Human bodies cannot be studied directly for multiple reasons, including ethical and safety issues. To better study human diseases, human stem cells are a better alternative than animal-derived cells to build functional organ-on-chip models. This is more often the case in the fields of drug discovery and toxicity tests as there is a rather big gap between what happens in human bodies versus in animals. Previous research in sepsis showed that molecular mechanisms differ significantly between mice and humans, despite their similar disease phenotype (Seok et al., 2013). For brain-on-chip models, microfluidic chips with specific shapes have been proposed to culture various types of neural cells. The two main types of cells used in neural models are primary cells and stem cells (Grainger et al., 2018). Compared to stem cells, primary cells represent the organs inside the body better but are harder to obtain and proliferate slower. Also, primary neural cells are usually generated by animals, instead of humans. On the contrary, stem cells can reproduce themselves indefinitely and differentiate into any cell type in a suitable environment (Takahashi and Yamanaka, 2006; Takahashi et al., 2007). Stem cells, especially induced pluripotent stem cells (iPSC), are derived from human somatic cells and exhibit humanoid morphology. Disease models can be improved by using stem cells from patients without ethical considerations.

Glial cells also play an essential role in neural activity since they support neural growth and biochemical transportation. Previous research indicated that glial cells could be related to many neurodegenerative diseases (Miller et al., 2004). For instance, astrocytes, the most abundant glial cells in the brain, regulate blood flow in the blood–brain barrier to keep the central neural system stable. Meanwhile, they are also quite sensitive to pathological triggers like infection and stroke. Regarding brain-on-chip models, co-culture of glial cells with neural cells can provide neurons with a better microenvironment and promote axon growth. Ahn et al. proposed a blood–brain barrier model by endothelial cells, pericytes, and astrocytes, as shown in Figure 2A (Ahn et al., 2020). The chip had two microchannels, combining a 2D endothelial monolayer with a 3D brain microenvironment, which mimicked the in vivo environment very well. This model was a promising tool for testing neural drugs and studying the blood–brain barrier in both physiological and pathological conditions.

Culturing neural cells in vitro helps us observe the developmental process of neural cells and the connections between neurons. Given that the brain is much more complicated than artificial NoN on a chip, it is impossible to be fully reconstructed in vitro. In addition, various neural cells, such as neurons, astrocytes, and microglia, are present in the human brain and interact with each other to play important roles in brain activity (Goshi et al., 2020). Artificial neural circuits help us understand the fundamental mechanisms of brain function but fail to achieve high-order functions such as emotion, memory, and recognition. In this respect, brain slides from animals can be an excellent alternative to establishing neural disease models as their structures are much closer to the human brain than planar NoN.

The organ-on-chip model has become a hot topic in recent years as it mimics what happens inside the body to a great extent, with a low-cost and easy-to-fabricate process (Benam et al., 2015; Kim et al., 2015; Xu et al., 2016; Zhang et al., 2009). In contrast to traditional cells cultured in a petri dish, real organs are usually cultured in 3D structures. As such, scaffolds are required to culture the cells in 3D structures to mimic organs in the body (Murphy et al., 2017; Zhang et al., 2019). Hydrogel is one of the most popular scaffold materials in the 3D cell culture due to its excellent biocompatibility and softness. Additionally, the mechanical properties of hydrogen are similar to those of human organs under certain conditions (Pek et al., 2010; Zhu and Marchant, 2011; Mahoney and Anseth, 2006). Other materials like fibrous and solid porous are also popular for cell culture use for various applications (Hayman et al., 2005; Cao et al., 2009). Liu et al. proposed a coaxial flow-focusing capillary-assembled microfluidic device to fabricate spherical hydrogel beads with cells in them, as shown in Figure 2B (Liu et al., 2020). Alginate, one kind of hydrogel that could be solidified with calcium iron, was used as the shell of the sphere. Cells inside the spheres could grow well due to the permeability, biocompatibility, and softness of hydrogel. It was a promising tool that could be used to culture brain organoids. Some other hydrogels, such as matrigel and collagen, are mainly used to coat the surface of culture dishes or chips to promote cell adherence and growth.

