Extracellular matrix proteins in construction and function of in vitro blood-brain barrier models

The blood-brain barrier (BBB) is a highly impermeable barrier separating circulating blood and brain tissue. A functional BBB is critical for brain health, and BBB dysfunction has been linked to the pathophysiology of diseases such as stroke and Alzheimer’s disease. A variety of models have been developed to study the formation and maintenance of the BBB, ranging from in vivo animal models to in vitro models consisting of primary cells or cells differentiated from human pluripotent stem cells (hPSCs). These models must consider the composition and source of the cellular components of the neurovascular unit (NVU), including brain microvascular endothelial cells (BMECs), brain pericytes, astrocytes, and neurons, and how these cell types interact. In addition, the non-cellular components of the BBB microenvironment, such as the brain vascular basement membrane (BM) that is in direct contact with the NVU, also play key roles in BBB function. Here, we review how extracellular matrix (ECM) proteins in the brain vascular BM affect the BBB, with a particular focus on studies using hPSC-derived in vitro BBB models, and discuss how future studies are needed to advance our understanding of how the ECM affects BBB models to improve model performance and expand our knowledge on the formation and maintenance of the BBB.


Introduction
The neuronal network in the central nervous system (CNS) is highly complex, requiring a specific microenvironmental composition to function properly (Abbott, 1992;Shao et al., 2021). Brain capillaries, which account for~85% of the~644 km cerebral vessels (Zlokovic, 2008;Sweeney et al., 2018), not only supply nutrients and oxygen and remove wastes to support the metabolic demands of the CNS, but also form a highly impermeable, regulated barrier that restricts the entry of most molecules in the bloodstream and actively transports specific classes of molecules in a polarized manner. This barrier is referred to as the bloodbrain barrier (BBB) and is essential to a healthy brain, with dysfunction linked to prevalent and debilitating diseases such as Alzheimer's disease (AD), multiple sclerosis (MS), stroke, and traumatic brain injury (Cirrito, 2005;Weiss et al., 2009;Shlosberg et al., 2010). Anatomically, the BBB consists of brain microvascular endothelial cells (BMECs) surrounded by brain pericytes embedded in a shared basement membrane (BM), and encased by astrocytic endfeet (Figure 1) (Abbott et al., 2006;Tornabene and Brodin, 2016). The BM is mainly composed of extracellular matrix (ECM) proteins that are secreted and deposited by BMECs, pericytes, and astrocytes.
Compared to microvascular endothelial cells (ECs) outside of the CNS, BMECs have no fenestra, few intracellular vesicles, and high expression of tight junction (TJ) and TJ-related proteins such as Occludin, Claudin-5, and ZO-1. These differences result in low rates of endocytosis and transcytosis, and continuous tight junctions that restrict paracellular permeability, respectively. BMECs also differentiate themselves from non-brain ECs by the expression of nutrient transporters and polarized efflux transporters such as glucose transporter (Glut-1), and P-glycoprotein (P-gp), respectively. The quality of the barrier formed by BMECs is typically measured by its low permeability to molecular tracers such as fluorescent dyes and other small molecules, its high transendothelial electrical resistance (TEER), and by the directional transport of specific transporter substrates.
The cellular neighbors of BMECs within the neurovascular unit (NVU), brain pericytes and astrocytes, also are crucial to a functional BBB. As early as the 1970s, Bradbury et al. suggested a special relationship between brain capillaries and astrocytes (Bradbury, 1979), soon after which in vivo and in vitro studies confirmed the importance of endothelial-astrocyte interactions for barrier properties in BMECs (Debault and Cancilla, 1980;Janzer and Raff, 1987). Building upon this knowledge, a standardized, reliable in vitro co-culture protocol was described in 1990 using primary rat astrocytes for improved barrier function in bovine BMECs (Dehouck et al., 1990). More recent studies described how pericytes stabilize capillary structures and upregulate barrier function in endothelial cells (Ramsauer et al., 2002;Hori et al., 2004;Dohgu et al., 2005). In 2010, two independent studies demonstrated how brain pericytes are required for both development of the embryonic BBB and maintenance of the adult BBB (Armulik et al., 2010;Daneman et al., 2010). More specifically, these studies demonstrated that BBB permeability was inversely correlated to brain pericyte coverage in developing mouse brains (Daneman et al., 2010) and that pericyte deficiency increased BBB permeability in adult mice (Armulik et al., 2010).
While these studies have provided insight into the role of the cellular components of the NVU in regulating the BBB, we have less of an understanding of how the BM regulates BBB formation and maintenance. Here, we first introduce the major components of the vascular BM and review studies demonstrating the importance of each component as it relates to the in vivo BBB. We then discuss the impact of ECM components on in vitro BBB models, focusing on models comprised of hPSC-derived cells, and briefly discuss how cell type-specific ECM could be important in BBB models. Finally, we describe how hPSCderived models can be used to explore the effects of brain BM on the BBB to improve these models, to advance our understanding of the BBB in development and disease, and to develop improved strategies to target therapies across the BBB.
2 Major ECM components and dynamics of the BM Biochemically, the vascular BM has been profiled on the protein level, primarily by targeted in situ immunostaining and western blotting (Barber and Lieth, 1997;Paulsson, 1992;Po€schl et al., 2004;Sixt et al., 2001;Song et al., 2017;Sorokin et al., 1994). These experiments have demonstrated that the primary constituents of the BM can be classified into four major types of ECM proteins: collagen IV, laminin, nidogen (also called entactin), and heparan sulfate proteoglycans (HSPGs), which include perlecan and agrin.

