Introduction

As a dynamic interface between the blood circulatory system and the central nervous system (CNS), the blood-brain barrier (BBB) maintains homeostasis and normal function of CNS by strictly regulating material exchange between the blood and brain (Palmiotti et al., 2014). The maintenance of BBB is mainly attributed to the expression of tight junctions (TJ) and adherent junctions (AJ) between adjacent brain endothelial cells as well as a variety of drug transporters. However, as a double-edged sword that protects CNS function, BBB also restricts the transport of some drugs from blood to brain, leading to poor CNS therapeutic effects and even CNS treatment failure (Banks, 2016).

Several in silico, in vitro, in situ, and in vivo methods have been developed to assess the permeability of drugs across BBB, but each method has its limitations (Hanafy et al., 2021). The in situ brain perfusion (ISBP) is considered the “gold standard” for assessing BBB permeability but there exist limits related to animal ethics. Moreover, ISBP is not suitable for human or high-throughput workflows. Thus, a suitable, accurate, and high-throughput in vitro BBB model is required to predict the permeability of drug candidates through BBB.

BBB is formed by neurovascular units (NVU), composed of neural (neurons, microglia, and astrocytes) and vascular components (vascular endothelial cells, pericytes, and vascular smooth muscle cells) (Potjewyd et al., 2018). The constant crosstalk and interactions among these cells contribute to the structural and signaling-based regulation of transcellular and paracellular transport, control of BBB permeability, and regulation of cerebral circulation (Arvanitis et al., 2020; Muoio et al., 2014; Potjewyd et al., 2018). Although brain microvascular endothelial cells (BMECs) are commonly utilized as an in vitro BBB model to assess drug permeability across BBB, in vitro mono-culture of BMECs tends to lose some unique characteristics without the support of other cell types. To overcome the drawbacks of mono-culture BBB models, a range of multicellular co-culture BBB models co-cultured with other NVU elements (such as astrocytes or pericytes) have been developed. Astrocytes are the most abundant glial cell type in brain (Clasadonte et al., 2017). Their terminal feet cover >80% of the surface of capillaries to form interdigitating coverage without slits (Mathiisen et al., 2010). In addition, astrocytes also contain several proteins related to the tight binding of the basement membrane. The in vitro multicellular co-culture BBB models composed of astrocytes, pericytes, and endothelial cells are considered to mimic the vascular structure and offer paracellular tightness in BBB (Ito et al., 2019; Nakagawa et al., 2009; Watanabe et al., 2021).

Neurons also play a crucial role in the NVU and may be involved in the regulation of BBB function. The maturation of BBB in mice considerably overlaps with the establishment of neuronal activity (Biswas et al., 2020). A study revealed that the conditioned medium collected from primary rat neurons also attenuated cell death caused by glucose-oxygen-serum deprivation (Lin et al., 2016). Furthermore, the primary rat neurons were reported to affect the differentiation and formation of BMECs (Savettieri et al., 2000; Schiera G. et al., 2003; Xue et al., 2013). Recent research reported that, compared with the double co-culture of hCMEC/D3 and 1321N1 (human astrocytoma cells), the triple co-culture of hCMEC/D3, 1321N1, and human neuroblastoma SH-SY5Y cells exhibited a higher transendothelial electrical resistance (TEER) (Barberio et al., 2022). A similar report showed that co-cultivation of RBE4.B (rat brain capillary endothelial cells) and neurons resulted in the lower permeability of [³H] sucrose than RBE4.B cells grown alone (Schiera Gabriella. et al., 2005). All these studies indicate that neurons, as the elements of the NVU, may be tightly connected to the formation and maintenance of BBB functions.

The aims of the study were: 1) to establish and characterize an in vitro triple co-culture BBB model consisting of human brain endothelial cells (hCMEC/D3), human astrocytoma cells (U251), and human neuroblastoma cells (SH-SY5Y); 2) to investigate whether neurons were involved in the formation and maintenance of BBB integrity and explore their underlying mechanisms. The integrity of BBB was evaluated by quantifying the leakage of both fluorescein and FITC-Dextran 3–5 kDa (FITC-Dex); 3) to validate the in vitro results through in vivo experiments. Finally, the in vivo/in vitro correlation assay (IVIVC) was analyzed to prove that the triple co-culture BBB model could better predict the BBB penetration of CNS drugs compared with the mono-culture BBB model.

Result

Establishment and characterization of the in vitro triple co-culture BBB model

TEER values were measured during the co-culture (Figure 1C). TEER values of the four in vitro BBB models gradually increased until day 6. On day 7, the TEER values showed a decreasing trend. Thus, a 6-day co-culture was used for subsequent experiments. The highest TEER values were observed in the triple co-culture BBB model, followed by double co-culture with U251 cells and double co-culture with SH-SY5Y cells BBB models, hCMEC/D3 cells mono-culture BBB model showed the lowest TEER values. We also measured the TEER values of U251 monolayer cells, and the results showed that U251 cells themselves also contributed to the physical barrier of the model (Figure 1D). The apparent permeability coefficients (P_app) values of fluorescein and FITC-Dex were measured to characterize the integrity of the four BBB models (Figure 1E, 1F). Consistent with the TEER values, the triple co-culture BBB model showed the lowest permeability of fluorescein and FITC-Dex, followed by double co-culture with U251 cells and double co-culture with SH-SY5Y cells BBB models. It was noticed that the monolayer of U251 cells itself also worked as a barrier, preventing the leakage of permeability markers, which may explain why the permeability of FITC-Dex in the double co-culture model with U251 cells is lower than that in the double co-culture model with SH-SY5Y cells (Figure 1G). Moreover, co-culture with SH-SY5Y, U251, and U251 + SH-SY5Y cells significantly enhanced the proliferation of hCMEC/D3 cells by 5.80%, 4.61%, and 10.41%, respectively (Figure 1H).

Figure 1.