As learned from the abovementioned messages, MEA is widely used to record signals from cells in a given small area. To fabricate MEA, gold is considered an ideal material because of its high conductivity and biocompatibility. For fluorescence immunoassay experiments that require high transparency of the substrate for better observation, indium tin oxide glass, as a combination of electrical conduction and optical transparency, is more often used in fabricating such microelectrodes. However, the relatively high impedance to obtain high transparency leads to high background noise and a low signal-to-noise ratio. Poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT: PSS) (Koutsouras et al., 2017; Kshirsagar et al., 2019; Susloparova et al., 2021) is a new interfacial material for modern bioelectronics and is better than ITO in some aspects, such as conductivity and flexibility. It is still facing some challenges like longevity and multichannel (Liang et al., 2021). As such, new materials have been developed in recent years to improve the electrodes, such as boron-doped diamond (McDonald et al., 2017) and graphene (Koerbitzer et al., 2016; Thunemann et al., 2018). Their excellent properties in better transparency and higher conductivity make them attractive for various applications in the near future.

For microfluidic chips, the most common material is polydimethylsiloxane (PDMS) because of its excellent elasticity, optical transparency, and biocompatibility (Wang et al., 2014). It also has high resolution on microstructure fabrication and has been widely used to build microchannels and microwells by molding. The basements are usually silicon wafers with microstructures on them. If the precision demand is not high, 3D-printed soft lithography is a good choice due to its low cost and ease of fabrication (Wu et al., 2017).

Micro-/nano-fabrication techniques such as photoetching and thin film deposition have been developing rapidly recently. Microfluidic systems are getting popular in the bioengineering field for neural degeneration disease modeling, neuroprotective mechanism modeling, and neural circuit establishment (Lassus et al., 2018). Single-cell manipulation can be achieved in the microfluidic chip, and microenvironment around cells can be controlled precisely, making microfluidic systems an up-and-coming tool to conduct neural study (Chen et al., 2011).

Commercially available planar MEAs are 8 × 8 square which meets the requirements of the majority of experiments for neural research (Furukawa et al., 2013; Rastegar et al., 2019; Gao et al., 2021b). However, 64 electrodes are not sufficient when studying a large-scale NoN study that involves complex connections between cells. Large-scale MEA recording systems have been developed as such to better study NoN over the past few years (Lott and Hoy, 2008; Laurent et al., 2009; Maccione et al., 2012). Due to the limitations on the size of microelectrodes and chips, it is infeasible to match each electrode with its own amplifier chain and digitizer. In this case, time-division multiplexing is widely used for high-throughput recordings (Berdondini et al., 2009). However, traditional time-division multiplexing has a low signal-to-noise ratio when applied to thousands of electrodes. David et al. proposed a CMOS electronics-based super large-scale MEA chip that contains 65,536 electrodes (Tsai et al., 2017). Compared to traditional methods, this large-scale MEA has high density, low noise, and, particularly, does not require antialiasing filters for each channel to overcome scaling limitations. Furthermore, thin-film-transistor (TFT) technology has been widely used in display fabrication and has also been used to fabricate MEAs in recent years (Shaik et al., 2017). Compared to the traditional homemade MEAs, although the circuit design could be more complex, it is easier to get and has stable quality. Some elaborate TFT electrode array chips can achieve cell manipulation via dielectrophoresis (Tixier‐Mita et al., 2019).

Noise could be an issue during the electrophysiology recordings since the electrical signals produced by cells are rather small. Except for the system noise, the majority of the noise comes from the change of the ion environment near electrodes (Hai et al., 2010; Mateus et al., 2019). Thus, increasing the contact area between cells and electrodes is a promising strategy to increase the signal-to-noise ratio. Abbott et al. proposed a nanoelectrode array coated with platinum-black, which not only increased the surface area but also lowered the impedance of electrodes (Abbott et al., 2020a; Abbott et al., 2020b). A cell–electrode interface model was also built, as shown in Figure 4C. Additionally, intracellular signals can be detected through membrane permeabilization, induced by a small faradaic current. In this way, tiny signals can be recorded, which is the key to investigating the synaptic connectivity between neurons. One individual electrode performs similarly to the patch clamp, but the entire electrode array is scalable, subminiaturized, and of high density, which is beyond the reach of the patch clamp. Cell stimulation can also be achieved in each electrode by a small current. The stable and precise recording makes network-wide mapping of neurons possible, which certainly promotes further research into the human brain ex vivo.

