Advances in microfluidic in vitro systems for neurological disease modeling

Neurological disorders are the leading cause of disability and the second largest cause of death worldwide. Despite significant research efforts, neurology remains one of the most failure‐prone areas of drug development. The complexity of the human brain, boundaries to examining the brain directly in vivo, and the significant evolutionary gap between animal models and humans, all serve to hamper translational success. Recent advances in microfluidic in vitro models have provided new opportunities to study human cells with enhanced physiological relevance. The ability to precisely micro‐engineer cell‐scale architecture, tailoring form and function, has allowed for detailed dissection of cell biology using microphysiological systems (MPS) of varying complexities from single cell systems to “Organ‐on‐chip” models. Simplified neuronal networks have allowed for unique insights into neuronal transport and neurogenesis, while more complex 3D heterotypic cellular models such as neurovascular unit mimetics and “Organ‐on‐chip” systems have enabled new understanding of metabolic coupling and blood–brain barrier transport. These systems are now being developed beyond MPS toward disease specific micro‐pathophysiological systems, moving from “Organ‐on‐chip” to “Disease‐on‐chip.” This review gives an outline of current state of the art in microfluidic technologies for neurological disease research, discussing the challenges and limitations while highlighting the benefits and potential of integrating technologies. We provide examples of where such toolsets have enabled novel insights and how these technologies may empower future investigation into neurological diseases.


| The importance of studying human cells for human disease
Animal models have provided significant insights into biological systems and are currently indispensable in the drug discovery pipeline, however, numerous important cellular and molecular differences between species should not be overlooked. In the context of the brain, there are significant species-specific differences in efflux transporters, tight junctions, and cell-cell signaling observed in brain endothelial cells (BECs, Warren et al., 2009), extensive interspecies variation between human and rodent astrocytes (including functional and morphological complexity, glutamate responses, and immune responsivity) (Chandrasekaran et al., 2016), and numerous differences noted in electrical activity (Beaulieu-Laroche et al., 2018), gene expression and morphology (Hodge et al., 2019) between homologous human and rodent neurons. As such the inclusion of human-derived cells, tissues, and patient samples is highly recommended to improve predictive power in dementia research (Vargas-Caballero et al., 2016). The reasons for the high failure rate of candidate CNS drugs are myriad, including an incomplete understanding of the exact mechanisms underlying human disease hampering model development. However, the considerable differences in human and animal model's cellular and molecular neurobiology clearly necessitates new more physiologically relevant methods to study human cells, in order to more effectively explore fundamental mechanisms and improve drug discovery and translation.

| Limitations of current in vitro methods
In vitro systems provide the opportunity to interrogate human cell biology, however, many cell culture approaches rely on immortalized or cancerous cells grown in non-physiological environments on 2D substrates such as functionalized glass or plastics. While providing a near limitless supply of cells for high-throughput screening, transformed or immortalized cell lines often show substantial drift in their transcriptional and epigenetic profiles over repeated passaging which may confound pharmacogenomics studies (Horvath et al., 2016;Nestor et al., 2015). Furthermore, non-physiological culture conditions promote genetic adaptions, and phenotypic changes. Common basal media formulations, such as Dulbecco's modified eagle medium have been optimized for rapid cell growth, often at the expense of maintaining cell identity. High glucose concentrations and fetal serum found in these media preparations have, for example, been shown to promote dedifferentiation of primary cell types toward fetal-like phenotypes (Morris, 1962). Oxygen levels are also rarely controlled in standard cell culture setups, often leading to far greater oxygen tension than would be present in tissues, impacting upon the redox environment and cell metabolism (Tiede et al., 2011). Plastic and glass substrates furthermore provide high mechanical stiffness, in the gigapascal range, while most in vivo environments are on the millipascal to kilopascal scale. With a growing understanding of mechanobiology and appreciation of the influence of substrate stiffness and the extracellular matrix (ECM) on cell behavior appropriate cell culture substrates are increasingly being recognized as a major factor in shaping in vitro cell phenotypes (Buxboim et al., 2010;Engler et al., 2006;Rauti et al., 2020). Thus, standard macroscale cell culture often fails to recapitulate the distinct microenvironments that constitute the Significance Microfluidic technologies have enabled new opportunities across a range of scientific fields, particularly neuroscience, where the ability to organize cells to mimic brain structures in vitro has provided novel insights into neurobiology. Once restricted to specialist laboratories and cross-disciplinary collaborations, systems are now commercially available and becoming an important tool in neuroscientist's investigational arsenal. Microfluidic systems are already being developed into models that recapitulate aspects of neurological diseases to improve drug discovery. This review surmises the current state-of-the-art in in vitro neurological disease modeling and may serve as guide to those wishing to explore microfluidic toolsets to expand experimental possibilities.
normal and diseased phenotypes that occur in vivo. Advances in microfluidic cell culture and "organ-on-chip" technologies have in contrast, allowed for an exploration of precisely tuned in vitro cell culture environments with greater physiological relevance.
Modeling the complexity of the human CNS in vitro, is undoubtedly hugely challenging. With limited ability to experimentally manipulate and probe human brains, along with translational gaps in animal neurological disease models, advanced microfluidic cell culture systems can serve as experimental tools to provide unique insights into human cellular function in a quasi-physiological context.

