Engineering three-dimensional microenvironments towards in vitro disease models of the central nervous system

Electrospinning is an easy, versatile and popular technique for producing tissue-engineered scaffolds. It creates fibres with ECM-like structures and enables one to easily tune physical properties such as fibre diameter and porosity [29, 30]. In a typical setup, a high voltage potential over the working distance creates an electrostatic force, which draws charged polymer threads to a collector. If the collector is a stationary plate, the deposited micro- or nanofibres form a mat with random orientations. On the other hand, a collector with a rotating wheel leads to fibres with orderly, parallel alignments. For neural tissue engineering, a uniaxially aligned fibrous network that orients cell growth is an important criterion to ensure accurate simulation [31, 32]. Xie et al found that the direction of neurite outgrowth can either be parallel or perpendicular to the nanofibre alignment, depending on parameters including fibre density, protein deposited on fibre surface, and surface properties of the supporting substrate [33].

In one study, Luo et al used electrospun polylactic acid (PLA) nanofibrous scaffolds for long-term culture of neuromuscular junction, a specialised synapse associated with neurodegenerative diseases. They co-cultured primary embryonic motor neurons from Sprague-Dawley rats and C2C12 cells in a random or aligned nanofiber configuration. While the conventional 2D glass substrate could only maintain the culture for 2 weeks, random and aligned PLA scaffolds had 7-week cell survival rates of 55% and 70% respectively [34].

In another study, Jakobsson et al used a semi-spherical array of metal needles as the collector for electrospinning. Without the compression of fibres in a typical setup, they obtained low-density electrospun poly-ε-caprolactone (PCL) fibrous scaffolds. The high porosity allowed for full 3D infiltration of neural cells. Using neural progenitor cells, they observed a highly integrated network, synaptogenesis, and extensive neurite outgrowth. Compared to the 2D culture, where neuronal cells grew on top of glial cells in separate layers, cells intermixed in 3D culture, which is observed in vivo [35].

Microfluidic technology tightly controls microlitres to picolitres of fluids within networks of micro-channels. Microfluidic chips are excellent in vitro models to study CNS degeneration and regeneration for three reasons [12]. Firstly, these platforms enable one to analyse cell secretions, transcriptions, and protein expressions at the single-cell level through designed compartmentalisation. As a result, one can study more in-depth with regards to myelination, neurite outgrowth, signal propagation, and neuronal networks [36, 37]. Secondly, such platforms can help probe cell–cell communication by co-culturing different CNS cells in one or more connected chambers [38–40]. Thirdly, microfluidics can accurately model or control the biophysical and biochemical cues of the microenvironment, so that one can monitor various cellular events post-injury or post-disease [41].

For example, Wevers et al used a microfluidic platform to culture induced pluripotent stem cells (iPSCs)-derived neural stem cells (NSCs) into 3D, ECM-embedded, neuronal-glial networks [42]. The platform had a 384-well microtiter plate format, with 96 tissue chips (4 wells per chip). The plated organoids were suitable for 3D analyses of biological processes such as cell differentiation, cell–cell interactions, cell–ECM interactions and related gene expression. Similarly, Wang et al created an organ-on-a-chip system to mimic blood brain barrier (BBB) for drug screening. They used human induced pluripotent stem cells to produce brain microvascular endothelial cells, and cocultured them with rat primary astrocytes. They used this microfluidic system to test permeability of various model drugs, and obtained data that were comparable to in vivo values [43].

3D printing is a technique that deposits materials in a layer-by-layer fashion to create 3D objects. With the help of computer-aided design software, one has the capacity to produce customisable hardware parts for rapid prototyping of CNS disease models. For instance, Johnson et al printed microchannels and compartmentalised chambers using PCL. They then deposited neurons, Schwann cells, and epithelial cells into separate chambers based on how cells are organised in vivo. The team proceeded to obtain insights about pseudorabies virus infection, thus highlighting the usefulness of this nervous system on a chip [44].

