Microfluidic high-throughput 3D cell culture

Non-clinical testing relies on in vitro cell culture systems with physiological relevance and high throughput^1,2, as well as animal models that mimic physiological conditions but pose challenges in terms of sample size and maintenance^3,4. Two-dimensional cell-based assays permit simple and large-scale experiments, such as toxicity assessments, but fall short in the comprehensive morphological and functional evaluation of organs or tissues. Alternatively, microfluidic-based 3D cell culture may delicately balance these considerations^5,6. Microfluidics-based systems can be engineered with diverse designs to generate valuable data on the morphological and functional properties of human organs. The biological composition of coculture conditions, physicochemical stimuli and biomolecule delivery allow the development of tissue- and organ-mimetics from human cells. Such systems enable high-throughput experiments under a range of optimized conditions^7,8.

‘High-throughput’ here refers to a quantitative approach to scientific research that involves generating and processing large amounts of data in a short timeframe⁹. Microfluidic cell culture systems can be manufactured to enable high-throughput, reproducible cell culture and mimic an in vivo 3D environment to generate reliable data for clinical, pharmaceutical and regulatory purposes. Such systems may be particularly useful for non-clinical drug testing, as supported by the FDA Modernization Act 2.0, which aims at reducing the use of animals in non-clinical testing if scientifically reliable in vitro models are available¹⁰. High-throughput microfluidic cell culture systems may thus aid in improving the preclinical prediction of clinical outcomes by providing a comprehensive and predictable map of cellular responses to drugs, thereby complementing current preclinical models¹¹.

In this Review, we outline the design of microfluidic cell culture devices, highlighting key prerequisites for specific applications, and examining future milestones in the field of high-throughput microfluidic cell culture systems.

Biological complexity can be increasingly implemented in cell culture studies in vitro; for example, the growing interest in immunotherapy is driving the exploration of the intricacies of the tumour immune microenvironment^12,13,14. In addition to characterizing morphological cellular aspects, functional properties of cells can be assessed in vitro, for example to study the behaviour of metastatic cancer cells traversing blood vessel walls or to assess the cytotoxic capacity of immune cells^{15,16,17,18,19,20}. Static 2D cell culture methods, including viability, proliferation and apoptosis assays, remain prevalent in preclinical studies owing to their scalability. However, they often fail to mimic in vivo biological complexity²¹. Alternatively, microfluidic 3D cell culture and coculture systems can be engineered in which human cells are arranged to mimic their in vivo environment more closely. In these systems, small channels and chambers allow control of the cellular environment by providing distinct chemical and mechanical stimuli. For example, the direction of fluid flow within the 3D extracellular matrix can be programmed to investigate the effect of interstitial flow on the angiogenic activities of vasculature²². Microfluidic systems also enable the culture of polarized epithelium at the air–liquid interface²³. Moreover, the stiffness modulus of the substrate or matrix can be modified to match the stiffness of a specific tissue²⁴.

However, throughput, customizability, manufacturability, reproducibility and design modifications remain challenging in 3D microfluidic cell culture systems. To overcome these limitations, processing methods such as laser processing and 3D printing can be applied to achieve designs in which cells can reconstitute in-vivo-like functions^25,26. Standardized microscale systems may further improve experimental throughput and reproducibility by integrating automation systems, such as liquid handlers^27,28 and high-content analysis systems^29,30,31. Such devices may offer a scalable, customizable and high-throughput platform for 3D cell culture studies that mimic in vivo biological complexity (Fig. 1a).

a, Timeline of the development of microfluidic 3D cell culture platforms. b, Different cell types, cell sources and materials can be used for microfluidic 3D cell culture platforms. Different cell formations can be patterned in biomaterials to achieve 3D structures. c, Microstructures, such as micropillar arrays, can be used for controlling cell and material location as well as fluid flow. Environmental conditions can be regulated through the integration of equipment that enables dynamic flow. In addition, automated procedures, including miniaturization and standardization, can be used. GF, growth factor; PDMS, polydimethylsiloxane.

Full size image

Rather than aiming at fully recapitulating the intricate complexity of human biology, the design of 3D cell culture devices should be based on their underlying purpose³². For example, the devices can be engineered to visualize events that are difficult to observe in vivo or to control specific cell phenotypes to investigate therapeutic interventions^33,34,35. In particular, 3D cell culture devices can serve as preclinical models for drug evaluation, from cell death assays to more complex phenotypic or functional tests (Box 1). However, developing disease models with a high level of biological complexity is crucial for understanding therapeutic mechanisms of action and advancing therapeutic development, which can be achieved by specific device and microfluidic designs as well as the appropriate choice of cellular culture³⁶.

Cell source

High-throughput 3D cell culture methods require efficient screening under multiple conditions. Primary cells offer valuable insights into patient-specific behaviour and response; however, their availability is limited, and their optimization for high-throughput applications can be challenging³⁷. By contrast, immortalized cell lines, such as established cancer cell lines derived from different tumour types, benefit from reproducibility and scalability, making them well suited for high-throughput screening in microfluidic systems. For example, cell lines can be sufficiently proliferated to study cancer biology and drug response at high throughput³⁸. Moreover, stem cells, including embryonic stem cells and induced pluripotent stem cells (iPS cells), enable the generation of patient-specific models to improve the biological relevance of high-throughput screening^39,40,41. Pluripotent stem cells have the ability to differentiate into various cell types and thus aid the investigation of developmental processes and tissue regeneration in a controlled microfluidic environment^42,43 (Fig. 1b).

