Multiscale engineering of brain organoids for disease modeling

Abstract

Brain organoids hold great potential for modeling human brain development and pathogenesis. They recapitulate certain aspects of the transcriptional trajectory, cellular diversity, tissue architecture and functions of the developing brain. In this review, we explore the engineering strategies to control the molecular-, cellular- and tissue-level inputs to achieve high-fidelity brain organoids. We review the application of brain organoids in neural disorder modeling and emerging bioengineering methods to improve data collection and feature extraction at multiscale. The integration of multiscale engineering strategies and analytical methods has significant potential to advance insight into neurological disorders and accelerate drug development.

Graphical Abstract

1. Introduction

Human brain development is orchestrated by sequential molecular and cellular steps to form complex anatomic structures and sophisticated functions. The hierarchical process is initiated by genetic blueprints and unfolds in conjunction with environmental modulators through broad spatial and temporal scales, such as cell cycles, cell-cell interactions, cell-extracellular matrix interactions, and morphogen gradients. Deconstructing and reconstructing the tissue architecture is crucial for understanding the mechanism of brain development and pathology. However, species-specific features of animal models and the inaccessibility of developing human brain tissues hinder the comprehension of human brain development and pathogenesis. Experimental models capable of recapitulating developing human brain features are urgently needed.

The rapid expansion of human embryonic stem cells (ESCs) [1,2] and induced pluripotent stem cells (iPSCs) [3] technology provides an unprecedented opportunity to investigate human brain development and neural disorders. The PSCs enable human neural cell generation, experimental manipulation, and analysis in vitro [1–3]. Initial studies mainly used 2D culture systems that lack three-dimensional (3D) architecture and diverse cell types. Over the past decade, a new model system, brain organoids, has been developed to replicate better the developing brain’s diverse cellular composition and cytoarchitecture [4,5].

The generation of brain organoids is categorized into unguided and guided methods[6]. Both methods start with embryoid body (EB) formation by aggregating PSCs. Unguided methods do not use patterning factors and reply on the self-organization of EBs under optimized culture conditions, which generates brain organoids with random, multiple positional identities [4]. Guided methods use patterning factors to orchestrate key signaling and morphogen pathways to generate region-specific organoids, such as the forebrain, midbrain, thalamus, hypothalamus, hippocampus, choroid plexus and cerebellum organoid [5,7–9]. These brain organoids successfully resemble developing brains in various aspects, including cellular diversity, microscale tissue architecture, and developmental trajectory [10]. They also recapitulate the widespread and dynamic switching of epigenetic modifications and cell fate restrictions during CNS development [11–13].

However, these models lack spatial-temporal control of the microenvironment, leading to batch-to-batch variability, constrained biological fidelity, and limited tissue- or organ-specific phenotypes and functions [6,14,15]. Bioengineering approaches have been integrated to overcome these challenges by providing more physiologically relevant microenvironments at molecular-, cellular- and tissue levels.

This review will discuss the latest engineering methods to control molecular, cellular, and tissue-level microenvironments to develop high-fidelity brain organoids and improve readouts of the brain organoids. We then highlight advances in building brain organoids for neural disorders, which uncover the pathogenic features at molecular, cellular, and tissue levels during brain development. The integration of neurobiology and bioengineering will benefit drug development for neural disorders.

2. Engineering molecular-level contents

Molecular cues, including intrinsic/extrinsic factors and ECM components, play a key role in stem cell renewal, differentiation, and organization. Integrating bioengineering approaches enables precise control of microscale orders and improves brain organoids’ physiological resemblance and reproducibility.

2.1. Soluble factors

2.1.1. Optimize soluble factor composition

PSCs’ self-organization is highly sensitive to the local cell niche. Modulating exogenous soluble factors in culture environments has been applied to influence PSCs’ viability, self-renewal and differentiation.

Automated high throughput screening is necessary to accelerate the discovery of new molecules that could improve brain organoid generation (Figure 1Ai). The Rock inhibitor Y27632 is frequently employed to enhance the viability of dissociated PSCs during EB formation [16]. But it does not eliminate cell death. The uncontrolled cell death during aggregation complicates the reproducibility of consistent EB sizes and quality, particularly when the same cell counts are used across different batches and PSC lines. An urgent need is to find more potent chemicals to improve cell survival at the initial stage of brain organoid generation. Yu Chen et al. [17] discovered a four-component small-molecule cocktail (chroman 1, emricasan, polyamines and tran-ISRIB) demonstrating improved iPSC viability than Y27632 after screening 15333 compounds with automated quantitative high throughput screening (qHTS). The screening system is also used to optimize the combinations and concentrations of the molecules, which eventually exhibited dramatic improvement in EB formation and neural differentiation of organoids. Similarly, Studer’s group developed a high-content imaging system to identify compounds that drive PSC-derived cortical neuronal maturation [18].

Figure 1. Multiscale engineering of molecular-, cellular- and tissue-level contents for high-fidelity brain organoids.

A. Engineering molecular contents including soluble factors and ECM. Efforts include (i) high throughput screening of the signaling factors, (ii-v) spatiotemporal control of the morphogen delivery, ECM optimization with (vi) natural decellularized brain ECM, (vii) synthetic materials with tunable properties and (viii) ECM library screening. Images ii, iii and iv were adapted from references [39] [41] and [42].

