One-day-old, 3–4-week-old (80 ± 10 g) and 6-month-old (360 ± 10 g) male SD rats were used in this study. Three biological replicates at each time point were used for sequencing. In one-day-old group, there are 10 rats per repetition, in 3–4-week-old and 6-month-old groups, there are 5 rats per repetition. Previous studies suggested sex-specific differences in tight junction protein expression that could affect BBB function (Weber and Clyne, 2021). To avoid the effect of gender on the blood-brain barrier, we mainly studied the maintenance and development of the blood-brain barrier in male rats. All animal experiments were approved by the Ethics Committee of Henan Provincial People’s Hospital, Zhengzhou University, China, and every effort was made to minimize the number of animals and their suffering. All animals were in a light-and temperature-controlled environment with access to food and water ad libitum. In order to minimize the subjective bias in the design, we adopt the principles of randomization and double blind.

Evans Blue (EB) extravasation was used to assess BBB permeability (Lobo et al., 2020). In brief, 2% EB (3 mL/kg, Sigma) was injected via the vein at various timepoints. After the EB was allowed to circulate for 2 h, the rats were anesthetized and then perfused transcardially with normal saline solution. To assess EB extravasation, brain cortices were isolated and weighed, homogenized in 1 ml of 50% trichloroacetic acid, and centrifuged at 10,000 rpm for 30 min. The supernatant was collected and mixed with an equal amount of ethanol. EB concentrations were determined spectrophotometrically at 620 nm. The permeability of the blood-brain barrier was evaluated by calculating the EB content according to the standard curve.

After a rat was anesthetized with isoflurane, the head was washed with 70% alcohol, and quickly separated from the body to remove the brain. Then, the thin slices of subdural meninges (SDM) on the dorsal surface of the brain were carefully removed using the dissecting microscope to validate the removal. The brain was cut in half horizontally and divided cautiously into cortex and internal structures, and dissected in cold phosphate-buffered saline (PBS). The separated cerebral cortex was properly cut into 1 mm³ pieces of tissue in order to facilitate enzyme-based digestion or the subsequent Dounce homogenization steps. The tissue pieces were incubated in type II collagenase (Gibco 17101015, 1 mg/mL) and 100 μL DNase1 to DMEM F12. Microvascular pellets were resuspended in a 20% BSA-V solution and centrifuged (4°C) for 8 min at 1,000 × g to remove the top layer, containing myelin and brain parenchymal cells. Then, the microvascular pellets were filtered with a 40 μm cell strainer and centrifuged at 1,000 × g for 2 min to obtain pure pink microvascular fragments. The microvascular fragments were directly sorted into 1 mL trizol (Invitrogen 15596018), and stored at −80°C with liquid nitrogen.

After the total RNA was extracted using a trizol reagent kit (Invitrogen, Carlsbad, CA, United States) according to the manufacturer’s instructions, all rRNAs were removed to retain the mRNAs. The remaining clean reads were further used in assembly and gene abundance calculation. The quality and integrity of total RNA were assessed using Nano Photometer spectrophotometer (IMPLEN) and monitored on 1% agarose gel. The purified mRNA was cut into short fragments using fragmentation buffer and reverse transcribed into cDNA with random primers. RNA concentration was measured with the Qubit RNA Assay Kit in a Qubit 2.0 fluorometer (Life Technologies). The ligation product was amplified and enriched PCR to construct a cDNA library template. The ligation products were size selected by agarose gel electrophoresis, PCR amplified, and sequenced using Illumina Novaseq6000 by Gene de novo Biotechnology Co., Ltd., (Guangzhou, China). Raw data of RNA sequencing were deposited in NCBI Sequence Read Archive database (PRJNA790676).