The paracellular barrier of BBB is associated with TJs, AJs, and transporters (Abbott, 2013). The mRNA levels of TJs (claudin-5, ZO-1, and occludin), AJs (VE-cadherin and β-catenin), and transporters (P-gp, BCRP, OCT-1, OCT-2, OAT-3, and OATP1A1) in hCMEC/D3 cells from the four BBB models were analyzed using qPCR (Figure 1I and 1J). Compared with hCMEC/D3 cell mono-culture model, double co-culture with SH-SY5Y, double co-culture with U251, and triple co-culture BBB models showed markedly increases in claudin-5, VE-cadherin, and BCRP mRNA expression. Expression of corresponding proteins was measured using Western blot (Figure 1K-1N). Notably increased claudin-5 expression was detected in double co-culture with SH-SY5Y cells and triple co-culture BBB models, while VE-cadherin expression was markedly increased in double co-culture with U251 cells and triple co-culture BBB models. Expression levels of other TJ proteins (ZO-1 and occludin) and AJ protein (β-catenin) were unaltered. The expression of BCRP was slightly affected by co-cultivation without statistical differences. Significant negative correlations were found between P_app values of fluorescein and the expression of claudin-5 or VE-cadherin. P_app values of FITC-Dex were also negatively correlated to the expression levels of claudin-5 or VE-cadherin (Figure 1O-1R). These results indicate that the decreased permeability of fluorescein and FITC-Dex mainly results from the upregulated expression of both claudin-5 and VE-cadherin.

Neurons and astrocytes upregulated the expression of claudin-5 and VE-cadherin by GDNF secretion

The hCMEC/D3 cells did not direct contact with U251 or SH-SY5Y cells in the double co-culture and triple co-culture BBB models, indicating that cell-cell interaction between U251, SH-SY5Y, and hCMEC/D3 cells relied on secreted active factors. To test this hypothesis, the effects of conditioned medium (CM) from SH-SY5Y cells (S-CM), U251 cells (U-CM), and co-culture of SH-SY5Y and U251 cells (US-CM) on the expression of claudin-5 and VE-cadherin in hCMEC/D3 cells were analyzed (Figure 2A). Both S-CM, U-CM, and US-CM markedly increased the expression of claudin-5 and VE-cadherin. US-CM showed the strongest induction effects on claudin-5 and VE-cadherin.

Figure 2.

To investigate which cytokines were involved in the promotion of hCMEC/D3 cell integrity by U251 and SH-SY5Y cells, the mRNA expression levels of various cytokines in these three types of cells were compared. The results showed that U251 or SH-SY5Y cells exhibited significantly higher expression levels of GDNF, nerve growth factor (NGF), insulin-like growth factor-1 (IGF-1), vascular endothelial growth factor (VEGF), basic fibroblast growth factor (bFGF), and transforming growth factor-β (TGF-β) compared to hCMEC/D3 cells (Figure 2B). Furthermore, the mRNA expression of GDNF, IGF-1, TGF-β, and bFGF in SH-SY5Y cells was higher than those in U251 cells In these cytokines, GDNF (Dong et al., 2018; Igarashi et al., 1999; Shimizu et al., 2012), bFGF (Shimizu et al., 2011; Wang et al., 2016), IGF-1 (Ko et al., 2009; Nowrangi et al., 2019), and TGF-β (Fu et al., 2021) have been reported to promote BBB integrity. Thus, the concentrations of GDNF, bFGF, IGF-1, and TGF-β in the CMs were measured (Figure 2C). The results showed that levels of GDNF, bFGF, and IGF-1 in S-CM and US-CM were significantly higher than CMs from hCMEC/D3 cells (H-CM) and U-CM, but levels of TGF-β in S-CM and U-CM were lower than those in H-CM. Interestingly, the level of IGF-1 in US-CM was remarkably lower than that in S-CM, indicating that U251 cells suppressed IGF-1 secretion from SH-SY5Y. The effects of GDNF, bFGF, IGF-1, and TGF-β on the expression of claudin-5 and VE-cadherin were investigated (Figure 2D-2G). Among the four tested neurotrophic factors, only GDNF induced the expression of claudin-5 and VE-cadherin in a concentration-dependent manner (Figure 2D). In contrast, a high level (5 ng/mL) of TGF-β slightly downregulated claudin-5 and VE-cadherin expression (Figure 2G). These results demonstrate that upregulation of claudin-5 and VE-cadherin expression by US-CM are attributed to secreted GDNF.

To provide additional verification of the deduction, anti-GDNF antibody was used to neutralize exogenous and endogenous GDNF in culture medium. Consistent with our expectation, the anti-GDNF antibody concentration-dependent reversed the US-CM-induced claudin-5 and VE-cadherin expression (Figure 2H). Furthermore, GDNF-induced upregulation of claudin-5 and VE-cadherin expression was also reversed by the anti-GDNF antibody (Figure 2I). The induction effects of US-CM and GDNF on the claudin-5 and VE-cadherin expression in hCMEC/D3 cells were also confirmed in primary rat brain microvascular endothelial cells (Figure 2J)

GDNF forms a heterohexameric complex with two GFRα1 molecules and two RET receptors to activate the GDNF-GFRα1-RET signaling (Fielder et al., 2018). The RET receptor tyrosine kinase inhibitor SPP-86 (Bhallamudi et al., 2021) and Src-type kinase inhibitor PP2 (Morita et al., 2006) were used to further investigate whether GDNF upregulated the expression of claudin-5 and VE-cadherin in hCMEC/D3 cells by activating the GDNF-GFRα1-RET signaling pathway. Both SPP-86 and PP2 markedly attenuated claudin-5 and VE-cadherin expression induced by GDNF (Figure 2K) and US-CM (Figure 2L).

