In vitro modeling of the neurovascular unit: advances in the field

Characterizing various forms of Mendelian dementia such as familial AD (FAD) mutations in the amyloid precursor protein (APP) or presenilin 1 (PSEN1) genes significantly improved our understanding of AD pathogenesis. Studies have also shown RAGE, and LRP balance plays a significant role in amyloid-beta transport into and out of the brain through the BBB [59]. A reduction in brain LRP levels may also play a role in amyloid-beta peptide accumulation [60]. Genetically engineered mice expressing the mutant genes for both APP and PSEN1 yielded essential insights into the mechanisms and consequences of amyloid deposition in the intact brain. However, to this day, the AD animal models have been unable to predict success in the clinic. For example, the Tg2576 model strain, which is based on a single FAD transgene, confirmed the relationship between amyloid pathology and impaired performance on cognitive tests in a preclinical experiment. Tg2576 mice have been improved or cured more than 300 times using different molecules. Yet none of these remarkable preclinical findings seem to transition to the clinical phase. It is now being argued that these disease models only helped address the early/asymptomatic stage of the disease [55].

Huntington’s disease (HD) is an autosomal dominant inherited invariably fatal disorder. It results in mutated Huntington proteins that aggregate in different neurons. Over the course of the disease, there is significant neuronal cell loss, and patients show typical phenotypic hallmarks such as cognitive deficits, personality disorder, and hyperkinetic movements [61]. Studies have found that the BBB is disrupted in an animal model of HD (R6/2). Corresponding morphological changes were also observed in post-mortem tissues from human patients [62, 63]. Similar findings, however, are yet to be seen in live human subjects. To date, over 25 transgenic rodent models of HD have been used, yet none are able to effectively reproduce the neurodegeneration features and disease progression patterns that has been clinically observed [57, 58].

Parkinson’s disease (PD) is a movement, and cognitive disorder with many pathologic hallmarks, including the formation of proteinaceous inclusions inside neurons, called Lewy Bodies and loss of dopaminergic neurons in the Substantia Nigra pars compacta [64]. PD has been found to be associated with multiple mutations. However, no single genetic anomaly was proven to be causing PD. The disease is thought to be linked to a range of polygenetic and environmental factors. 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) is a lipophilic compound and crosses the BBB easily and was found to induce symptoms and pathology of PD in animal models [65]. However, the injury-induced models mimic nigrostriatal dopamine deficiency but do not recapitulate the slow, progressive, and degenerative nature of the disease in humans. Whereas in clinical trials, the interventions are usually administered over a prolonged period of time, putative neuroprotective agents were being delivered at similar doses and schedules as an acute Parkinson’s disease-like lesion, which was induced in the typical underlying animal studies [66].

Stroke incidence increases with age, and patients commonly have other comorbidities that might increase their stroke risk, which complicates the clinical progression and affect the functional outcome. Up to 75% of acute stroke patients have hypertension, and 68% have hyperglycemia as comorbidities [67]. On the contrary, only 10% of focal ischemia studies used animals having hypertension, and less than 1% used animals had induced diabetes. The preclinical studies also are mostly performed in young animals. This latter is because achieving appropriate infarct volumes in older animals as well as animals that have additional disease conditions using conventional techniques (such as middle cerebral artery occlusion) is quite tricky and prone to high variabilities. The animals used in different stroke models are usually young and mostly males. More than 95% of the studies were done on rodents. Larger animals that would be biologically closer to humans are rarely used [68]. Most preclinical studies also fail to acknowledge the delay between the identification of symptoms of stroke and the start of treatment, which is typically a couple of hours for most stroke patients. Also concerning the post-stroke assessment of BBB integrity as well as in other conditions where the integrity of the barrier might be compromised, gadolinium-based studies of microvascular integrity performed have shown limited sensitivity to detect low levels of extravasations across the BBB [69]. The facts discussed above highlight the urgent need to utilize technology that would better represent the disease conditions in humans.