B. Engineering cellular contents, including genetic/epigenetic engineering of cells and geometric control of cell organization via 2D micropattern, micro scaffold and sectioning. Images ii and iv were adapted from references [134] and [137].

C. Engineering tissue-level contents, including (i) cell migration between brain regions, (ii) neural circuits formation, and (iii-vii) vascularization.

In guided brain organoid models, small molecules or growth factors are added to the differentiation medium to mimic the in vivo morphogen signaling events and differentiate the PSCs towards specific neural populations. In the developing CNS, neural induction and neuronal differentiation are achieved by the regulation of cell-cell signaling of ectodermal cells through bone morphogenic protein (BMP), fibroblast growth factor (FGF) and Wingless and Int (WNT) signaling pathways. Other signaling pathways, such as the Notch signaling pathway, also contribute to the proliferation or differentiation of neural progenitor cells. The rostral-caudal axial patterning is mainly driven by FGF, retinoic acid (RA), and WNT. The interaction of the opposing concentration gradients of downstream effectors, Emx2 and Pax6, is essential for anterior-posterior patterning of the early neocortex, and disruptions on the gradient ultimately alter neocortex formation [19–21]. The FGF and WNT signaling mediated the spatiotemporal events in intrinsic anterior-posterior axial patterns. Combining these morphogen gradients along the anterior-posterior axis further drives the formation of the forebrain, midbrain and hindbrain [22]. The dorsoventral axis formation is driven by the morphogens Sonic hedgehog (SHH) and bone morphogenetic protein (BMP). SHH levels guide ventral brain patterning and the formation of various neurons (dopaminergic neurons of the ventral forebrain/midbrain and motor neurons of the spinal cord) depending on the spatial organization of progenitor cells and their interactions with nearby inductive factors [21].

The combination of patterning factors has been applied to generate various region-specific brain organoids, including the forebrain [23–25], midbrain [26,27], thalamus[28,29], hypothalamus [30], hindbrain [31], cerebellum [32], striatum [33], hippocampus [34], choroid plexus [35], and spinal cord [36]. Generally, these protocols harness the prior knowledge of spatiotemporal morphogen expression patterns and iteratively modulate morphogen pathways to yield desired neural populations [37]. For example, applying SHH drives the PSCs toward ventral neural cell fates [23,25]. Combining SHH with WNT and FGF8 yields midbrain dopaminergic neurons, while combining SHH, WNT, and RA generates spinal motor neurons. The synergetic effects of morphogens on regional patterning and neuron specification have been systematically screened by Neal et al. [38], who employed 14 morphogen modulators and their combinations for brain organoid generation. With multiplexed RNA sequencing and single-cell references mapping to the human fetal brain, the authors built an in vitro atlas of human neural cell differentiation, guiding combinatorics and morphogen timing for generating specific neural cell types.

2.1.2. Engineering soluble factor spatiotemporal dynamics

Instead of submerging the culture in a homogeneous culture medium in a dish, microfluidics provides more spatial-temporal control of the soluble factors (Figure 1Aii–iv). For region-specific brain organoids, EBs are skewed to develop into neuroectoderm with SMAD inhibitors, followed by patterning factors to drive the specification of regional identities. The concentrations of the patterning factors oscillate along with the daily medium change. Microfluidics provides optimal control of the concentration dynamics of both exogenous and endogenous factors. Programmed perfusion controls the timing of soluble factor addition or withdrawal, therefore controlling the cellular processes in a stage-specific manner. Optimal extrinsic signal modulation can be achieved for desired cell lineage specification and differentiation by tuning the frequency of media delivery. Giovanni et al. reported a microfluidic approach for differentiating PSCs on a chip with periodic medium delivery with stage-dependent frequency to optimize the cell lineage specification [39]. The iPSCs were cultured in straight microfluidic channels with an automatic pump system. By tuning the perfusion frequency, the authors could direct the germ layer commitment of the PSCs under spontaneous differentiation or achieve more functional terminal differentiated cells with induction factors.

In addition to the temporal delivery of cues, spatial delivery may be necessary to produce brain organoids that recapitulate the asymmetry and regional heterogeneity of developing CNS. Current brain organoids are flooded with homogenous culture medium. Yet, concentration gradients of soluble factors are very common in vivo to guide cellular migration and differentiation. It is challenging to apply factors in a gradient manner in conventional culture ware, but this can be easily achieved via microfluidic gradient generators [40–43]. The simplest gradient generator design is a “T” or “Y” junction consisting of two channels, with fluid inputs of different concentrations of the target factor merging into a central channel. The target factor diffuses at the interface between the laminar streams and generates a tunable and predictable gradient pattern (Figure 1Aiii). An alternative microfluidic gradient generator uses branching networks of serpentine channels (Figure 1A iv). The input streams are serially diluted into separate channels and merged into a large central channel, forming a concentration gradient. Both design paradigms have induced axial patterning in PSC-derived neural tube models through opposing SHH and BMP4 gradients [41] or WNT gradients [42]. PSCs exposed to the gradients of morphogens differentiated into neural tube-like tissues with in-vivo-like axial patterning [41,42].