The contributions of GDNF-induced claudin-5 and VE-cadherin expression to TEER and permeability were investigated using hCMEC/D3 cells mono-culture BBB model. GDNF significantly increased TEER values (Figure 2M, 2O) and decreased the permeability of fluorescein and FITC-Dex (Figure 2N, 2P), which were almost abolished by SPP-86 or PP2. Furthermore, treatment with SPP-86 or PP2 completely reversed the increased TEER values (Figure 2Q, 2S) and decreased permeability of fluorescein and FITC-Dex (Figure. 2R, 2T) in the triple co-culture BBB model. These results indicate that neurons but also astrocytes upregulate claudin-5 and VE-cadherin expression in hCMEC/D3 cells by secreting GDNF. Subsequently, GDNF induces claudin-5 and VE-cadherin expression by activating GDNF-GFRα1-RET signaling.

GDNF induced claudin-5 and VE-cadherin expression of hCMEC/D3 by activating the PI3K/AKT and MAPK/ERK pathways

GDNF exerts its biological activities by activating several signaling pathways, including the phosphatidylinositol-3-kinase (PI3K)/ protein kinase B (AKT), mitogen-activated protein kinase (MAPK)/ extracellular regulated kinase (ERK), MAPK/ c-Jun N-terminal kinase (JNK), and MAPK/ p38 pathways (Fielder et al., 2018). The effects of the PI3K/AKT, MAPK/ERK, MAPK/JNK, and MAPK/p38 pathway inhibitors LY294002 (Figure 3A), U0126 (Figure 3B), SP600125 (Figure 3C), and SB203580 (Figure 3D), respectively, on GDNF-induced claudin-5 and VE-cadherin expression in hCMEC/D3 cells were investigated. GDNF increased claudin-5 and VE-cadherin expression, accompanied by the phosphorylation of AKT (p-AKT) and ERK (p-ERK). However, it did not stimulate the phosphorylation of JNK and p38. SPP-86, LY29002, and U0126 significantly suppressed GDNF-induced claudin-5 and VE-cadherin expression, while SP600125 and SB203580 had almost no effect on GDNF-induced claudin-5 and VE-cadherin expression. GDNF-induced phosphorylation of AKT and ERK was also markedly attenuated by the anti-GDNF antibody (Figure 3E). Similarly, US-CM remarkably upregulated the expression of claudin-5, VE-cadherin, p-AKT, and p-ERK, which were also markedly reversed by SPP-86, LY29002, U0126, or anti-GDNF antibody (Figure 3F-3J). These findings indicate that GDNF induces the expression of claudin-5 and VE-cadherin in hCMEC/D3 cells by activating both the PI3K/AKT and MAPK/ERK pathways.

Figure 3.

GDNF upregulated the claudin-5 expression in hCMEC/D3 cells by activating the PI3K/AKT/FOXO1 pathway

Claudin-5 is negatively regulated by the transcriptional repressor forkhead box O1 (FOXO1) (Beard et al., 2020). FOXO1 is also an important target of PI3K/AKT signaling. FOXO1 phosphorylation results in FOXO1 accumulation in the cytoplasm (Zhang et al., 2011) and lowers its level in the nucleus. Here, we investigated whether GDNF-induced claudin-5 expression is involved in FOXO1 nuclear exclusion. As shown in Figure 4A, both GDNF and US-CM significantly enhanced FOXO1 phosphorylation (p-FOXO1). Similarly, GDNF and US-CM increased the levels of phosphorylated and unphosphorylated FOXO1 in the cytoplasm and decreased the levels of nuclear FOXO1 (Figure 4B). Whether FOXO1 was involved in the GDNF-induced regulation of claudin-5 and VE-cadherin expression was investigated in hCMEC/D3 cells transfected with FOXO1 siRNA. Silencing FOXO1 significantly decreased FOXO1 levels in both whole-cell lysates and nucleus of hCMEC/D3 cells (Figure 4C), demonstrating FOXO1 silencing efficacy. Consistent with our expectation, silencing FOXO1 upregulated the expression of claudin-5 rather than VE-cadherin expression (Figure 4D). In contrast, high levels of FOXO1 were observed in both whole-cell lysates and nucleus of hCMEC/D3 cells that were transfected with plasmids containing FOXO1. Meanwhile, FOXO1 overexpression resulted in a decrease in both basal and GDNF-induced claudin-5 expression (Figure 4E), consistent with the known role of FOXO1 on claudin-5 expression.

Several reports have demonstrated that nuclear localization of FOXO1 is modulated by multiple pathways, including the PI3K/AKT (Tang et al., 1999) and MAPK/ERK (Asada et al., 2007) pathways. The effects of LY294002 and U0126 on GDNF-induced FOXO1 phosphorylation were measured in hCMEC/D3 cells. LY294002, but not U0126, significantly reversed the GDNF-induced alterations in total p-FOXO1, cytoplasmic p-FOXO1, cytoplasmic FOXO1, and nuclear FOXO1 (Figure 4F). Furthermore, LY294002 reversed the upregulation of claudin-5 expression induced by FOXO1 siRNA and GDNF (Figure 4G).

Figure 4.

GDNF upregulated VE-cadherin expression in hCMEC/D3 cells by activating the PI3K/AKT/ETS1 and MAPK/ERK/ETS1 signaling pathways

