Reconfigurable Microphysiological Systems for Modeling Innervation and Multitissue Interactions

Abstract

Tissue-engineered models continue to experience challenges in delivering structural specificity, nutrient delivery, and heterogenous cellular components, especially for organ-systems that require functional inputs/outputs and have high metabolic requirements, such as the heart. While soft lithography has provided a means to recapitulate complex architectures in the dish, it is plagued with a number of prohibitive shortcomings. Here, concepts from microfluidics, tissue engineering, and layer-by-layer fabrication are applied to develop reconfigurable, inexpensive microphysiological systems that facilitate discrete, 3D cell compartmentalization, and improved nutrient transport. This fabrication technique includes the use of the meniscus pinning effect, photocrosslinkable hydrogels, and a commercially available laser engraver to cut flow paths. The approach is low cost and robust in capabilities to design complex, multilayered systems with the inclusion of instrumentation for real-time manipulation or measures of cell function. In a demonstration of the technology, the hierarchal 3D microenvironment of the cardiac sympathetic nervous system is replicated. Beat rate and neurite ingrowth are assessed on-chip and quantification demonstrates that sympathetic-cardiac coculture increases spontaneous beat rate, while drug-induced increases in beating lead to greater sympathetic innervation. Importantly, these methods may be applied to other organ-systems and have promise for future applications in drug screening, discovery, and personal medicine.

1. Introduction

Microphysiological systems (MPS) are of immeasurable value to the investigation of fundamental biology and to offset the need for in vivo animal models for drug discovery, efficacy, and toxicity assays.^[1] However, many in vitro models fail to fully capture the complex cell–cell interactions, 3D microenvironments, structural organization, and vascularization of multicellular organ systems.^[2] Previous efforts toward establishing compartmentalized MPS rely on the use of microposts or microtunnels and silicon-based materials,^[3] which have limited utility in some research and commercial settings.^[4] Toward overcoming these limitations, our previously reported “cut and assemble” manufacturing technique was expanded to exploit the meniscus pinning effect via GelPins to compartmentalize 3D cell-laden materials within a tailorable MPS.^[5]

2. Results and Discussion

GelPins are thin layers of material within our custom microfluidic devices that facilitate the formation of discrete, yet contiguous structures in either the x–y and/or z planes (Figure 1A). This manufacturing technique, in combination with photocrosslinkable hydrogels, enables rapid (hours) and inexpensive (<$2 in materials per MPS) prototyping of custom,^[5] multilayer, 3D MPS with nearly limitless geometric configuration possibilities (Figure 1B–D). As a proof of concept, GelPins were leveraged to establish the first biomimetic in vitro model of the cardiac sympathetic nervous system (SNS) for investigating the effects of sympathetic activity and increase beat rate on cardiomyocytes (CMs) and innervation, respectively. Despite the nervous system’s critical role for tissue development and functional control, innervation is often overlooked in most tissue engineering applications due to the complexities of biofabricating heterogeneous cell populations and tissue structures.^[6] Therefore, these devices representative a paradigm shift to enable modeling of multitissue interactions and innervation within flexible MPS for fundamental mechanistic discoveries and drug screening applications.

Figure 1.

Development and characterization of compartmentalized microtissue devices. A) Photograph of organ-chip model and cross-section schematic demonstrating innervation in the basal layer of the device. B) Vector illustrations, materials, cut settings (denoted as #% speed, #% power) and thicknesses of the layer within a typical device. C) Photocrosslinkable hydrogel precursors are loaded to the GelPin and are crosslinked with visible light in situ to form a contiguous gel layer with discrete compartmentation. D) Photographs of colored gel loaded devices demonstrating a wide range of possible microtissue configurations for studying different tissue interactions. Exploded layer-by-layer assemblies are available in the supplemental material. E) COMSOL modeling showing the meniscus pinning effect with a floating GelPin and the pressure needed to overcome this barrier. F) Colorcoded depth map of the entire microtissue thickness demonstrating the neurons (beta III tubulin) cross both above and below the GelPin into the cardiac microtissue compartment. Scale = 100 μm. G) Normalized absorbance of blue food coloring diffusing through hydrogels of different thicknesses sandwiched between two membranes over several hours (n = 3).