To date, various in vitro BBB models have been developed and characterized in terms of barrier tightness, expression of BBB specific proteins, and usefulness for physiological and pharmacological studies [3, 81]. Traditionally the transwell systems utilizing immortalized endothelial cells were not producing tight barriers and had low TEER values. Higher molecular weight markers such as FITC dextran were used regularly for such systems since the lower molecular weight markers would pass through relatively quickly. As discussed earlier, animal models do not always recapitulate the human BBB physiology or a disease condition (including pathological characteristics such as onset, progression, etc.). Unavoidable interspecies differences are likely to play a major role. For example, works done by Terasaki et al., using QTAP technology clearly demonstrated that the mouse BBB is different from the human BBB [82]. Furthermore, intraspecies variabilities can equally impact data reproducibility thus affective the translational relevance of the results [83]. Tentative in vitro models based on immortalized human brain endothelial cells have also been proposed in static or dynamic conditions [40, 84]: these models recapitulate at least some of the expected characteristics of the BBB, like high level expression of BBB-expressed receptors and transporters. Therefore, “humanizing” the in vitro BBB platforms using human-derived cells, may circumvent the translational limitations of current models and provide a complementary tool to support existing in vivo approaches. However, one of the major limiting factors to reproduce a “humanized” in vitro model to study neurological disease or disease state includes the availability of human brain tissues for cell isolation. This is quite difficult to procure and mostly originate from either post-mortem specimen (including fetal tissue) or patient-derived surgical resections. This latter, although may not represent the actual conditions in the native state.

Although most in vitro BBB models currently available are based on primary cultures of cerebral endothelial cells, endothelial progenitor cells (EPCs) might be other alternatives that present a specific phenotype of EC [85]. In these systems, cord blood-derived hematopoietic stem cells are used to generate a reproducible and stable human BBB model where the cells are initially differentiated into endothelial cells (ECs). Following the differentiation process, these ECs are then prompt to develop BBB properties by co-culture with pericytes [86].

These models were found to form much tighter barriers that can mimic human BBB integrity. However, there are still challenges ahead for developing the in vitro model of BBB due to different designs of models and quantitative protocols. Key features of recent microfluidic and 3D printed models of BBB has been summarized in Table 1.

Although cell culture-based in vitro models are useful tools to study the regulatory mechanisms modulating the physiology and function of the BBB as well as assess the passage and transport mechanisms of putative brain-penetrating drugs, reproducing the BBB properties in its entirety remains a significant and still unresolved challenge [87]. Different approaches have been used to mimic the BBB in vitro, and this includes static and dynamic (flow-capable) platforms, as well as the use of different cell types such as primary cells, immortalized cell lines, and, more recently, stem cells. In addition to using different cell types, cell cultures for BBB modeling have grown in structural complexity ranging from basic monocultures to multiple culture systems such as co-culture and tri-culture settings [17, 19]. The transwell system is one of the most commonly used tools as a BBB in vitro model. Although being very user-friendly and relatively easy to setup (Transwells also offer moderate scalability and high throughput screening—HTS—capabilities [40, 41]), there are substantial limitations inherent to the use of these platforms that need to consider. These include the two-dimensional structure, absence of enabling endothelial exposure to shear stress, and “edge effects” where areas of the transwell walls surrounding the membrane are intrinsically very permeable [41, 44, 45] (see Fig. 2a).

Further technological advances were then introduced to enable exposure of the BBB endothelium to physiological shear stress (SS), whereas SS modulates endothelial morphology [88], but also their function and physiological responses paving the way toward the development of the so-called “Dynamic Models.” Artificial hollow fiber constructs made of thermoplastic polymers such as polysulfone, polypropylene, etc., were initially used for the construction of brain microvessels and other CNS vascular beds [89] where EC-glia co-culture could be arranged to mimic the spatial and topographical distribution of these cells in situ resembling the anatomy of brain microvessels [90]. Under these controlled hemodynamic conditions combined with exposure to glial cells, ECs acquire more stringent BBB properties than those observed static platforms, including high TEER [40], cell polarization, and expression of specialized transporters [90] and efflux systems [40, 91] (see Fig. 2b).

However, the enthusiasm for the add on advantages brought by this dynamic in vitro-BBB (DIV-BBB) platform is hampered by additional drawbacks inherent to its design and construction. Since the model relies on the use of capillary-like tubes structures surrounded by a larger enclosure, no practical way exists to monitor the cells cultured on or within the artificial microvessels. The relatively large diameter of the capillaries compared to proper brain microvessels was more representative of larger vascular beds like distal pre- and post-capillary segments. The DIV-BBB platform required a relatively large volume of reagents and high quantities of cells (on the magnitude of > 10⁶) for culture initiation, thus affecting the cost, and the model per se does not have high throughput screening (HTS) capabilities. Furthermore, the system setup was quite complex, requiring a significant amount of time, resources, and technical skills than conventional platforms (e.g., Transwells) [91, 92]. Moreover, the most dynamic and realistic in vitro BBB model, microfluidic devices, requires specific equipment and technical skills mostly confined to the lab environment that developed the platform. These constraining factors, unfortunately, limit the adoption and further development of these models for basic and translational research [87] (see also Fig. 2c).