Collagen
Collagens are proteins with a characteristic triple helix of three α-chains. There are more than 20 types of collagens. Type I collagen, the most abundant and well-studied collagen, is found in bones, tendons, skin, ligaments, cornea and many interstitial connective tissues (Gelse, 2003). Type IV collagen is mainly found in basement membranes, and six different collagen IV α-chains have been identified. Collagen IV, with mainly α1 and α2 chains (encoded by COL4A1 and COL4A2), is synthesized by BMECs, pericytes, and astrocytes (Webersinke et al., 1992;Tilling et al., 2002;Stratman et al., 2009;Vanlandewijck et al., 2018). Collagen IV can bind to α1β1 and α3β1 integrins which are expressed by BMECs and astrocytes (Baeten and Akassoglou, 2011).
Collagen IV is a major structural component of vascular BMs throughout the body. Global knockout of COL4A1/2 results in lethality as early as embryonic day 10.5 (E10.5) in mouse models (Po€schl et al., 2004). COL4A1 mutations are also correlated with cerebrovascular and neurological diseases such as ischemic stroke, intracerebral hemorrhage (ICH), and porencephaly in human families (Gould et al., 2006(Gould et al., , 2005. Conditional deletion of COL4A1 in mouse BMECs or pericytes caused fully penetrant ICH, incompletely penetrant porencephaly, and macroangiopathy, demonstrating a central role of collagen IV in vascular defects and brain damage (Jeanne et al., 2015).
Like collagen IV, laminin is a major structural component of vascular BM, and most global knockouts of its subunits (e.g., α5, β1, γ1) are embryonically lethal. Recently, viable genetic mouse models revealed that laminin contributes to BBB integrity. In 2014, Yao et al. (2014) showed that conditional deletion of laminin γ1 in mouse neural progenitor cells (and thus astrocytes differentiated from these cells) resulted in BBB breakdown and spontaneous ICH. Menezes et al. (2014) showed that global knockout of laminin α2 in mice resulted in defective BBB with increased permeability to Evans blue. , Gautam et al. (2019 showed that conditional deletion of laminin α5 in mouse endothelial cells had little effect under homeostatic conditions, but resulted in elevated BBB permeability after ICH was induced by intracerebral injection of collagenase. In 2020, Gautam et al. (2020) reported that conditional knockout of pericyte-derived laminins in mice resulted in mild BBB breakdown during aging. In short, laminins are important to the BBB, especially for maintaining BBB integrity.
Global knockout of both nidogen isoforms in mice is perinatally lethal (Bader et al., 2005), although mice lacking nidogen-2 showed no overt abnormalities (Schymeinsky et al., 2002). In mice with global knockout of nidogen-1, the thickness of brain capillary BM was significantly reduced compared to the wild type. In some cases, the brain capillary BM was completely absent and the BMECs appeared swollen in the cerebral cortex, which suggests possible functional defects of the CNS (Dong et al., 2002). However, in a later study, the authors reported no significant defects or abnormalities in the CNS of nidogen-1-null mice and concluded that there was no significant difference between mutant mice and wild type mice in terms of the exclusion of Evans blue by the BBB (Vasudevan et al., 2009).
Global knockout of perlecan leads to embryonic lethality in mice (Arikawa-Hirasawa et al., 1999;Costell et al., 1999). , Nakamura et al. (2019 reported that even though perlecan deficiency does not appear to affect the BBB under normal conditions, more BBB leakage and larger infarct volumes were detected in conditional perlecan-deficient mice after transient middle cerebral artery occlusion. Moreover, primary human BMECs and brain pericytes attached to substrates coated with either full-length or the C-terminal domain V of perlecan (perlecan DV) in vitro, and the use of perlecan DV coating promoted PDGF-BB-induced pericyte migration in an in vitro wound healing assay.
Global knockout of agrin is also embryonically lethal in mice (Gautam et al., 1996). Conditional knockout of agrin in mouse endothelial cells resulted in increased brain vascular accumulation of β-amyloid (Aβ), which is linked to AD pathology (Rauch et al., 2011). During chick and rat development, agrin was detected in association with brain microvessels around the time when the BBB was formed (Barber and Lieth, 1997). Taken together, nidogen and agrin have been linked to BBB formation and function, although their mechanisms of action are unclear, while perlecan has been suggested to support BBB maintenance and repair via pericyte recruitment following ischemic stroke.