The development of MPS has become increasingly important for study of in vitro neuronal cocultures.^[7] Traditionally, researchers have utilized microtunnels and/or microposts to establish compartmentalization within microfluidic devices for 2D and 3D cultures, respectively.^[3] However, fabrication of these technologies relies on high-resolution, often soft lithography, based manufacturing which has limited their use in many research settings as well as presents difficulties to manufacture at-scale using standard practices.^[4] Further, the majority of these devices are fabricated using silicone-based materials, which can adsorb small molecules and are difficult to manufacture, reducing their applicability for pharmacological drug screening at the commercial scale.^[4,8] Toward overcoming these limitations, we have previously reported on the development of a novel “cut and assemble” manufacturing technique as an alternative to traditionally used, soft-lithography fabrication approaches (Figure 1B).^[5] While continuing to improve in resolution, laser engravers are not able to fabricate free-standing features such as microtunnels and/or micropost to enable robust compartmentalization as their silicon counterparts. Therefore, we developed and integrated micrometer-scale GelPins within our devices to enable establishment of temporary liquid–air interfaces, that once polymerized, formed discrete hydrogel regions (Figure 1B,C). Phaseguides have been previously reported as a small ridge on bottom of a microfluidic channel that act as capillary pressure barriers for filling dead angles, mixing, and preventing gel from flowing into an adjacent channel.^[9] However, this technology has been predominantly limited the separation of a single homogenous cell laden hydrogel and perfusion channel^[10] or three parallel channels to create two discrete cell laden or extra cellular matrix components and corresponding path for perfusion.^[11] Herein, we demonstrate the potential for GelPins to create multiple interfacing tissue compartments without the reliance of a parallel perfusion channel. Dissimilar to both microtunnels and microposts that occupy a significant percentage of the interface between compartments and can interfere with cell migration,^[12] GelPins function to maintain discrete compartmentalization for a wide range of 3D geometric patterns without the need for significant physical barriers (Figure 1D and Figures S1A–F and S2A–F, Supporting Information). Architectures are not limited by wafer dimensions, but rather can be robustly expanded via GelPins in the x–y plane and indefinitely in the z direction through our layer-by-layer approach.

Often, constraints on diffusion limit 3D cell culture system dimensions.^[13] Previously, meniscus pinning technologies for 3D cell culture required that the adjacent channel be reserved for medium flow to provide the microtissue with nutrients, limiting the complexity of tissue geometries as well as heterogeneity of hydrogel formulation, including encapsulated cell populations and formation of chemical or mechanical gradients that could ultimately be achieved.^[14] Additionally, this design approach may lead to chemotactic effects due to the directionality of nutrient supply.^[15] For example, cancer cells have been shown to migrate across collagen-Matrigel interfaces toward adjacent medium channels due to paracrine signaling.^[16] However, it is impossible with these techniques to decouple the orientation of nutrient gradients to elucidate if cell migration/neurite extension is nutrient, material/mechanical properties, or target cell-dependent. To overcome these limitations we have included, 1) a membrane layer above and/or below 3D cultures allow homogenous media diffusion through the z axis and 2) a floating Gelpin, positioned in the basal channel of a bilayer device, allow multiple 3D tissue compartments in the x–y plane. COMSOL modeling was used to confirm that the positioning of a floating GelPin would allow the GelPin to remain active as temporary capillary pressure barriers, requiring ≈750 Pa to be overcome (Figure 1E).

To establish continuous microtissue with discrete compartmentalization, cell-laden, precursor hydrogel solutions were loaded sequentially and gelled in situ within the basal channel of the device. Here, the use of photocrosslinkable hydrogels improves experimental throughput by enabling each hydrogel region to be crosslinked in situ in <60 s per compartment (Figure 1C). In our previous work, visible light photocrosslinked GelMA hydrogels were shown to support the 3D culture of primary cardiac cell populations and peripheral neurons.^[17,18] Furthermore, to maintain ion channel expression, membrane resting potential, and depolarization characteristics to that of intact tissue in vitro, SNS neurons must be cultured within a 3D microenvironment.^[19] Here, we established cardiac microtissues on-chip, using a 7.5% (w/v) GelMA hydrogel. A 50/50 mixture of enrichened CMs and adherent cardiac cells (aCCs) from the left and right ventricles was prepared with a 20 m cells mL⁻¹ concentration in a GelMA precursor solution.^[17] This solution was carefully loaded into the device so that it filled to the GelPin and gelled in situ with 10 mW cm⁻² 405 nm light for 60 s. Adjacent to these tissues, primary postganglionic SNS neurons from the superior cervical ganglion (SCG) similarly loaded and gelled in situ at 5 m cells mL⁻¹. A representative depth-coded, immunofluorescent image of the cardiac SNS MPS demonstrates neurites extend toward the target cardiac cells, both above and below the floating GelPins, crossing the gel interfaces while maintaining physiological compartmentalization of neurons’ somas and cardiac cell bodies (Figure 1F). Further, on-chip culture supported similar viability results previously reported using more traditional culture techniques. ^[17,20] Using a live/dead assay 7 d postencapsulation, cardiac cells and SNS neurons were shown to have 89.5% ± 1.4% and 67.5% ± 6.6% viability, respectively.