Integration of high-throughput 3D cell culture methods with disease modelling enables the creation of reproducible assays, for example by coculturing patient-derived cells with specific genetic modifications⁴⁴. Such disease models may recapitulate patient-specific conditions, offering a powerful platform for studying disease mechanisms and evaluating therapeutic interventions. Personalized cellular models may further provide valuable insights into disease pathogenesis and expedite the development of targeted treatments. For example, pluripotent stem cells obtained from a donor can be differentiated into all major cell types forming the blood–brain barrier (BBB), including astrocytes, pericytes and brain microvascular endothelial cells⁴⁵, which can be cultured in a microfluidic platform. This personalized BBB model can provide data on electrical signalling and BBB permeability. Similarly, a microfluidic high-throughput system can be applied to study the effect of mesenchymal stromal cells from multiple patients on vascularization²⁹, enabling the analysis of the heterogeneity of patient-derived cells for cell-based therapies, based on specific gene expression and morphological characteristics.

However, the main priorities for fresh samples from a patient are typically histology and RNA analysis for diagnosis^46,47. Consequently, even with the use of microfluidics, in vitro modelling remains challenging owing to the limited number of available patient samples. In addition, the use of patient-derived cells requires standardized processing, careful selection of media for cocultures and appropriate cell-to-cell ratios. Therefore, the choice of cell source, ranging from immortalized cell lines to patient-derived cells, plays a pivotal role in achieving desired outcomes.

Organoids and cellular spheroids

The human microphysiological environment can be approached by using organoids or cellular spheroids (Fig. 1b). Organoids are 3D structures derived from stem cells or tissue progenitors that can partly replicate the cellular diversity and spatial organization of specific organs^47,48. These models may be applied to study organ development, disease pathology and drug responses^49,50. Cellular spheroids, compact aggregates forming spherical structures, lack the distinct tissue development seen in organoids. However, they are easier and more cost-effective to construct, making them valuable for 3D modelling as tissue-level objects⁵¹. Thus, organoids and cellular spheroids offer complementary approaches to mimic the microphysiological environment in vitro^52,53. In particular, their integration with microfluidics enables the recreation of physiological fluid flow and incorporation of vascularization to increase nutrient exchange and oxygenation^26,30,54. This dynamic environment not only improves the fidelity of the model but also aids studies on organ development and responses to stimuli.

Cellular spheroids or organoids can be cultured in microfluidic systems using separate microfluidic channels or compartments for each type of spheroid or organoid^30,55. These channels or compartments enable controlled flow of culture media or other fluids, creating dynamic microenvironments for spheroids or organoids²⁶. Furthermore, microfluidic valves or pumps can be incorporated for precise regulation of flow rates, allowing dynamic control of nutrient availability, oxygen concentration and other environmental cues⁵⁶. Automated systems or pneumatic controllers can be used to efficiently load spheroids or organoids into microfluidic channels in parallel or compartments. For example, to model the biological mechanisms of vascularization around tumour spheroids, endothelial cells and fibroblasts that make up the stromal environment can be cocultivated, mimicking angiogenesis in the tumour microenvironment. Such a platform can be used to study endothelial regulation of drug transport, as it allows endothelial cells to migrate into tumour spheroids and form vascularization^57,58. In addition, such microfluidic systems can be designed to enable the exchange of media or other fluids between channels or compartments, enabling communication and interaction between cocultured spheroids or organoids.

However, the integration of organoids into microfluidic systems presents several challenges. Diverse media are required for different organoids, as each organoid may have specific nutritional and growth factor requirements. To accommodate multiple reagents, microfluidic-based automation systems can be implemented⁵⁹. In addition, coculture conditioned media can be collected for scalable assays, such as cell viability assays, mass spectrometry, or protein detection and quantification (for example enzyme-linked immunosorbent assay (ELISA)). Moreover, cell viability and function may be impacted by dynamic flow conditions and cell confinement, making long-term culture of organoids in microfluidic systems challenging. Therefore, culture conditions need to be precisely controlled to preserve organoid integrity and function. For example, to achieve vascularization, vascular precursors can be derived from human pluripotent stem cells to recreate organotypic and developmentally coregulated blood vessels⁶⁰ for the formation of cerebral organoids and neurovascular cocultures. Interestingly, as assessed by neuronal nuclear antigen (marker for mature neurons) and doublecortin (marker for neurogenesis) staining, the developmental kinetics between cerebral organoids (devoid of vasculature) and cocultures (harbouring both vascular and cerebral organoids) differ, indicating an accelerated differentiation process in the cocultures. Here, culture conditions were optimized over a 30-day culture period to ensure organoid growth and differentiation, and integration of 3D-printed microfluidic cell culture channels allowed the establishment of a perfusable vascular network within the organoids.

A microfluidic organoid culture system that incorporates a brain extracellular matrix with brain-specific cues enables the reproducible development of brain organoids⁶¹ by providing dynamic flow, thereby offering a standardized protocol for neural development. This high-throughput platform allows 3D culture of organoids for up to 75 days. However, variability among organoids, both within and between batches, may negatively affect reproducibility.

Patient-derived cells

Patient-derived cells allow the investigation of patient-specific cellular responses, but sources are often limited^62,63. The use of microfluidic systems has the potential to enable the development of patient-specific models from limited samples as well as rapid quantitative assessments across multiple conditions, thereby aiding in personalized, informed decision-making within limited timelines. For example, iPS cells can be generated from the blood samples of patients using reprogramming techniques that convert adult cells into pluripotent stem cells. These iPS cells can then be differentiated into specific cell types, such as neurons, cardiomyocytes or hepatocytes, and cultured in microfluidic devices to create patient-specific disease models and drug screening platforms. This approach enables the development of personalized models using cells directly derived from blood samples^45,64,65,66. Similarly, ascitic fluid, which accumulates in the peritoneal cavity under certain conditions, such as cancer or liver disease, contains tumour cells and other cells. Ascites is a valuable source for isolating cells that can then be cultured in microfluidic devices to generate patient-specific models, in particular, for studying tumour biology and drug responses, and for identifying personalized therapies^67,68. Ascites is obtainable in high quantities, compared with other samples, and can also contain a diverse range of components, from tumour cells to exosomes. Notably, exosomes derived from ascites of people with cancer harbour genetic mutation information and exhibit specific homing to tumour cells from the same donor in microfluidic settings⁶⁹.