| ADVANTAG E S OF MI CROFLU ID I C CELL CULTURE SYS TEMS OVER TR ADITIONAL CELL CULTURE
Microfluidic technologies provide the ability to precisely engineer micron to nano-scale architecture to shape the physical and chemical microenvironment, distilling brain physiology into manageable functional units, relevant to a specific research question (e.g., neuronal circuits, blood-brain barrier (BBB) and neurovascular unit (NVU) models). Such microphysiological systems (MPS) have already provided unique insights into mechanisms underpinning brain physiology and are now being utilized to explore pathophysiological mechanisms and novel interventions in neurological diseases.
Microfluidic systems offer several advantages over traditional macroscale cell culture (for in depth reviews see (Halldorsson et al., 2015) and (Tehranirokh et al., 2013)). The ability to shape physical and chemical microenvironments with a high level of precision has allowed for new possibilities to mimic physiological environments and to manipulate and integrate different cell types (Figure 1).

| Directing cells through microphysiological architecture
Microfluidic devices provide a physical architecture at the cellular and subcellular scale, which may be used to direct, pattern or confine cells for the precise investigation of cell to cell interactions, paracrine signaling, biomolecular analysis, automated image analysis and axonal guidance and trafficking.
Geometries tailored for precise positioning of cells can be used to study contact-based signaling such as stem cell differentiation, myelination, immune responses, and synaptic activity. This is exemplified by the work of Frimat et al. in developing a single cell culture or co-culture microfluidic device for studying juxtracrine signaling (Frimat et al., 2011)  This ability to create cell-scale architecture and precisely define physical barriers is especially powerful in enabling a more faithful representation of the spatial organization and structure of the CNS.
Taylor et al. engineered a microfluidic multicomponent device that aligned the growth of neurites from a somal compartment to a neurite compartment through the use of subcellular sized axon guidance channels, thus creating a tool to manipulate neuronal microenvironments in a highly precise manner (Taylor et al., 2003). Developments on this device format have since been used to direct in vitro neuronal connectivity and create simplified circuits to enable numerous biological insights as discussed later. While microfluidic device geometries can provide direct physical barriers to cells, they can also be used to direct flows, deposition of molecules, surface substrate patterning, and gel formation. For instance microfluidic devices can be used to create miniature stamps or templates for patterning of surface adhesion molecules or deposition of cells on glass coverslips for customized cell patterning (Chiu et al., 2000) (Figure 1c). Channel geometries can also pattern three-dimensional (3D) gels, taking advantage of surface tension to retain gels within set boundaries (Huang et al., 2009) (Sharma et al., 2019), or even to assemble and align fibrous structures of hydrogels using flows to dictate the nanogel structure and guide subsequent cell growth (Kim et al., 2017).

| Precise control of flow and chemical environment
By taking advantage of the unique physics of flow at the microscale, microfluidics can provide precise control of shear stress experienced by cells, along with chemical gradients, to mimic dynamic environments found in vivo. Microfluidic control of fluidic shear has enabled new insights into mechanotransduction, for example revealing physical forces as a pleiotropic modulator of the endothelial cell phenotype, influencing BEC, tight junction expression, and transcellular permeability (Brown et al., 2019). Since laminar flow prevails in most microfluidic flow regimes, mixing is largely dictated by molecular diffusion, providing the possibility for spatial and temporal chemical environments within the same cell culture platform for localized chemical stimulation even to the scale of a single cell (Horayama et al., 2016). Kim et al. demonstrate the power of this capability, creating a fully programmable and automated cell culture array to test the pairwise combination of drugs at the different concentrations on cell cultures grown in parallel (Kim et al., 2012) (Figure 1a), helping to address the challenge of scaling screening processes with combinatorial treatments where the availability of cells is limited, such as with primary cells. In the context of the CNS microfluidic cell culture devices have enabled production of chemical gradients to provide insights into microglial chemotaxis (Cho et al., 2013), endothelial migration, angiogenesis, and vasculogenesis (Barkefors et al., 2008;Kuzmic et al., 2019) and neurite outgrowth and guidance mechanisms (Romano et al., 2015). Microfluidic control of chemical milieus has also been extended to the gaseous microenvironment, such as oxygen levels and other bioactive gaseous compounds such as nitric oxide, carbon monoxide, and hydrogen sulphide (Wu et al., 2018). Spatiotemporal chemical control has also been achieved at sub-millisecond resolution and femtoliter volumes, allowing for the investigation of extremely fast reactions such as precise ligand binding dynamics (Chiang & West, 2013) and nano/femtoliter chemical release has allowed for investigations using "artificial chemical synapses" (Peterman et al., 2004).

| Integration of analysis
The use of MPS provides the additional benefit of enabling integration of monitoring systems, such as transepithelial/endothelial resistance (TEER), impedance and electrophysiological systems.
Such techniques often allow for real-time, long term, label-free, and relatively non-invasive monitoring of cell culture conditions and cellular processes.
Gene expression and protein translation changes may occur on the scales of seconds to minutes, with phenotypic or morphological changes occurring on the scale of minutes to hours. Current cell assays are rarely capable of encompassing all of these timescales, gathering continuous data that can report on cell function under different conditions. Organ-on-chips and MPS have the potential to push the boundaries of on-board sensing devices that continuously monitor cell metabolism, phenotype, and growth to allow collection of multiparametric data over hours and days if not weeks (Modena et al., 2018). F I G U R E 1 Advantages of microfluidic cell culture systems. (a). Control of fluid flows used to provide precision cell positioning (Frimat et al., 2011) (image reproduced from Frimat et al., 2011 LoC with permission from The Royal Society of Chemistry). (b). Patterning of protein disposition on surfaces using microfluidic stenciling and transfer printing to control patterning (Chiu et al., 2000)