The limitation of microfluidics is the dimension incompatibility with the brain organoids. The morphogen gradient pattern becomes unstable as the channel size increases, and the interaction between the 3D culture and the flow becomes non-negligible [44]. This challenge could be resolved by integrating a pseudo morphogen signaling center into one pole of the developing brain organoid (Figure 1Aiv) [45–47]. Gustav et al. successfully generated dorsal-ventral axial patterned forebrain organoids by incorporating a small cluster of inducible SHH-expressing hiPSCs in the EBs. The titratable expression of SHH is controlled by doxycycline and forms the SHH protein gradient in the developing forebrain organoids [46]. Incorporating optogenetic control enabled tighter spatial-temporal control of the SHH gradients [45,47].

2.2. Extracellular matrix (ECM)

The brain ECM primarily comprises hyaluronan, heparan sulfate proteoglycan, chondroitin sulfate proteoglycan, tenascins, collagen, laminin, and fibronectin. During brain development, the ECM supports the attachment, migration, differentiation, neurite outgrowth, and maturation of neuronal cells [48,49]. Selecting proper ECM is crucial to generating brain organoids closely resembling the developing brain.

Matrigel is the most widely used ECM for current brain organoid generation protocols [4,50,51], either as a scaffold for organoid expansion or a medium additive. The Matrigel derives from the murine basement membrane and is enriched with 2000 unique proteins [52]. However, the undefined chemical composition, lack of brain ECM components, and batch-to-batch variability make it undesirable for building high-fidelity and reproducible brain organoids.

Decellularized ECM extracted from brain tissue is a promising substitute for Matrigel by providing a more physiologically relevant niche for brain organoid culture (Figure 1Avi) [53–55]. Brain organoids generated with a mixture of decellularized porcine brain ECM, laminin, and type 1 collagen exhibited mature neurons and astrocytes with reduced astrogliogenesis. However, access to decellularized brain ECM is very limited, and the decellularization usually contains host cell residues, which makes it impossible for large-scale application and translational medicine.

In contrast to undefined hydrogels, chemically defined hydrogels are more uniform and, therefore, improve the reproducibility of brain organoid generation. The chemically defined hydrogels are either from natural ECM components or synthetic polymers. Hyaluronic acid (HA) and modified HA have been used for neural differentiation of PSCs [56,57] and cerebral organoid generation [58]. The HA-based hydrogels have promoted the differentiation of PSC towards neural fates or proved equivalent to Matrigel. Synthetic hydrogels provide more room for customization and control, including bioactive molecule/ECM component conjugation, ligand density, stiffness, and degradability [59–62]. These modifications could mimic the ECM signals in neural progenitor self-renewal, neuronal migration, and cerebral cortex folding. Adrian et al. developed tunable synthetic extracellular matrices with peptides conjugated poly (ethylene glycol) (PEG) hydrogels to create a controlled 3D microenvironment [62]. The hydrogel is tunable for matrix stiffness, degradation, and ECM composition to promote neurogenesis of PSC and brain organoids, remarking the orthogonal multimodal engineering of chemical and physical properties (Figure 1A vii). The incorporation of nanomaterials into the interstitial spaces of polymetric matrixes could help modulate the mechanical properties of ECM locally or globally [63,64]. If the nanoparticles do not impede the polymer sidechain crosslinking, they contribute to an increase in the local stiffness of the ECM. Conversely, the resulting stiffness is reduced if this incorporation disrupts the sidechain crosslinking process [65]. Spatiotemporal modulation of the mechanical ECM niche could be achieved through nanomaterials responsive to exogenous stimuli, such as magnetic, optical, acoustic and electric fields [66]. The control of ECM chemical cues, such as morphogen gradient, could also be achieved by control-released nanomaterials. Integrating conductive nanomaterials, such as polypyrrole nanomaterials [67], carbon nanofibers [68], metal oxide nanomaterials, and semiconductive nanomaterials [69], into the ECM or scaffolds could enhance its conductive properties, thereby improving the electrophysiological activity of neurons within brain organoids.

Cell-ECM interactions during brain development are crucial for neural stem cell differentiation, migration, and maturation [70–76]. The complex spatial and temporal dynamics of ECM compositions restrict quantitative and rational material design for the brain organoids. High throughput technologies and data science have great potential to address this challenge. ECM libraries have been used for PSC culture, differentiation, and organoid generation [62,77–79]. Array screening enabled systematic manipulation and optimization of various backbone compositions and multimodal features such as stiffness, degradability, and biochemical factors (Figure 1Aviii).

3. Engineering cellular-level contents

At the cellular level, hundreds of billions of neural cells are generated during development, with tightly regulated survival, formation of synaptic connections, and function to create the mature architecture of the human brain [80]. Neural cell generation, migration, morphological and functional specialization, and selection for maturation are temporally and spatially controlled [20]. Cell division and differentiation occur at different rates depending on location, as different neural progenitors are involved in symmetric and asymmetric division during neurogenesis, leading to histologically and spatially distinct brain regions. Distinct cell populations from different regions interact to determine neural development. For example, projection neurons derived from radial glial cells migrate to the cortical plate. Each of the six cortical layers is developed through temporally controlled morphological and migratory processes [81], ultimately creating an inside-out structure with the earliest neurons in the deepest layers and the later neurons in the superficial layers.[82] Brain organoids have successfully recapitulated the cellular diversity, microarchitecture, and developmental trajectory of the early-stage embryonic brain [11,12,83–90]. Engineering the intrinsic properties of cells and cell organization improves the robustness and reproducibility of brain organoid generation. It facilitates their application in disease modeling, drug screening, and differentiation mechanism study.