E26 oncogene homolog 1 (ETS1) is a transcription factor that binds to the ETS-binding site located in the proximal region of the VE-cadherin promoter, hence regulating the expression of VE-cadherin (Lelièvre et al., 2000; Luo et al., 2022). Activation of the PI3K/AKT (He et al., 2023; Hui et al., 2018) and MAPK/ERK (Watanabe et al., 2004) pathways was reported to upregulate the expression of ETS1. The previous results showed that GDNF induced VE-cadherin and claudin-5 expression in hCMEC/D3 cells by activating the PI3K/AKT and MAPK/ERK pathways. Therefore, we hypothesized that GDNF modulated ETS1 levels to promote VE-Cadherin and claudin-5 expression through PI3K/AKT and MAPK/ERK signaling pathways. Both US-CM and GDNF significantly increased total (Figure 5A) and nuclear ETS1 expression (Figure 5B). LY294002 and U0126 markedly attenuated GDNF-induced total (Figure 5C) and nuclear ETS1 expression (Figure 5D). To further confirm the involvement of the PI3K/AKT/ETS1 and MAPK/ERK/ETS1 pathways in GDNF-induced VE-cadherin expression, ETS1 in hCMEC/D3 cells was knocked down using ETS1 siRNA. ETS1 silencing saliently declined the expression levels of total (Figure 5E) and nuclear (Figure 5F) ETS1 in hCMEC/D3 cells, demonstrating silencing efficacy. In ETS1 silencing hCMEC/D3 cells, GDNF no longer induced the expression of total (Figure 5E) and nuclear (Figure 5F) ETS1. Moreover, ETS1 silencing substantially downregulated VE-cadherin expression and attenuated GDNF-induced VE-cadherin expression (Figure 5G and 5H), while having minimal impact on both basal and GDNF-induced claudin-5 expression (Figure 5G and 5I).

Figure 5.

Brain GDNF deficiency increased BBB permeability partly due to the impairment of claudin-5 and VE-cadherin expression

To further demonstrate the positive effects of GDNF on BBB maintenance, GDNF in mice brains was knocked down via intracerebroventricular injection of AAV-PHP.eB packaged with Gdnf short hairpin RNA (shRNA) (Figure 6A). Knockdown efficiency was confirmed through Western blotting (Figure 6B). Consistent with in vitro results, GDNF knockdown greatly downregulated the expression of claudin-5 and VE-cadherin in the brains of mice (Figure 6B). The integrity of BBB was assessed by examining the brain distributions of fluorescein and FITC-Dex. The results showed that specifically knocking down brain GDNF little affected plasma levels of fluorescein (Figure 6C) and FITC-Dex (Figure 6F), but significantly elevated the concentrations of fluorescein (Figure 6D) and FITC-Dex (Figure 6G) in the brains, leading to notable increases in the brain-to-plasma concentration ratios of two probes (Figure 6E and 6H). These alterations were consistent with the decline in claudin-5 and VE-cadherin expression. In addition, significant reductions in the levels of p-AKT (Figure 6I), p-ERK (Figure 6J), p-FOXO1 (Figure 6K), and ETS1 expression (Figure 6L) were observed in the brains of Gdnf knockdown mice.

Figure 6.

The triple co-culture BBB model better predicted the permeabilities of drugs across BBB

This study demonstrated that the developed triple co-culture BBB model displays better BBB characteristics compared with the hCMEC/D3 mono-culture BBB model. In this study, 18 drugs were utilized to further investigate the superiority of the developed triple co-culture BBB model over the hCMEC/D3 mono-culture BBB model. The P_appof the 18 drugs from the apical to the basolateral side based on the hCMEC/D3 mono-culture (P_{app, Mono}) and triple co-culture (P_{app, Triple}) BBB models are measured and listed in Table 5. The results showed that P_{app, Triple}values of all tested drugs were lower than the P_{app, Mono} values. Significant differences were observed in 14 out of 18 drugs. The predicted permeability coefficient-surface area product values (PS) of the tested drugs were respectively calculated based on their P_{app, Mono} values (PS _{Pre, Mono}), and P_{app, Triple} values (PS _{Pre, Triple}). The predicted PS values were further compared to their corresponding observations (PS _obs). The results showed that the predictive accuracy of PS _{Pre, Triple} was superior to PS _{Pre, Mono}. Except for verapamil, amitriptyline, fluoxetine, and clozapine, the predicted PS _{Pre, Triple}values of the other 14 drugs in the triple co-culture BBB models were within the 0.5–2 folds of PS _obs(Figure 7B). However, in the hCMEC/D3 mono-culture BBB model, only 7 predicted PS _{Pre, Mono} values were within the 0.5–2 folds range of their observations (Figure 7A).

Figure 7.

Discussion

The main findings of the study were to successfully develop an in vitro triple co-culture BBB model consisting of hCMEC/D3, U251, and SH-SY5Y cells and to confirm the involvement of neurons in BBB maintenance as well as the possible mechanisms. Co-culture with U251 and/or SH-SY5Y cells markedly promoted the TEER of hCMEC/D3 cells and reduced the leakage of fluorescein and FITC-Dex due to the upregulation of claudin-5 and VE-cadherin expression.

The roles of claudin-5 and VE-cadherin in the maintenance of BBB function have been demonstrated (Dejana et al., 2008; Hashimoto et al., 2023; Li W. et al., 2018; Ohtsuki et al., 2007). It was reported that Cld-5–deficient mice exhibited BBB impairment, allowing the transport of small molecules (<800 D) across BBB (Nitta et al., 2003). In contrast, claudin-5 overexpression significantly restricted the permeability of inulin across the conditionally immortalized rat brain capillary endothelial cell monolayer (Ohtsuki et al., 2007). Moreover, the claudin-5 expression in hCMEC/D3 is much lower than in human brain microvessels (Eigenmann et al., 2013; Ohtsuki et al., 2013; Weksler et al., 2013), which may cause the low TEER values. VE-cadherin, a major member of the cadherin family in endothelial cells, is also required for BBB integrity (Dejana et al., 2008; Li W. et al., 2018). The absence of VE-cadherin resulted in faulty cell-to-cell junctions and disrupted distribution of ZO-1 (Sauteur et al., 2017; Tunggal et al., 2005). Strongly negative correlations between P_app values of fluorescein or FITC-Dex and the expression of claudin-5 or VE-cadherin further demonstrated the importance of claudin-5 and VE-cadherin in BBB integrity.