TA B L E 1 (Continued)
While many microfluidic chips are optically clear allowing for fluorescence assays, which are increasingly becoming automated, growing attention has focused on electrical systems that monitor properties of cells in a non-invasive manner. Pioneering work by Giaever and Keese explored how electrical impedance spectroscopy could be used to measure cell adhesion, differentiation, and even micromotion (Giaever & Keese, 1984;Lo et al., 1995). Further developments in this vein have led to devices integrated with traditional cell culture formats (multi-well plates and Transwell devices) for measuring TEER (Wegener et al., 2004). Examples of microfluidic chips with integrated sensing are becoming more widespread. There are two main types of device-those which integrate electrodes above and below cells grown on a porous membrane (analogous to a Transwell system) and those which integrate electrodes directly into the surface on which the cells are grown (Benson et al., 2013). In the case of the BBB, the Transwell device is potentially more relevant, as both the TEER of the endothelial cells, along with the ability of molecules to be transported across the barrier are interesting complementary parameters (Griep et al., 2013;Park, Mustafaoglu, et al., 2019). Neuronal firing can also be usefully interrogated using multi-electrode arrays, which pick up on action potentials at given frequencies. Maoz et al. have even demonstrated integration of both multi-electrode array and transepithelial electrical resistance onto a single device for simultaneous measurements of cellular electrical activity and barrier function .
Although at present such arrays cannot access intercellular electrical activity (e.g., picked up via patch clamp), monitoring neuronal firing in mono-culture or even in organoids or spheroids can be extremely valuable for drug discovery. Given the numbers of neurons present in any given network, leveraging the power of microelectrode arrays (MEAs) within a microfluidic chip is hugely beneficial  ( Figure 1d,e) particularly when coupled with the ability to accurately pattern and position cells (Soscia et al., 2017). For a detailed review of sensor integration into MPS please see Kilic et al. (2018).

| MI CROFLU ID I C TOOL S TO S TUDY NEURONAL FUN C TION
In recent years, microfluidic technologies have increasingly been used in neuroscience research, to extend the experimental capabilities for studying axonal guidance and transport, synapse formation and function, and neuronal function across subcellular to network levels (Neto et al., 2016). Through precise control of microenvironments, MPS offer new opportunities to create in vitro physiologically and pathophysiologically relevant models.

| Directing neurons: Guiding and compartmentalization
In 1977 as axonal response to injury (Taylor et al., 2005), myelination (Park et al., 2009), synaptic formation and function (Shi et al., 2013), probing the direct and indirect response of neuronal cultures to chemical stimuli (Robertson et al., 2014), neurite growth ( Other methods to direct neuronal outgrowth include direct micromanipulation (Magdesian et al., 2017) as well as use of gradients of neurotrophic factors using spatio-temporal fluid control (Millet et al., 2010;Taylor et al., 2015) (Figure 2c). Micro-contact printing may also be used to pattern surface proteins promoting neuronal outgrowth in defined circuits (Offenhäusser et al., 2007) (Kim et al., 2017). However, with the greater physiological relevance provided by directing 3D neuronal circuits, this also provides an added layer of complexity in probing neuronal function and analysis workflows.

| Electrophysiological probing neuronal function
The above outlined methods for creating neuronal circuits in vitro are now increasingly being integrated with MEAs for electrophysiological interrogation, combining the advantageous spatial cell patterning capabilities and precise control of chemical cues provided by microfluidics with electrical stimulation and recording of neuronal networks. Throughout the years, several methodologies have been developed, from manual whole cell patch clamping in standard two-chamber microfluidic devices (Jokinen et al., 2013), to tubeless devices based on capillary forces for drug delivery (Resto et al., 2017), to microfluidic planar patch clamp systems (Xu et al., 2014) with automated drug delivery (Yuan et al., 2016). However, it is the integration of microfluidics with MEA substrates, possibly due to the planarity of both technologies, that has produced higher-throughput and in-

| Toward networks
Neural circuits are anatomical connections within the brain with far-reaching implications for many neural functions. As an example, the hippocampus, known to be involved in spatial navigation, consists of multiple sub-regions which are interconnected in a predefined, mostly unidirectional, way (e.g., mossy fibers) (Strange et al., 2014). Similarly, the midbrain is another example of a source of projection neurons (Caggiano et al., 2018;Tovote et al., 2016). Dopaminergic neurons in this area project to areas across the brain, including the striatum and multiple cortical areas.
The majority of studies carried out on such specific brain regions have been performed ex vivo using brain slices. However, this approach has obvious limitations due to undesired axotomy, uncertainty on cell topological connections and short experimentation times due to cell death. Therefore, in vitro models where different cell populations can be grown and connected "ad hoc" present simplified but valuable tools to investigate network function as well as neuropathological conditions. Microfluidic technologies have been employed to create human brain-mimicking neuronal circuits for the study of cortico-striatal networks using calcium imaging (Lassus et al., 2018), to model neuroprotective mechanisms (Samson et al., 2016), to study structure-function relationship by precise neurite guidance achieved by electric fields [77]), to highlight brain region-specific cell identities (Kamudzandu et al., 2019), physiology and function (Dauth et al., 2017), to create 3D structured circuits (