Liquid biopsy is a non-invasive method for obtaining patient-derived samples, such as circulating tumour cells, cell-free DNA and exosomes, from blood or other body fluids. These samples can be collected through blood draws and processed for downstream use in microfluidic cell cultures. Microfluidic devices can be designed to capture and culture these circulating cells or genetic material, allowing the development of patient-specific models for cancer diagnosis, monitoring of treatment responses and investigation of disease progression^70,71.

Endoscopy is a minimally invasive procedure that allows the visualization and sampling of tissues from various organs, such as the gastrointestinal tract, respiratory tract or urinary tract⁷². These tissue samples are collected during endoscopic procedures and can be used for microfluidic cell culture for the design of patient-specific models to study organ-specific diseases, drug responses and personalized therapies. In addition, microfluidic cell culture methods can be applied for other types of patient-derived samples, such as biopsy specimens, fine needle aspirates or surgical resections^73,74.

Extracellular matrix

The hydrogel material used in microfluidic cell culture devices determines the 3D cell culture environment. Hydrogel materials differ by biocompatibility, mechanical properties and mimicry of the composition and structure of the extracellular matrix (ECM)⁷⁵ (Fig. 1b). For example, collagen and gelatin hydrogels are biocompatible and similar to native ECM, making them suitable for conditions in which cell–matrix interactions are crucial. Alternatively, alginate hydrogels offer adjustable mechanical properties and easy crosslinking, for example to screen the impact of mechanical signals on cell behaviour. In addition to natural hydrogels, synthetic polymers, such as polyethylene glycol and polyacrylates, can be used in microfluidic-based 3D cell cultures^76,77,78. These synthetic hydrogels typically have adjustable mechanical properties and benefit from high stability, and the ability to integrate diverse functional groups, thereby granting precise control over cell–matrix interactions. However, they may have to be modified to improve cell adhesion or include biochemical cues⁷⁹.

In microfluidic platforms, ECM hydrogels can be aligned along defined microstructures, allowing the creation of diverse 3D scaffolds. Beyond providing structural support, ECM hydrogels can instruct cellular function and morphology to create organ and tissue models^80,81. Unlike in tissue engineering, hydrogels in microfluidics exhibit a distinct conditional gelation process at small volumes, requiring control in the presence of cells. For example, fibrin gelation relies on thrombin-induced enzymatic cleavage of fibrinogen, leading to the formation of fibrin monomers that undergo spontaneous polymerization⁸². By contrast, physiological pH and temperature conditions are often sufficient for collagen gelation, although specific conditions may vary based on the collagen source and type⁸¹. Matrigel, a complex matrix derived from mouse sarcoma cells, undergoes gelation through temperature-dependent interactions of extracellular matrix proteins, creating a 3D matrix⁸³. Alginate gelation is induced by calcium ions forming crosslinks between alginate chains⁸⁴. Importantly, suspending cells in a hydrogel may alter its physical properties, which should be considered when choosing the hydrogel⁸⁵. Rapid gelation can be a hurdle to achieving experimental reproducibility; for example, uniformly filling the cell culture channel may be difficult^86,87. Open microfluidics can be used to induce rapid fluid behaviour, such as spontaneous capillary flow^88,89, to form 3D scaffolds in precisely defined regions before transitioning to a gel state. This spontaneous process does not require intervention and can be automated, improving throughput and reproducibility^88,90,91.

Hydrogels can also be modified with various cues to regulate cell behaviour in high-throughput 3D cell culture techniques. Incorporating cell adhesion ligands, such as RGD peptides, can promote cell attachment and spreading⁹² and thus alter cell responses. Additionally, the introduction of growth factors, such as cytokines and growth factor mimetics, allows precise control over cell actions, including proliferation, differentiation or migration^93,94. Mechanical signals, such as stiffness gradients or shear stress, can also be introduced to study the impact of mechanical forces on cell behaviour in a high-throughput manner⁹⁵. Beyond serving as a structural support for 3D cell culture, the ECM plays a crucial role in facilitating cell attachment to reconstruct membrane structures, such as the endothelium or epithelium. In microfluidic devices, tissue interfaces can be established by separating two or more microchannels with a porous membrane, typically composed of synthetic and biologically inert materials, which can be coated with ECM molecules, such as collagen type I and fibronectin, to increase cell adhesion^96,97,98. For example, in a placenta model, coating of the hydrogel interface with fibronectin enables the adhesion of endothelial cells and extravillous trophoblasts⁹⁹. To preserve ECM hydrogel functionality during storage, the hydrogel can be coated with polydopamine to maintain ECM and polymer surface functionality¹⁰⁰.

Incubation milieu

Microenvironmental control is crucial to ensure the reproducibility and reliability of experimental results. Microfabrication techniques allow the integration of sensors and actuators into the microfluidic system, enabling real-time monitoring and control of key parameters¹⁰¹. Microfluidic valves and pumps can be used to precisely regulate the flow rates of culture media or other fluids, creating dynamic environments that can mimic physiological conditions^102,103. For example, a dynamic culture environment is important to maintain the self-renewal capacity and multilineage differentiation potential of single haematopoietic stem cells¹⁰³ in microfluidic cell culture arrays, thereby enabling long-term culture. Thus, dynamic culture environments may preserve the properties of stem cells and prevent cellular senescence or spontaneous differentiation.