3.1. Engineering cells:

The PSCs and PSC-derived neural cells can be genetically engineered to control stem cell differentiation and/or integrate artificial genetic elements (Figure 1Bi). For the former purpose, CRISPR/Cas9 systems have shown enormous potential for basic and clinical application. It can precisely introduce gene modifications into endogenous gene loci in PSCs to study the causation of single gene mutations or to rescue single disease-causing mutations [91–97]. CRISPR-Cas9s could be used to generate the genotype in normal brain organoids or as a rescue measure in patient-derived organoids associated with single gene mutations.

Lineage tracing is a technique used to study the developmental history of cells, allowing researchers to map the fate of individual cells and their progeny during development or differentiation. In the context of brain organoids, lineage tracing can help elucidate the differentiation pathways, cell fate decisions, and the organization of different cell types within the organoid. Combining CRISPR-Cas systems with the Cre-loxP system or optogenetics could further restrict the gene manipulations to subpopulations or specific cell lineages. The insertion of inheritable genetic markers, such as a barcode or fluorescent proteins [47,97–103], can be analyzed with single-cell RNA seq and single-cell microscopy to decipher single stem cell lineages during cell fate specification. Lineage tracing in brain organoids can provide valuable insights into the developmental processes that shape the formation and organization of neural tissue. This knowledge can help refine organoid protocols, identify novel cell types or signaling pathways, and advance our understanding of brain development, function, and diseases.

Genetic engineering could also help integrate biosensors in the brain organoid for neural activity monitoring. Engineering the GCaMP family sensor enables the detection of cellular Ca²⁺ dynamics with single-spike sensitivity in neuronal cell types [104–109]. Similarly, transmembrane voltage protein sensors have been used to reliably detect action potentials with single-cell resolution [110–114]. Neurotransmitter indicators have also been developed to detect synaptic activities among different neuron types (e.g., glutamatergic, GABAergic and dopaminergic neurons) [115–118]. Coupled with neuronal subtype-specific promoters or enhancers [119–122], the expression of biosensors can be restricted to specific subclasses of neurons, allowing for selective activity recording of specific neuronal subtypes.

The epigenetic variability among iPSC lines has been a major factor causing the variation in the differentiation capacity to specific lineages, which hurdles their application in disease modeling, developmental research, and drug screening [123–125]. Scientists have tried to use epigenetic engineering to convert iPSCs/ESCs to totipotent stem cells to improve the robustness of pluripotency [126–129]. Mingzhu Yang et al. converted mESC to totipotent-like cells by the combination of DOT1L, KDM5B and G9a/GLP inhibitors through remodeling the pericentromeric heterochromatin and reestablishing H3K4me3 domains [126]. The mouse iPSCs have also been reverted to totipotent stem cells with a defined chemical cocktail [128]. Though there is no report on the successful conversion of human PSCs to totipotent stem cells, the concept of manipulating pluripotency via epigenetic engineering is established. As a proof-of-concept, Hongkui Deng’s group managed to reprogram human somatic cells to iPSCs with well-defined small molecules, which synergistically target cell signaling pathways and epigenetic modifiers [130]. This approach has great potential for standardization of hiPSC generation and applications in translational medicine and drug development.

3.2. Engineering cell organization:

Micropatterns have been applied to trigger self-organized patterning in 2D cultured hPSCs [131,132], as the colony geometries could manipulate the paracrine signaling in early germ layer development. The same paradigm has recently been applied to 3D organoid culture to guide neural cell organization, polarization, and lumenogenesis (Figure 1Bii), akin to neural tube morphogenesis in vivo [133,134]. To generate these micropatterned 3D cultures, hESCs/iPSCs were seeded on micropattern and differentiated to neural lineage with SMAD inhibitors. The culture thickened over time and formed layered structures akin to the epidermis and neuroepithelium. In contrast to brain organoids, which usually contain multiple disarrayed neural epithelium clusters, the neuroepithelium would fold inward and fuse, forming a single-lumen structure resembling the developing neural tube. Since the spatial self-organization within colonies is preserved in detaching aggregates [131], the 2.5D culture could be a new way to generate brain organoids with improved morphological resemblance to the developing brain.

The geometry constraint could be applied in a 3D manner with micro scaffolds for PSCs aggregation [68,135,136] (Figure 1Biii). The PSCs will attach to the PLGA microfilaments, forming elongated aggregates instead of spherical ones. The engineered EBs have higher surface area to volume ratio, exhibiting enhanced neuroectoderm formation and improved neural differentiation. However, the shapes of EBs are random and heterogeneous. Future improvements of the scaffold, including optimizing materials and shape control, will potentially address this issue.

As mentioned above, reducing the thickness at one dimension of the brain organoids improves the oxygen penetration of the brain organoid without vasculature. Microfluidic and microfabrication approaches could enable homogenous and long-term control of the geometric shape of the brain organoids (Figure 1Biv). For example, Eyal et al. inserted EBs into a microfluidic chamber with a height of 150 μm [137]. The EBs were differentiated in hydrogel while sandwiched by a cover slip and a porous membrane for medium exchange. The surface wrinkles started to appear five days after insertion. This model provides a unique platform to study brain folding and lissencephaly modeling.