Neurons and astrocytes, as important components of the NVU, may be involved in the formation and maintenance of BBB function. Several reports showed that co-culture with astrocytes CC-2565 or SC-1810 enhanced the TEER of hCMEC/D3 cells (Hatherell et al., 2011), whereas co-culture of RBE4.B cells with primary rat neurons and astrocytes showed a lower permeability of [³H] sucrose than mono-culture of RBE4.B cells (Schiera Gabriella. et al., 2005). Similarly, co-cultured with hCMEC/D3 cells and 1321N1 (astrocytes) or 1321N1+SH-SY5Y cells possess higher TEER values and lower permeability to Lucifer yellow than hCMEC/D3 alone. Compared with double co-culture of hCMEC/D3 and 321N1 cells, the triple co-culture of hCMEC/D3, 1321N1, and SH-SY5Y cells showed higher TEER values (Barberio et al., 2022). Our study also demonstrated that co-culture with U251 and/or SH-SY5Y cells significantly lowered BBB permeability and upregulated VE-cadherin or claudin-5 expression in hCMEC/D3 cells.

Next, we focused on the molecular mechanisms by which neurons and possibly astrocytes upregulated the expression of VE-cadherin and claudin-5 in BMECs. Co-culture with SH-SY5Y cells significantly upregulated claudin-5 and VE-cadherin expression in hCMEC/D3 cells. In the double co-culture with SH-SY5Y cells or triple co-culture BBB models, hCMEC/D3 cells were not in direct contact with SH-SY5Y cells, indicating that the interaction between SH-SY5Y and hCMEC/D3 cells depended on the release of some active compounds. It was consistent with the above deduction that the S-CM also markedly induced claudin-5 and VE-cadherin expression. Different from S-CM, U-CM mainly upregulated the expression of VE-cadherin and just had a slight impact on claudin-5 expression. In general, neurons but also astrocytes secrete some neurotrophic factors (Lonka-Nevalaita et al., 2010; Sweeney et al., 2019) that could contribute to the maintenance of structural stability of BBB. High levels of bFGF, GDNF, IGF-1, and TGF-β were detected in U-CM, S-CM, and US-CM. Further research showed that only GDNF induced VE-cadherin and claudin-5 expression in hCMEC/D3 cells in a concentration-dependent manner, and the anti-GDNF antibody attenuated claudin-5 and VE-cadherin expression induced by US-CM or GDNF. These results indicated that neurons upregulated claudin-5 and VE-cadherin expression through GDNF secretion. Levels of GDNF in S-CM were higher than those in U-CM, which seemed to partly explain why S-CM has a stronger promoting effect on claudin-5 than U-CM. The roles of GDNF in BBB maintenance and the regulation of claudin-5 and VE-cadherin expression were further confirmed using brain-specific Gdnf knockdown C57BL/6J mice. Consistent with our in vitro results, brain-specific Gdnf silencing greatly increased BBB penetration of fluorescein and FITC-Dex, accompanied by the downregulation of claudin-5 and VE-cadherin expression.

Generally, GDNF activates several signal transduction pathways, such as the PI3K/AKT and MAPK signaling pathways (Fielder et al., 2018) by forming a heterohexameric complex with two GFRα molecules and RET receptors. It was also reported that GDNF improved BBB barrier function due to the activation of MAPK/ERK1 (Dong et al., 2018) and PI3K/AKT (Liu D. et al., 2022) signaling. Consistent with previous reports, we found that signaling inhibitors SPP-86, PP2, LY294002, and U0126 markedly attenuated US-CM– and GDNF-induced claudin-5 and VE-cadherin expression in hCMEC/D3 cells, inferring that GDNF promoted claudin-5 and VE-cadherin expression via activating both the PI3K/AKT and MAPK/ERK pathways.

Signal transduction pathways control gene expression by modifying the function of nuclear transcription factors. The nuclear accumulation of FOXO1 negatively regulates claudin-5 expression (Beard et al., 2020; Taddei et al., 2008). FOXO1 is an important target of the PI3K/AKT signaling axis (Zhang et al., 2011), and AKT-induced phosphorylation of FOXO1 results in cytoplasmic FOXO1 accumulation and decreases nuclear FOXO1 accumulation (Zhang et al., 2011). From these results, we inferred that GDNF-induced claudin-5 expression in hCMEC/D3 cells may be involved in the activation of PI3K/AKT/FOXO1 pathway. Similarly, both US-CM and GDNF increased cytoplasmic p-FOXO1 and decreased nuclear FOXO1 in hCMEC/D3 cells, which was reversed by LY29002 rather than U0126. Roles of FOXO1 in GDNF-induced claudin-5 were verified through silencing and overexpressing FOXO1 in hCMEC/D3 cells. FOXO1 silencing enhanced claudin-5 but not VE-cadherin expression, accompanied by a decline in total and nuclear FOXO1. In contrast, FOXO1 overexpression significantly decreased claudin-5 expression. In hCMEC/D3 cells overexpressing FOXO1, GDNF lost its promotion effect on claudin-5 expression. It was noticed that U0126 attenuated the GDNF-induced upregulation of claudin-5 but had minimal impact on GDNF-mediated FOXO1 phosphorylation and the decline of nuclear FOXO1. In Sertoli cells, it was found that testosterone stimulated claudin-5 expression by activating the RAS/RAF/ERK/CREB pathway (Bulldan et al., 2016). The transcriptional regulation of claudin-5 by CREB was confirmed in bEnd.3 (mouse brain endothelial cell). CREB overexpression significantly increased both gene and protein expression of claudin 5. In contrast, depletion of CREB decreased claudin-5 expression in gene and protein levels (Li Y. et al., 2022). However, another report showed that in human lung microvascular endothelial cells, U0126 attenuated phosphorylation of ERK and lipopolysaccharide-stimulated claudin-5 damage, indicating activation of MAPK/ERK pathway impaired rather than promoted claudin-5 expression (Liu Y. et al., 2019). Thus, the real mechanisms that GDNF-induced activation of the RET/MAPK/ERK pathway promotes claudin-5 expression need further investigation.