As a branch of 3D printing, 3D bioprinting utilises biological ingredients (e.g. biomaterials incorporated with viable living cells) as the ink to build functional 3D tissue constructs [45]. With precise spatiotemporal control over cell and biomaterial distributions, 3D-bioprinted objects can have accurate, complex and even personalised features that imitate the fine shape and architecture of natural tissues [46, 47]. While this technique has shown promising results in treating a range of brain-related injuries and disorders, it also possesses the capacity to produce normal or diseased tissues for cell behaviour studies [48]. For example, Lozano et al printed brain-like structures with discrete layers as an in vitro model to study cortical cell survival and axonal development [49]. Specifically, primary cortical neurons were encapsulated within a peptide-modified hydrogel, gellan gum-RGD to create a three-layer construct that included two cellular layers (top and bottom) and an acellular middle layer. After culture for 5 d, they observed that a neuronal network was formed in the brain-like structure, and extensive axons penetrated into the acellular layer. These results not only validated the versatility of bioprinting to control cell and ECM organisation for constructing a complex and viable 3D cell-containing construct, but also highlighted the possible reproduction of a more accurate 3D in vitro brain-like microstructure that might increase our understanding of neurological diseases and injuries.

AD is the most common form of dementia and is characterised by a gradual decline in cognitive and executive functions. AD affects 5.5 million people in the United States alone and with an ageing population, cases are predicted to increase dramatically [115, 116]. There is no curative treatment, and symptomatic management is limited [117–119]. The two major pathological hallmarks of AD are β-amyloid plaques and the accumulation of neurofibrillary tangles [120, 121]. The 'amyloid hypothesis' posits that the excessive accumulation of β-amyloid peptides results in neurofibrillary tangles composed of hyperphosphorylated tau, which subsequently accumulates in axons, dendrites, and cell bodies. These will ultimately cause neuronal death [122–124]. This hypothesis has only recently been proved in an in vitro setting. Commonly used cell models do not reflect the complexity of AD as it is still challenging to simultaneously incorporate various cellular components (e.g. axons, synapses) and AD-specific pathological proteins (e.g. tau) [125].

3D cultures can recapitulate the natural CNS microenvironment to promote neuronal maturation and increase tau formation, which is essential for reconstituting tauopathies such as AD [126]. In one study, Zhang et al cultured neuroepithelial stem cells derived from human iPSCs, with RADA-16 self-assembling peptides (SAPs). Upon adding cations, the SAPs self-assembled into a 3D hydrogel (figure 2(A)) [127]. Both P21-activated kinases (PAKs; a critical link in mechano-transduction pathways) and drebrin (an actin-stabilising protein in the brain) had high degrees of expressions in these 3D hydrogels, but not in 2D culture models. They postulated the reason to be related to the soft mechanical surroundings (stiffness of a few Pascal), which resembles brain tissue much more than conventional 2D culture dishes. The addition of Aβ oligomers, which are thought to be directly linked to the pathogenesis of AD [128], attenuated the expressions of both proteins. This was observed in the 3D SAP matrix, but not in 2D culture systems, further proving the superiority of the 3D model [100]. However, it did not evaluate the aforementioned 'amyloid hypothesis' by investigating whether excessive accumulation of Aβ oligomers would lead to neurofibrillary tangles composed of aggregated tau proteins.

Zoom In Zoom Out Reset image size

Subsequently, Choi et al provided experimental validation of this hypothesis (figure 2(B)) [129]. In their study, they used ReNcell VM human neural precursor (ReN) cells embedded in a Matrigel scaffolds and supplemented with neuronal growth factors. They first demonstrated the accumulation of high levels of both β-amyloid and phosphorylated tau. Then, they attenuated tauopathy by inhibiting β-amyloid generation. The findings thus provided a possible causal link between these two pathological processes. The scaffold material had high levels of brain ECM proteins, providing appropriate bioactive and architectural cues for optimal cell differentiation. Furthermore, compared to 2D culture conditions, 3D culture displayed dramatically increased neuronal and glial differentiation, β-amyloid deposition and subsequent levels of tau isoforms. The authors postulated the reason to be bioactive cues and secreted β-amyloid diffusing into a large volume of media in 2D conditions with subsequent removal upon media changes. A 3D culture, on the other hand, would limit diffusion and thus accelerate neuronal and glial differentiation and aggregation of β-amyloid.