For mature brain organoids, geometric confinement could help improve cell survival and functional maturation. Like organotypic slice culture, organoids have been sectioned into thin slices to bypass the diffusion limit [138–140] (Figure 1Bv). The sliced organoids were collected on filter inserts and cultured at the air-liquid interface in a transwell plate. This approach has improved the oxygen supply for mature brain organoids and sustained neurogenesis in long-term culture, leading to expanded cortical plates with upper and deep cortical layers. The neuronal survival and axon outgrowth increased dramatically, yielding thick axon tracts with long-range projections.

4. Engineering tissue-level contents

Brain formation depends on the integration of diverse cell types, which originate from various brain regions and other germ layers, such as yolk-sac-derived microglia and mesoderm-derived vascular cells. Cell migration, such as glia, interneurons and oligo progenitor cells migrate across brain regions to shape the brain [70]. Brain development also establishes interconnected circuits across the whole brain and the other tissues/organs in the organism. Emerging co-culture approaches have been developed to recapitulate the intricate tissue-level interactions related to brain construction.

4.1. Engineering tissue-tissue interaction

Though the unguided brain organoid consists of multi-regional identities of the brain, their localization and interactions are very stochastic and difficult to reproduce. The development of guided brain organoids led to emerging approaches to generate brain assembloids for controlled assembly of regionalized brain organoids and other cell types [141].

Initially, the assembloids were used to model neuron migration by fusing dorsal and ventral forebrain organoids [23,142,143] (Figure 1Ci). The model recapitulates the GABAergic interneurons’ migration from the ventral forebrain to the dorsal forebrain. Interactions between other regions have been established with the assembloid approach, such as cortex-thalamus [28], cortex-striatum [33], and hypothalamus-pituitary gland [144]. These models recapitulated the unique axonal pathways specific to brain regions. For example, the reciprocal projection in thalamocortical assembloids and unidirectional projections from cortical to striatal organoids.

Functional assembly of multiple regionalized organoids and micro-tissues has proven feasible. A triple-parts assembloid of human cortico-spinal-muscle organoids showed functional circuit formation [145] (Figure 1Cii). In this assembloid model, corticofugal neurons project and connect with spinal organoids, and the spinal motor neurons connect with the muscle tissue. The muscle contraction could be robustly controlled via optogenetic stimulation of the cortical organoids. Functional retinol-thalamo-cortical assembloid has been established as well [146]. In the future, more complex assembloids could be generated to study the circuit modulation and dysfunction with patient cells.

The assembloids are usually generated by manually fusing the organoids, which showed limited scalability and poor control of the organoid position. An engineered platform was developed to improve scalability, automation, and precise manipulation of the organoids. 3D printing of organoids enables the control of spatial distribution and interactions. Organoids can be suspended in bio-inks and printed continuously [147,148] or individually handled by aspiration [149,150]. However, the former method cannot control the position of individual organoids, and the aspiration might damage organoids. Julien et al. improve the 3D printing technology to enable spatial control of brain organoids by magnetic levitation [151]. The new platform consists of iron-oxide nanoparticle-laden hydrogels and a magnetized 3D printer, allowing controlled movement of brain organoids without damage. This platform holds great potential for constructing larger 3D cultures using organoids as building blocks. The orientation of fusing organoids is also important, especially for building an aligned axis. Zheng et al. designed a hexagonal acoustic microfluidic chip to handle brain organoid fusion [152]. By regulating the acoustic dynamic field in the hexagonal chip with suspended organoids, the author can move and rotate the organoids without direct contact and damage. This method enabled the alignment of organoids with the desired orientation.

4.2. Engineering vascularization:

The simplest method to vascularize brain organoids is through the co-culture of brain organoids and vascular endothelial cells, such as HUVECs [153] and hPSC-derived ECs [154] (Figure 1Ciii). The ECs are self-assembled into vasculature-like structures inside the organoids. A more mature and functional vascular network can be achieved by transplanting the organoids to animals, where the vascular network can connect to the host circulation system. Instead of spiking in ECs exogenously, ECs could be co-differentiated alongside the neural cell differentiation in one brain organoid (Figure 1Civ). Bial et al.[155] and Tyler et al.[156] discovered that overexpression of ETS transcription factor ETV2 could drive neuroectodermal epithelial cells toward endothelial fate. Bial et al. engineered a doxycycline-inducible ETV2 expressing hESC line for brain organoid generation [155]. The chimeric EBs consisted of ETV2-expressing hESCs and normal hESCs successfully differentiated into spheroids containing neuroectodermal-derived neural cells and mesodermal-derived endothelial cells. The endothelial cells formed a complex vascular-like network under neural differentiation conditions.

In addition to vascular cells, vascular microtissues have also been used for brain organoid vascularization. Similar to assembloids, mesodermal progenitor spheroids [157] and mesoderm-derived vessel organoids [158–161] have been used for brain organoid vascularization (Figure 1Cv). The mesodermal organoids contain various vascular cells, including vascular endothelial cells, smooth muscle cells, fibroblasts and pericytes. These cells formed complex vascular structures, including interconnected endothelial lumens enveloped by mural cells and vessel ultrastructural junctions. When the brain and vascular organoids are fused, the vasculature will penetrate the neural epithelial layers in the brain organoids, mimicking the brain vascularization progress in vivo. However, the penetration depth of the vasculature is minimal without physical cues such as hemodynamic flow. Lewis’s group designed and 3D printed a perfusion chamber connected to an aligned inlet and outlet for high-flow rate perfusion [162]. A thick layer of hydrogel was cast in the chamber, and the kidney organoids were embedded in the gel near the center of the chamber. Their result indicated that high shear stress facilitated angiogenesis in the kidney organoids. The vasculature demonstrated deep penetration into the renal tissue and formed a perfusable network with openings connected to the flow system [162].