After the total RNA was extracted from different development of brain cortex microvessels using a trizol reagent kit (Invitrogen, Carlsbad, CA, United States), expression levels of ocln, cldn5, tjp1 genes were compared using real-time quantitative PCR (qRT-PCR). Perform three to six repetitions for each sample (5–6 rats/repetition). cDNA (500 ng) was synthesized from obtained RNA samples using a reverse transcription kit (HiScript II Q RT SuperMix for qPCR, Vazyme). Quantitative PCR reactions were prepared using ChamQ Universal SYBR qPCR Master mix (Vazyme) and specific primers (Table 1) on the ABI 7500 system (Applied Biosystem). qRT-PCR thermal cycle reaction started at 95°C for 5 min, followed by 40 cycles of 95°C for 10 s, 60°C for 30 s. Relative gene expression in ocln, cldn5, tjp1 was calculated using the 2−ΔΔCt method (Livak and Schmittgen, 2001).

Rats were sacrificed under anesthesia through transcardiac perfusion with cold phosphate-buffered saline (PBS). We extracted rat cerebral cortical microvessels (groups of 5–6 rats with three replicates each), weighed and homogenized the microvessels in RIPA lysis buffer. Samples were homogenized, and protein concentrations were determined by bicinchoninic acid (BCA) protein assay (Thermo-Fisher, Waltham, MA, United States). The proteins were separated on 10% SDS–PAGE gel and transferred to nitrocellulose membranes (Invitrogen, United States). After three times of washing in TBS with 0.1% Tween-20 (TBST), the membranes were blocked in TBST with 5% skimmed milk for 2 h at room temperature. The membranes were incubated overnight with primary antibodies at 4°C and then further incubated with HRP-conjugated secondary antibodies (1:1,500) for 1 h at room temperature. The following primary antibodies were used: rabbit anti-occludin (1:100, Thermo Fisher Scientific, #711500); rabbit anti-claudin 5 (1:100, Thermo Fisher Scientific, # 34–1600); rabbit anti-zo-1 (1:100, Thermo Fisher Scientific, #402200) and rabbit anti-caveolin-1 (1:1,000, Cell Signaling Technology, #3238). The secondary antibodies were used: goat anti-rabbit IgG (1:1,500, Beyotime Biotechnology, #A0208). Western Bright ECL solution was used to develop the blots. Chemiluminescence antibody signals were documented using Chemidocit 510 image system (Analytik Jena United States) and Vision Works software (Media Cybernetics, Rockville, MD, United States), using exposure settings and quantified for band densitometry. Each membrane was then probed for β-actin (1:2,000, Beyotime Biotechnology, #AF5003) as endogenous loading controls. The antibody used to probe for occludin, claudin5, zo-1 and caveolin-1 detect the same bands as reported in previous publications. Relative protein expression was quantified by measuring band densitometry. Images for the blots of a protein of interest and their corresponding β-actin were selected for analyses and presentation based on a predetermined upper grayscale value to assure linear densitometric values. Images were analyzed using the lane and band functions of the ImageLab™ software. For presentation purposes, the grayscale range was set between 0 units and the predetermined upper grayscale value for the protein of interest and corresponding β-actin prior to exporting as an image file. Post-export modifications of the images were limited to cropping to the regions of interest.

After the right hemisphere was divided into 1-mm³ sections, some tissue close to the bregma was selected and further cut into 1 × 1 × 1 mm tissue blocks. The tissue blocks were incubated with 2.5% glutaraldehyde for 6 h, dehydrated, and embedded in epoxy resin. Ultrathin sections were cut at 60 nm thickness and observed under an electron microscope (Hitachi TEM system, Tokyo, Japan).

Comparisons between multiple groups were made by one-way ANOVA followed by Dunnett’s post hoc test. Data normality was assessed using the Shapiro–Wilk test. The data were presented as mean ± SD, and the boxplots show the maximum and minimum values. Data were considered significant when *p < 0.05, ^**p < 0.01, ^***p < 0.001 and ^****p < 0.0001.

Cortical microvessels have an important regulatory function for the maintenance of the brain microenvironment. To further understand the functional status of the cortical microvessels in different development periods, we enriched the cortical microvessels and then performed RNA sequencing. To isolate microvessels from the rat cerebral cortex, as shown in Supplementary Figure 1, we adapted and modified the protocol initially described by Zlokovic (Wu et al., 2003; Lee et al., 2019; Ujifuku et al., 2020). In brief, using type II collagen digestion, resuspension and 40 μm cell filter filtration, we enriched cortical microvessels. After the enrichment of microvascular endothelial cells from the cerebral cortex, we isolated the RNA for sequencing. The cortical microvascular samples showed high levels of RNA from endothelial cell genes with minimal levels of RNA from neuronal and glial genes. The level of mural cell genes in the cerebral microvascular samples was low but detectable, and estimated to be < 2.0% of the RNA in the 3–4 weeks cerebral microvascular samples (Figure 1D).