Microfluidics aims to enhance current 3D disease models to create in vivo-like dynamic microenvironments. In one study, Park et al used a 3D culture-based microfluidic device as an AD model, and exposed neurospheroids to interstitial flow. Compared to static cell cultures, the dynamic culture conditions yielded larger spheroid sizes, had a greater neurite extension and an enhanced neural progenitor cell differentiation. This was likely due to the continuous supply of fresh medium and removal of metabolic waste through the flow. Additionally, exposure of neurospheroids to β-amyloid by an osmotic micropump remarkably decreased their viability compared to when they were exposed under static conditions (figure 2(C)) [130]. The reason was believed to be the interstitial flow enabling deeper infiltration of β-amyloid into the neurospheroids, causing more neurons to undergo apoptosis. The addition of dynamic flow further adds to a more realistic in vitro brain model. This model was further enhanced by the same group to include microglial cells capable of inducing a neuroinflammatory microenvironment [131]. Using a microfluidic triculture system involving neurons, astrocytes and microglia resulted in microglial recruitment and secretion of neurotoxic soluble factors with subsequent loss of neurons and astrocytes. It was identified that chemotactic microglial recruitment depended on Aβ accumulation and upregulation of chemokine (C–C motif) ligand 2/monocyte chemoattractant protein 1 (CCL2/MCP-1), both factors are known to be upregulated in human AD brains.

Despite recent advances, the design of 3D in vitro models must account for some important limitations. For instance, the self-assembling nature of some 3D AD models makes them difficult to control, resulting in marked differences in cell microenvironments [127] and batch-to-batch variations [132]. Furthermore, the use of iPSC-derived neurons to model AD holds several limitations: (a) their relative developmental age does not consider the fact that AD usually develops in adults beyond their fifth decade [133], and (b) the variability within iPS cell lines and resulting genetic and epigenetic variations of subsequent clones make it difficult to develop reproducible in vitro models [134].

PD is the second most common neurodegenerative disorder, characterised by bradykinesia, rigidity, resting tremor, and deteriorating cognitive function [135]. An abnormal accumulation of the α-synuclein protein in the form of toxic Lewy bodies results in irreversible degeneration of dopamine-producing neurons originating in the substantia nigra pars compacta. PD has no cure, and current treatment options are mainly limited to dopamine replacement strategies, which are complicated by long-term movement-related side effects [136, 137]. Understanding the complex molecular pathophysiology of PD, it is crucial to developing effective therapies. However, it is nearly impossible to extract live neurons from PD patients, let alone make representative in vitro models to study the disease [138].

Stem cell-based in vitro models mimicking PD have thus been developed using cell cultures and synthetic scaffolds [139]. For example, iPSCs derived from PD patients have been differentiated into dopaminergic neurons exhibiting signs of PD pathophysiology in vitro. Up until recently, most in vitro studies have relied on 2D cultures to demonstrate dopaminergic differentiation of stem cells. This, as previously described, fails to fully mimic the microenvironment in a living organism. As a result, 3D SAP scaffolds composed of amino acid sequences have been developed to improve survival and differentiation of NSCs [140, 141]. SAP scaffolds could support cell viability and induce much higher differentiation of murine embryonic stem cells into dopaminergic neurons, compared to 2D plates (41.5% ± 3.4% versus 8.3% ± 1.4%) [142]. This was thought to be due to the differences in spatiotemporal distribution of nutrients, cell surface receptors and other molecular signals in 3D culture environments. In particular, a higher surface exposure to midbrain patterning factors in the 3D SAP system was considered to be the crucial differentiating factor between 2D and 3D systems.

Although animal-derived stem cells are useful, their experimental results cannot be fully translated into human setting. Brito et al cultured human midbrain-derived neural progenitor cells (hmNPCs) of foetal origin. Under stirred culture conditions, the hmNPCs aggregated into neurospheres and differentiated towards a dopaminergic lineage [143]. The same group further developed this dynamic culture system for more efficient dopaminergic differentiation and neuronal maturation, making it suitable as a 3D in vitro PD model [144]. They exposed their neurospheres to the bioactive signalling factor cyclic adenosine monophosphate (cAMP). This led to significantly upregulated dopaminergic markers since cAMP is thought to promote the differentiation, maturation, and survival of midbrain dopaminergic neurons [145, 146]. These studies demonstrate the relevance of 3D in vitro neurospheres in elucidating PD pathogenesis. By taking advantages of neuronal precursor cells to self-organise into 3D structures [19, 147], one can mimic the fundamental processes of brain development [148]. This provides significant advantages in recapitulating physiological and pathophysiological mechanisms of brain function and dysfunction. Nevertheless, the foetal origin of the neural progenitor cells limits their clinical relevance in modelling PD associated with old age.