Above the basal channel is an apical medium flow channel separated by a porous membrane to facilitate circulation-like diffusion of nutrients into the microtissue (Figure 1A). The uniform diffusion of nutrients throughout the microtissue limits formation of chemotactic gradients in the x–y plane typical of conventional devices with flanking medium channels.^[21] To assess nutrient diffusion through the microtissue in the basal channel, the microfluidic chips were redesigned so that samples could be collected below the basal hydrogel compartment layer (Figures S1G and S2G, Supporting Information). Here, we demonstrate a detectable concentration of blue food coloring in the basal flow channel within 1 h for devices with a 120–380 μm thick hydrogel and approaching an equilibrium over 48 h. As expected, the permeability decreases from 4.29 to 3.99 to 3.35 × 10⁻³ mm s⁻¹ for hydrogels with thickness of 120, 250, and 380 μm, respectively (Figure 1G). However, in cell-laden microtissues it is expected these values will change over time as the cells start to proliferate, degrade the hydrogel matrix, and deposit cell derived extracellular matrix.

The layer-by-layer approach further allows additional tissue complexity in the z axis. As an example, the membranes used to separate the apical flow channel and the basal gel microtissue may be seeded with endothelial cells to mimic an innervated arterial tissue cross-section if desired (Figure 2A and Figure S3A, Supporting Information). With a flow rate of 0.06 mL h⁻¹, the average fluidic shear stresses over the bare membrane was calculated to be an average of 9 × 10⁻⁵ dyne cm⁻² using COMSOL (Figure 2B). To confirm the validity of the simulation, the averaged fluidic shear stress was compared to an existing approximation for flow between two plates and found to be within 9% compared to the prediction. Based on each MPS design’s unique geometry and flow rate parameters, using this simulation, fluidic shear stresses may be calculated and adjusted to promote endothelial monolayer formation.^[5] However, while membranes seeded with endothelial cells may be used to help mimic vasculature-like diffusion, membranes fail to fully recapitulate the tissue-vessel iterations and act as additional diffusion barriers.

Figure 2.

Enhancing the functionality of microphysiological models with vasculature-like features and instrumentation. A) Schematic and immunofluorescent image demonstrating how endothelial (red, CD31) may be cultured atop a membrane with smooth muscle cells (green, α actinin) below to mimic a 3D cross-section of innervated arterial tissue. The immunofluorescent image is a maximum intensity projection of the entire microtissue tissue layer. Arrows point to neurons (purple, neurofilament-200) crossing from the neural compartment to the smooth muscle cell compartment. Scale = 100 μm. B) COMSOL model showing the shear stresses on the membrane of a device can be simulated computationally based on flow rate. C) Layer-by-layer assembly showing how the placement of GelPin slats may be used to establish vasculature-like flow on-chip and COMSOL modeling showing the meniscus pinning effect with GelPin slats with different spacings to tune the percentage of tissue–tissue or tissue/medium interactions. Devices shown contain fluorescent beads to visualize each layer. D) Photograph of an organ-chip model reversible mounted to a multielectrode array and representative raster plot of extracellular potentials across electrodes from beating CMs cultured within a hydrogel on-chip. E) Photograph of an organ-chip fitted with carbon electrodes and measurement of beat rate with applied electrical stimuli (n = 1, **p < 0.05). F) Overview schematic and photograph of organ-chip fitted with ferrule couples to light stimulation. G) Percentage of light that passes through each layers of the devices at different wavelengths (n = 3, **p < 0.05).

As an alternative to the inclusion of membranes to define boundaries in the z axis, GelPins may be orientated as slats to create a vertically separated compartment system (Figure 2C). COMSOL simulations were used a priori to confirm GelPins with a 1:2 and 1:3 width to spacing ratio and a gap of 400 and 600 μm, respectively, functioned as a pressure barrier in the z-direction (Figure 2C). Single micrometer scaled gel layers may be crosslinked both above and below the GelPin slats layers to form contiguous stacked hydrogel structure. The use of GelPins slats enables micrometer-scale tissue layers to be established (Figure 2C), optimal for developing culture systems with striated tissue layers like the intestinal lumen, dermis, and retina. Furthermore, in addition to tissue–tissue interfaces, GelPin slats may also enable different tissues to share soluble components without being in contact with one another. By incorporating a flow channel between two GelPin slat layers that contain microtissue on either side, cells have access to nutrients in a biomimetic context in vitro.