Other ECM components of the vascular BM
Unlike capillaries and postcapillary venules, arterioles and venules are not surrounded by pericytes embedded in endothelial BM, but instead vascular ECs are surrounded by endothelial BM, interstitial matrix, then smooth muscle cells and their BM, in that order. The interstitial matrix consists mainly of collagen I and collagen III, together with other minor components such as decorin, biglycan, fibronectin, and vitronectin (Thyberg et al., 1990;Dufourcq et al., 1998;Yousif et al., 2013). However, it is noteworthy that most of our understanding of the composition of the vascular BM surrounding arterioles and venules is based on studies of vessels from organs other than the brain. Profiling the composition of the BM in larger vessels in the brain will be important to identify similarities and differences to the BM in other vessels in the body.

Developmental dynamics of the BM
The composition of the vascular BM during development is dynamic, guiding formation and maturation of vessels. For instance, based on in situ hybridization, capillary ECs did not express LAMA5 in embryonic and newborn mouse brains, but LAMA5 was found in mouse brain capillary ECs~4 weeks after birth (Sorokin et al., 1997). LAMA4 expression on the other hand, was detected by in situ hybridization in mouse brain capillary ECs as early as embryonic day 13 . Thus, laminin composition appears to be developmentally dynamic in brain with LAMA4 expressed throughout development but LAMA5 induced in mature capillaries. Other than laminin, the details of the developmental dynamics of brain-specific vascular BM and their role in regulating BBB induction remain largely unclear.

Changes in BM in neurological degenerative diseases
AD is the most common form of dementia, and one of the pathological hallmarks of AD is cerebral amyloid angiopathy (CAA), the abnormal accumulation of Aβ in cerebral blood vessel walls. Recent hypotheses propose a combination of genetic factors and vascular factors, such as BBB dysfunction (Nelson et al., 2016), in AD pathogenesis. For instance, BMECs and pericytes are found to be impaired in terms of their Aβ-clearing abilities in the early stage of AD (Gorelick et al., 2011;Montagne et al., 2018). At the same time, various studies using mouse models and post-mortem human tissues have demonstrated BM thickening in AD brains (Mancardi et al., 1980;Claudio, 1995;ZAROW et al., 1997;Bourasset et al., 2009;Gama Sosa et al., 2010;Merlini et al., 2011;Lepelletier et al., 2017), thus it is hypothesized that BM thickening may play a role in compromising overall Aβ clearance and exacerbating Aβ accumulation. However, variable results have been reported regarding the changes in specific BM components enriched or depleted using transgenic AD model mice and/or post-mortem human tissues. For instance, Hawkes et al. (2013), Hawkes et al. (2012) reported increased collagen IV in the BM of mouse models of AD, while Bourasset et al. (2009) and Mehta et al. (2013) reported decreased collagen IV.
The second most common neurodegenerative disease, Parkinson's disease (PD), is also linked to BBB dysfunction (Li et al., 2014;Booth et al., 2017) and to thickening of brain capillary BM (Farkas et al., 2000;Bertrand et al., 2008). The number of string vessels (collapsed BM without endothelium) was found to be significantly increased in PD patient samples (Yang et al., 2015). In addition to AD and PD, BM changes coincided with BBB dysfunction or breakdown in amyotrophic lateral sclerosis (ALS) (Garbuzova-Davis et al., 2007;Coatti et al., 2017;Yoshikawa et al., 2022). However, similar to AD, it is unclear how specific BM components change as a result of ALS. For instance, while Wiksten et al. (2007) reported increased laminin in the BM of ALS patients, Liu et al. (2011) reported decreased laminin.
Unlike AD and PD which were associated with BM thickening, degradation or dissolution of the BM was found following stroke in mice, rats and baboons (Hamann et al., 1995;Fukuda et al., 2004;Hamann et al., 2004;Kwon et al., 2009;Katsu et al., 2010). Most stroke studies reported that collagen IV, laminin, perlecan and agrin were degraded in both animal models and human postmortem tissues (Hamann et al., 1995;Horstmann et al., 2003;Vosko et al., 2003;Fukuda et al., 2004;Solé et al., 2004;Gu et al., 2005;McColl et al., 2008;Rosell et al., 2008;Baumann et al., 2009;Lee et al., 2011). However, some studies reported increased abundance of collagen IV and laminins following ischemic stroke in rodents (Anik et al., 2011;Ji and Tsirka, 2012). In addition to AD, PD and stroke, MS was also found to be associated with changes in NVU BM structure and composition. Specifically, BM in MS lesions was irregular and discontinuous, and BM components including laminins and HSPGs were abnormally deposited into the ECM of the CNS white matter in MS patients (van Horssen et al., 2006(van Horssen et al., , 2005. Future studies may be able to address how BM composition changes during disease and uncover mechanisms by which ECM components affect neurodegenerative disease pathologies using disease models built with patient-derived induced pluripotent stem cells (iPSCs) or genetically modified hPSCs, in conjunction with animal studies and human postmortem samples.
3 Use of BM ECM components in modeling the BBB in vitro 3.1 Collagen IV