| HE TEROT YPI C CELL CULTURE SYS TEMS: TOWARD ORG AN -ON -CHIP MODEL S
Historically, the neuron has provided the central focus for studies of neurological diseases. Recent years have, however, seen a conceptual shift from this neuro-centric view to one that emphasizes the importance of multidirectional interactions between all brain cell types, including neurons, glia, vascular, and perivascular cells. This concept of the NVU, (formalized at the 2001 NINDS Stroke Progress Review Group meeting) has emerged as a new paradigm from which to investigate brain physiology and pathobiology and has revealed commonalities of pathogenic processes across many neurological diseases, such as glial activation and BBB compromise. As such, numerous efforts have been made to replicate the heterotypic cellular interactions and physio-chemical environment that underpin the BBB and NVU in vitro, to enable a greater insight into neurological diseases.

| Blood-brain barrier models
The BBB is a dynamic and highly selective conduit that maintains the unique chemical environment of the brain (Langen et al., 2019).
This barrier provides a considerable obstacle in delivering neuropharmaceuticals to the brain, while its disruption has been implicated in the pathophysiology of a number of diseases, including Alzheimer's, multiple sclerosis, epilepsy, stroke, and Parkinson's disease. Early attempts to replicate this structure in vitro involved the culture of BECs on Transwells™ under static conditions to allow the study of transcellular and paracellular permeability. Such systems have since allowed for a dissection of the components that make up the BBB providing an insight into the roles of astrocytes, pericytes, and the ECM in BBB maintenance. While the mainstay for such investigations has been traditional Transwell assays, microfluidic technologies are now emerging that provide a greater level of control over microscale architecture and physiochemical cues, such as flow, that regulate BBB function.
The significance of the in vivo microenvironment in maintaining BEC properties was highlighted by observations that proliferating rat cerebral capillary fragments gave rise to confluent monolayers ex vivo but their endothelial characteristics were lost over time in culture (Panula et al., 1978). Early blood-brain interphases were generated by co-culturing primary BEC and brain parenchymal such as the bi-layered Emulate chip (Table 1)  were able to study receptor-dependent transcytosis of nano-particle coupled drug delivery (Ahn et al., 2020).
Compartmentalized models with BEC monolayers are disadvantaged by artificial barrier interphases that limit direct neurovascular coupling. These systems often consist of cell monolayers that do not recapitulate 3D ECM and heterotypic cell interactions, also using rectangular channel cross sections that result in poor cell coverage and flow dead zones at the corners. Other microfluidic designs have focused on recapitulating the 3D NVU architecture to preserve cell associations by using cells embedded in gels. One approach to compartmentalize a 3D matrix has been through the use of surface tension to pin a gel at a change in channel geometry such as with phase guides or micropillars (Figure 3c), as incorporated in the Syn-BBB, B 3 C BBB models, or the commercially available MIMETAS OrganoPlate ® (Table 1). These allow for the controlled dimensions of the vascular channel to be maintained while allowing 3D culture and F I G U R E 4 Approaches to NVU and BBB models. (a) Macroscale approaches to developing BBB models typically consist of cell monolayers (left) and Transwell systems (center and right), the relatively low surface area to volume ratio (typically ~10 cm −1 ) can result in poor gas exchange and distance between heterotypic cells can result in a large dilution of paracrine signaling factors. Microfluidic systems typically have surface area to volume ratios of ~800 cm −1 and show rapid gas exchange, however, rapid accumulation of waste products and consumption of nutrients necessitates flow. Multiple approaches to dynamic microfluidic BBB models have been made. (b) Two layer systems, utilize a membrane to separate blood and brain compartments, such as in the Emulate chip, similar to a transwell system but incorporating flow and with more physiological cell-extracellular fluid ratios. (c). Devices using pillars or phase guides on one edge (as in the Mimetas OrganoPlate ® , the SyM-BBB (Prabhakarpandian et al., 2013) and SynVivo's SynBBB) pin a cell laden gel allow for a lumen with one region in direct contact with parenchymal cells in 3D. (d). Fully 3D defined lumens can be created in a gel using sacrificial (Golden & Tien, 2007) or removable moulds (Bouhrira et al., 2020) or by taking advantage of Saffman-Taylor instability such as (Herland et al., 2016). Homogenous WSS or TEER measurements are, however, difficult to achieve due to the unpredictable geometry of vessels formed.

| Neurovascular unit models
The NVU represents a functional unit of the brain. By extending BBB models to include neurons, microfluidic in vitro systems have been able to better model the NVU, demonstrating physiologically relevant BBB integrity, cell-cell interaction and paracellular signaling and metabolic coupling.