Lastly, while traditional organ-chip models offer simple and robust means to recapitulate the architecture of the in vivo microenvironment of different human tissues in vitro, extensive postsample processing and analysis is required to understand cellular functions and responses.^[22] Therefore, to facilitate real-time data acquisition and record the electrophysiology of excitable cell populations such as cardiac and neural cells, GelPin MPS may be reversibility mounted to commercially available multielectrode arrays or include custom arrays or sensors at different layers (Figure 2D and Figure S3B, Supporting Information). CM beating on-chip was monitored in real time via a multielectrode array and cardiac microtissues showed robust beating coordination across electrodes. Furthermore, devices may be modified to incorporate electrodes for electrical stimulation of specific cell populations on-chip (Figure 2E and Figure S3C, Supporting Information). CMs were paced at 2 Hz irrespective of sympathetic innervation using pulsatile electrical stimulation via carbon electrodes in real-time. However, applied electric fields cannot be used to modulate neural activity independent of voltage sensitive cardiomyocytes.^[23] To overcome this limitation, neurons can be genetically modified to express opsins for photostimulation, and devices may be fitted with ferrule connectors so that specific regions of the chip may be illuminated with light for optogenetic stimulation (Figure 2F and Figure S3D, Supporting Information).^[23,24] Minimal light attenuation through each layer was measured for wavelengths commonly used for optogenetic applications (Figure 2G and Figure S4, Supporting Information).

MPS models, developed herein, may disrupt the current clinical pipeline and may be used in place of in vivo systems for investigating multitissue interaction and innervation. In vivo, the SNS increases heart rate in response to an external stressor as a part of the body’s innate “flight-or-fight” response via junctions with cardiac pacemaker cells in the heart.^[25,26] Previously, animal models have been used to investigate cardiac SNS dysfunction,^[27] but due to their inherent complexities, variability, and compounding cardiac-autonomic control systems, researchers established in vitro models to simply and control conditions.^[23,28,29] These 2D static culture models focused primarily on compartmentalization and the hierarchical structure of the neurocardiac axis,^[28] rather than a more physiologically 3D tissue microenvironment, which may be necessary to mimic cardiac innervation in vivo.^[30] Cardiac cellular morphology and phenotype are heavily influenced by the microenvironment and may be lost during the longer-term in vitro cultures needed to study innervation.^[13,31] Further, the local microenvironment’s physical, chemical, and cellular properties are highly correlated with the rate of neurite extension.^[32] Therefore, it is critical to mimic native tissue structure and composition to faithfully design in vitro models that can translate to in vivo systems. Nevertheless, although these interactions are essential for cell development and function, mimicking complex static 3D environments in vitro may also limit nutrient transport or have other compounding effects such as necrotic cell death.^[33]

Toward modeling sympathetic innervation in the absence of the parasympathetic activity and neurohormonal regulation, a cardiac SNS MPS were prepared as described above (Figure 1) and beat rate and degree of beating coordination were quantified.^[17] Immunofluorescent images of cardiac SNS MPS showed that SNS neurites extend toward CMs, across the GelPin, and exhibit colocalization of neuron and cardiac cell markers with synapsin I, a presynaptic marker (Figure 3A). Cardiac cells in proximity to SNS neurites have a ≈28% increase in beat rate (p < 0.05), with a ≈7% decrease in beating coordination compared to controls (p < 0.05, Figure 3B,C). However, looking closer at the distribution of these data revealed not only a broader but also a right/positive skew in the beat rate statistics for areas of high colocalization in the cardiac compartment. Toward better understanding these results, original data files were mined to plot beat rate data from each unique cell within the recording area as a histogram (Figure 3D). These results further exacerbated the findings but also show the presence of a subpopulation of beating CMs that were likely responsible for this skewed result (Figure 3D, inlay). Therefore, it is probable that only those CMs that are establishing junctions with or are in close proximity to sympathetic neurites are beating at the increased rate. This hypothesis was further corroborated by the decrease in coordinated contractions observed in SNS innervated cultures, as select CM populations may be beating at different rates because of the presence of a neuroeffector cue. Further evidence is needed to demonstrate the formation of mature synaptic junctions and determine if the increase in beat rate and decrease in coordination was a result of a physical sympathetic neurocardiac coupling or some other chemical signal.