ETS1 is a member of the ETS family that plays an important role in cell adhesion, migration, and blood vessel information. ETS1 binds to an ETS-binding site located in the proximal region of the VE-cadherin promoter, controlling VE-cadherin expression (Lelièvre et al., 2001). Several studies have demonstrated the role of ETS1 in the regulation of VE-cadherin expression. For example, IFN-γ and TNF-α impaired BBB integrity by decreasing ETS1-induced VE-cadherin expression (Luo et al., 2022). ETS1 silencing reduced VE-cadherin expression in umbilical vein endothelial cells (Colas-Algora et al., 2020). In contrast, ETS1 overexpression induced VE-cadherin expression in mouse brain capillary endothelial cells and fibroblasts (Lelièvre et al., 2000). As expected, ETS1 silencing resulted in a decrease in the expression of VE-cadherin in hCMEC/D3 cells, but claudin-5 expression remained unaffected. Additionally, ETS1 silencing removed the inductive effect of GDNF on VE-cadherin expression while unaffecting the upregulation of claudin-5 induced by GDNF. The activation of the PI3K/AKT (He et al., 2023; Hui et al., 2018) and MAPK/ERK (Watanabe et al., 2004) pathways promotes ETS1 expression. Our findings also demonstrated that US-CM and GDNF significantly increased total and nuclear ETS1 levels, which were eliminated by signaling inhibitors LY294002 and U0126. Both our results and previous research provide evidence that GDNF upregulates ETS1 expression via the activation of PI3K/AKT and MAPK/ERK signaling to promote VE-cadherin expression.

The present study showed that the characteristics of the in vitro triple co-culture BBB model were superior to those of the hCMEC/D3 mono-culture BBB model. The hCMEC/D3 mono-culture and triple co-culture BBB models were used to try to predict the PS values of 18 drugs by comparing them with their observations. The prediction success rate (14/18) of triple co-culture BBB model was greater than that of hCMEC/D3 mono-culture BBB model (7/18). However, poor predictions were observed for verapamil, amitriptyline, fluoxetine, and clozapine, which may partly be due to inaccuracies in their f_{u, brain} values. These four drugs are high protein-binding drugs. Due to the methodological discordance and the limitations of the historic devices for these drugs, the f_u ≤ 0.01 maybe with low confidence and accuracy (Bowman et al., 2018; Di et al., 2017). For example, f_{u, brain}values of amitriptyline from different reports display large differences (0.002 (Sanchez-Dengra et al., 2021), 0.0071 (Weber et al., 2013)). The same large differences of f_{u, brain}values were also found in fluoxetine (0.0023 (Maurer et al., 2005), 0.004 (Summerfield et al., 2007), 0.00094 (Liu X. et al., 2005)), and clozapine (0.0056 (Bhyrapuneni et al., 2018), 0.014 (Cremers et al., 2012), 0.0094 (Summerfield et al., 2007)).

In summary, the developed triple co-culture BBB model outperformed the mono-culture or double co-culture BBB models, mainly attributing to the fact that neurons and possibly astrocytes upregulated claudin-5 and VE-cadherin expression by secreting GDNF. GDNF upregulates claudin-5 expression by activating the PI3K/AKT/FOXO1 and MAPK/ERK pathways, and the activation of PI3K/AKT and MAPK/ERK signaling also promotes ETS1 to modify VE-cadherin expression. Additionally, the developed in vitro triple co-culture BBB model accurately predicted the in vivo BBB permeability of CNS drugs. This suggests the potential of our triple co-culture BBB model for utilization in CNS candidates screening during the drug development process (Figure 8).

Figure 8.

Materials and Methods

Cell culture

Rat brain microvascular endothelial cells were isolated from Sprague-Dawley rats (male, 7–10 days old, Sino-British Sippr/BKLaboratory Animal Ltd, Shanghai, China) as the described method (Ji et al., 2013; Li M. S. et al., 2013) and cultured in Dulbecco’s Modified Eagle Media (DMEM)/F12 (#12500-039, Gibco, Carlsbad, CA, USA) containing 10% fetal bovine serum (FBS) (#10100147C, Gibco, Carlsbad, CA, USA) and 62.5 µg/mL penicillin and 100 µg/mL streptomycin (SunShine Biotechnology Co., ltd. Nanjing, China). Then, hCMEC/D3, U251, and SH-SY5Y cells were cultured in DMEM/F12 containing 10% FBS, 62.5 µg/mL penicillin and 100 µg/mL streptomycin.

Establishment of the triple co-culture model

The establishment process of the triple co-culture model is illustrated in Figure 1A. SH-SY5Y cells were seeded at a density of 4.5 × 10⁴ cells/cm² in plates and differentiated with 10 μM retinoic acid (Sigma-Aldrich, St. Louis, MO, USA) for 72 h. The differentiated SH-SY5Y cells were cultured in the fresh DMEM/F12 medium containing 10% FBS. U251 cells were seeded at 2 × 10⁴ cells/cm² on the bottom of Transwell inserts (0.4 µm pore size, SPL Life Sciences, Pocheon, Korea) coated with rat-tail collagen (Corning Inc., Corning, NY, USA). Next, the inserts were suspended in plate wells containing the culture medium after 5 h of incubation. After 24 h of incubation, hCMEC/D3 cells were seeded on the apical side of the inserts at 3 × 10⁴ cells/cm² and cultured for another 48 h. Then, the inserts seeded with U251 and hCMEC/D3 cells were suspended in plates seeded with differentiated SH-SY5Y cells and co-cultured for another 6 days. The culture medium was replaced every 24 h.

Three other types of BBB in vitro models, namely the hCMEC/D3 cell mono-culture model, the double co-culture with SH-SY5Y cells BBB model, and the double co-culture with U251 cells BBB model were also established (Figure 1B). TEER was periodically measured using a Millicell-ERS (MERS00002) instrument (Millipore, Billerica MA, USA) to monitor cell confluence and development of tight junctions.