In-silico models have also been developed to estimating structure–activity relationships for the BBB permeation of drug compounds based on their physicochemical properties [93‐95]. Although these models are inexpensive, less time consuming, and high throughput screening methods for novel compounds in the drug discovery process, results obtained using computer simulation must be verified by in vivo experiments [96].

An organoid is an in vitro organotypic preparation consisting of various cells grown together under appropriate conditions to generate a miniature artificial version of an organ of interest, including the brain [97]. BBB organoids consist of human primary brain endothelial cells, astrocytes, and pericytes [98] assembling under low-adherence conditions into a multicellular structure resembling the blood–brain barrier [99]. One of the important characteristics of this model is that within the organoid, each of the cell types is directly in contact with one another, which plays a pivotal role in maintaining BBB integrity as well as function [100]. It has also been demonstrated from various studies that organoid shows better characteristics of BBB, which includes enhanced tight junctions, adherens junctions, and efflux pump expression compared to more traditional static culture systems. Additionally, lack of paracellular permeability, high drug efflux, and receptor-mediated transcytosis infer a realistic barrier function to the organoids [98]. Furthermore, using an organoid in vitro BBB model consisting of ECs, astrocytes, and pericytes, a recent study reported high expression of TJs proteins, VEGF dependent permeability, receptor-mediated transcytosis of angiopep-2, as well as activity of efflux pumps.

On the other hand, the transwell model exhibited a lower level of BBB regulatory proteins, and the differentiation between the transportation of angiopep-2 and a control peptide was not possible. In this experiment, two different detection approaches, namely confocal fluorescence microscopy and MALDI mass spectrometry imaging, were used as a screening tool [98]. More recently, Pham et al. have developed a human vascularized cerebral organoid model utilizing a patient’s iPSCs to study the BBB under normal and pathological conditions in health and disease [101].

To sum up, organoid BBB models seem to offer several advantages over other conventional in vitro BBB platforms, including but not limited to HTS due to ease of culture, simplicity, low requirement of reagents, and miniature size [87]. The throughput of this model can be further improved through integration with automated microscopy and robotics-assisted mass spectrometry technologies. Cost-effectiveness and reproducibility make this model more acceptable and attractive to implement in research as well. Considering the advantages offered by BBB organoids-based in vitro models, it is evident that the technology provides an efficient method for studying drug transport through the BBB and practical support for the development of brain-targeting drugs for the treatment of CNS diseases [98].

Nevertheless, organoids have some limitations, as well. The issues of inter-samples variability and high processing time are not negligible. One of the critical drawbacks of organoids is the absence of essential types of cells, including glia, microglia, oligodendrocytes, vasculature, etc. Moreover, the timescale of human development, lack of microglia, regional inputs, and myelination may hinder neurons maturation, thus limiting its utilities for specific disease models [87, 102]. Furthermore, methods of barrier function analyses from two dimensional cultures will have to be adjusted to the analysis of 3D organoid structures. Few attempts have been made to develop this methodology to assess the permeability of a compound in a 3D organoid and so far, these methods have been primarily developed for measurements of the barrier permeability in intestinal organoids [103].

Focusing on the drawbacks of conventional organoid in vitro BBB model, recently, the development of a 3D spheroid model of BBB has been reported. This model more closely mimics the human brain tissue since it is comprised of six cell types found within the brain cortex. These cell types include human brain microvascular endothelial cells (HBMEC), human pericytes (HBVP), human astrocytes (HA), human microglia (HM), human oligodendrocytes (HO) and human neurons (HN), with endothelial cells enclosing the brain parenchymal cells.

Induced pluripotent stem cells (iPSC) were used to derive the cells which could help narrow the gap in achieving an ideal in vitro BBB model for clinical applications in studies aimed to understand neurological diseases. Interaction between BBB and adjacent brain cells provides a platform to evaluate the ability of a novel drug to cross the BBB and its effect on microglia, oligodendrocytes as well as neurons, which is crucial for studying neurodegenerative conditions including amyotrophic lateral sclerosis, multiple sclerosis, stroke, Alzheimer’s disease. High cell viability was found to be maintained up to 21 days in the spheroid model containing six cell types, which is useful in evaluating long term effects of drug toxicity [104]. Expression of P-gp and GLUT-1 proteins were also identified as these proteins have a pivotal role in expelling unwanted chemical from the brain tissue and transport glucose into the brain tissue respectively and abnormalities in these proteins lead to different diseases [104‐107]. Expression of tight junctions, adherens junctions, and proteins associated with adherens junction was also identified to avert the free paracellular diffusion of substances into the brain parenchyma. Additionally, it was also found that tight and adherens junction protein localization was disrupted by hypoxia, thus supporting the usefulness of this model to study ischemia. BBB selectivity was also evaluated by measuring the effect of mercury ions in brain parenchyma by assessing cell viability [104].