Figure 3.

Quantification of beat rate and sympathetic ingrowth in on-chip cardiac microtissues. A) Immunofluorescent image demonstrating synapsin I positive staining in a cardiac SNS organ-chip model (saracomeric alpha actinin, red; neurofilament-200, purple; synapsin I, white; DAPI, blue). Arrows point to the synapsin I positive junctions between cardiac cells and neurons. Scale = 10 μm. B) Degree of local SNS ingrowth significantly increases beat rate and C) decreases coordinated contractions compared to controls (n = 31–41, **p < 0.05). D) Binned individual beat rate data reveal the presence of a cardiac subpopulation with a high degree of innervation as shown by the skewed moving averages. Inlay represents the offset between innervated and noninnervated beat rate data. E) Isoproterenol (ISO) increases beat rate for both innervated and noninnervated systems as compared to controls (n = 25, p < 0.05). F) Representative immunofluorescent images of the cardiac microtissue showing sympathetic innervation (purple, neurofilament-200) in response to ISO compared to control and their quantification. G) Differences in beat rate were attenuated between innervated and noninnervated cultures, but significantly increased compared to controls when organ-chips were treated with ISO (n = 6–10, **p < 0.05).

Following a heart attack, ischemic cell death and myocardial remodeling preferentially leads to increased sympathetic activity and innervation, and a greater risk of arrhythmia and heart failure.^[25,34] While there have been improvements in the symptomatic management of cardiovascular disease using beta-blockers to inhibit SNS activity systemically,^[35] there are currently no lasting treatments for dysautonomia due an incomplete understanding of SNS pathophysiology. We hypothesized that sympathetic hyperactivity and ingrowth are a consequence a reduced cardiac output, therefore increased SNS innervation may be halted by pharmacologically increasing beat rate. To demonstrate the utility of our device and investigate this hypothesis, organ-chips were treated with 1 × 10⁻⁶ m isoproterenol (ISO) from day 7 to 14 and innervation quantified using Neurolucia (Figure 3E,F). ISO was selected because it binds exclusively to the beta-adrenergic receptors on cardiac cells to increase beat rate and should have no independent effect on neurons. MPS were treated starting on day 7 so that cardiac cells would have sufficient time to remodel their environment before made to beat more rapidly. Quantification of these experiments demonstrated that a drug-induced, approximately twofold increase in cardiac beat rate promotes ≈400% more sympathetic over a 2-week period despite the same initial seeding densities (p < 0.05), disproving our original hypothesis. ISO was found to have no effect on neurite outgrowth in monocultures of sympathetic neurons (Figure S5, Supporting Information), therefore the increase in sympathetic ingrowth within cardiac compartments was likely a direct impact of CM changes. Taken together with our previous results, this suggests the presence of a feedforward mechanism, where faster heart rate leads to more sympathetic innervation, which in turn may then lead to further increases in heart rate.

3. Conclusion

In summary, the MPS presented here may provide versatile platforms to study the role of the nervous system in various tissues in addition to multitissue interactions and perfusable tissue structures. The microphysiological model of the cardiac SNS described here demonstrates the utility of this technology for complex MPS designs without traditional fabrication methods. Using this in vitro platform, we then demonstrated that in the absence of additional cardiac regulators, increased cardiac stress may lead to a lasting remolding of the cardiac SNS. By developing microphysiological models of the neurocardiac axis, we will be better able to understand the underlying cellular mechanism that lead to dysfunction of the autonomic nervous system and may help to elucidate novel therapeutic targets.

4. Experimental Section

Device Fabrication:

Organ-chip devices were fabricated using a modified approach to previously described methodology.^[5] A laser engraver system (Epilog Zing 16, Epilog Laser) cut and shaped flow paths in double-sided adhesive tape (966 adhesive, 3 m), polymethyl methacrylate (PMMA, McMaster-Carr) and polyethylene terephthalate (PET, McMaster-Carr) sheets, and a polycarbonate (PC) track etched membrane (GE Healthcare) using material specific speed and power settings (Figure 1B). Each individual layer was then assembled layer-by-layer into a complete device with the aid of precut alignment holes and a custom alignment jig.