In addition to using pump-driven perfusion flow to improve the physiological relevance of microfluidic 3D cell cultures, hydrostatic perfusion flow can be used. For example, reservoirs symmetrically positioned from a porous hydrogel with embedded cells can be filled with culture fluid at different heights. Culture flows through the hydrogel, then reaches an equilibrium. This approach can be applied to mimic physiologically relevant interstitial flow, for example to trigger an angiogenic response of endothelial cells²². Similarly, the sprouting morphogenesis of lymphatic endothelial cells can be recreated through flow that is generated by a volume difference in a prolymphangiogenic factor cocktail¹⁰⁴; here, the cocktail, comprising essential components, such as vascular endothelial growth factor-A (VEGF-A), VEGF-C, basic fibroblast growth factor and sphingosine-1-phosphate, is stimulated by hydrostatic pressure-driven flow, which is directed to aligned lymphatic endothelial cells, inducing sprouting morphogenesis and shedding light on the mechanisms underlying lymphatic vessel formation.

Temperature control elements, such as heaters or thermoelectric devices, can be incorporated into microfluidic systems to maintain a constant temperature, which is essential for maintaining cell viability and functionality. Cellular behaviour is also affected by changes in pH, and therefore pH sensors and controllers can be implemented to monitor and adjust the pH of the culture medium¹⁰⁵. In addition, gas-permeable membranes or microchannels can be designed to regulate gas concentrations, such as oxygen and carbon dioxide levels, to replicate physiologically relevant conditions^106,107.

Efficient and scalable experimentation in high-throughput 3D cell culture models necessitates automation and parallelization of microenvironmental control. Microfluidic systems allow automated handling and manipulation of cells, hydrogels and culture media by incorporating robotic systems or pneumatic controllers to streamline cell seeding, hydrogel fabrication and culture medium exchange. In particular, pumpless models eliminate the need for external pumps, allowing simultaneous high-throughput processing of multiple samples^108,109,110 (Fig. 1c).

The scalability and throughput of 3D cell culture platforms can be improved by device design, fabrication methods and implementation of automation as well as data collection and analysis (Table 1). In particular, the cell culture platform, in which experiments are conducted and samples are generated for analysis, should be carefully designed to improve throughput, including arrangement of cells, incorporation of physiological factors, control of the environment, and compatibility with automated equipment.

Physiological parameters

Microfluidic 3D cell culture systems can incorporate the physiological characteristics of an organ or tissue, such as mechanical, biochemical or electrical parameters.

Mechanical parameters

The breathing motions of the lung can be modelled in a lung-on-a chip platform¹¹¹ (Fig. 2a) by stretching a porous membrane on which lung epithelial cells and microvascular endothelial cells are cultured on either side, thereby mimicking the expansion–contraction motion of lung alveoli. Stretching is achieved by applying vacuum to channels on both sides of the microchannel, in which the membrane vertically divides the endothelial and epithelial layers. The applied strain levels, ranging from 5% to 15%, closely match the normal strain observed in alveoli within the whole lung in vivo. A small airway reconstructed with microfluidics demonstrates the behaviour of primary human neutrophils in response to inflammatory factors under physiological flow and shear stress (1 dyn cm⁻²)¹¹².

a, Physiological parameters can be integrated in microfluidic 3D cell culture platforms, including mechanical parameters, such as periodic stretching motion on a porous membrane¹¹¹, fluid flow^{113,114,208,209}, interstitial flow^115,116 and shear stress, mimicking the blinking motion of the eyes¹¹⁷; biochemical dynamics, including cell–cell interactions^119,120 and biochemical gradients²⁰; and electrical components, such as electrical stimulation^123,124, transendothelial electrical resistance¹²⁶ and measurement of electrical signalling between neuron cells^127,128. b, Various levels of 3D blood vessel models. In monoculture and coculture models, micropillar arrays or microbump structures can be used to separate channels from adjacent channels^22,135. High-throughput models can be produced by injection moulding⁶. Rail structures exploiting spontaneous capillary flow can increase experimental throughput¹³⁷. A tumour spheroid can be loaded on a capillary network in a polydimethylsiloxane (PDMS)-based device⁵⁷. A microfluidic channel design integrated with a 384-well plate allows the generation of 128 vascularized organoids in a compact layout³⁰. Multiorgan platforms can be designed by connecting cell cultures from different organs with microfluidic channels¹³⁹. In addition, endothelial barriers can be integrated to study crosstalk between cells mediated by blood vessels¹⁴⁰.

Full size image

Flow generators in microfluidic devices allow stem cell maturation towards a specific subtype¹¹³ and organoid maturation^26,114. For example, flow can be generated through a hydrogel block that partitions a microfluidic device into multiple channels^115,116. Interestingly, the fluid force of interstitial fluid can modulate endothelial sprouting¹¹⁶, and biochemical signalling mediated by interstitial fluid can regulate the angiogenic response of endothelial cells²². The mechanics of the human ocular surface can be modelled¹¹⁷ with a hydrogel eyelid that moves back and forth as tear fluid is dispensed over a 3D dome-shaped ocular surface, which consists of corneal and conjunctival epithelial cells. The pressure gradient exerted by this engineered eyelid (approximately 1.9 kPa) is within the physiologic range (0.3–7 kPa) of upper eyelid pressure. A miniaturized cardiac scaffold can mimic the formation and cyclic contraction of a ventricular chamber¹¹⁸. Integrated with sensitive miniaturized valves, this microfluidic chip can recapitulate ventricular fluidic function, demonstrating a complete pressure–volume loop, thereby mimicking the physiological behaviour of the heart. Importantly, scaffold structures can be tuned for biological relevance by adapting spatial organization.