Microfluidic devices have been used for generating perfusable vasculature [163,164]. The endothelial cells on the chip undergo angiogenesis and anastomosis to form perfusable capillary networks in hydrogel-filled chambers. However, the limited dimension of the microfluidic channels makes it challenging to host larger brain organoids and vascular networks. Instead of generating capillaries on microfluidics, 3D printing has been applied to generate larger perfusable vessels [165,166] (Figure 1Cvi). The vascular lumens were generated by dissolving the sacrifice materials or co-axial extrusion and further lined with ECs. 3D printing combined with other orthogonal engineering strategies could enable programmable rebuild of higher-order cerebral architectures. Recently, Mark et al. successfully generated large programmable vascularized 3D neural tissue via bioprinting [167]. Using doxycycline-inducible ETV2-expressing hiPSCs and wild-type hiPSCs, cells can differentiate into divergent cell types using the same differentiation protocol. The composition and position of the ETV2-expressing hiPSCs and wildtype hiPSCs were precisely controlled by the multi-nozzle bioprinter and simultaneously differentiated into vascular endothelial cells and neural cells with the desired pattern.

So far, transplantation has been the most effective method to vascularize brain organoids. Brain organoids with [154] or without [168] prebuilt vasculature have been transplanted into host animals and developed functional vasculature by angiogenesis and anastomosis (Figure 1Cvii). The vasculature was connected to the host’s circulation system to introduce blood flow within the deep layer of the brain organoids. When implanted into the brain, human brain organoids have developed extensive synaptic and vascular connections with the hosts and exhibited increased viability and maturation [168,169]. Optogenetic manipulation of the integrated cortical organoids has enabled the control of the reward-seeking behavior of the host animal. Though transplantation approaches introduced xenogeneic components into the developing human brain organoids and may compromise their resemblance to the human brain, they represent a humanized animal model and provide a unique opportunity to study circuit-level phenotypes in pathogenesis.

5. Engineering techniques to improve data collection and feature extraction

Brain organoids exhibit unprecedented advantages in modeling human brain development and neural disorders, yet the intricate cell diversity and 3D environment make it challenging to extract and analyze the biological features. Diverse engineering methodologies and tissue engineering techniques have been deployed to amass detailed, quantitative data across molecular, cellular, and tissue dimensions (Figure 2).

Figure 2. Bioengineering techniques to improve data collection and feature extraction.

A. Multi-omic techniques enable genomic, transcriptomic and proteomic feature profiling of brain organoids at the molecular level.

B. Microscopic techniques and patch clamps enable the detection of various cellular, architectural and functional phenotypes of brain organoids. (i-iii) high-resolution microscopy for the detection of neuron morphology, neuronal migration and corticogenesis in developing brain organoids. (iv) patch clamps for recording electrophysiology of individual neurons. (v, vii) 3D whole-mount imaging of brain organoids with light-sheet microscopy allows for cell type distribution assay, volumetric measurement and topology detection. Images iii, v, and vi were adopted from references [25], [189], [190].

C. MEA for electrophysiology detection of brain organoids. Various formats of MEAs were developed for the detection of neural activity in brain organoids: (i) regular 2D MEA; (ii-iv) engineered 3D MEAs to adapt the geometry of brain organoids, including (ii) a high-density MEA probe on a shank [195]; (iii) kirigami MEAs [194]; (iv) Shell MEAs [191].

The transcriptomes vary spatiotemporally along with the human brain development. With the rapid development of high-throughput next-generation sequencing (NGS), many transcriptomes have been obtained at different stages of human brain development. Single-cell RNA sequencing (scRNA-seq) has become prominent due to its ability to precisely identify cell populations, uncover regulatory gene relationships and track cell lineages [170]. scRNA-seq has been widely applied to organoid research to track temporal gene expression through organoid development and maturation [171], to define and resolve cell diversity [15], to validate models against human or model organism sequences, and to reveal gene interactions in proliferation and differentiation [85]. Using scRNA-seq in conjunction with other technologies can help establish a deeper understanding of neurodevelopment and neuropsychiatric disease pathophysiology.

While scRNA-seq is changing the study of genomes and transcriptomes, advances in mass spectrometry are improving proteomic analysis. Mass spectrometry allows the identification and quantification of proteins in cultures and delivers a map of protein expression. Stable isotope labeling with amino acids in cell culture (SILAC) uses isotopic atoms to generate small differences in peptide mass that can be used to quantify protein changes by mass spectrometry. SILAC minimizes the handling bias and increases the reproducibility for quantification of the differential proteome [172]. Another novel proteomics method using label-free protein samples was used to characterize the effects of the hallucinogenic molecule 5-MeO-DMT on cerebral organoids, revealing a downstream protective function against cell death and neurodegeneration in 3D organoids but not in 2D cerebral cultures [173]. Label-free quantification is based on precursor ion intensity rather than on isotopes or tags. Therefore, it can be easily applied to any sample and scaled up. Mass spectrometry proteomics can be utilized to quantify protein abundance and study post-translational modifications and protein localization and interaction.