From top to bottom, a typical device consisted of a 4.5 mm PMMA top layer that features 4 mm diameter fluidic inlets and outlets for the apical follow channel in addition to 1 mm diameter gel filling ports to provide access to the basal channel of the device. For the compartmentalized microtissue and multilayer tissue devices, below the top layer was the apical flow channel that connects the fluidic inlets and outlets and covers the area above the microtissue. This layer was comprised of a 1.5 mm PMMA sheet between ≈60 μm thick double-sided adhesive tapes on either side and again features gel loading holes. A PC membrane with cutouts to access the basal gel compartment was attached to the bottom of the flow channel and functions as the top to the gel compartment. The basal gel compartment was a stack of several flow paths prepared with 3 m tapes and a PET that form a contiguous channel with discrete areas for gel loading. Each 3 m tape layer was cut into the desired shape of the microtissue and connected to each gel filling port with a 1 mm channel. For the compartmentalized devices, a PET GelPin layer was positioned in-between 3 m tape gel channels and features identical cutouts to these layers, but with individual strips of material with a final width of ≈200 μm remaining wherever a gel interface will later be established. However, for the multilayer tissue devices, a PET GelPin slats layer functions to separate each layer in the z-direction and features similar cutouts to the gel layer, but with ≈200 μm wide strips of material spaced ≈600 μm apart remaining. Below this GelPin slat layer are additional gel and GelPin slat layers to increase the total number of tissue layers that may be established on-chip.

Unlike the aforementioned designs, below the top PMMA layer in the perusable vascular models was a gel layer comprised of one or more ≈60 μm 3 m tapes with the desired microtissue size and shape. Below this upper gel layer was the first PET GelPin slat layer and then a ≈60 μm thick 3 m flow path that connects to the fluidic inlets and outlets. A second GelPin slat layer was attached below this centrally located fluidic channel and functions to constrain the second gel in the lower gel channel. This lower channel was the mirror image of upper gel channel and was again comprised of one or more ≈60 μm 3 m tapes. Lastly, a No. 1 glass coverslip (22 mm × 40 mm) was attached below each of these gel layer and serves as the base of all these devices. Prior to use, chips were pressed in a drill press and stored under vacuum at vacuum at 50 °C overnight to eliminate outgassing induced bubble formation, then UV sterilized for 600 s on each side to prevent contamination.

Diffusion Characterization:

Compartmentalized microtissue chips were modified for permeability studies to facilitate quantitative measurements of nutrient diffusion within the hydrogel. In brief, each layer was prepare as outlined above, but with two additional through-holes placed in-line with the four alignment holes in each layer. The height of the gel compartment was varied via addition or removal of 3 m tape layers. To enable sampling from below the gel layer, the glass coverslip was replaced with a membrane containing the four alignment holes and the two additional through-holes. Below this second membrane, a basal flow channel was prepared using a 1.5 mm PMMA sheet sandwiched by a layer 3 m tape and cut so that the flow path reached to two added through-holes while covering the entirety of the gel area. The device was sealed with a 1.5 mm PMMA sheet.

To measure nutrient diffusion on-chip, the hydrogel layer was first filled with a GelMA precursor solution to photocrosslinked in situ. GelMA precursor solutions were prepared at 7.5% w/v in deionized water containing 0.5% w/v lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP, Biobots). Once fully loaded into the device via the gel filling ports, the GelMA solution was photopolymerized with visible light (405 nm light-emitting diode, QUANS) for ≈45, 60, and 90 s (0.25 s μm⁻¹ of hydrogel thickness) at 10 mW cm⁻². Once the hydrogel layer was established, the apical flow channel was filled with a 1:100 solution of blue food coloring in deionized water, while the basal flow channel was filled with pure deionized water. Diffusion through different GelMA concentrations and chip thicknesses was quantified by collecting samples from the basal flow channel at different time points and measuring the transmittance over time with a plate reader (Synergy HT, BioTek). A minimum of three independently prepared devices for each gel thickness were used to quantify diffusion over time.

Computational Fluid Dynamics Modeling:

Finite element method computational fluid dynamics was conducted using COMSOL Multiphysics version 5.5 to model both the gel seeding and fluidic shear stress in the flow chamber of the microfluidic system. In both cases, results were validated for mesh independence. The precursor gel solutions were assumed to be roughly similar in flow characteristics to water at room temperature (density = 997.7 kg m⁻³, dynamic viscosity = 0.89 cP). The culture media for the fluidic shear stress model was adjusted for a higher temperature (37 °C) and the presence of serum based on available product notes (density = 993.6 kg m⁻³, dynamic viscosity = 0.97 cP). No slip conditions were set on all boundaries unless otherwise noted. Inlet flow rate was set to 0.25 μL s⁻¹ (entered as velocity of 0.25 mm s⁻¹) and 1 μL min⁻¹ (flow rate) for the gel seeding and fluidic shear stress simulations, respectively, with the outlet rate determined by ambient pressure.