Not only electrical signals can be used for neural recording and stimulation, but optical signals and chemical signals can also represent the activities of neurons. Through optogenetic modification, inserting light-sensitive genes into cells, cells can be stimulated by different colors of light (Yoo et al., 2020). Meanwhile, neural activity can be reflected by fluorescence intensity in some aspects. In addition, the optogenetic technique has already been used in vivo several years ago to control the brains of mice, creating the most famous mouse that has been featured in countless top journals (Nabavi et al., 2014). The fundamental principle of chemical recording is that neurons release neurotransmitters when they communicate with each other. Various biosensors have been proposed to monitor these chemicals in real time (Su et al., 2020). Also, some chemicals can be used to stimulate neurons. Amanda et al. proposed a bioelectronic neural pixel to stimulate neurons by chemicals and record their activities simultaneously with PEDOT:PSS electrodes because it has both conductivity and permeability (Amanda et al., 2016). A cation exchange membrane was used to deliver chemicals, and gold was used to conduct electrical signals. This bioelectronic neural pixel is a potential platform to study the effects of different neurotransmitters on neurons.

Furthermore, culturing cells on MEAs for the long-term is essential for NoN development, as axons and connections between cells take time to grow. The major obstacle to long-term culture is infection from the environment, which can be minimized with the use of a clean environment and standard operation. Previous research demonstrated that hyperosmolality, caused by medium evaporation, is the second reason for neuron apoptosis (Potter and DeMarse, 2001). Many devices for long-term culture of neural cells have therefore been equipped with their own gas supply system or membrane-sealed chamber to avoid evaporation.

Kramer et al. fabricated a lollipop-shaped container using dual hydrogels to develop a myelinated peripheral nerve model, as shown in Figure 5A (Khoshakhlagh et al., 2018; Kramer et al., 2020). Neural cells and gliocytes are co-cultured in a spheroid microplate to form a cell sphere that will be moved to the circle part of the container. Axons will grow through the channel in a few weeks. In order to mimic the human brain to a great extent, researchers also applied human iPSC to build a peripheral nerve model (Sharma et al., 2019). It is a desirable platform to study neurodegenerative diseases and perform drug-related tests. After four weeks of culture, a patch clamp was used to monitor the electrophysiology of their nerve model, including nerve conduction velocity and amplitude, which are gold standards for evaluating chemotherapy-induced peripheral neuropathy (Brewer et al., 2016; Stagg et al., 2016). Meanwhile, the nerve model was challenged by six types of drugs with or without neuropathic properties, and the results were as expected, demonstrating its reliability in drug tests.

It is nearly impossible to decrypt the brain even with modern state-of-the-art technologies due to the complexity of our brain. In this case, guiding neurites by microtunnels between two isolated microfluidic regions provides an alternative way to study the NoNs (Jean-Michel et al., 2011; Renault et al., 2015; Samson et al., 2016a; Samson et al., 2016b; Renault et al., 2016). Such type of microfluidic device can be used to study synaptic plasticity, disease processes, and communication between different brain regions that are related to certain neural diseases. For example, in a generalized seizure, epileptic seizures spread to the whole brain, while in a focal seizure, the seizures stay within a certain small area (Fisher et al., 2017). The effects of the central nervous system on other tissues, such as muscles and blood vessels, can also be studied on the chip by guiding axons to germinate in the corresponding tissues through microchannels while preventing cell migration (Gilbert-Honick et al., 2020). However, the development of this technology is limited due to the complicated operations of soft lithography, which generally includes the pouring of liquid PDMS into a micro-structured wafer and tearing it off manually once solidified. Such a process can damage the microstructures and limit the design options, as structures in 3D and with large aspect ratios cannot be molded.