Primary culture models
In vitro studies often use collagen-coated surfaces to culture primary BMECs or purify for primary BMECs via selective adhesion. In 1986, one of the first established protocols for in vitro culturing of primary animal BMECs used a rat tail collagen coating which mainly consists of collagen I (Freshney, 1986;Abbott et al., 1992). In 1998, Tilling et al. (1998 found that Transwell inserts coated with collagen IV, laminin, fibronectin, or 1: 1 (mass ratio) mixtures of any two of these proteins all significantly Frontiers in Chemical Engineering frontiersin.org increased TEER across the barrier formed by primary porcine BMECs, compared to the rat tail collagen coating. Since then, many studies adopted collagen IV or collagen IV + fibronectin (C4/Fn) coatings to culture primary BMECs. For example, in 1999in , Igarashi et al. (1999 cultured primary porcine BMECs on collagen IV-coated plates to study how glial cell-derived neurotropic factor affects barrier function. Perrière et al. (2005) cultured primary rat BMECs on collagen IV-coated Petri dishes in 2005, andCalabria et al. (2006) cultured primary rat BMECs on C4/Fn-coated surfaces including Transwell inserts in 2006 to study how puromycin can be used to purify primary BMECs in in vitro cultures. Other studies followed a protocol published in 2013 (Navone et al., 2013) where primary human and mouse BMECs were purified by selective adhesion to collagen I-coated flasks and cultured on collagen I-coated surfaces, such as Transwell inserts (Chan et al., 2018;Fan et al., 2019;Zhong et al., 2020).

Stem cell-based BBB models
In the last decade, an increasing number of studies have used hPSC-derived cells to build human in vitro BBB models for their human origin, high scalability, and physiological barrier properties. It was first reported in 2012 that cells possessing some key functional characteristics of BMECs can be derived from hPSCs. These characteristics include expression of tight junction proteins, nutrient transporters and polarized efflux transporters, the ability to form a barrier with high, physiological TEER and low passive permeability comparable to that of primary animal BMECs, and responsiveness to astrocyte-and pericyte-derived cues (Lippmann et al., 2012;Canfield et al., 2019;Di Marco et al., 2020). These hPSCderived BMEC-like cells (hPSC-BMECs for short) were generated with the following steps ( Figure 2A): hPSCs were seeded and expanded in hPSC culture medium on plates coated with Matrigel, a complex ECM protein mixture (largely laminin, collagen IV and nidogen) derived from Engelbreth-Holm-Swarm mouse sarcomas (Hughes et al., 2010). The medium was first switched to unconditioned medium which drives simultaneous differentiation of neural cells and BMEC-like cells, simulating the microenvironment of the developing brain. The medium was then switched to one that selectively expands endothelial cells, and the mixed population was re-plated onto C4/Fn-coated surfaces to purify the endothelial cells.
Since the initial success of generating hPSC-BMECs, numerous modifications have been made to the differentiation protocol, but the vast majority of the published protocols use the same ECM coatings: Matrigel for the 1st phase of the differentiation, and C4/Fn for the 2nd phase ( Figure 2A) (Lippmann et al., 2014(Lippmann et al., , 2012Hollmann et al., 2017;Qian et al., 2017;Neal et al., 2019). The authors of the first hPSC-BMEC publication chose C4/Fn coating because it is "commonly used for primary BMEC culture" (Lippmann et al., 2012). However, the C4/Fn coating may not optimally support all BMEC phenotypes. For example, Nakakura et al. (2021) reported in 2021 that fibronectin can maintain functional fenestra in rat ECs from fenestrated capillaries (i.e., leaky capillaries, as opposed to highly impermeable, barrierpossessing capillaries in the brain), which suggests that the use of fibronectin may not be the best choice for BBB models. However, at the same time, recent in silico analyses of RNA sequencing results indicate that human brain pericytes express FN1 (the gene encoding fibronectin) at a higher level than mouse brain pericytes (Gastfriend et al., 2021a), thus future studies are needed to uncover speciesspecific effects of fibronectin on the BBB formation and/or maintenance.
C4 is not only used in ECM coatings to aid in purifying endothelial-like cells from a mixed cell population during the 1st  (Lippmann et al., 2012). (B-C) Schematics of experimental setups for studies on how laminin isoforms affect hPSC-BBB models. It was found that laminin isoforms 221 and 511 can replace or supplement Matrigel in hPSC-BMEC differentiation phase 1 (Aoki et al., 2020) and C4/Fn in phase 2(Motallebnejad and Azarin, 2020), respectively, for improved hPSC-BMEC characteristics such as higher TEER and lower permeability.