Biochemical parameters

The biochemical properties of human tissues can be replicated as crosstalk between different cell types in a 3D environment and as a gradient of chemical cues, oxygen or nutrients (Fig. 2a). In microfluidic platforms, cell–cell interactions can be investigated by positioning different cell types in discrete but connected hydrogel blocks. For example, a 3D angiogenesis model can be built using a micropillar array²² that separates the middle channel, which contains an acellular fibrin gel, from the side channel, which contains a fibroblast-containing fibrin gel in which fibroblasts can release growth factors. The micropillar array is attached to endothelial cells at the far side of an acellular matrix. The endothelial cells then form lumenized sprouts towards fibroblasts, following the direction of the released growth factors. Here, dynamic conditions can be achieved either by integrating an external pump or by using hydrostatic pressure¹¹⁹. Similarly, a bump structure, working as a capillary valve, enables the coculture of endothelial cells with astrocytes and pericytes, resulting in tight endothelial cell junctions that recapitulate the blood–brain barrier¹²⁰.

By controlling cell configuration or chemical cues, chemical gradients can be recreated. For example, the hypoxic environment in the core of a solid tumour can be reconstituted in a 3D cell culture device consisting of a lumen lined with endothelial cells and a long cancer cell region (15 mm)²⁰, which includes a gradient of nutrients, waste and O₂ from the lumen to its distal region. This microfluidic tumour-on-a-chip model allows the evaluation of the role of tumour environmental stress on natural killer cell exhaustion. A radial chemokine gradient can be generated in a donut-shaped device to trigger radial angiogenesis, mimicking the radial cornea or retina¹²¹. Both radial and one-directional chemical gradients can be genererated¹²² by incorporating a middle rail with a through-hole at the centre and higher rails on either side of the middle rail. Tumour spheroids loaded into the through-hole in the middle rail then generate a radial chemical gradient to recruit microvessels from the surrounding vessel network. This device can also be used for modelling angiogenesis by attaching endothelial cells to the side of the acellular hydrogel matrix underneath the middle rail and positioning cells underneath the opposite side of the rail to evaluate angiogenic capacity. This cell layout generates a unidirectional chemical gradient, aiding the comparison of the length and number of angiogenic sprouts.

Electrical parameters

Microfluidic cell culture devices also allow the investigation of tissue responses to electrical stimulation (Fig. 2a). For example, electrical signalling plays a crucial role in the maturation of heart tissue¹²³. A 3D heart tissue model can thus promote tissue maturation through electrical stimulation¹²⁴. Here, a hydrogel solution with cardiomyocytes is casted using a mould that contains carbon rods. A second mould with two carbon wires is then guided through the hydrogel, allowing the wires to be linked at either end¹²⁵. This wire design not only provides electrical stimulation for tissue maturation but also enables non-invasive measurement of Ca²⁺ transients and force of-action. Electrode-integrated devices can also be used to measure electrical signalling to analyse key functions of tissues, such as transendothelial electrical resistance and electrical signalling of neural circuits. For example, electrodes can be integrated in the top and bottom channels of a microfluidic device to measure the electrical resistance of an epithelial layer cultured on a porous membrane sandwiched between the two channels¹²⁶. This device can mimic the blood–brain barrier and has been applied to verify enhanced barrier function in hypoxic condition⁶⁶. Similarly, nerve cells can be grown on electrode arrays to study electrical signalling of neural networks; for example, the soma and axon can be compartmentalized in polydimethylsiloxane (PDMS) channels to mimic unidirectional signalling^127,128.

Cell configurations

Cell patterning methods

Cells can be patterned in physiologically relevant layouts and spatially separated to study their chemical interactions. In particular, microstructures can be integrated in microfluidic devices to guide cells to their intended locations. Importantly, the patterning strategy affects experimental throughput. For example, gaps smaller than a typical cell can be used to confine cells in a channel, for example to place hepatic cells in a channel between adjacent nutrient channels, which allows loading of hepatic cells in high concentrations through draining of media to the nutrient channel¹²⁹. The narrow gap also serves as an endothelial-like barrier in sinusoid space, aiding the diffusion of nutrients. Similarly, a narrow channel preventing the soma of a neuron from passing through the channel allows the axon to grow, resulting in alignment of neurons in a microfluidic device to enable biochemical analysis of axonal fractions¹³⁰. Size-based confinement can also be achieved using porous membranes; for example, to mimic the alveolar–capillary interface, a channel can be separated into two channels by a porous membrane with a pore size smaller than the cell size. Endothelial cells grown on one side of the membrane and epithelial cells grown on the other side of the membrane can then establish tight junctions, while their transmigration through the membrane is prevented.

In addition, hydrogels containing cells can be patterned in 3D within microfluidic devices. Microstructures, such as micropillar arrays, bump structures and rail structures, can be introduced to position multiple types of cells in 3D hydrogel blocks. By implementing hydrophobic surfaces, capillary pressure valves can be designed, as a higher capillary pressure is required for liquids to proceed over a narrow gap with a hydrophobic surface than over a wide gap. This prevents hydrogel fluid leakage through the gap until the pressure reaches the maximum capillary pressure in the gap^90,131,132. When fluid is injected into a microfluidic channel that has micropillar arrays with narrow gaps or a bump structure forming a narrow gap with a bottom surface, the fluid proceeds through the channel; however, fluid fronts at narrow gaps do not overflow into the adjacent channels. Such structures, known as capillary burst valves, can guide hydrogels containing cells to specific channels, while enabling chemical crosstalk between cells.