We performed RNA-seq following the cortical microvascular purification at three different developmental stages for the neonatal (1-day-old), adolescent (3–4-week-old) and adult (6-month-old) groups with three biological replicates per group (5–10 rats per biological replicate). The RNA-seq generated an average of 5.9 billion reads per sample, and more than 93% of the reads were mapped to the rat genome. The cortical microvascular samples showed high levels of RNA from endothelial cell genes with minimal levels of RNA from neuronal and glial genes. The mural cell genes had low but detectable level, which was estimated to be < 2.0% of the RNA in the 3–4 weeks cerebral microvascular samples (Figure 1D). We used principal component analysis (PCA), hierarchical clustering and heat map based on Pearson’s correlation analysis between different samples to make transcriptome-wide, unbiased comparisons between neonatal, adolescent and adult groups. In PCA, adolescent and adult datasets also clustered together along the principal component (PC1, explaining 94% of the variance; and PC2, explaining 2.9% of the variance), while weaker clustering along the principal component (PC1, explaining 94% of the variance; and PC2, explaining 2.9% of the variance) was observed for the neonatal datasets (Figure 1A). Altogether, in hierarchical clustering, these results suggested that neonatal datasets have more variability than adolescent and adult samples, potentially attributable to differences in genetic background, or brain regions which are harder to separate (Figure 1E). In the heat map, we observed a high level of measurement consistency among biological replicates, although neonatal datasets had more variability (Figure 1B). The heatmap demonstrated that “3–4 weeks-3” was intermediate between 1-day-old groups and other 3–4-week groups. 3–4 weeks old rats were used at this time point, “3–4 weeks-3” rats were 3 weeks old rats, “3–4 week-2” and “3–4 week-1” were 4 weeks old rat, which may lead to differences in intra-group. The transcriptome data revealed that the neonatal group was significantly different from the adult group, however, there were small differences in the transcriptome data between the adolescent group and the adult group (Figure 1B). To quickly find common and unique genes among different groups, we performed Venn diagram analysis. There were 11,427 overlapping genes among three groups in the Venn diagram, and the most unique changes were observed in the neonatal group (Figure 1C).

In order to compare the changes in gene expression among each development stage, we defined the numbers of upregulated genes (log2 ratio > 2.000; F < 0.05) and downregulated genes (log2 ratio < −2.000; F < 0.05) at each stage. We used a histogram to analyze the number of upregulated and downregulated genes in each comparison group. As seen in Figure 2A, the cortical BBB related genes of rats in the neonatal group and the adolescent group are quite different. Moreover, compared with the neonatal group, the number of downregulated genes in rats of the adolescent group was higher than the number of upregulated genes. However, the BBB-related genes had little change between the adolescent group and the adult group, which indicates that there is no significant difference in the BBB between adolescence and adulthood. Surprisingly, the number of upregulated genes was more than that of the downregulated genes during BBB maturation. Venn diagram analysis also was conducted to find the gene expression changes during development. We observed that the expression of 39 genes remained elevated with age, such as Cxcl12, which plays an important role in development (Figures 2B,C). Furthermore, there were 68 genes whose expression levels continued to decrease, such as Ttr (transthyretin) (Figures 2D,E). As a new transporter of nanoparticles for receptor-mediated transcytosis in rat brain microvessels, transthyretin may contribute to neonatal administration to penetrate the BBB (Kim et al., 2015). These differences indicate several well-known neonatal-specific mechanisms that are different from those in adults, which reflects the adaptation to a neonatal-specific environment. The heat maps illustrate the transcript abundance in biological triplicates among three microvascular groups for the 10 highest confidence in 6 months-enriched (Figure 2C) and 1 day-enriched (Figure 2E).