Frontiers in Chemical Engineering
frontiersin.org phase of hPSC-BMEC differentiation, but also supports the continued culture of differentiated hPSC-BMECs during the 2nd phase. It was reported that hPSC-BMECs can also be purified and cultured until the formation of confluent monolayers on collagen I-coated glass surfaces for simultaneous imaging analyses and permeability assays (Ruano-Salguero and Lee, 2018). The final cell products expressed Occludin, Claudin-5, Glut-1 and Mfsd2a. Permeabilities of sodium fluorescein (NaFL) and IgG measured through confocal microscopy of hPSC-BMECs cultured on collagen I-coated glass surfaces were comparable to previously reported values from hPSC-BMECs cultured on C4/Fn-coated Transwell inserts (Ruano-Salguero and Lee, 2018). It is possible that this system may be further improved in terms of physiological relevance if ECM components found in the brain vascular BM are used in conjunction with collagen I gels, for example, if the collagen I gel was coated with collagen IV and/or laminin.
In contrast to 2D studies where hPSC-BMECs are cultured on C4/ Fn-coated plastic surfaces or porous polymer membranes, recent studies aiming to construct 3D brain microvessels often use collagen I gels as the scaffolds, adjust and optimize gel stiffness, then coat the gel with physiologically relevant BM components. In pilot 2D experiments preparing for 3D microvessel construction, Katt et al. (2018) noticed that gels with a higher collagen I concentration, thus higher stiffness, led to better cell coverage after seeding hPSC-BMECs. Katt et al. (2018) also found that the collagen I gel fabricated with or without LN/nidogen before coating the gel with C4/Fn had no effects on cell coverage. Eventually, Katt et al. (2018) chose to construct 3D microvessels by seeding hPSC-BMECs into cylindrical collagen I gels coated with C4/Fn ( Figures 3A, B). The resulting microvessels expressed ZO-1 and exhibited much lower permeability to Lucifer yellow (LY) than in previous studies using microvessels formed by human umbilical vein endothelial cells (HUVECs). In 2019, two additional studies used the cross-linker genipin to adjust the stiffness of a collagen I gel used to generate the cylindrical structure. Grifno et al. (2019) coated the 3D structure with Matrigel before seeding hPSC-BMECs, and the resulting microvessels expressed Claudin-5 and exhibited low permeability to LY.  coated the collagen I gel with C4/Fn before seeding hPSC-BMECs. The resulting hPSC-BMEC vessels expressed BBB markers such as Claudin-5 and Glut-1, and exhibited low permeability to LY, Rhodamine 123 (R123), and 10 kD dextran compared to vessels formed with HUVECs. These two studies reported consistent results of low permeability across hPSC-BMECs cultured in 3D vessels built with collagen I gels coated with Matrigel or C4/Fn ( Figures 3A, B).
Recent studies have also demonstrated effectiveness of C4/Fn coatings on other substrates and scaffolds in hPSC-BMEC models. For instance, in a study published in 2022, electrospun fiber mats from gelatin and/or poly-ε-caprolactone (PCL) solutions were synthesized, and then the fiber mats were coated with C4/Fn before seeding hPSC-BMECs. This study showed that NaFL permeability was significantly lower for hPSC-BMECs cultured on PCL fiber mats coated with C4/Fn compared to those cultured on Matrigel-coated Transwell inserts (Rohde et al., 2022). Another study published in 2022 used the capillary lane of a MIMETAS OrganoPlate ® 3-lane (a device with three divided lanes, namely, capillary, middle (hydrogel), and brain lanes) coated with C4/Fn ( Figures 3C, D). BMECs cultured in this system expressed BBB markers such as Claudin-5, ZO-1, Glut-1 and Breast Cancer Resistance Protein (BCRP) and exhibited low permeability to LY. Notably, the authors found that ABCG2 (gene encoding BCRP) expression in BMECs was 3.3-fold higher in their 3D system compared to 2D monolayer culture measured by qRT-PCR, and that transport of BCRP substrates in their 3D system was comparable to that in previous in vivo rat studies (Kurosawa et al., 2022). It is possible that these systems could be further improved and lead to better in vitro BBB models if coatings alternative to C4/Fn were tested and optimized.
Collagen gels have also been used to embed cells for co-culture experiments to model cellular interactions in the NVU. For example, in 2019, hPSC-derived mesodermal pericytes were embedded in Frontiers in Chemical Engineering frontiersin.org 06 collagen I gels placed onto Transwell inserts, then hPSC-BMECs were cultured on the gel. There were no significant differences in barrier tightness measured by TEER with and without embedded pericytes. However, the authors found that the presence of hPSC-BMECs caused the pericytes to migrate further away from the gel surface compared to pericytes in gels without BMEC coverage, demonstrating that hPSC-pericytes respond to cues derived from hPSC-BMECs (Jamieson et al., 2019). While brain pericytes are indeed embedded in vascular BM in vivo [ Figure 1], the use of brain vascular BM components (e.g., collagen IV instead of collagen I) to embed primary human pericytes or hPSC-derived brain pericytelike cells  may provide an in vitro platform that better models BMEC-pericyte interactions.