Considering the characteristics and features of this model, we can state that, human cortex spheroid in vitro BBB model may provide some additional advantages over conventional organoids and can be a suitable platform for studies related to drug discovery, disease modeling, and neurotoxicity. However, further structural consideration to identify the production and proper deposition of extracellular matrix proteins of BBB and analysis of effects on individual cell types to evaluate cell-specific function is needed, which will make this model more acceptable to study different neurological diseases [104]. A more comprehensive description of CNS-related organoid models, including similar techniques, protocols, use, and limitations, has been recently published by Pacitti et al. elsewhere [101].

One of the significant hurdles in the field of in vitro modeling that current technologies have been trying to overcome is that of providing a quasi-physiological microenvironment to promote the BBB development of realistic physiological properties and responses to endogenous as well as exogenous stimuli [108, 109]. 3D in vitro tissue models are currently available for a variety of organs and tissues (including muscle, bone, liver, and cardiac tissues) [110]. The use of this technology is now making a modest appearance in cerebrovascular and BBB research, as well.

In the specific case, the brain microcapillaries are grown on self-polymerizing extracellular matrix protein (ECM) scaffolds where the BBB cellular components can develop close interactions with one another while being exposed to trophic factors along quasi-physiological biochemical gradients (see Fig. 3). High-resolution confocal microscopy and/or other 3D imaging techniques such as multiphoton microscopy and optical coherence tomography can be used to monitor the dynamic changes of cells cultured in the 3D ECM microenvironments.

However, the adoption of this technology in BBB research is still limited due to the complexity of developing an in vivo-like matrix architecture, where the omission of even minor ECM constituents can potentially alter the matrix property and thus architectural assembly processes of the microvasculature. These platforms are currently confined to basic research with relatively low translational/pharmaceutical appeal due to several factors including high complexity, lack of HTS capabilities, consistent reproducibility, etc.

In recent years, in the field of biomedical research, microfluidics has emerged as a promising alternative due to its high throughput, automation capabilities, and low fabrication and operation costs [111]. However, current microfluidic devices have relied on multi-step lithographic processes, which are time-consuming and complicated. In order to solve this critical issue, currently, 3D-printing (additive manufacturing) is becoming an alternative approach to microfluidic fabrication with complex architectures, avoiding multi-step processing with a wide range of materials [112‐114]. In fact, 3D printing, as digital fabrication technology, is a process of adding elements to fabricate objects from 3D model data, layer by layer, enabling precise construction of complex objects directly from a computer-aided design (CAD) software [115]. Indeed, 3D printed microfluidic technology provides researchers with several advantages over traditional fabrication techniques including the ability to build channels with unprecedented shape and complexity that are uniform and reproducible at minimal operating cost and time (reduced from weeks to a few hours), product complexity, reduction of user error, precisely controlled size, interconnectivity and geometry, flexibility and throughput [112, 116‐118].

A similar procedure is used in most 3D printing processes for manufacturing solid structures from digital designs. Briefly, the intended product is digitally rendered in 3D with computer-aided design (CAD) software, and then 3D designs are converted to the stereolithography file format (STL), describing the external surface of a 3D model [116]. The data is then further sliced into a build file of 2D layers and sent to the 3D printing machine [115]. Raw materials such as thermoplastic polymers (including acrylonitrile butadiene styrene—ABS, polylactic acid—PLA, polyamide—PA, and polycarbonate PC), natural polymers, and biocompatible synthetic polymers [114, 116, 119‐123] are processed into filaments or granules. Then binder solutions are added and solidified automatically in a layer-by-layer manner to produce the desired product. After printing, products may require polishing, drying, sintering, or other post-processing steps. Unprinted materials will also be harvested and recycled for continued use in the printing process [124]. Various printing techniques have been utilized for microfluidic applications. The leading 3D set up processes for microfluidic systems are 3D Printed Transfer Molding (PTM), Fused Deposition Modeling (FDM), Stereolithography (SLA), Direct Ink Writing, and Selective Laser Sintering (SLS) [124‐126].

Recently, several new techniques have been developed for 3D printing, including Poly jet, digital light processing (DLP), liquid deposition modeling (LDM), and fiber encapsulation additive manufacturing (FEAM) [127‐130]. Although these methods have more material selections or less processing time, only a few studies have been conducted to assess the viability of these techniques to manufacture microfluidic systems, due to their high cost and complexity compared to traditional 3D printing methods. A more comprehensive description of 3D printing techniques, protocols, use, and limitations has also been recently published by Sivandzade et al. elsewhere [22].