| Alzheimer's disease
The Using microfluidic devices, several studies have analyzed the different species that can propagate from cell to cell. Tau dimer, trimer, or oligomer aggregates as well as short fibrils, but not monomers or long fibrils, can be taken up by neurons and transported both anterogradely and retrogradely (Usenovic et al., 2015;Wu et al., 2013).
Tau oligomers, but not monomers, also induce an increase in aggregated and phosphorylated Tau, alongside neurite retraction, loss of synapses, aberrant calcium homeostasis, and imbalanced neurotransmitter release (Usenovic et al., 2015). Another study shows that phosphorylated high-molecular-weight Tau, although very low in abundance, is taken up, transported in axons, and transmitted through synapses to connected neurons (Takeda et al., 2015). Tau pathology is propagated via synaptic activity (Wu et al., 2016), possibly through exosomes , as well as non-synaptic mechanisms (Calafate et al., 2015). Studies using MPS have contrib-

| Stroke
Stroke, caused by a disruption of blood flow to the brain, is the largest cause of adult disability. However, despite significant research efforts, with over 1,000 neuroprotective therapies tested pre-clinically, no neuroprotective treatment has yet been approved (O'Collins et al., 2006). Despite significant differences between human and rodent metabolic rates, inflammatory responses and neurobiology, methods to test stroke therapeutics using human In future, such techniques could be used to delineate the cellular and molecular mechanisms that confer selective neuronal vulnerability of human hippocampal CA1 neurons observed in stroke (Bartsch et al., 2015) to potentially reveal novel therapeutic targets.
Many of these microfluidic systems rely on the gas permeability have not yet been applied to model stroke.
In addition to providing controlled hypoxia, microfluidic devices have also allowed for more detailed study of specific pathological processes that underlie the delayed and progressive neuronal damage which follows cerebral ischemia. Initial energy crisis results

| Parkinson's disease
Parkinson's disease (PD) is a neurodegenerative disorder affecting over 6 million people worldwide (Dorsey et al., 2018), characterized by loss of dopaminergic neurons in the substantia nigra involving the accumulation of intracellular protein inclusions comprised largely of alpha-synuclein (α-syn). Whist the exact mechanisms of PD remain to be fully understood, microglia activation, mitochondrial dysfunction, oxidative stress, and chronic neuroinflammation have all been demonstrated to play pathological roles (Dexter & Jenner, 2013).
Toxin-based models of PD replicate many of the known mechanisms of cell death, however, the progressive nature of the disease and dopaminergic selectivity is not mirrored. Drugs that have proved successful in combating toxin-induced cell death have yet to translate into neuroprotective therapies for PD. Genome-wide association studies have informed the design of genetic models of PD, however, there are as yet no animal models that fully recapitulate PD pathology (Dexter & Jenner, 2013). While a current lack of in-depth mechanistic understanding of the human disease pathobiology hinders in vitro modeling, microfluidic systems have been used to allow interrogation of specific pathological events that characterize the human disease.
While some studies have used MPS to explore neurotoxin models of PD, such as concentration gradients of 6-hydroxydopamine (Seidi et al., 2011), the main focus of microfluidic systems has been in building a more detailed understanding of α-syn inclusion formation and trafficking, through taking advantage of precise control over chemical environments and compartmen- show that increased α-syn oligomers disrupt axonal integrity and impair axonal mitochondrial transport (Prots et al., 2018). Other microfluidic systems have been used to investigate the interplay between different cell types under pathological conditions. Fernandes et al. developed a microfluidic platform with two cell culture chambers connected by three channels and equipped with integrated pneumatic valves for precise temporal control of cell treatment and diffusion or perfusion between compartments.
Using this system Fernandes et al. were not only able to observe the release and spread of GFP tagged α-syn between H4 neuroglioma cells, but were able to co-culture with N9 microglial cells demonstrating increased levels of reactive oxygen species in H4 cells cultured in the presence of activated N9 cells (Fernandes et al., 2016). While such systems predominantly use 2D cultures it is increasingly being recognized that 3D architectures are required to recapitulate in vivo like gene expression profiles (Baker & Chen, 2012) and to reproduce more physiological equilibration and transport of soluble factors (Ramanujan et al., 2002). cell culture OrganoPlates, that in the absence of stressors, 2D cell culture systems fail to exhibit robust endophenotypes (Bolognin et al., 2018). In 3D cultures of LRRK2-G2019S neurons, compared to isogenic wild-type lines, showed time-dependent dopaminergic degeneration, altered mitochondrial morphology, and enhanced cell death all in the absence of exogenously administered stressors (Bolognin et al., 2018). Such high-throughput 3D assays may thus allow for more effective screening of therapeutics for PD.

| Multiple sclerosis
Multiple sclerosis (MS) is a chronic inflammatory demyelinating disease affecting more than 2 million people worldwide (Wallin et al., 2019). Commonly used experimental systems to study MS  (Mehling et al., 2015). This system allowed for the division of T cell subsets in relation to migration speed, and subsequent analysis using microfluidic droplet digital-PCR to match cell behaviors to RNA profiles. While yet to be realized, combining primary T cell culture and chemotactic studies with BBB models or in vitro myelination models, such as those previously described, could be used in future study of MS related disease mechanisms and potential therapeutics.