Laminin
Isolating and purifying specific laminin isoforms from tissues has been difficult or impractical, and production protocols for recombinant laminins have a relatively short history (~10 years) (Miyazaki et al., 2012) compared to laminin-containing ECM mixtures such as Matrigel (30+ years) (Kleinman et al., 1986;Guzelian et al., 1988;Fridman et al., 1990;Taub et al., 1990). This results in considerably higher costs of purified laminin isoforms compared to Matrigel. As discussed in Section 3.1.2, most studies developing and optimizing hPSC-BMEC differentiation protocols focused on the composition of the culture media and parameters such as the seeding density of hPSCs, with little regard to the composition of the ECM coatings (Lu et al., 2021a;Yan et al., 2021). Due to limited access to purified laminins and lack of consideration of the ECM in hPSC-BMEC differentiation, effects of specific laminins in in vitro BBB models were not systematically studied until recently. In the past few years, researchers hypothesized that the use of specific laminins can influence performance of in vitro BBB models given the physiological roles of laminins in the NVU BM. Specifically, Aoki et al. (2020) compared coating 2D cell-culture surfaces with Matrigel, Fn, vitronectin, LN221, LN411, and LN511 during the 1st phase (days 0-8) of hPSC-BMEC differentiation ( Figure 2B). All groups were re-plated onto the standard C4/Fn-coated surfaces for the 2nd phase of differentiation. The authors found that hPSC-BMECs differentiated on LN211 and those on vitronectin had significantly higher TEER compared to those differentiated on Matrigel. hPSC-BMECs differentiated on LN221 also exhibited significantly lower permeabilities to FD4 and LY and significantly higher P-gp and BCRP activities, measured by accumulation of their substrates R123 and Hoechst, respectively. The authors speculated that LN211 coating might be optimal for the 1st phase of hPSC-BMEC differentiation because the laminin α2 subunit plays a role in BBB formation, as demonstrated by previous in vivo study conducted by Menezes et al. (2014).
In another study published in 2020, Motallebnejad et al. hypothesized that the use of laminin coating might be beneficial for later phases of hPSC-BMEC differentiation, since there exists evidence for a switch from fibronectin-mediated signaling during development to laminin-mediated signaling during maturation in mouse CNS (Milner, 2002). The authors compared LN511, LN411, and C4/Fn coatings following a shortened (1-2 h incubation instead of up to several days) 2nd phase of hPSC-BMEC differentiation ( Figure 2C), while keeping the preceding differentiation substrates unchanged from standard protocols (i.e., Matrigel for the 1st phase and C4/Fn for the selective purification). The authors found that LN511, but not LN411, improved expression and localization of Occludin, Claudin-5, ZO-1, and VEcadherin. The use of LN511 resulted in a more activated phenotype in hPSC-BMECs (e.g., significantly lower expression of ANGPT2, MMP1, MMP9, FN1 and LAMA5 measured by mRNA abundance, and more prominent migration in wound healing assay). The use of LN511 also enhanced responses of hPSC-BMECs to shear stress under dynamic flow conditions: Analysis of phase contrast images revealed a significant increase in cell elongation measured by aspect ratio of the cells; fluorescent staining for F-actin demonstrated formation of stress fibers which were not present in the static culture, and immunocytochemistry analysis showed increased expression of Claudin-5, ZO-1, and VE-cadherin compared to cells cultured in static conditions (Motallebnejad and Azarin, 2020). Few studies building 3D models of the BBB using hPSC-BMECs have incorporated laminins in the scaffolds. Katt et al. (2018) compared various ECM coatings on collagen I gels and found no significant differences in cell adhesion or coverage between coatings with one, two, or three components among collagen IV, Fn, and laminin.
NVU cell types other than BMECs differentiated on specific laminin isoforms were also found to differ in their BMEC barrierinducing capacities. For instance, in 2019, hPSC-astroglia differentiated on human LN521 were found to induce barrier properties in hPSC-BMECs (e.g., lower NaFL permeability and higher VE-cadherin expression level) more than those differentiated on murine laminins (Delsing et al., 2019), demonstrating that BM components can also affect the BMEC barrier indirectly through associated NVU cells.