Cells can also be arranged in a high-throughput manner using microfluidic devices with spontaneous capillary flow along narrow gaps with hydrophilic surfaces. For example, a rail structure can be added to the well walls of microwell plates (with 24, 96 or 384 wells), positioned hundreds of micrometres away from the bottom surface. A hydrogel solution dispensed in the wells then flows spontaneously along the corner of the well and into the region underneath the rail structure^89,91. This layout is called open microfluidics, because both sides of the hydrogel under the rail are exposed to air¹³³. The same approach can be applied to pattern multiple cell-containing hydrogels by adding rail structures at different distances from the bottom surface. Hydrogel solutions preferentially fill narrower gaps owing to capillary force, and therefore the first injected hydrogel is located underneath the lower rails. Following crosslinking of the first hydrogel, a second hydrogel solution fills the higher rails without invading the first hydrogel region. This method is based on spontaneous capillary flow and thus does not require an injection port or precise injection into a port. The patterning of the hydrogel solution ensures consistency, mitigating experimenter error and enhancing throughput, regardless of the dispensing location within the well.

Experimental throughput and biological complexity

Biological complexity can be increasingly implemented in microfluidic cell culture platforms, without compromising throughput, thereby providing an alternative to animal models for certain applications¹³⁴. For example, an angiogenesis model can be designed in a microfluidic PDMS device (Fig. 2b); here, a hydrogel solution is injected into the middle channel, followed by culture of human umbilical vein endothelial cells on the hydrogel⁸⁶. Micropillar arrays between the channels guide the hydrogel solution to fill the middle channel without leaking into the adjacent channels. The same channel layout can be applied for the coculture of endothelial cells and cancer cells to mimic a vascularized tumour microenvironment¹³⁵.

High throughput can be achieved using injection moulding^6,136 and by designing a well-plate format to allow automated liquid dispensing as well as readouts by plate readers and motorized imaging systems. However, hydrogel patterning mediated by micropillar arrays is sensitive to injection pressure, and therefore the hydrogel solution needs to be gently and manually injected, resulting in low experimental throughput. To increase throughput, rail structures with hydrophilic surfaces can be applied^91,121; here, the capillarity of the hydrophilic surfaces causes the hydrogel solution to fill the area underneath the rail structure. Such capillary flow notably reduces limitations associated with the dispensing position and pressure, thereby improving operational efficiency. In this approach, sequentially dispensed hydrogels fill the rail structure at different heights (with the lower height difference from the bottom being filled first), allowing hydrogels embedded with different cells to be aligned side by side. The assembled hydrogels allow a variety of cell coculture models, which is useful for engineering vascularized tissues, such as cancer vascular models¹³⁷.

In addition, spheroids or organoids can be integrated into such vascular models. Following the generation of a 3D vascular network in a microfluidic device with a rail structure, a preformed cell spheroid or organoid can be loaded onto a through-hole^55,57, allowing positioning of the vascular network and spheroid in the same layout in a high-throughput fashion¹³⁸. Similarly, a microfluidic platform that contains a PDMS layer to integrate microchannels in a 384-well plate can be applied for generating vascularized organoids³⁰. Although low-throughput, this platform can generate 128 vascularized organoids in a single plate.

Multiorgan interactions can be modelled in devices that connect cell cultures from different organs in separate microchannels¹³⁹. For example, a plug-and-play multiorgan chip, which is the size of a microscope slide, allows four-organ vascularization using 3D printing¹⁴⁰. In addition, endothelial-cell-lined channels or endothelial cell layers on transwell structures can be introduced to study cellular crosstalk via blood-vessel-mediated circulation using robotic fluid control²⁸.

Materials and fabrication

Reproducible manufacturing of high-throughput cell culture platforms is important for their use as non-clinical platforms. Microfluidic platforms are typically made of PDMS, which is biocompatible, chemically inert, gas-permeable and transparent¹⁴¹. PDMS-based microfluidics are typically fabricated by soft lithography, which allows the design of various microstructures to control fluid behaviour, but this approach is limited by low processing times and low degrees of freedom in manufacturing¹⁴². Alternatively, non-ablative direct laser processing²⁵ enables the creation of 2D and 3D PDMS structures starting from a PDMS monolith, substantially reducing processing times. This approach allows the iterative optimization of biology-driven designs by seeding cells one layer at a time; for example, stratified squamous epithelia can be seeded this way for the engineering of skin models.

3D printing is a flexible, fast and precise method for fabricating microfluidic cell culture devices, allowing rapid prototyping, customization and integration of multiple components, as well as the use of various biomaterials¹⁴³. It often requires surface treatment to aid the release of PDMS from the 3D-printed mould. For example, a self-assembled monolayer (SAM) coating¹⁴⁴ can be applied: here, molecular layers spontaneously form on surfaces through chemical interactions and exhibit low surface energies and anti-adhesive properties. The cleaned 3D-printed mould is first exposed to a silane vapour, such as trichloro (1H, 1H, 2H and 2H-perfluoro-octyl) silane, until the silane molecules bind to the surface and form a monolayer, aiding the detachment of PDMS¹⁴⁵. Alternatively, the unmodified 3D-printed surface can be used for cell culture; however, this requires the deposition of a thin layer of parylene to make the surface compatible with cell culture^146,147,148. The absorption of small molecules by PDMS remains a major issue¹⁴⁹, preventing the analysis of small-molecule concentrations in PDMS-based devices.