| Brain cancer
Primary brain tumors refer to a heterogeneous group of tumors which originate within the CNS, ~75% of which are gliomas (Lapointe et al., 2018). While only accounting for ~3% of all new adult cancers, in children brain cancers are the most common form of solid tumors (Logun et al., 2018;Ostrom et al., 2014). Brain cancers lead to a high mortality rate and the unique microenvironment of the brain, particularly the restrictive nature of the BBB, make these cancers notoriously difficult to treat. MPS have already enabled some unique insights into the heterogeneous cell-cell and cell-environment interactions that influence tumorigenicity, such as investigations by Lei et al demonstrating that nerve bundles can provide biophysical support for directional cancer cell migration (Lei et al., 2016).
An important benefit of MPS in cancer research is the ability to integrate patient-derived biopsies and tumor cells to provide direct translational relevance. By integrating bio-printing with on-chip culture, Yi et al. not only mimic heterotypic cellular interactions in a relevant brain-derived ECM 3D microenvironment with a hypoxic tumor core but also were able to reproduce clinically observed patient-specific resistances to chemoradiation and temozolomide using patient-derived cells (Yi et al., 2019). Future work might additionally enable investigations of immune responses, potentially offering an advantage over current xenograft animal models, which require immune compromised animals.
The use of vascularized models has allowed the study of both anti-angiogenic and anti-cancer therapies (Sobrino et al., 2016) as well as in brain-specific models, for instance to assess strategies to enhance drug delivery through the BBB (Bonakdar et al., 2017) and to delineate pathways of brain metastasis. Indeed most malignant brain lesions are secondary brain tumors metastasized from other organs. With metastasis occurring in up to 30% of adults who have a malignant primary tumor at another site, with the highest incidence being in lung cancer (30%-50% of patients) (Weller et al., 2015). A number of studies have made use of BBB MPS to model brain metastasis, providing insights such as the role of astrocytes in restricting cancer cell extravasation (Xu et al., 2016). Such systems might in future take advantage of establish panels of brain metastatic cell lines to investigate brain tropism and mechanisms that enable cancer cells to surmount the BBB (Valiente et al., 2020).
As is discussed in section 7.1 the ability to connect multiple MPS which mimic different tissue microenvironments opens up the opportunity to investigate organ-organ interactions and may be used to shed light on cancer metastasis. This approach has already been applied to lung-brain metastasis. Yi et al. demonstrate this through the connection of a lung on chip model incorporating flow and cyclic stretch, upstream of three separate organ specific 3D culture of cells (Astrocytes, osteoblasts, and hepatocytes) demonstrating the importance of lung stromal cells on epithelial-mesenchymal transition and metastatic capacity (Yi et al., 2019).
For in depth reviews on how microfluidics models are not only advancing understanding of the brain tumor microenvironment but also cancer cell extravasation, diagnostics and drug efficacy screening see (Logun et al., 2018), (Cai et al., 2020), and (Coughlin & Kamm, 2020).

| Traumatic brain injury
Traumatic brain injury (TBI) is not only a leading cause of mortality and morbidity in adults but is also a risk factor for the future development of neurodegenerative diseases such as AD and PD (Gupta & Sen, 2016).
The initial mechanical shear, stretching, and compression, resulting from inertial forces induced during trauma such as motor vehicle accidents, falls, and sporting injuries, not only causes immediate physical injury to brain tissue but also sets in motion a range of protracted structural and biochemical changes including; anterograde degeneration, mitochondrial dysfunction, and secondary injury from excitotoxicity and inflammation (Yap et al., 2017). Recently CNS microfluidic models have been applied to TBI research to allow a more precise and targeted control of mechanical injury than can be produced in traditional in vitro models. A number of these developments have been made using neuronal cultures in axon guidance devices, taking advantage of high spatial precision to study axonal injury and white matter damage. One method to induce mechanical injury which has been used both in brain slices and with primary cell cultures, is by making use of flexible device materials (such as PDMS) and a pressurized pneumatic channel to deform the axonal channels mimicking the stretching of axons caused by shear deformation during TBI (Dollé et al., 2013;Fournier et al., 2014;Yap et al., 2017). Other approaches such as vacuum aspiration (Taylor et al., 2005), chemical treatment, laser-based axotomy , and electro-mechanical shear in 3D cultures (LaPlaca et al., 2005) have all been explored and the benefits and limitations of each of these approaches have been reviewed elsewhere (Shrirao et al., 2018). While these systems provide reproducible and precise injury, they have yet to fully recapitulate the many pathogenic events that follow TBI, such as inflammatory responses and the corresponding swelling and tissue hypoxia. Instead the current benefit of current MPS in this case, is the ability to study a specific mechanism within precisely defined conditions.

| FUTURE DE VELOPMENT AND CHALLENG E S
The desire in the pharmaceutical industry is to access models predictive of target efficacy and drug safety and the dose of drug needed to achieve therapeutic benefit in the clinic. A lack of efficacy and safety are the two most significant reasons why drugs in clinical trials fail to progress (Harrison, 2016). Looking to the future, the hope is that new human CNS models in development, are better able to predict target efficacy and safety, compared to historically used models, reducing clinical attrition and late-stage failure.
In the past, targets have been identified using models as a starting point, with a strong dependence on animal models. The pharmaceutical industry is now taking advantage of advances in genetics and computational biology, to interrogate large data sets from human populations, enabling target identification by linking genetic variants to human disease (Nelson et al., 2015), for example, LRRK2 to PD (Tolosa et al., 2020). The next step is target validation, to generate experimental evidence that therapeutic modulation of the target will provide efficacy in disease. physiologically relevant insights, such as BBB-brain metabolic coupling (Maoz et al., 2018) and brain metastasis (Yi et al., 2019).