MS is a chronic inflammatory disorder of the CNS, characterised by immune cell infiltration, demyelination, axonal degeneration, and gliosis [149]. Myelin is essential for the efficient transmission of action potentials throughout the nervous system. Following demyelination, for unknown reasons, only limited remyelination occurs [150], characterised by a smaller diameter of myelin [151]. It is therefore crucial to develop reproducible in vitro models to study how demyelination happens, and what biochemical and biophysical cues can induce better remyelination in MS.

For instance, Harrer et al exposed murine organotypic cerebellar slice cultures (OSCs) to demyelinating factors to characterise their in vitro regenerative abilities [152]. This ex vivo system was found to be useful for evaluating therapeutic strategies to prevent demyelination and enhance remyelination in MS patients. However, disadvantages like loss of myelin during staining and questionable data caused by insufficient diffusion of demyelination-inducing agents through whole brain slices remained. For these reasons, Vereyken et al developed 3D whole brain spheroid aggregates in vitro, in which all cell types of the CNS were represented [153]. They exposed the single cell suspensions of rodent embryonic brains to constant rotational forces on a gyratory shaker. These continuous mechanical cues supported the spherical aggregation of embryonic brain cells into 3D spheroids. Once cultured for 4 weeks, myelin formation was detected throughout the whole spheroid. Exposure of spheroids to lysophosphatidylcholine (LPC), an agent used to induce demyelination [154], led to isolated myelin breakdown while sparing axons and causing little astrogliosis. They achieved remyelination thanks to OPCs present within whole brain spheroids, which provide mitogenic cues [155] and trophic factors [156] that were required for myelin production and unaffected by LPC. Furthermore, LPC toxicity could be attenuated by compounds such as simvastatin, which supports process extension and OPC differentiation. These results demonstrate that this 3D in vitro model can model demyelination and investigate interventional strategies.

HD is an incurable hereditary neurodegenerative disorder, affecting around 5–7 individuals out of 100 000 (lower in Asian countries) [157]. It is inherited in an autosomal dominant fashion, typically manifesting between the ages of 35–55 with changes in personality, cognition, and motor skills [158]. The clinical course of HD is progressive for over many years, ultimately leading to severe brain atrophy and death [159]. For individuals affected with HD, there is an expansion of CAG repeat region within the huntingtin (HTT)-encoding genes, resulting in aggregates of polyglutamine [160].

Many stem cell-based 2D in vitro HD models have been developed [161–163], but a limited number of relevant physiological models exist. Similarly, animal models, for the aforementioned reasons, are increasingly falling out of favour. Therefore, researchers have developed more suitable in vitro systems to elucidate HD pathophysiology. For example, Zhang et al used suspensions of self-aggregating HD-iPS cells to generate NSCs [164]. Within this 3D system, the HD-NSCs differentiated into striatal neurons containing the same CAG expansion found in the HD patient from whom the iPS cell line was established. Such differentiated cells could serve as a human HD cell model to analyse its pathophysiology or for drug screening. Despite some success in developing stem cell-based 3D in vitro cultures, most knowledge about the molecular pathways of HD still comes from analyses of HD mouse models or post-mortem HD tissue. With the help of patient-derived iPSCs, new avenues of elucidating the pathophysiology of HD will be available in the future.

