A and GLUT1; and BBB-related transcription factors such as ZIC3, FOXF2, and SOX17 were upregulated. Further supporting the HDAC2-mediated developmental regulation of the Wnt pathway, in HDAC2 ECKO, several Wnt target genes, including AXIN2 and APCDD1, were activated in E-17.5 ECs after HDAC2 deletion. Key genes that are downregulated include APOE, PSEN1 and VEGFB.

To identify the potential of inhibiting HDAC2 in activating BBB and angiogenesis-related genes in adult ECs, and to obtain a detailed picture of CNS EC genes regulated by MS-275, a class-I HDAC inhibitor, we performed mRNA-seq analysis of MS-275-treated primary adult cortical ECs in culture and compared with the control. MS-275 treatment upregulated 48% of genes, downregulated 34% of genes, and did not alter the expression of 18% of genes.

Differentially expressed genes were grouped into six relevant categories, and all the categories showed a significant no.of differentially regulated genes (Fig-7D). Fig-7E illustrates the potential of MS-275 in regaining BBB, angiogenesis, Wnt, and transcription factors that are differentially regulated during development (E-13.5 vs Adult).

Pharmacologically Induced Epigenetic Changes Are Reversible and Partially Translate to Human Vessels We next assessed whether the gene expression changes acquired by adult ECs following MS-275 treatment were reversible. To this end, we treated adult ECs with MS-275 for 48 hours and collected ECs at 48 hours of treatment and seven days after treatment. As previously shown, CLDN1 was significantly upregulated and CLDN5 was downregulated considerably after 48 hours (Fig-8A). While the expression of CLDN1 and CLDN5 in adult ECs returned to normal in ECs collected seven days after the treatment (Fig-8A).

Finally, we examined if human vessels show similar reactivation when treated with MS-275. To this end, the human brain vessels collected from epilepsy surgery irrespective of age and gender. We tested three genes with MS-275 treatment: CLDN1, CLDN5, and AXIN2. We found that MS-275 treatment significantly activates the expression of CLDN1 and AXIN2 compared to the control, with no difference in CLDN5 expression (Fig-8B).