Three-dimensional-printed soft lithography provides a better option to fabricate complex microfluidic structures, as PDMS can be extruded everywhere on the chip with high resolution. Kajtez et al. proposed a prototype composed of several chambers which are connected by microtunnels (Kajtez et al., 2020). Gasket ink and compartment ink were developed for extrusion in a master mold to generate tall “walls” between chambers, as shown in Figure 5B. Human neural stem cells were cultured on the microfluidic chip to study Parkinson’s disease via modeling of nigrostriatal pathways by guiding the growth of dopaminergic projections through microchannels. Compared with traditional lithography methods, 3D-printed soft lithography has high precision and high resolution and greatly simplifies the fabrication steps. These advantages make it widely applied for printing the framework for cell growth and cell communication (Achille et al., 2021). Three-dimensional bioprinting was developed on the basis of 3D printing. Cells and scaffolds are generally made of polymers and extruded from a nozzle and cultured in 3D (Elalouf, 2021; Jamee et al., 2021). Many organs, such as bone (Arrigoni et al., 2017; Kim et al., 2021; Alcala-Orozco et al., 2022), lung (Hu et al., 2021), and skin (Gao et al., 2021c), have already been generated by 3D bioprinting, showing plenty of potential in the future. Cell delivery (Hwang et al., 2022) and tumor modeling (Samadian et al., 2021) can also be achieved by bioprinting. Today, 3D printing is still in its initial stage and requires much effort to improve, especially in cell co-culture technique, bioink properties, and fabrication methods. Nevertheless, the great benefits of flexible design and rapid prototyping will certainly ensure a bright future for 3D bioprinting in the coming years.

Except for building neuron models in microfluidic systems, the system can even perform like a neuron after elaborate design. Levi et al. proposed a microfluidic neuron that is similar to a biological neuron (Levi and Fujii, 2016). Each microfluidic neuron had a chamber representing an intracellular environment. Chambers were connected by microtunnels, representing axons that transmit signals. The exchange of irons was controlled by Quake valves and selective ion permeable membranes in microtunnels. Thus, the membrane potentials were formed and measured by integrated electrodes. It could be a promising tool to explore in neuromorphic engineering.

Large-scale MEAs can monitor the neural activities of the whole NoN, but it remains challenging to attain high resolution and to record potential changes in a single cell. Decreasing the size of the electrode is an alternative solution to achieve high-resolution recording. In this way, noise can also be reduced as the entire surface of the electrode is covered by cells, which avoids any potential negative effects from iron in solution. Inspired by the morphology of a synapse, Wijdenes et al. proposed a novel bio-mimicking and high-resolution planar MEAs for long-term neural recordings, as shown in Figure 6A (Wijdenes et al., 2016). Several electrodes were applied to record one single cell simultaneously, increasing the resolution by 14 times as compared to the traditional planar electrode. Additionally, the edge of each electrode was modified with gold nanostructures to increase the contact area between neurons and electrodes. This nano-edge structure prevented current leakage into the environment, which reduced the noise and obtained high-quality signals.

In order to further improve the signal quality and the sensitivity of MEAs, 3D microstructures can be modified on the top of the electrodes, such as nanostraws (VanDersarl et al., 2012), nanowires (Robinson et al., 2012), and nanopillars (Xie et al., 2012). The main advantage of 3D structured electrodes is that intracellular action potential can be detected, which reflects the information of ion channels and synapses of a single cell (Mechler et al., 2011). Although some of the aforementioned planar electrodes can record intracellular signals by electroporation (Abbott et al., 2020a; Buzsáki, 2004), they may disturb spontaneous cell activities at the same time as the membrane is damaged. Also, such electrodes require complicated circuit design and operation. In this case, 3D structured electrodes provide a harmless, simple way to measure high-quality exocellular and intracellular potential changes. Liu et al. generated a 3D high-density nanowire array to record both exocellular and intracellular signals (Liu et al., 2017a). Electrical circuit models were built, as shown in Figure 6B. Planar circuits were made of Ni and patterned by photolithography and electron beam lithography on top of a transparent sapphire substrate. A Ni mask and plasma etch technology were used to generate the Si nanowire with a proper height of 10 µm and SiO₂ being deposited as the insulating layer. Human iPSC were cultured on the nanowire and induced into neurons. Nanowire electrodes under the soma entered into cells during the growth phase to record intracellular signals; electrodes near neurites were able to record exocellular signals. Casanova et al. fabricated nanowire probes on a chip through a CMOS-compatible large-scale fabrication process based on conventional lithography tools (Casanova et al., 2018). It could achieve a high surface-to-volume ratio and have a large signal-to-noise ratio. Compared to planar electrodes, 3D-empowered electrodes led to improved signal quality with reduced background noise and increased signal amplitude.