Epilepsy encompasses a variety of syndromes which predispose the affected individual to generating an abnormal, transient discharge of neurons in the brain, or a seizure [172]. Depending on the brain region involved, the effects of seizures range from changes in cognition, convulsions and unusual sensations. Realistic in vitro disease models of epilepsy are useful for toxicity testing of new pharmacological agents and elucidating underlying mechanisms of seizure generation. Additionally, in vitro models enable the discovery of side effects of the tested drugs early on, thus increasing efficacy and reducing cost. Unfortunately, current models heavily rely on expensive animal models, rendering their results possibly untranslatable to the human scenario. Ideally, an in vitro model of epilepsy should employ cell types found in the human brain capable of forming functional neurophysiological networks and displaying seizure-like activity. It should further be amenable to high-throughput drug screening [173]. Currently available in vitro seizure models including their advantages and limitations are summarised in table 1.

Autism spectrum disorder (ASD) is a complex and heterogeneous neurodevelopmental disorder, which is usually defined as a cluster of symptoms including deficits in social communications, interactions, and restricted, repetitive patterns of behaviours. In some cases, ASD can cause cognitive delay. Little is known about the pathophysiology of ASD [174]. Animal models still cannot fully recapitulate such disorders, due to the broad spectrum of behavioural phenotypes and the inherent difficulty in recreating human-like behavioural patterns in rodents. During differentiation of normal human neural progenitor cells, many ASD-related genes and signalling pathways are highly co-expressed indicating the neurodevelopmental origin of ASD [175, 176]. Thus, relevant models should focus on the early disease development during which genes and environmental factors may play a significant role.

In a recent study, Mariani et al developed 3D neural cultures in organoids derived from iPSCs to explore neurodevelopmental variations in patients with severe ASD [177]. While they did not identify a single underlying genomic mutation, gene network analyses showed an upregulation of genes involved in cell proliferation, neuronal differentiation, and synaptic assembly. Organoids developed from individuals with ASD displayed an accelerated cell cycle and overproduction of GABAergic inhibitory neurons, which are hypothesised to be an underlying cause of ASD.

Rett syndrome (RTT), while not technically part of ASD according to symptomatology, shares commonalities with ASD in its early stages [178, 179]. Maria et al isolated fibroblasts from patients with RTT symptoms, and infected them with retroviral reprogramming vectors [180]. After culture for 2–3 weeks, compact iPSC colonies appeared in the background of fibroblasts, and then were manually picked up and transferred to Matrigel. They dissociated the cell clusters and plated them onto low-adherence culture dishes for 5–7 d to obtain embryoid bodies (EBs) for neural differentiation. The formed EBs were then transferred and plated on poly-ornithine/laminin-coated dishes. After a week of culture, EB-derived rosettes became visible and were collected for use. The RTT-iPSCs maintained the ability to undergo X-inactivation and generate proliferating NSCs and functional neurons. This model has the potential to recapitulate the early stages of some neurodevelopmental diseases. It may also be a promising tool for disease diagnosis, drug screening, and personalised treatment for RTT and ASD.

Thus far, EBs are one of the most common methods of modelling neurodevelopmental disorders in vitro. However, it remains questionable whether these models can reflect neurodegenerative diseases accurately enough to draw clinically relevant conclusions. Much more research is required to progress from the current stages of infancy to technically mature models.

Microcephaly is another neurodevelopmental disorder characterised by a reduction in brain size. Currently, the underlying causes and mechanisms of microcephaly are poorly understood [181], and there is still no available treatment. To address this issue, human iPSC-derived cerebral organoids have been applied in the study of microcephaly [27, 182]. For example, Lancaster et al developed an in vitro model using fibroblast-derived iPSCs taken from a patient with severe microcephaly [27]. These iPSCs self-assembled into 3D EBs and were embedded into droplets of Matrigel to provide an in vivo-like scaffold (figure 4). Continuous spinning acted as a mechanical cue to ensure optimal nutrient absorption, resulting in the rapid development of brain tissues within 20–30 d. Compared to control EBs, those derived from microcephalic iPSCs demonstrated smaller neural tissues with only a few progenitors, and exhibited premature differentiation into neurons. This phenotype could be reversed by reintroducing CDK5RAP2, a protein that causes premature neuronal death when mutated and is associated with microcephaly. This model recapitulates some fundamental mechanisms of mammalian neurodevelopment and could be used in the future to develop interventions to prevent the development of microcephaly in utero.

Zoom In Zoom Out Reset image size