In vitro BBB permeability study

The culture medium was removed from the apical and basolateral sides of the inserts and washed twice with preheated Hank’s balanced salt solution (HBSS). Fresh HBSS was then added to both the apical and basolateral chambers. After 15 min of preincubation, HBSS in the apical and basolateral chambers was replaced with HBSS containing FITC-dextran 3–5 kDa (FITC-Dex) (Sigma-Aldrich, St. Louis, MO, USA), fluorescein sodium (Sigma-Aldrich, St. Louis, MO, USA), or other tested agents and blank HBSS, respectively. Next, 200 μL aliquots were collected from the basolateral chamber after 30 min of incubation at 37 °C. The concentrations of the tested agents in the basolateral chamber were measured.

The apparent permeability coefficients (P_app, cm/s) of the tested agents across the in vitro BBB model were calculated using the equation (Tavelin et al., 2002):

where S is the surface area of the insert membrane (0.33 cm² for 6.5 mm inserts, 4.46 cm² for 24 mm inserts), Q is the transported amount of the tested agents transported from the donor chamber to the receiver chamber for 30 min (1800 s), and C₀ is the initial concentration of the tested agents in the donor chamber.

The quantification methods of prazosin, verapamil, lamotrigine

Prazosin, verapamil, and lamotrigine (Aladdin, Shanghai, China) were analyzed by high-performance liquid chromatography (Shimadzu, Kyoto, Japan) with YMC-Triart C18 column (5 μm, 150 × 4.6 mm, YMC America Inc., Allentown, PA, USA). Prazosin and verapamil were detected using the RF-20A fluorescence detector. Lamotrigine was detected using the SPD-20A ultraviolet detector. Samples were centrifugated at 12000 rpm for 10 mins, then 150 μL supernatant was taken and used for analysis. The run temperature was set at 40 ℃, the injection volume was 20 μL and the flow rate was 1 mL/min. Initial concentrations in donor chamber and other chromatographic conditions of drugs are summarized in Table 1.

Table 1

The quantification methods of clozapine, venlafaxine, bupropion, amantadine, carbamazepine, fluoxetine, amitriptyline, gabapentin, midazolam, risperidone, olanzapine, mirtazapine, metoclopramide, doxepin, donepezil

All compounds were purchased from Aladdin (Shanghai, China). Except for prazosin, verapamil, and lamotrigine, the other compounds were analyzed by using liquid chromatography-mass spectrometry (Shimadzu, Kyoto, Japan) with YMC-Triart C18 column (5 μm, 150 × 2.0 mm, YMC America Inc., Allentown, PA, USA). Each sample was mixed with 10 µL internal standard. Then 1 mL extraction was added to each sample. The samples were vortex vibrated on the oscillator for 10 min, and then centrifuged at 4 ℃ and 12000 rpm for 10 mins. The supernatant solvent was evaporated with nitrogen flow, then redissolved with 100 µL 40 % (v/v) acetonitrile, and centrifuged at 4 ℃ and 15000 rpm for 10 min. The supernatants were injected into LC-MS for analysis. The injection volume of each sample was 5 µL. The mass charge ratio, extraction, and initial concentrations in donor chamber of drugs were summarized in Table 2.

Table 2

Cell proliferation analysis

On day 6 after co-culture, hCMEC/D3 cells were fixed with 4% paraformaldehyde for 15 min and washed with phosphate buffered saline (PBS) for three times. Next, the fixed cells were blocked with 5% goat serum for 2 h and washed with PBS for four times. The blocked cells were incubated with DAPI (Invitrogen, Carlsbad, CA, USA) and washed with PBS for four times. Cell numbers were counted using Cytation5 (BioTek, Winooski, VT, USA).

Reverse transcription and Quantitative real-time PCR (qPCR)

Total RNA of cells was extracted using RNAiso Plus reagent (Takara Bio Inc. Otsu, Shiga, Japan) and reverse transcribed using HiScript III RT SuperMix (Vazyme, Shanghai, China) as the described method (Yang et al., 2023). Paired primers were synthesized by Tsingke Biotech Co., Ltd (Beijing, China), and their sequences were listed in Table 3. The SYBR Master Mix was purchased from Yeasen (Shanghai, China). Then, qPCR was performed on the Applied Biosystems QuantStudio 3 real-time PCR system (Thermo Fisher Scientific, Waltham, MA, USA). The mRNA levels of related genes were normalized to ACTB or GAPDH using the comparative cycle threshold method.

Table 3

Western blotting analysis

Whole-cell and tissue lysates, nucleoprotein, and cytoplasmic protein were prepared using RIPA Lysis Buffer (Beyotime, Shanghai, China) as the described method (Wu et al., 2021). Proteins were separated through sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred onto nitrocellulose or polyvinylidene difluoride membranes. The membranes were blocked with 5% skim milk and incubated with corresponding primary antibodies at 4 ℃ overnight. After being washed with Tris-buffered saline Tween buffer, the membranes were incubated with secondary antibodies (Cell Signaling Technology, MA, USA) at 1:3000 dilution: Anti-mouse IgG, HRP-linked Antibody (#7076), Anti-rabbit IgG, HRP-linked Antibody (#7074). Protein levels were visualized using a highly sensitive ECL western blotting substrate and a gel imaging system (Tanon Science & Technology, Shanghai, China).

Preparation of conditioned medium

Conditioned medium (CM) of U251 cells (U-CM), SH-SY5Y cells (S-CM), or co-culture of U251 and SH-SY5Y cells (US-CM) were prepared. U251 cells were seeded at the top of the insert membrane and suspended on 6-well plates seeded with differentiated SH-SY5Y cells to co-culture U251 cells with SH-SY5Y cells. The medium was collected every 24 h. The CMs were subsequently used for hCMEC/D3 cell culture after filtrating with 0.2 μm filters for 144 h. The levels of glia-derived neurotrophic factor (GDNF), basic fibroblast growth factor (bFGF), insulin-like growth factor-1 (IGF-1), and transforming growth factor-β (TGF-β) in CMs were measured using corresponding ELISA kits according to the manufacturers’ instructions.