Nidogen, perlecan, agrin, and other ECM proteins
There has been very little exploration of the roles of nidogen, perlecan, agrin, or other ECM proteins using in vitro BBB models. , Katt et al. (2018 found that the addition of agrin, but not perlecan, to the C4/Fn coating on collagen I gels significantly increased coverage of hPSC-BMECs. However, the addition of agrin and laminin to C4/Fn coating also significantly decreased TEER of the final hPSC-BMEC monolayers. This effect was speculated by the authors to be agrin/ laminin-mediated enhancement of focal adhesion formation to the detriment of barrier function. , Qian et al. (2017 found that Synthemax or recombinant human vitronectin coating can be used to replace Matrigel for hPSC culture and the 1st phase of hPSC-BMEC differentiation (Figure 2A), and that the resulting cells expressed key proteins such as Claudin-5, Occludin, Glut-1, and P-gp, similar to those differentiated on Matrigel. More systematic and quantitative comparisons are needed to examine the effects of incorporating nidogen/agrin/ perlecan and other ECM proteins into in vitro BBB models.

Decellularized ECM and the BBB
A complementary strategy to systematic comparison of single or combinations of ECM components is the use of decellularized Frontiers in Chemical Engineering frontiersin.org ECM, which is typically obtained using detergent-containing wash buffers to remove cells from confluent culture, leaving a layer of ECM deposited by those cells on the cell culture surfaces. In 2007, Hartmann et al. (2007) cultured primary porcine BMECs on decellularized ECMs derived from porcine brain pericytes, mouse astrocytes, porcine aorta ECs, and porcine BMECs. Both pericyte-ECM and astrocyte-ECM increased TEER, while aorta-ECM decreased TEER, compared to the BMEC-ECM control. In 2016, Zobel et al. (2016) cultured primary porcine BMECs on decellularized ECMs derived from porcine brain pericytes, spinal cord astrocytes, and porcine BMECs. BMECs cultured on pericytederived ECM exhibited significantly higher TEER than those cultured on ECMs derived from astrocytes or BMECs. Zobel et al. (2016) also generated layered decellularized ECMs by sequentially culturing and removing astrocytes followed by pericytes (AP-ECM), or pericytes followed by astrocytes (PA-ECM). Both AP-ECM and PA-ECM were found to induce TEER to a greater extent compared to double layers of BMEC-ECM. These studies suggest that decellularized ECMs derived from NVU cell types could enhance BMEC phenotypes in in vitro BBB models. Compared to using surfaces coated with specific ECM proteins or protein mixtures, decellularized ECM has the advantage of better recapitulation of the complex structure and composition of ECM in vivo, providing appropriate chemical and mechanical cues for cell function from specific cell types (i.e., cell types used for generating decellularized ECMs are cell types from which chemical and mechanical cues are derived). However, the use of decellularized ECM has some disadvantages as well, including being largely uncharacterized, thus not directly providing insights regarding molecular mechanisms, having larger batch-to-batch variability, and source material availability issues that complicate large-scale experimentation.

Perspectives and conclusion
The vascular BM is an indispensable part of the BBB. Major BM components in the healthy adult NVU, including collagen IV, laminin, TABLE 1 ECM coatings used in in vitro hPSC-BBB models. Summary of ECM coatings used in recent in vitro hPSC-BBB models, compared to the established protocol described in section 3.1.2 and illustrated in Figure 2A.