We identified the signaling and biological pathways altered during brain development at each stage by DAVID Bioinformatics (Figure 3). The pathways which were altered in each stage, as well as unique pathways, were shared. Examples of upregulated pathways in the 3–4-week-old group were “regulation of multicellular organismal process,” “immune system process,” “regulation of response to stimulus,” “single-multicellular organism process,” and “regulation of immune system process,” as identified by Gene Ontology (GO). This showed that the ability of the BBB to adapt to the environment gradually increases with age. Strikingly, in the 6-month-old group, the top 3 upregulated GO pathways were “immune system process,” “immune response,” and “regulation of response to stimulus.” Downregulated pathways were the most unique to “cell projection organization,” “cell projection morphogenes,” “cell part morphogenes,” and “nervous system development” both in the 3–4-week-old and 6-month-old groups. The most unique upregulated Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways in the 3–4-week-old and 6-month-old groups were “cytokine-cytokine receptor interaction.” We observed that the top 3 downregulated KEGG pathways were “nicotine addiction,” “axon guidance” and “cell cycle” both in the 3–4-week-old and the 6-month-old groups.

Principally, maintaining the integrity of the BBB is one of the essential functions of microvessels in the brain. We further analyzed the gene modules related to BBB function. During development, BBB demonstrates some important differences, particularly in transport mechanisms, many of which have only recently been described. Transcytosis in microvascular endothelial cells can be divided into two categories: receptor-mediated transcytosis (RMT), in which the ligand binds to the receptor to mediate endocytosis, and non-selective adsorbent transcytosis (Villaseñor et al., 2017). The two main endocytic pathways of the BBB are the clathrin-mediated pathway and the caveolin-mediated pathway (Villaseñor et al., 2017). We further observed the changes of RMT receptors, clathrin adaptors and caveolin-adaptors during the development of BBB. First, over time, the expression of some RMT receptors and clathrin junctions increases with age, however, that of the remaining genes decreases with age (Figure 4A). To investigate the correlation of RMT receptor genes, we performed a correlation heat map analysis. We observed scarb1 expression parallel to other correlates, such as slc2a1, bsg, ldlr, igf1r, lepr, insr, forming a co-expression module that suggests common orchestration. The expression of this scarb1 module was mutually exclusive with respect to that of tfrc, lrp8, slc3a2, slc7a5, and Ager (Figure 4B). Interestingly, in the correlation analysis of clathrin-related genes, we found the same phenomenon. We observed Cltb expression compared with other correlates, such as Rin3, Hip1r, Hip1, forming a co-expression module. The expression of this Cltb module was mutually exclusive with respect to the expressions of epn1, ap2a2, ap2s1 (Figure 4C). This showed that there were different dominant RMT receptors and clathrin adaptors between the neonatal and the adult group, which may provide novel therapeutic opportunities for CNS drug delivery. Next, we investigated the expression of cav1, which encodes the principal component of caveolae. The result showed that it trended upward with age (1 day vs 3–4 weeks *P = 0.0138, df = 6; 1 day vs 6 months ^**p = 0.0074, df = 6), to a mechanism of BBB development (Figure 4E). Meanwhile, mfsd2a expression decreased (1 day vs 3–4 weeks *P = 0.0126, df = 6; 1 day vs 6 months ^**p = 0.0083, df = 6), the product of which suppresses caveolae formation (Figure 4D). We next separated the cerebral cortex microvessels of rats at three time points and extracted the proteins for Western blot analysis. We found that the expression of cav-1 protein gradually increased (1 day vs 3–4 weeks ^**P = 0.0012, df = 6; 1 day vs 6 months ^***p = 0.0006, df = 6) (Figures 5A,E). We further used TEM to observe the ultrastructural changes of the BBB. We found that vesicle density in the endothelium was significantly increased in the 3–4-week-old and 6-month-old groups (Figures 4F,G). These results indicated that endothelial cell transcytosis was increased during the development of the BBB. In several stroke models, an increase in the number of endothelial cell transcytosis was observed by transmission electron microscopy, along with an increased uptake of the tracer by endothelial cells (Lossinsky and Shivers, 2004). This upregulation of endothelial cell transcytosis increased BBB permeability. Injection of EB was used to assess blood-brain barrier permeability in rats at three time points. We found that the permeability of the blood-brain barrier to EB increased (1 day vs 3–4 weeks *P = 0.0334) with age and the cortex of 6-months-old rats was visually permeable to EB (Supplementary Figure 3).