Gu et al. designed an elaborate scalable field-effect transistor (FET) array that can switch between 2D and 3D as shown in Figure 6C (Gu et al., 2021). The process of transformation is accomplished by compressive buckling technique, a pre-strained elastomer substrate was used to buckle the 2D electrode into 3D. It was able to record the signals both from planar NoN and 3D tissues. Due to the fact that the distance between two electrodes was smaller than the length of cardiomyocytes, it could also record the intercellular signal conduction. They found that the conduction velocity of intracellular signals in cardiomyocytes is about five times the conduction velocity of intercellular signals. It is a very promising platform to study the signal conduction pathways and screen drugs.

Despite the limited damage of 3D electrodes on cells, some effects may still happen to cell electrophysiology. Mateus et al. proposed a mushroom-liked 3D electrode array to further improve electrophysiology recordings (Mateus et al., 2019). The mushrooms were wrapped in cells so that, on the one hand, signal quality was much higher than that of planar electrodes; on the other hand, only very limited noise was obtained from the environment since all the surface of the electrode was covered by cells. Despite their capability of measuring exocellular signals only, 3D electrodes were totally non-invasive and performed better than most planar electrodes. Topography effects on neural cells were demonstrated in this study as well. Besides, this chip is compatible with commercial MEA data acquisition systems, which shows excellent application value.

Synapse plays an important role in neural communication since it can transmit the excitatory impulses produced by the neuron body to other neurons or effectors (Lines et al., 2017). MEAs can record most electrical activities from somas, but signals from axons are hard to measure due to the limited diameters of axons. In this case, the microfluidic device is a powerful tool to enhance the recording of signals from neurites (Narula et al., 2017). Shimba et al. modified microtunnels on top of MEAs to limit the neurites elongation (Shimba et al., 2019). Laminin and poly-L-ornithine (PLO), which can promote the growth of neurons, were coated on the surface of tunnels to induce the neurites. Each tunnel contains a microelectrode inside to detect the neural activity form neurites, as illustrated in Figure 6D. Compared to neurons on conventional MEAs, neurons on this advanced chip achieve longer neurites elongation. Recording of neural activities was significantly enhanced, as a result of significantly increased number of neurites per electrode. Importantly, neurites in the microtunnels can sustain for more than 1 year, demonstrating high potential of neural long-term culture and monitoring.

Neurodegenerative disease models can be potentially established on the MEA-microfluidic combined chip, as it can record signals from neurons and manipulate cells simultaneously. Interestingly, as only neurites can be induced to go through the microtunnels, the combined chip is a powerful tool to study communication between neurons and body cells. For example, scientists have classified several kinds of epilepsy, and one of them is called focal epilepsy, indicating that only part of the brain, rather than the whole area of it, suffers from seizures (Shorvon, 2011; Fisher et al., 2017). In order to study the connection between brain regions with and without seizure, Anssi et al. designed a platform that contains three compartments with microtunnels to connect them (Pelkonen et al., 2020). The microtunnels prevented liquid exchange, and therefore liquids were isolated within their own compartment. Human iPSC were cultured in all three compartments and induced to neurons to represent different brain regions. Neurons in three compartments were connected by axons through microtunnels. The convulsant and kainic acid were added to one compartment to mimic focal epilepsy, while the other two compartments were not treated with any drug. In general, the combination of MEAs with a microfluidic device empowers MEAs with excellent performance and amazing functions.