ECM coatings
Substrate Timing Applications/Results Source

Motallebnejad and
Azarin (2020) Laminin 521 Cell culture plates hPSC-astroglia differentiation hPSC-BMECs co-cultured with astroglia differentiated on human LN521 had higher TEER and lower permeability to NaFL than BMECs co-cultured with astroglia differentiated on murine laminin mixtures Delsing et al. (2019) Synthemax or vitronectin Cell culture plates hPSCs and hPSC-BMEC Phase 1 (Figure 2A; instead of Matrigel) Final cell products expressed Claudin-5, Occludin, Glut-1, and P-gp, similar to those differentiated on Matrigel Qian et al. (2017) Frontiers in Chemical Engineering frontiersin.org nidogen, perlecan, and agrin, were all found to play roles in BBB development and/or organism viability based on animal studies with targeted deletions. Initial in vitro BBB models were typically comprised of primary BMECs or hPSC-derived BMEC-like cells cultured on surfaces coated with Matrigel, collagen I, collagen IV, fibronectin, and mixtures of these ECM proteins, and the resulting models capture key BBB phenotypes including barrier formation and transporter activities. More recent models have cultured hPSC-BMECs on specific combinations and isoforms of BM components that are informed by neurovascular development, such as laminin 511 following collagen IV-fibronectin mixture. These combinations improved BBB phenotypes, including elevated barrier function and/or enhanced responses to shear stress in hPSC-BMECs, resulting in improved in vitro human BBB models (Table 1). In this review, we focused on how hPSC-BMEC models are affected by different ECM components since the hPSC-BMEC model is currently the only in vitro human BBB model with both a physiologically tight barrier (e.g., measured by TEER greater than one thousand Ω·cm 2 ) and organotypic transporter activities (Aazmi et al., 2022), but admittedly, current hPSC-BMEC models have significant limitations. For example, it was found that hPSC-BMECs exhibit epithelial transcriptional signatures and low expression of endothelial genes (Lippmann et al., 2020;Workman and Svendsen, 2020;Lu et al., 2021bLu et al., , 2021a. Introducing endothelial-specific transcription factors ETV2, ERG, and FLI1 to hPSC-BMECs induced expression of endothelial transcripts, but also diminished BBB phenotypes, including barrier properties (Lu et al., 2021b). Generating hPSC-BMECs that both possess endothelial identity and recapitulate BBB phenotypes is a critical, current roadblock to employing hPSC-derived models of the BBB, and identifying appropriate ECM substrates for differentiation and culture of these cells could play an important role in improving the fidelity of hPSC-BMECs. As a complementary approach to inducing endothelial gene expression in existing hPSC-BMEC models, efforts are underway to impart BBB phenotypes to naïve or "generic" endothelial cells (ECs) (e.g., Lian et al., 2014;Patsch et al., 2015) by expressing transcription factors (Roudnicky et al., 2020) or treating the cells with small molecules (Gastfriend et al., 2021b). These strategies can only induce a limited subset of BBB properties in generic ECs; for example, the resulting cells do not form physiologically tight barriers and lacked efflux transporter activity. In addition, the role of ECM in stimulating BBB specification in these ECs has not yet been studied and could be a fruitful path forward. While recent reviews of in vitro BBB models have thoroughly summarized how different models compare and can be used to study BBB/NVU function (Erickson et al., 2020;Linville and Searson, 2021;Yan et al., 2021;Aazmi et al., 2022), none has recognized the role of ECM in hPSC-derived BBB models, indicating a continued need for ECM exploration in the BBB modeling field.
Moving forward, a better fundamental understanding of the composition, organization and dynamics of the BBB BM during development, health, and disease is needed to guide in vitro BBB model development. Notably, while NVU-specific transcriptomic data provides some insight into cell type-specific synthesis of ECM components, quantitative proteomics specific to the NVU BM is largely lacking, making it difficult to screen and rationally design biomimetic ECM mixtures for BBB modeling. Thus, future studies that profile and analyze the brain vascular BM composition and dynamics would accelerate design of defined mixtures of ECM components that are more physiologically relevant and/or provide improved functional phenotypes of BMECs. Although better proteomic data may inform optimal ECM coatings for in vitro BBB models, such "bottom-up" methods (i.e., building mixtures with individual components) could still be difficult due to challenges in sourcing and scaffolding specific proteins and protein isoforms. Future studies should also utilize decellularized matrices from primary or hPSC-derived cells to elucidate how BM components produced by different cell types affect BBB properties. Proteomic profiling of the decellularized matrices followed by "top-down" methods (e.g., selective depletion of components via gene disruption or protein blocking) may also provide insight into the roles of various ECM molecules in regulating BBB function. In addition to the BM on the abluminal side, the endothelial glycocalyx (EG) on the luminal side of BMECs is also comprised of a mixture of ECM molecules (e.g., proteoglycan proteins and glycosaminoglycan chains) synthesized by BMECs and has been found to be important for regulating the BBB (Kabedev and Lobaskin, 2018;Jin et al., 2021;Zhao et al., 2021). Although it would be challenging or impractical to artificially deposit ECM molecules on BMECs to simulate the EG in in vitro BBB models, future studies should assess the ECM secreted by BMECs on the luminal side to benchmark how well the EG is recapitulated in in vitro models, and to make appropriate improvements.
Both the BBB and BM change during development and under disease conditions, but we have insufficient understanding about what changes occur and the mechanisms by which these changes regulate developmental or disease progression. As discussed in previous sections, a few studies have explored how different laminin isoforms used at certain time points of the experiments affect the in vitro human BBB models, but none has investigated effects of dynamic changes in laminin and/or other ECM coating composition that may be physiologically relevant for BMEC cultures. An hPSC-derived BBB model incorporating BM dynamics would be a powerful tool to assess the roles of ECM in human BBB development and neurodegenerative diseases and to develop new ECM-based strategies to restore BBB function during disease.

Publisher's note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.