As the correlation between mRNA and protein expression levels is uncertain [174], it is important to combine NGS and mass spectrometry methods to obtain a full picture of neuropsychiatric disease mechanisms. Combinations of NGS and mass spectrometry have established high-resolution baseline information and genes responsible for vital cellular events in human development [175]. Convergent functional genomic approaches to integrate transcriptomics, proteomics, and methylomics data have been proposed to compare molecular signatures between patients and controls to obtain insight into disease mechanisms [176].

At the cellular/tissue level, high-resolution microscopy is frequently used to profile the cellular/subcellular morphology, cell migration and laminar structures of brain organoids [23,26,138,139](Figure 2Bi–iii). Labeled with fluorescent cell type-specific markers, it allows for evaluation of cell type population, cortical layer development, neuron maturity etc. Microfluidic chips with micro-grooved channels are efficient in probing the axons independently of the cell body and dendrites [177,178]. The small width of the microgroove channels blocks the entrance of the cell body, while the dendrites are filtered out by the growth property variations that axons grow faster and longer. This platform allows for parallel studies of axonal transport with high-resolution live imaging systems, such as bright field, epifluorescence and confocal microscopy. The incorporation of microelectrode sensors on the substrate further expands its capability for monitoring electrophysiological activity [179]. The integration of brain organoids will require some modifications of the chip design, for example, enlarging the cell culture channel to allow for brain organoid attachment. Patch clamp electrophysiology (Figure 2Biv) has been applied to brain organoids to detect changes in postsynaptic firing, dopaminergic activity, and resting potentials and to verify neuron maturity [27,118,180,181].

Light-sheet microscopy is an example of innovative imaging applied to neuroscience. The large scale of light-sheet microscopy allows population recordings that extract neural network and pattern information from large swaths of neurons and smaller, more specialized populations within the image [195]. Lattice light-sheet microscopy (LLSM) uses Bessel-shaped rather than Gaussian light beams to reduce photobleaching and toxicity while maintaining high spatiotemporal resolution and allowing imaging of hundreds of planes per second [113]. This advancement has proved useful for visualizing systems from single cells to cell-cell and cell-matrix interactions to embryogenesis. As such, LLSM provides a new avenue for organoid imaging. Coupled with tissue-clearing techniques, The light-sheet microscope has proven to be a powerful tool for whole-mount 3D imaging of brain organoids [182–189]. Tissue-cleared brain organoids with genetic labels or exogenous staining could provide 3D spatial information such as cell type distributions, cell type population, cell-cell spatial correlation, and ventricular size (Figure 2Bv).

LLSM also allows for fast three-dimensional image acquisition of living tissues. LLSM coupled with adaptive optics has been employed to capture 4D images of endocytosis in intestinal organoids, demonstrating that LLSM can be applied to cerebral organoids for efficient and precise 4D imaging [111,196,197]. These new approaches have been recently applied to understand the impact of PTEN mutation on the macrocephaly phenotype of ASD patients [190] (Figure 2Bvi).

Multielectrode arrays (MEAs) can monitor synchronicity in neuron spiking via correlation of spike times (Figure 2C); emerging MEAs built with high-density electrodes and soft, flexible materials significantly improve the accessibility and sensitivity for brain organoid neuronal activity detection [191–195]. For example, Xiao et al. [194] developed basket-like flexible electronics to accommodate brain organoids or assembloids in suspension. The electronics enable long-term organoid integration and are compatible with microscopy.

Increasing computational power has been applied to studying neural cell behavior. Image-based profiling screens for cell phenotype at single-cell resolution can accurately predict the effects of molecules on cells for drug discovery applications [196]. The weaknesses of image-based cell morphological profiling, including low image quality and the complexity of neuronal arborization, have been mitigated by using an iterative back-tracking neuron reconstruction algorithm [197].

Machine learning (ML) and artificial intelligence (AI) can be incorporated into the HTS workflow to improve efficiency and accuracy [198–201]. ML/AI facilitates automated imaging-based phenotyping and enables large dataset handling. They can extract complex morphological feature vectors from 2D or 3D images [202,203], which is unnoticeable to humans. The granular profiling makes it possible to identify potential candidates. Extracted features are then clustered and scored to discover relationships of the perturbations. ML/AI can build predictive models that use existing data to predict the outcomes of new experiments, therefore optimizing the candidate selection. ML/AI has also been applied to sequencing-based techniques to assist data analysis, including annotation of cell clusters, sequence elements, epigenetic, proteomic or metabolomic data [204–207]. Generative AI has been successfully integrated into the scRNA-seq data analysis pipeline and exhibits remarkable accuracy for cell annotations and integration of multi-omic data [208,209].

6. Application of brain organoids in disease modeling

Brain organoids can capture certain aspects of the tissue architecture and cell type diversity of the human brain through multiple developmental stages. These features brought unique opportunities to recapitulate the disease phenotypes at multiple scales, including abnormalities ranging from impaired signaling pathways, defect cell type differentiation, and architectural deformation (Figure 3). Brain organoids successfully modeled volumetric and progenitors’ deficits in the developing human embryonic brain following exposure to the Zika virus, which could not be recapitulated by 2D models [4,25,210]. These studies provided the first direct human tissue-based evidence for a causal connection between the Zika virus and microcephaly. These models discovered that AXL inhibitors block the Zika receptor, which could potentially prevent Zika virus-induced microcephaly.