| Incorporating iPSC technologies
Obtaining relevant human cells for CNS studies is particularly challenging, not only due to the scarcity of living tissue, (such as from epilepsy surgery) but also due to the post-mitotic nature of neurons, limiting their expansion in vitro. The ability to reprogram terminally differentiated somatic cells into stem cells by induced expression of pluripotency factors (OCT3/4, SOX2, c-Myc, Klf4) (Aasen et al., 2008;Takahashi et al., 2007;Takahashi & Yamanaka, 2006) has increased the availability of stem cells.
Developmentally inspired protocols now allow the production of human brain cells which carry the unique genetic sequence of the adult somatic cell, thus enabling investigation of disease-specific phenotypes such as in Schizophrenia Brennand et al., 2015;Marchetto et al., 2011;Wen et al., 2014), Parkinson's (Sánchez-Danés et al., 2012, and Alzheimer's diseases (di Domenico et al., 2019;Julia et al., 2019). The differentiation of mature functional iPSC-derived cortical neurons is, however,, requires prolonged cultivation times, with maturation occurring over a period of months (Kirwan et al., 2015;Odawara et al., 2016), as neural progenitor cells differentiate into neurons and astrocyte lineage cells, that participate in spontaneous network activity (Kirwan et al., 2015;Odawara et al., 2016;Shi et al., 2012). Single-cell RNA-seq data from human fetal and adult brain confirms that iPSC-derived cortical neurons are highly similar to primary cell neurons and with extended time in culture develop a more adult phenotype (Handel et al., 2016). Protocols have also been established for derivation of specific neuronal identities including GABAergic , glutamatergic (Cao et al., 2017), and dopaminergic (Mahajani et al., 2019) neurons, while novel methods of rapid neuronal induction and maturation have also been developed by inducing lineage-determining transcription factors . Methods for differentiating other key brain types including astrocytes (Lundin et al., 2018), microglia (Hasselmann & Blurton-Jones, 2020), and pericytes (Faal et al., 2019) have also been developed and many of these protocols have already been integrated into MPS (Appelt-Menzel et al., 2017;Prots et al., 2018;Usenovic et al., 2015). While integration of iPSC technologies may have the potential to greatly improve translational relevance for these models, as outlined in reviews by A. Sharma et al. (2020) and Pasteuning-Vuhman et al.
(2020), to be widely adopted, they need to be carefully characterized both phenotypically and functionally. Without careful phenotypic characterization, cell identity of differentiated cells can be mistaken. This has been recently suggested in a preliminary report that provides single cell sequencing and immune-staining evidence that a widely used brain endothelial differentiation strategy in fact produces neuroectodermal epithelial cells that form tight junctions reminiscent of those seen with BECs (Lu et al., 2019).
Thus robust characterization and reliable methods are required for iPSC technologies to succeed in delivering on their potential to significantly improve CNS disease modeling.

| Advancing organoid models
Recent developments in 3D culturing of human iPSCs into neural "organoid" tissues offer a promising horizon to explore (Camp et al., 2015;Kelava & Lancaster, 2016;Pasca et al., 2015;Qian et al., 2016). Various region-specific brain organoids have already been described, including retinal, whole-brain, cortical, forebrain, and midbrain (Eiraku & Sasai, 2012;Eiraku et al., 2011;Kadoshima et al., 2013;Lancaster et al., 2013;Lindborg et al., 2016;Marton & Pasca, 2016;Pasca et al., 2015;Qian et al., 2016). Such organoid cultures show great promise to extend the possibilities of in vitro neurological research but are beyond the scope of this review (for reviews please see (Wang, 2018) and (Setia & Muotri, 2019)). One notable limitation of organoids is the lack of vascularization, leading to a necrotic core of cells at sizes above ~300 µm due to the limitations of oxygen and nutrient diffusion. Combining organoid cultures with microfluidic techniques to provide perfusion  has shown to improve viability. Exploiting microfluidic techniques that promote angiogenesis to create perfusable vessels may allow further advances in organoid models (Shi et al., 2020). The use of microfluidic technologies is already advancing organoid culture, as reviewed by .

| Validation
One of the most important factors in the successful uptake of new in vitro models is validation and robust data showing improvements against existing models with greater relevance to the clinical disease. Emerging models are often described as "validated" by using existing drugs. This is valid for benchmarking new drugs against established drugs for safety and toxicology stud-

| Materials
Plastic has been the material of choice for biologists for almost 50 years and as such biologists are familiar with the capabilities.
Some microfluidic devices manufacturers have opted for the use of thermoplastics, while in other cases limitations in machining and properties have necessitated the use of other less familiar materials.
The vast majority of microfluidic laboratories experimenting with new possibilities through rapid prototyping use PDMS due to its ease of fabrication, low cost production, optical clarity, gas permeability, and relative biocompatibility. PDMS is also used for its high elasticity, as in the Emulate Chip which enables stretch to be used to mimic breathing cycles (Huh, 2015) and vascular stretch (Sato et al., 2019). PDMS is, however, highly hydrophobic with a low surface energy that, unless treated, resists biological coatings and may also absorb small molecules such as drugs, which can have signifi-