Injection moulding, typically applied for plastic components, also allows the production of microfluidic cell culture devices^150,151. Injection moulding involves the injection of molten plastic into a cavity formed by two metal moulds. The plastic is then allowed to cool and ejected as a solid part, resulting in the final device. However, fabricating microscopic features smaller than 0.5 mm remains challenging. 3D printing can be used to optimize design parameters for subsequent injection moulding, which enables the manufacturing of microfluidic cell culture devices with high precision and reproducibility. Notably, injection moulding can be used with cell-compatible materials, such as polystyrene¹⁵², and materials with transparent optical properties, such as cyclic olefin copolymers¹⁵³. Importantly, these materials show only minimal hydrophobic small-molecule absorption. Injection moulding further enables the production of devices with volumes ranging from 500 to 30,000 pieces, mould costs ranging from US$3,000 to US$4,000, and minimal device-to-device variability¹⁵⁰. Consequently, the integration of 3D printing and injection moulding shows great potential for the large-scale production of microfluidic cell culture devices.

Scalability and standardization

PDMS-based 3D cell culture systems typically possess a flat shape, with dimensions of around 3–5 cm in side length and a height below 1 cm, because of the limitations of SU-8-based photolithography and the requirements for efficient injection into microfluidic channels. Although this configuration allows the incorporation of additional components, it also results in large devices, affecting tissue function and experimental analysis. Large devices necessitate large incubators and assay equipment, introducing logistical challenges. Further miniaturization may allow the design of arrays with a higher number of devices. In addition, microfluidic designs may be engineered that conform to standards set by the Society for Biomolecular Screening¹⁵⁴ to ensure compatibility with automated equipment, such as automated liquid handling devices¹⁵⁵, transepithelial electrical resistance measurement instruments^28,156, plate readers and automatic imaging systems¹³⁷. Ultimately, fully automated systems, originally designed for 2D in vitro models, may also be achieved for microfluidic 3D cell culture devices.

Microfluidic 3D cell culture devices may serve as non-clinical screening models for drug development. However, device and model design, quality control and readout quantification will need to be standardized^2,157. For example, common equipment, such as well plates, should be used to harmonize devices across laboratories and industry. Plate readers, automated dispensers, high-content screening equipment and dynamic incubation shakers conform to a common format, emphasizing the importance of considering compatibility of microfluidic designs with existing analytical instruments. The ultimate goal is to create high-throughput cell culture platforms that balance compatibility, standardization and flexibility¹⁵⁸.

Automation

Cell seeding, media exchange and maintenance of physiological parameters limit throughput in microfluidic 3D cell culture devices. In particular, the injection of cell-containing hydrogels is time-sensitive, because the physical parameters of hydrogels change during crosslinking. Hydrogels are also sensitive to temperature and humidity. Therefore, robotic hydrogel dispensing in controlled environments may improve reproducibility and yield of cell seeding. Bioprinting allows precise hydrogel dispensing and environmental control, for example, to seed cells in a microfluidic device to engineer an airway model¹⁵⁹. In this platform, organoid arrays can be printed in standardized well plates and the entire process can be automated, including hydrogel equilibration, cell seeding, media replacement and drug exposure, to enable high-content image-based phenotyping of patient-derived tumours⁵.

To model tissue function, mechanical or chemical stimulation as well as media circulation and exchange is required. For example, hydraulic pressure can be applied by filling culture media in pipette tips inserted into the port of the device to generate flow^160,161. The pressure decreases as the medium flows into the tips with lower hydraulic level; thus, medium needs to be added to generate continuous flow. Alternatively, pump systems or automatic liquid handlers can be used, in particular for multiorgan models. In such models, chemical components shared between organ models need to be scaled to account for organ size. Automatic fluid delivery systems can be applied to distribute the flow from one organ model to another^{27,28,162,163}; for example, a cell culture module that generates cyclic stretching motion, media flow and replacement has been commercialized for lung and gut models¹⁶⁴. Such automated processes can improve experimental throughput and reproducibility.

Large-scale data acquisition

In high-throughput cell culture models, a number of conditions can be tested and various types of data can be generated in a single experiment (Table 2). Various imaging techniques, such as fluorescence, confocal and live-cell microscopy, provide dynamic visual information about cell dynamics, morphology, mobility and important functional markers in real time¹¹¹. Additionally, biosensors and multielectrode arrays can continuously monitor cellular parameters and electrophysiology, particularly in neurons¹⁶⁵ and cardiomyocytes^166,167. Bioluminescence and Förster resonance energy transfer allow real-time tracking of intracellular activities¹⁶⁸. However, some assessments require post-experimental cell extraction, such as mass spectrometry and liquid chromatography–mass spectrometry, which offer detailed molecular insights¹⁶⁹, and genomic, transcriptomic and epigenomic sequencing to extract genetic information¹⁷⁰.

These data acquisition approaches allow real-time testing across various conditions and personalized therapy^5,171. For example, the apoptotic response of individual patient-derived tumour organoids in response to drugs and treatment can be assessed in real time⁵, enabling rapid correlation with clinically relevant data. A number of data readouts can be obtained from a single microfluidic cell culture device, including multiomics, morphology and sensing data^172,173. However, interpreting complex biological responses solely from a single data source remains challenging.

High-throughput data analysis with deep learning

Interpreting multiomics readouts from microfluidic devices requires the combination of heterogeneous data sources. Early, intermediate and hierarchical integration methodologies have been developed to minimize the loss of information, control the complexity of data and avoid reliance on existing biological datasets in generating insights. Data integration strategies require the design of corresponding deep-learning-based network frameworks. Methodologies for integration serve the dual purpose of consolidating multiomics data and improving the detection of subtle dynamics within cellular microenvironments.

The integration of deep learning techniques with high-throughput data analysis is particularly relevant for imaging data; for example, based on the high-throughput analysis of images for cell segmentation, detection and morphology, DeepCE, which is based on a graph neural network model, can predict how drugs can alter gene expression by matching compounds to patients’ symptoms¹⁷⁴. Similarly, deep learning architectures, such as convolutional neural networks (CNNs), enable the extraction of informative indicators from imaging data. For example, artificial intelligence can be applied for image analysis of a bone-on-a-chip platform to monitor cell–cell interactions and the effects of drug treatments on bone cells¹⁷⁵. To avoid manual screening of organoids, a deep neural network can be applied to track organoids throughout the entire culture process at high throughput. This deep learning model provides an automated and accurate method for tracking organoid growth and morphology, which is essential for monitoring the development and responses of organoids in microfluidic cell culture systems¹⁷⁶.