Gel seeding was modeled via 2D laminar flow physics coupled with the level set method on a simplified 2D geometry representing a cross-section of the gel chamber and GelPin. The surface tension force (N m⁻²) over the course of the simulation was calculated by combining the x and y surface tension components, plotting them as intensity gradients, and—using imageJ—determining, then multiplying together, the meniscus length (m) and the average surface tension per volume (N m⁻³). To further demonstrate the predicative power of this approach, GelPin slats were simulated a priori to determine whether the meniscus pinning effect could enable stacking of gel compartments in the Z direction. Simplified 2D geometries for both the GelPin slats with/without a vascular-like compartment were prepared and flow physics simulated.

Fluidic shear stress (𝜏_w) was simulated via 3D laminar flow physics on an exact replica geometry and determined as the viscosity (μ) multiplied by the sum of the respective gradient of velocity (v) and transpose of velocity for the plane occupied by cells and flow direction on an x, y, z axis orientation (Equation (1)). A single fluidic shear stress was calculated by averaging the values along the bottom surface of the flow channel.

$${\tau }_w=\mu \left(\frac{\partial {\nu }_x}{\partial z}+\frac{\partial {\nu }_z}{\partial x}\right)$$ (1)

To confirm the validity of the simulation, the averaged fluidic shear stress was compared to an existing approximation for flow between two plates.^[36]

$${\tau }_w=\frac{6\mu Q}{w{h}^2}$$ (2)

Primary Cell Isolation:

Primary CMs and aCCs from the ventricles, and primary postganglionic sympathetic neurons from the SCG were isolated from 2 d old (p2) Sprague–Dawley neonatal rats (Taconic Biosciences) using established protocols approved by Northeastern University’s Institutional Animal Care and Use Committee (19–0104R).^[17,20] In brief, rat pups were euthanized by decapitation and isolated tissue kept on ice in Hibernate-A (BrainBits) while tissue from every pup was collected. Once all tissue was harvested, SCGs were enzymatically dissociated sequentially in collagenase I (305 units mg⁻¹ in Hank’s Balanced Salt Solution (HBSS), Gibco) for 60 min and then in 0.5x Trypsin in HBSS for 15 min. Partially dissolved ganglia was then further broken up via mechanical trituration with a fire polished pipette. Dissociated sympathetic neurons were counted, then cryopreserved with 10% dimethyl sulfoxide (Fisher) at 1 × 10⁶ cells mL⁻¹ for use on-demand. The following day, cardiac tissue was dissociated by serial collagenase II (305 units mg⁻¹ in HBSS, Gibco) digestions at 37 °C after an overnight incubation in 0.5% v/v trypsin in HBSS overnight at 4 °C. CMs were purified from aCCs via differential attachment in which any unattached cells after 1 h were considered enriched CMs. CMs and aCCs were counted and seeded within 1–2 h following enrichment.

Cell Loading and Culture:

To establish an innervated cardiac microtissue, dissociated sympathetic neurons and cardiac cells were cocultured within a GelMA hydrogel on-chip. GelMA derived from fish gelatin was synthesized as previously reported.^[17] GelMA precursor solutions were prepared at 7.5% w/v in complete culture medium (Dulbecco’s modified Eagle medium with l-glutamine supplemented with 10% v/v fetal bovine serum, 1% v/v penicillin-streptomycin, 25 ng mL⁻¹ nerve growth factor, and 10 ng mL⁻¹ glial cell derived neurotrophic factor) containing 0.5% w/v LAP (Allevi). Dissociated sympathetic neurons and a 1:1 ratio of enriched CMs and aCCs were pelleted and resuspended in the GelMA precursor gel solution at a density of 5 × 10⁶ and 2 × 10⁷ cells mL⁻¹, respectively. Cardiac-laden precursor gel solutions were carefully loaded into the cardiac microtissue compartment via the corresponding fill port, so that a liquid–air interface formed at the GelPin. The cardiac-laden solution was then photocrosslinked in situ with visible light (405 nm, 10 mW cm²) for 60 s (0.25 s μm⁻¹ of hydrogel thickness). Following gelation of the cardiac-laden hydrogel, the neuron-laden solution was loaded into the microfluidic device and similarly photopolymerized adjacent to the cardiac microtissue to form a contiguous hydrogel construct with discrete cellular compartmentalization. Complete culture medium was perfused into the organ-chips via the fluidic inlet and exchanged daily over a 2-week period to allow for innervation of the adrenal compartment.