We performed transcriptome analysis on 47 TJs at three time points. The heat map in Figure 6A shows the changes in TJs expression at these three time points. We further used the Venn diagram to select 16 differentially expressed TJs genes (Figure 6B). We found that tjp1, esam, jam2, and ocln were gradually upregulated during the development of the BBB, while cdk4 and magi1 were gradually downregulated during the development of the BBB in Supplementary Figure 2. The tjp1 gene encodes zo-1, which plays a key role in the development of the BBB. Knockout of the tjp1 gene as early as E8.5 resulted in developmental defects, extensive neuroectodermal tissue apoptosis, vascular development failure, and embryonic death. We used qRTPCR to verify the expression of tjp1, ocln and cldn5 in the cortex microvessels for further verification (Figures 6C–E). We next separated the cerebral cortex microvessels of rats at three time points and extracted the proteins for Western blot analysis. The results showed that, with the development of the BBB, the expression of TJs such as occludin (Figure 5C, 1 day vs 3–4 weeks ^***p = 0.0009, 1 day vs 6 months ^***p = 0.0002), claudin5 (Figure 5D, 1 day vs 3–4 weeks ^***p = 0.0004) and zo-1 (Figure 5F, 1 day vs 3–4 weeks ^**p = 0.0013; 1 day vs 6 months ^***p = 0.0003) gradually increased and stabilized (Figure 5B). We further observed under the electron microscope that TJs are short and simple in neonatal brain microvascular endothelial cells. In adult brain microvascular endothelial cells, the particle density rises, and a significant increase in the correlation between TJs and the protoplasmic surface (P-Face) can be observed, which represents BBB continuing to mature (Figure 6F). Cav1–deficient (cav-1−/−) mice had significantly higher degradation of TJ proteins such as occludin, claudin 5, zo-1, and proteolytic activity of matrix metalloproteinase as compared to Cav-1 + / + mice. Conversely, the re-expression of cav-1 in Cav-1−/− mice restored TJs protein expression and reduced matrix metalloproteinase proteolytic activity (Choi et al., 2016). Interestingly, the correlation heat map analysis of transcytosis and TJ gene expression showed that cav-1, cldn5 and tjp1 genes have a strong correlation (Figure 5G). We speculate that under physiological conditions, cav-1 may interact with TJs to further promote the formation and stability of the BBB.

The BBB is located at the level of microvascular endothelial cells in the CNS, which includes capillaries, pre-capillary arterioles and post-capillary venules (Northrop and Yamamoto, 2015; Francisco et al., 2020). Previous gene profiling studies of brain vessels had several limitations, including the use of whole vessel fractions that contain many different cell types, the lack of description of the cerebral endothelial isolation procedures, and separation of the entire brain instead of specific areas (Enerson and Drewes, 2006; Richard et al., 2010; Munji et al., 2019). The vasculature is heterogeneous throughout the central nervous system, and the expression of endothelial junction proteins (occludin, claudin-5) is higher in the white matter (WM) compared to the gray matter (GM) (Gross et al., 1986; Wilhelm et al., 2016). In this paper, we combined enzymatic digestion purification with 40 μm filter to generate a transcriptional profile of highly purified microvascular endothelial cells from the cerebral cortex, and provide a new resource for understanding gene expression in the developing and adult BBB. We used pasteurized pipette mechanical homogenization combined with short-time enzymatic digestion, which can ensure tissue separation without affecting cell viability. These transcripts encode TJ molecules, transporters, signaling cascades, and proteins of unknown function, which together form the highly specialized interface during the development of BBB. The BBB transcriptome database is especially useful for understanding the molecular control of central nervous angiogenesis and the BBB permeability in the developing brain.