Neutralization of GDNF with Anti-GDNF Antibody

Exogenous and endogenous GDNF in the medium was neutralized with anti-GDNF antibody (#AF-212-NA, R&D system, Minneapolis, MN, USA). 0.25, 0.5, 1.0 μg/mL anti-GDNF antibody was added into the US-CM or medium containing 200 pg/mL GDNF, and then the medium was preincubated at 4 ℃ for 1 h. The hCMEC/D3 cells were incubated with 200 pg/mL GDNF or US-CM containing anti-GDNF antibody or not for 6 days, and then cell lysate was collected for western blot. The medium was replaced every 24 h.

Transfection of hCMEC/D3

Here, hCMEC/D3 cells were plated in the plates or culture dishes at 6 × 10⁴ cells/cm² and transfected with 10 nM of negative control or human FOXO1 and ETS1 small interfering RNA (siRNA) (Tsingke Biotechnology, Beijing, China) for 12 h using Lipofectamine™ 3000 (Invitrogen, Carlsbad, CA, USA) reagent according to the manufacturer’s instructions. Cells were then incubated with a medium containing GDNF for 72 h. The siRNA sequences of human FOXO1 and ETS1 were summarized in Table 4.

Table 4

FOXO1 Overexpression by plasmids

The plasmids encoding FOXO1 (EX-Z7404-M02) were constructed by GeneCopoeia (Rockville, MD, USA). The hCMEC/D3 cells were plated in plates or dishes at 6 × 10⁴ cells/cm². They were subsequently transfected with 1 µg of negative control or plasmids encoding FOXO1 6 h using Lipofectamine™ 3000 reagent according to the manufacturer’s instructions. Transfected cells were then incubated with the medium containing GDNF for 72 h.

Animals

C57BL/6J mice (male, 4–5 weeks old, 16–18 g, 12 mice) were obtained from Sino-British Sippr/BKLaboratory Animal Ltd (Shanghai, China). Mice were maintained in groups under standard conditions with free access to food and water. Animal studies were performed in accordance with the Guide for the Care and Use of Laboratory Animals (National Institutes of Health) and approved by the Animal Ethics Committee of China Pharmaceutical University (Approval Number: 202307003).

Brain-specific Gdnf knockdown and evaluation of BBB permeability

Mice were randomly divided into control (shNC) and Gdnf silencing (shGdnf) groups (6 mice each group). The shRNA sequence of mice Gdnf was listed in Table 4. The 2 × 10⁹ viral genome each of pAAV-U6-shRNA (NC2)-CMV-EGFP or pAAV-U6-shRNA (Gdnf)-CMV-EGFP (OBio Technology, Beijing, China) were injected into the bilateral lateral ventricle area (relative to the bregma: anterior-posterior –0.3 mm; medial-lateral ±1.0 mm; dorsal-ventral –3.0 mm) through intracerebroventricular (i.c.v) infusion. Three weeks following i.c.v injection, BBB permeability and expression of corresponding targeted proteins were measured in the mice.

A mixture of FITC-Dex (50 mg/kg) and fluorescein sodium (10 mg/kg) was intravenously administered to experimental mice. Thirty minutes after the injection, the mice were euthanized under isoflurane anesthesia, and brain tissue and plasma samples were obtained quickly. The concentrations of FITC-Dex and fluorescein in the plasma and brain were measured as previously described (Li P. et al., 2022; Zhou Y. et al., 2019). No blinding was performed in animal studies.

The prediction of drug permeability across BBB using the developed in vitro BBB model

The P_app values of 18 drugs – prazosin, verapamil, lamotrigine, clozapine, venlafaxine, bupropion, amantadine, carbamazepine, fluoxetine, amitriptyline, gabapentin, midazolam, risperidone, olanzapine, mirtazapine, metoclopramide, doxepin, and donepezil – across the hCMEC/D3 cells mono-culture and triple co-culture models were measured. The predicted in vivo permeability-surface area product (PS, μL/min/g brain) values across BBB were calculated using the following equation:

where VSA is the luminal area of the vascular space of brain, which was set to 150 cm²/g (Fenstermacher et al., 1988), and f_{u, brain} is the unbound fraction of brain.

The published in vivo brain permeability values were unified to observed PS (PS _Obs) by multiplying by VSA equal to 150 cm²/g. The PS _Obsand f_{u, brain}of these 18 drugs were summarized in Table 5. If PS _Pre values were within 0.5–2.0 folds of observations, the prediction was considered successful.

Table 5.

Statistical analyses

All results are expressed as mean ± S.E.M. Data were analyzed using GraphPad Prism software version 8.0.2 (GraphPad Software, La Jolla, CA, USA). P < 0.05 was considered statistically significant. The unpaired Student’s t-test or unpaired t-test with Welch’s correction for two-group comparisons, and one-way ANOVA followed by Fisher’s LSD test or Welch’s ANOVA test was used for multiple group comparisons. The simple linear regression analysis was used to examine the presence of a linear relationship between two variables.

Acknowledgements

The authors would like to give special thanks to the support of the China Pharmaceutical University Pharmaceutical Animal Laboratory Center.

Funding information

Author Contributions

Lu Yang: Conceptualization, Methodology, Investigation, Data curation, and Writing–original draft preparation; Zijin Lin: Data curation, Formal analysis; Ruijing Mu: Visualisation, Software; Wenhan Wu: Formal analysis; Hao Zhi: Data curation; Xiaodong Liu: Supervision, Conceptualization, Project administration, Funding acquisition; Hanyu Yang: Supervision, Conceptualization, Funding acquisition; Li Liu: Supervision, Conceptualization, Project administration, Funding acquisition, Resources. All authors reviewed and approved the final version of this manuscript.

Declaration of interests

The authors declare no competing interests.

Data availability

All data are available in the manuscript and supporting files; source data files for western blots have been provided for all figures.