With the same approach, an innervated arterial tissue on-chip was established by replacing the cardiac cell-laden gel with a smooth muscle cell-laden gel with 1 × 10⁷ cells mL⁻¹ and by seeding ≈5 × 10⁶ human umbilical vein endothelial cells on top of the membrane. A 50/50 mixture of endothelial cell growth medium-2 and complete culture medium was exchanged daily for a 2-week period to allow for innervation into the smooth muscle cell compartment and for the endothelial cells to form a monolayer atop the membrane prior to imaging.

Electrical Pacing:

MPS were redesigned to incorporate 1.5 mm diameter carbon rods on either side of the hydrogel compartment 2 cm apart. Each rod was carefully attached to a platinum wire and placed on-chip so that it was completely submerged in medium. Pulsatile electrical stimuli (2 Hz, 50% duty square wave, 0 V offset, 10 V peak-to-peak) was applied by connecting the ends of each wires to a function generator (Agilent). Video microscopy was used to measure the beat rate in response to electrical stimulation.

Cardiac Beating Quantification:

Cardiac beat rate and beating coordination (degree of coordination, DoC) was quantified on cell-by-cell basis using video microscopy as previously reported.^[17] In brief, phase contrast videos from organ-chips were captured at 30 frame s⁻¹ using a Zeiss Axio Observer with a 20x object and incubation chamber (37 °C and 5% CO₂). One video from the cardiac compartment was taken on each side of the device for both control and ISO conditions. The sum differences in frame-to-frame pixel intensity for each identified region of interest were used to quantify beat rate. After artifact rejection, the average beat rate and DoC were reported. A modification to the original algorithm was written to output individual beat rates for each beating cell in the field of view (≈15 cells per video) to better identify subpopulations. A minimum of two videos per device were taken from more than six devices from three independent preparations across multiple days in culture for each condition.

Immunostaining:

14 d postseeding, organ-chips were fixed with 4% paraformaldehyde for 45 min at room temperature. Chips were then washed thrice using phosphate buffered saline (PBS) with a 10–15 min wait between rinses. After excess paraformaldehyde was removed, cells were permeabilized with 0.1% triton X-100 in PBS for 30 min at room temperature. Triton X-100 was similarly washed away with PBS, before samples were blocked with 2.5% goat serum in PBS overnight at 4 °C. After overnight blocking, samples were incubated for an addition 12–24 h with 1:10 000 chicken anti-neurofilament-200 (Abcam, ab4680), 1:400 mouse antibeta III tubulin (Invitrogen, 480011), 1:200 rabbit antisynapsin I (Abcam, ab64581), 1:100 mouse anti-CD31 (Invitrogen, MA5–13188), and/or 1:200 mouse antisarcomeric alpha actinin (Abcam, ab9465) in newly prepared blocking solution at 4 °C. Following primary antibodies, chips were washed as described previously and secondary antibodies Alexa Fluor 488, 546, and 647 IgG (Molecular Probe) with various species reactivity prepared at 1:1000 in 2.5% goat serum were applied overnight at 4 °C to visualize each sample. Cell nuclei were visualized with 1:1000 4′,6-diamidino-2-phenylindole (DAPI) solution in PBS. Excess secondaries were rinsed away with sequence washing steps and chips were imaged on an inverted fluorescence microscope (Zeiss Axio Observer Z1).

Degree of SNS Ingrowth:

Total neurite ingrowth into the cardiac microtissue compartment was quantified with Neurolucida. To fully capture the innervation in each plane, maximum intensity projections of the entire device were used to measure total neurite ingrowth within the cardiac microtissue compartment using semi-automated 2D tracing tools. Only neural projects extending past the GelPin into the cardiac microtissue compartment were traced, and the degree of innervation was reported as the average total outgrowth for each condition. A minimum of six devices from three independent preparations were quantified for each condition.

Statistical Analysis:

Due to primary cell and animal derivation, beat rate data were first normalized to controls in Microsoft Excel. Statistical significance for all data was then calculated using GraphPad PRISM 7 using a student t-test or a one-way analysis of variance. Data were further checked for normality using the Shapiro–Wilk test. Error bars represent the mean ± standard deviation of measurement.