Previous studies have confirmed that the brain environment sends signals to brain endothelial cells to form the BBB, but the identity of these signals is largely unknown (Richard et al., 2010). Enriched KEGG pathway analysis in the neonatal groups captured some of the key events, such as “nicotine addiction,” and “axon guidance.” Consistent with our results, axon guidance was found as a key player during the developmental period of the human neocortex (Liu et al., 2016). Enriched KEGG pathway analysis in the 3–4-week-old group captured some of the key events, such as “cytokine-cytokine receptor interaction” or “mapk signaling pathway.” Cytokines play important roles in the physiological response to inflammation and neuroregeneration (Banks and Kastin, 1988; Ba Nks and Kastin, 1996; Pan and Kastin, 2001), and may be also important in the further maturation and maintenance of the BBB. Wnt/beta-catenin signaling is a key regulator of cerebral angiogenesis and BBB formation and maintenance; wnt ligands activate Frizzled receptors and Lrp5/6 complexes on the endothelium in the developing brain (Liebner et al., 2008; Zhou et al., 2014; Sweeney et al., 2019). In this study, the wnt and frizzled genes were involved in the cancer pathway, and these genes may further play a role by activating this pathway.

The BBB function continues to mature after birth, but the exact time window is still elusive and may depend on the species (Saunders et al., 2013; Halliday et al., 2016). In the mature stage, the mammalian BBB is stabilized by a highly specialized perivascular structure (Zhen et al., 2015), while the fundamental mechanisms underlying the formation and maintenance of the BBB remain largely unknown. Early research identified two structural features of the microvascular endothelium that underlie the restrictive properties of the BBB: special TJs and low rate of transcytosis (Farquhar and Palade, 1963; Reese and Karnovsky, 1967; Haseloff et al., 2015). Transcytosis in microvascular endothelial cells can be divided into two categories: receptor-mediated transcytosis (RMT), in which the ligand binds to the receptor to mediate endocytosis, and non-selective adsorbent transcytosis (Villaseñor et al., 2017). We found that the RMT required to maintain the BBB were different between the neonatal period and the adult period. Slc7a5, which transports branched chain amino acids through the barrier, was highly expressed in the neonatal period, while slc2a1, which transports glucose, was highly expressed in adulthood (Vivo et al., 1991; Tãrlungeanu et al., 2016). This may indicate that the substances needed to maintain BBB were different between development stages. Cav-1 regulates BBB permeability, and studies have shown that the increased number of endothelial caveolae and the rate of transcytosis were the reasons for the hyperpermeability of the BBB in the early stage of MCAO (Knowland et al., 2014). Phoneutria nigriventer spider venom-induced BBB opening is associated with a high expression of caveolae frame-forming cav-1 and the process of internalization and enhanced transcytosis (Soares et al., 2014). Therefore, it is considered that cav-1 can increase the BBB permeability via caveolae-based transcytosis. In this study, we found that, as the BBB matures, the expression of cav-1 and the caveolae-based transcytosis increased. Besides, we observed increased permeability to EB in rat cerebral cortex at 3–4 weeks and 6 months. Surprisingly, macroscopically increased EB permeability was seen in rat cortex at 6 months. A recent study found that cav-1-mediated vesicle transcytosis was significantly increased in 19-months-old mice compared with 3-months-old mice (Yang et al., 2020). This study suggests that an increase in permeability may occur in 6-month-old rats or earlier. Next, we assessed changes in tight junctions during blood-brain barrier development. We found the density of TJ particles increases, the inter-endothelial junctions become more extensive, interconnected and complexed, suggesting continued maturation of the BBB. While conventional wisdom has emphasized the importance of tight junctions, recent advances have highlighted the regulation of transcytosis as a crucial factor in limiting BBB permeability under physiological conditions and after pathological assaults (Daneman et al., 2010). Consistent with our findings, in the process of rat cortex development, tight junctions are continuously improved, but endothelial transcytosis increases, and the permeability of the blood-brain barrier increases. Our results suggest that under physiological conditions, low vesicle transcytosis is playing an important role in maintaining the integrity of the blood-brain barrier.

In summary, we have generated a comprehensive dataset describing the transcriptome of the developing BBB, which will provide a valuable resource for understanding the development and function of this crucial barrier, as well as its role of modulating CNS function.