| Increasing data through integration of realtime monitoring
Multi-parametric sensing is highly desirable, to generate highcontent data from cell systems. In this context, the extensive developments on electrochemical enzymatic sensing of glucose and lactate, primarily in the field of blood sugar monitoring for diabetics, are now being applied to MPS and would be a useful addition to CNS models. Indeed glucose and lactate production, along with CO 2 production and O 2 consumption yield a detailed picture of cell metabolism, and have been incorporated into MPS to monitor cell growth (Prill et al., 2014). Oxygen and glucose consumption has been usefully linked to neuronal firing and can even predict burst activity such as that seen in epileptic seizures (Ivanov et al., 2015). pH sensing may also be useful, which in many cases is a substitute for monitoring lactic acid production. Functionalization of electrodes with redox enzymes or other suitable biorecognition elements, can in principal allow sensing of a wide variety of analytes, although requiring careful design of the functionalization strategy and materials/reagents.
A major challenge in integration of sensing technology comes when the cell systems become more 3D to provide a more natural environment for the cells. In terms of electrodes, currently, the majority of systems rely on flat, rigid electrodes. New technologies are nevertheless enabling flexible polymeric devices which can be tailored to adapt to tissues in a more seamless manner (Kalmykov et al., 2019;Pitsalidis et al., 2018). Much of this development is being pushed by the neural interfacing community, faced with the sizeable challenge of implanting electrodes into brain tissue for durable and non-invasive neuronal recording. Lessons learnt from this community will inevitably translate into better electrical recording devices in vitro, adapted to complex biology. Transparency of the next-generation electronic materials translates into an added advantage in vitro, in terms of imaging cells in devices (Curto et al., 2017).
Overall, the advantages obtained by electrophysiological and realtime monitoring systems are compensated by higher costs and by the increased complexity of fluidically sealing the microfluidic device.

| Increasing throughput and integrating automation
The throughput of models for the lead discovery can range from 1000s for antibody screening to 100,000, even 1,000,000s, for a  (Parrish et al., 2018). This last example and many other MPS models require complex bespoke perfusion systems and holders to operate the devices, which may limit up take or restrict the user to a single type of MPS, thus raising the issue of standardization within the field.

| Standardization
Standardization of microfluidic neurological models, and microfluidics in general presents a huge challenge, due to the inherently interdisciplinary nature of the field and the fast moving pace of innovation.
Failure to standardize may have massive implications to the growth of the field and widespread uptake of emerging technologies, however, poorly devised standards could stifle innovation and technological optimization (Blind, 2016). This cross-pharma collaboration, created to facilitate data sharing and expedite uptake and impact of MPS, will also likely have an impact on how MPS and Organ-on-Chip technologies are standardized.
Standards that emerge will likely exist on multiple levels, reflecting the multidisciplinary nature of the field, and may include: fluidic and electrical interconnects, dimensions, materials, media, flow rates and cells, and culture protocols. Any standards that emerge should, and often do, follow currently accepted bioscience and pharmaceutical industry standards such as glass slide standard dimensions (Emulate chips and IBIDI channel slides), ANSI/SLAS multi-well plate footprint (The OrganoPlate ® from Mimentas and CellASIC ® ONIX Microfluidic Plates from Merck), and the Luer Lock interface (IBIDI channel slides). Increasing standardization from the Organ-on-chip, microfluidic community must not stifle creativity and innovation but improve compatibility to facilitate and simplify collaboration and integration, accelerating innovation, and widespread adoption.

| CON CLUS IONS AND CONS IDER ATIONS FOR E ARLY ADOP TER S
As we have detailed in this review, MPS have provided an array of new opportunities to study neurological disease and new technologies are rapidly developing. The initial foray into microfluidic technologies for disease modeling has been spearheaded through close collaborations between engineers, chemists, and biologists, and while many of these pioneers are now offering expertise in their platform as outsourced research and disease modeling services (such as Aracari Biosciences Inc. and Hesperos Inc.), a number of devices have already entered the market targeting biologists as end users (Table 1) (Walker et al., 2004;Young & Beebe, 2010) should be well understood, along with an appreciation of the effects of the resulting shear and the implications that the high surface area to volume ratio and material properties have on the microenvironment. For instance, in case of PDMS devices, surface adsorption of hydrophobic dyes and drug molecules or bulk absorption of nutrients from media can confound results and while high gas permeability allows rapid O 2 exchange, water vapor permeability can result in changes in concentration of media constituents unless high humidity is maintained. Thus appropriate coatings or pre-incubation with media or serum to saturate absorption prior to cell culture, should be considered.
Shear stress can be used to enhance barrier functions of BBB models and increase phenotypic relevance of endothelial cells.
However it should be noted that shear can easily be introduced where it is not intended, such as neuronal cultures, which can be highly sensitive to shear. As an example, a hydrostatic head of 10 mm feeding a channel of 100 μm high 500 μm wide and 10 mm long can introduce brief shear stress of over 2 dyne/cm 2 , while in vivo shear of cells in contact with cerebrospinal fluid is approximately 0.01 dyne/ cm 2 with small increases having previously been shown to influence cell behavior (Park et al., 2017). A background knowledge and appreciation of fluid dynamics would benefit those adopting such studies.

DECL AR ATION OF TR ANS PAREN C Y
The authors, reviewers and editors affirm that in accordance to the policies set by the Journal of Neuroscience Research, this manuscript presents an accurate and transparent account of the study being reported and that all critical details describing the methods and results are present.

ACK N OWLED G M ENTS
We gratefully acknowledge support from the UK Organ-on-a-Chip Technologies Network (www.organ onach ip.org.uk/), which is funded by UKRI via the Technologies Touching Life Scheme (Grant reference MR/R02569X/1). As members of the Organ-on-a-Chip Technologies Network from diverse fields in both academic and industrial sectors, the authors of this review were brought together in an initiative to surmise the key development, challenges, and unmet needs within the field.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/jnr.24794.