Evidence has shown that hippocampal slices induced the onset of epilepsy spontaneously after several days of culture ex vivo (Dyhrfjeld-Johnsen et al., 2010). Jing et al. cultured the organotypic hippocampal slices in a petri dish to study the effects of the culture solution on spontaneous epileptogenesis (Liu et al., 2017b). A tungsten microelectrode was placed in a hippocampal slice to record extracellular field potential, and different culture solutions were added to the dish. Results showed that improvements in culture medium could prevent apoptosis of the brain to some extent, but seizures always happened. However, one electrode cannot detect the spikes in different parts of the hippocampal slice. Two years later, the author designed a microfluidic MEA-combined chip with simplified linear electrodes made by PDMS to better record the seizure of hippocampal slices (Liu et al., 2019). Compared to the traditional tissue slides on MEAs (Scott et al., 2013), this integrated chip has six culture wells on one plate to achieve high-throughput drug discovery and better control the environmental variants.

During the recording process on MEAs, thick brain slides (>500 µm) may face issues of lack of oxygen, nutrition, and waste removal. These issues prevent us from studying the chronic toxicity of drugs and the subsequent progression of epilepsy. To address them, Killian et al. designed a long-term perfusion and imaging microfluidic device on the basis of MEAs, as shown in Figure 7A (Killian et al., 2016). Perforated MEAs were used to record signals from brain slices and to perfuse the culture medium from the bottom. Meanwhile, culture solution was extracted, with a volume equal to perfusion, to keep the environment stable. Compared to culture in dish methods, the tissue slides survive much longer and facilitate long-term experiment conductance.

An organoid, a 3D cultured cell sphere, has characteristics between a mature brain and a planar NoN. Research demonstrated that neurons in 3D culture could have longer axons than in 2D culture, indicating that the function of neurons is limited in planar NoNs (Kong et al., 2021). In contrast to the real brain, which has blood vessels to provide nutrition for neurons, the organoid is made of neural cells only. Thus, one big issue in organoid culturing is the lack of oxygen in the center. Although some research shows that co-culture organoid with vasculature can promote the maturation and differentiation of neurons, the oxygen issue still remains (Isshiki et al., 2020). Thus, the organoid is much more complex than planar NoN due to its 3D structure but simpler than the human brain due to its limited functions. MEA was used to record the neural activities on the surface of organoid. The perfusion microfluidic devices previously introduced can also be used for culturing organoids and prolonging their survival time (Killian et al., 2016).

Although no technique is available yet to record the neural activities inside an organoid, 3D MEAs hold great potential in studying the organoid. Guihua et al. proposed a microneedle with five cellular-scale electrodes to detect glutamate concentration and cell activity in the brain simultaneously (Xiao et al., 2021). Five electrodes can be regarded as detecting one individual region of the brain because their positions are very close to each other. From the authors’ perspective, such a microneedle could also be inserted into the organoid to record both neural signals and glutamate concentration inside the brain. One new challenge would be how to insert the needle into the organoid without damaging cells. Micro-probe arrays are widely used in vivo to detect the cell activity of one particular region of the brain (Zhang et al., 2020). Motlagh et al. fabricated a high-density 3D pyramid-shaped MEA intended to record signals in our brain, as shown in Figure 7B (Ghane Motlagh et al., 2016a). In their follow-up study, polyethylene glycol, a biocompatible polymer, was coated on the surface of electrodes to improve the biocompatibility (Ghane-Motlagh et al., 2016b). If an organoid was cultured in these 3D MEAs and wrapped well around the needles, signals inside the 3D NoN could be tracked precisely and serve us better to understand more about the secrets of the human brain.

Organoids can also connect to each other through a robust fascicle consisting of axons (Kawada et al., 2017). It could be a platform for drug screening, even better than 2D NoN. Kirihara et al. proposed a microfluidic chip to build a cerebral tract model connecting two cortical regions by using hiPSC (Kirihara et al., 2019). The two spheroid chambers on the chip, each with a 2 mm diameter, were used for organoid culture, and they were connected by a microchannel. Axons could grow through the channel and connect two organoids, as shown in Figure 7C. Importantly, two organoids were electrically connected through the axon fascicle, which provides a platform for compound evaluation and drug screening.