Figure 3.

Disease modeling with brain organoids recapitulates key phenotypic features at molecular, cellular, and tissue levels.

Empowered by the patient-derived hiPSCs, brain organoids have been used for modeling neural disorders (detailly reviewed by the reference [211]), including neural degenerative diseases, neurodevelopmental disorders, autism spectrum disorder (ASD) [212,213], lissencephaly [214], Rett Syndrome [108], Miller-Dieker Syndrome [214,215], schizophrenia [216], virus [4,25,210], and environmental risk [25,217,218]. Filtering the inter-individual genetic variation is a big challenge when using these models to decipher the molecular mechanism. A large sample size could help reduce the “background” noises but significantly increase the cost and workload. A balanced solution is to use gene-edited hPSCs for organoid generation, which is suitable for studying the loss of function of single gene mutations. Recent advances in CRISPR-cas9 toolboxes have facilitated long genomic sequence manipulation [219,220], presenting significant potential for modeling diseases with chromosomal abnormalities. Another solution is to use a multi-donor brain organoid model - Chimeroids [221]. The Chimeroids reaggregate neural stem cells or committed progenitor cells from multiple single donor organoids and co-develop into mature organoids for phenotyping instead of generating diseased brain organoids from each patient-derived iPSCs. The growth biases among donor cell lines are diminished, and each donor produces all cell lineages of the cerebral cortex. The Chimeroids preserve donor-specific differences in response to toxins and can be used as a scalable platform for personal medicine.

7. Conclusions

Integration of bioengineering and neurobiology will advance our understanding of the pathophysiological defects underlying neural disorders. They also represent a new, robust platform for disease modeling and drug screening. Although it may be daunting to effectively integrate these two fields to tackle these challenges, we have identified essential tasks that can help direct the progress, including identifying microenvironmental and genetic factors that direct long-term healthy neurodevelopment, optimizing parameters to recapitulate neurodevelopment in vitro and collecting/analyzing data from in vitro models and in vivo counterparts (Figure 4). As we move closer to creating a robust brain organoid model, this goal will encourage the close collaboration of neuroscientists and engineers to overcome the challenges in neural disorder diagnosis and treatment.

Figure 4. Integration of neurobiology and bioengineering to improve disease modeling and drug screening.

Induced pluripotent stem cell (iPSC) technology has advanced disease modeling and drug development in micromedicine and macromedicine based on cellular and molecular analyses of patients. iPSCs contribute to micromedicine through the analysis of individual patients and macro medicine through clinical trials for the categorization of multiple patients. Importantly, iPSCs could allow for more precise clinical trials by identifying subsets of diseased patients who optimally respond to the drugs of interest, thereby boosting success rates by pre-selecting these drug responders. Next Generation Sequencing can aid in identifying these genetic mutations/variants leading to differing patient responses.

Table 1.

Pros and Cons of ECMs in brain organoid generation

	Pros	Cons
Matrigel	• commercially available • validated application in brain organoid generation • easy to use	• derived from tumorous tissue • chemical composition not well-defined • xenogeneic • batch-batch variation residual growth factors
Decellularized ECM	• native brain acellular niche • commercially available	• limited access to healthy human brain • cross-species variations in ECM architecture and composition • tedious to extract • costly
Synthetic	• customizable chemical and physical properties • spatial-temporal control of cell-ECM interactions • well-defined composition	• require prior knowledge of chemistry, physics, material science and biology for synthesis • iterative trial and errors process for optimization

List of Abbreviations:

2D

two-dimensional

3D

three-dimensional

4D

four-dimensional

5-MeO-DMT

5-methoxy-N,N-dimethyltryptamine

AI

artificial intelligence

ASD

autism spectrum disorder

BP

bipolar disease

CNS

central nervous system

EB

embryonic body

ECM

extracellular matrix

ECs

endothelial cells

ESCs

embryonic stem cells

GABA

Gamma-aminobutyric acid

GCaMPs

genetically encoded calcium indicators

HA

hyaluronic acid

HD

Huntington’s disease

hPSC

human pluripotent stem cells

HUVECs

human umbilical vein endothelial cells

iPSC

induced pluripotent stem cells

KDM5B

Lysine Demethylase 5B

LLSM

lattice light-sheet microscopy

mDA

midbrain dopaminergic

MDS

Miller-Dieker syndrome

MEA

multielectrode array

mESC

mouse embryonic stem cells

Meso

mesodermal

ML

machine learning

NGS

next generation sequencing

PEG

poly (ethylene glycol)

PLGA

Poly-lactic-co-glycolic acid

PSC

pluripotent stem cells

PTEN

Phosphatase and tensin homolog

qHTS

quantitative high throughput screening

RGCs

radial glial cells

SARS-CoV2-i

severe acute respiratory syndrome coronavirus 2 infection

scRNA-seq

single-cell RNA sequencing

SCZ

schizophrenia

SD

Sandhoff disease

SHH

sonic hedgehog

SILAC

stable isotope labeling with amino acids in cell culture

SMAD

suppressor of mothers against decapentaplegic; family of transcription factors

sMDD

severe major depressive disorder

TS

Timothy syndrome

WNT

wingless-related integration site

ZIKV-i

zika virus infection