Graph convolutional networks can be applied for the high-throughput analysis of 3D biomedical imaging data, for example to monitor angiogenesis. These networks can detect the mesoskeletal structure of microvasculature and can segment complex structures, such as vascularized tumour microenvironments¹⁷⁷. DeepVesselNet provides a deep learning architecture that can process 3D angiographic volumes for precise vessel segmentation, centreline prediction and bifurcation detection, which is crucial for analysing vascular structures in microfluidic devices¹⁷⁸. A simplified U-Net-based convolutional neural network can be used for blood vessel segmentation in retinal images, demonstrating robust performance across different databases¹⁷⁹. Generative adversarial networks enable label-free virtual staining of bright-field images, thereby offering a non-invasive approach for detecting and predicting iPS cell colony formation and the automated selection of iPS cell colonies from bright-field microscopic images¹³⁹. In addition, these networks allow virtual histological staining of unlabelled tissue autofluorescence images¹⁸⁰. Similarly, Angio-Net, using a conditional generative adversarial neural network model, integrates quantitative angiogenesis assays from high-throughput microfluidic cell culture platforms into its framework¹⁸¹. The deep convolutional neural network Deep-STORM provides superresolution image reconstruction from dense fields of overlapping emitters, enabling precise, fast and parameter-free analysis of cellular structures at the nanoscale¹⁸². Residual channel attention networks can restore and enhance volumetric time-lapse fluorescence microscopy data to improve the denoising and resolution of images, enabling the visualization and analysis of cellular processes without photobleaching over extended time periods¹⁸³.

Various algorithmic tools for image analysis, including ImageJ¹⁸⁴ and BioImage Model Zoo¹⁸⁵, can provide image processing and analysis capabilities for the detailed visualization and assessment of complex biological structures in microfluidic devices. The 3DCellSeg framework enables 3D cell segmentation, which is particularly important in histopathological image analysis for cancer diagnosis and grading¹⁸⁶. Mesmer, a deep-learning-enabled segmentation algorithm trained on TissueNet, automates the extraction of key cellular features, such as subcellular localization of protein signals, thereby assessing cell lineage information in multiplexed datasets to quantify cell morphology changes¹⁸⁷. HE2RNA, based on a CNN model, can predict RNA-Seq expression profiles from whole-slide images of tumours without expert annotation, generating heatmaps for the spatial visualization of gene expression to improve prediction performance for specific molecular phenotypes. This approach may also enable the prediction of gene expression levels from histological images of 3D cell culture models for diagnosis and treatment response predictions^188,189.

High-throughput 3D microfluidic cell culture models can be designed with high biological relevance, for example by implementing iPS cells and organoids, as well as physiologically relevant culture conditions^190,191. Such devices can be designed by various methodologies, including 3D printing, laser processing and injection moulding^138,192,193. For example, a 3D hybrid micromesh-assisted bioprinting method combines micromesh scaffold structures, which were 3D-printed by digital light projection, with sequential hydrogel patterning¹⁴⁸. This approach allows precise 3D liquid patterning, eliminating the need for external pumps or valves, while managing fluid removal. 3D tissue morphogenesis can also be spatially guided by localizing different hydrogels at different parts of a device by providing a unit-based scaffold that can trap hydrogel solutions inside designated units. This MultiCUBE¹⁹⁴ approach allows spatial organization of biomolecular compositions and physical conditions of hydrogels, as well as the relative position of biological samples, and may well be applied for multiorgan models.

Many designs and operations of 3D microfluidic cell culture systems primarily focus on recapitulating physiological tissue architectures and functions, often placing less emphasis on testing assays and readouts. For these systems to become standard tools for drug and non-clinical testing, aligning with OECD testing guidelines, it is crucial to consider the compatibility of systems with common laboratory instruments for data acquisition and downstream analyses. For example, microfluidic cell culture devices enable various assessments, including phenotypic^19,170, genomic^195,196, proteomic^197,198 and histological analyses¹⁹⁹. However, multiomics analyses remain challenging. For example, extracting hydrogels from microfluidic devices to access tissues or cells may cause damage¹⁴². Therefore, device designs must accommodate such analyses. For example, attachment techniques or materials that allow reversible assembly may be considered. In addition, open microfluidic-based modular devices can facilitate access to cells or tissues. If the interfaces within these devices are high-throughput, samples subjected to the same experimental conditions can be acquired simultaneously to provide sufficient sample volume for multiomics analysis.

In addition to biological validity, practical use of 3D microfluidic cell culture models should be considered — that is, the level of required complexity. For example, a cell culture model in drug development should provide reliable responses to new drug candidates and function in a high-throughput manner, but may not have to mimic an entire organ or tissue. Therefore, complexity and throughput may have to be balanced (Box 2). Moreover, device productivity, experimental throughput and data analysis efficiency need to be improved for a specific context of use¹⁵⁷ (Box 3).

Microfluidic 3D cell culture has reached a technology readiness level, and by further integrating miniaturized microfluidics with physiological parameters, including controlled flow, standardized microfluidic systems may be engineered that operate at the scale of 96-well and 384-well plates, enabling real-time monitoring. In addition, modular device structures may aid the retrieval of cells or tissues from the device for high-content analysis. High-throughput microfluidic 3D cell culture models may thus provide important data for non-clinical testing, offering a complementary tool to animal models in modelling human health and disease.