Despite the potential advantages provided by 3D printing fabrication processes in BBB modeling, the technology is not yet mature, with several limitations that still hinder its widespread adoption. For example, the lack of high-throughput 3D-bioprinted tissue models for research makes this technology not yet suitable for drug discovery and toxicology studies. The complexity of the tissue to be reproduced increases exponentially the complexity of the technical challenges that need to be overcome. These include conjugating multiple elements such as fabrication materials, cell types, cell distribution as well as loading of the necessary biological factors to maintain cell viability, and construction of the tissue scaffold itself. However, to advance this up and coming technology, any further will require the integration of multiple fields of research, including engineering, biomaterials science, cell biology, physics, and medicine.

An additional variable to consider (and a potential challenge) for the development of viable and clinically relevant in vitro model is the type of cells used in the setup. More specifically, whether these are primary cells, cell lines, or induced pluripotent stem cells (which are then differentiated into the desired phenotype), each type of cell has advantages and disadvantages, which are analyzed below.

Animal and human-derived cell lines have been developed as biological surrogates for BBB modeling. One attractive feature of these cells over primary cultures is their relative affordability and (to some extent) their capability of retaining their differentiating properties over multiple passages. Only a few immortalized human endothelial cell lines (HCMEC/D3 [84], HMEC-1 [131], TY08 [132], hBMEC [133], and BB19 [134]) have been developed and reported in BBB modeling. Highly purified populations of cultured human brain (human brain microvascular endothelial cells—HBMEC), unfortunately, are quite expensive if acquired from commercial sources. Unfortunately, the high level of technical skill necessary to isolate these cells as well as long term viability after few passages makes their use limited to few laboratories in the field. Isolation of these cells from the native brain tissue also requires advanced technical skills (isolation and purification processes are labor-intensive), time, and a viable source. Human specimens provide features specific to a variety of neurological etiologies that otherwise would be near impossible to recapitulate in cell lines or animal brain-derived (rodent, porcine, or bovine) primary cultures. However, primary cells may provide an attractive alternative in a sophisticated set-up. Regarding Spheroid, hBMEC, pericytes, and astrocytes spontaneously form into a multicellular spheroid in co-culture under low-adherence conditions and self-assemble into a modular organization that resembles the BBB [98]. Needless to say, spheroid provides each cell type to interact with one another, which has been reported to play a pivotal role in the maintenance of BBB integrity and function [99]. In the microfluidic platform, hBMEC might be superior due to its shallow cell requirement.

Even though HBMECs are not yet a realistic (cost-effective) alternative to the use of cell lines for industrial (pharmaceutical) screening (such as testing permeability, toxicity, etc.) and testing of novel drugs, they have unmatched value for basic and translational research. Next to these primary cells, recent advancement in the field of BBB modeling has brought forward the use of human iPSCs [135, 136]. In recent years, with improvements in understanding of differentiation pathways, induced Pluripotent stem cells (iPSCs) are now regularly differentiated to different cell types for co-culture models in microfluidic devices as well as 3D printed models [137]. 3D structure, cell–cell interaction, and the exposure of shear stress result in better barrier function compared to conventional transwell models. iPSCs-based BBB models are the first human BBB models with in vivo like paracellular barrier properties [138]. Thus, these cells may hold enormous potential for the development of preclinical disease models as well as species-dependent differences [139]. The unique advantage of using iPSCs is their ability to generate a representative model of patients from which they have been originated [140]. Additionally, recent advances in gene editing techniques provide exciting opportunities in disease modeling using iPSCs, although these methods still have hindrances like epigenetic reprogramming and loss of patient-specific epigenetic signature [141]. iPSCs technology has some major limitations so that the risk of tumor formation is present due to the use of viral infections and low efficiency of reprogramming during the production of these cells. [142]. On the other hand, an existing limitation of the model is the narrow experimental window provided by iPSC-derived cells, which generally tend to de-differentiate quite rapidly under in vitro culture conditions. Moreover, the cell differential procedure depends upon various random and permanent insertion of transcription factors. Overall, iPSCs- could be applied in the construction of 3D models such as spheroids or organoids in order to support the development of a brain vasculature within these models [139]. Other future 3D models based on iPSCs might use cultivation in or on plastic scaffolds or hydrogels with defined 3D structures. Further development in the field and specifically in the cell differentiation processes and culture stabilization will undoubtedly further the availability and use of these cells in this and other fields of research.