Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Oct 10.
Published before final editing as: Cell Biomater. 2026 May 8:100459. doi: 10.1016/j.celbio.2026.100459

Limb-on-a-Chip: An All-Hydrogel Platform for Scalable and Reproducible Engineering of Functional Neuromuscular Tissues

Laura Schwendeman 1,2, Angel Bu 1,2, Sonika Kohli 1, Ferdows Afghah 1, Tamara Rossy 1,*, Ritu Raman 1,*,**
PMCID: PMC13652111  NIHMSID: NIHMS2174474  PMID: 42857256

SUMMARY

Diseases or injuries that impact neuromuscular tissues severely impact human health, motivating the development of in vitro models of the motor control system. Current platforms either have limited reproducibility or require complex fabrication protocols that preclude high-resolution imaging and scalable functional analysis. We have developed a one-step method for fabricating an all-hydrogel Limb-on-a-Chip that addresses key challenges of current systems by enabling reproducible and scalable manufacturing of neuromuscular tissues compartmentalized into “spinal cord” and “limb” chambers. Co-cultures of motor neurons and skeletal muscles within this platform enable longitudinal tracking of neuromuscular junction formation and visualizing changes in muscle contraction in response to motor neuron stimulation. We demonstrate tissue-wide recordings of muscle force and single cell-resolution measurements of muscle calcium activity. Our accessible method for fabricating reproducible in vitro neuromuscular models compatible with high-resolution imaging and functional readouts provides a powerful tool for investigating the neuromuscular interface in health and disease.

Keywords: tissue engineering, biomaterials, organ-on-a-chip, microphysiological systems, 3D printing, neuromuscular disease

ETOC

Schwendeman, Bu, et al. have designed and developed an accessible and scalable method for growing motor neurons and skeletal muscle in an all-hydrogel chip that supports the formation of functional neuromuscular tissues. The methods described demonstrate functional innervation and contraction of neuromuscular cultures and provide researchers with a tool for studying neuromuscular maturation and functionality at both the cellular and tissue level.

Graphical Abstract

graphic file with name nihms-2174474-f0001.webp

INTRODUCTION

Pathologies that affect the neuromuscular system, namely skeletal muscles and the motor neurons that control their activity, have a severe negative impact on human health, mobility, and quality of life.1 Neuromuscular pathologies impact people across the age spectrum and can stem from a variety of causes, ranging from inherited muscular dystrophies to traumatic injuries to neurodegenerative diseases.2,3 In vitro models of the neuromuscular interface have emerged as useful tools for investigating the underlying mechanisms of disease onset and progression, as well as to develop new therapeutic approaches that preserve or restore function.4–6

The key functional unit of the neuromuscular interface is the motor unit (a motor neuron and the downstream muscle fibers it controls), and several studies have established the ability to generate functional motor units in vitro. Among the most well-established approaches for generating motor units in vitro are protocols that rely on self-assembly.7 For example, several studies have demonstrated the ability to form functional neuromuscular junctions within 3D organoids derived from stem cells, including both self-organizing neuromuscular organoids8,9 and “assembloids” of motor neuron spheroids fused with skeletal muscle spheroids.10,11 While these approaches enable forming multiple independent and functional motor units, the large size of these tissues (~2–5 mm diameter) limits compatibility with high-throughput imaging of neuromuscular junction architecture or function.12 Moreover, these tissues are not compatible with functional readouts of contractile force. To combat this drawback, researchers have demonstrated that 2D co-cultures of skeletal muscle and motor neurons enable readily visualizing neuromuscular junction formation and activation.12–15 However, delamination of muscle monolayers from underlying substrates makes it difficult to preserve such 2D cultures over several weeks in culture, and often preclude functional readouts of contractile force.16–19 Overall, the stochastic processes that drive formation and maturation of both the self-assembled 2D and 3D systems described above make it difficult to reliably place, stimulate, and quantitatively monitor independent motor units within such in vitro neuromuscular models.20

By contrast, organ-on-a-chip approaches for engineering neuromuscular tissues enable fabricating systems with more reproducible architectures, comprising defined physical compartments including 3D motor neuron spheroids and aligned 3D muscle tissues.20–22 Stimulation of motor neurons using chemical or optogenetic techniques enable triggering downstream muscle contraction.23,24 Importantly, existing organ-on-a-chip models couple 3D muscle tissues to flexible cantilevers, enabling quantitative readouts of force.25 While such platforms thus improve reproducibility of tissue architecture, their reliance on 3D tissues limits the ability to precisely visualize individual motor units, or quantify the activity of individual muscle fibers, similar to 3D neuromuscular organoids. Moreover, implementing such systems require multi-step microfabrication protocols, introducing barriers for broader adoption. The current state-of-the-art26 thus motivates developing an in vitro platform that combines the precise tissue patterning and force readouts of 3D organs-on-a-chip with the ease of high-resolution imaging offered by 2D self-organizing co-cultures.

We have developed a 2D compartmentalized platform that addresses this challenge and enables scalable and reproducible engineering of the neuromuscular interface. Our device design takes inspiration from the native spinal cord in humans and other mammals. In vivo, motor neuron soma are housed within the spinal cord, and project axons outwards to innervate skeletal muscle tissue in the periphery, with multiple motor units working together to control a downstream limb (Figure 1). Likewise, in our spinal cord inspired-device, motor neuron soma are isolated in separate compartments, and connected via tapered micro-channels to a downstream mm-scale 2D muscle tissue. The fabricated tissue thus represents a Limb-on-a-Chip, composed of multiple independent units. A key advantage of our platform is that it uses precision engineering to compartmentalize motor neurons and skeletal muscle, mimicking the reproducible architecture of 3D organs-on-a-chip, while retaining a 2D morphology at the neuromuscular interface, similar to self-organized 2D co-cultures. Additionally, the open-well format of our device enables co-culturing muscles and neurons in the same media, in contrast to microfluidic devices that fluidically separate the two cell types. Promoting biochemical crosstalk between muscles and motor neurons is known to promote tissue maturation, highlighting the benefit of leveraging an open-well format.27,28 Furthermore, unlike traditional organ-on-a-chip devices, which rely on multi-step microfabrication protocols to fabricate complex microfluidic structures, our Limb-on-a-Chip relies on a 1-step hydrogel molding protocol that leverages a 3D printed stamp to precisely and rapidly pattern an all-hydrogel device. In this study, we first present and validate the reproducibility of our Limb-on-a-Chip fabrication protocol, and demonstrate the ability to culture contractile muscles and motor neurons in separate compartments. We also provide evidence for neuron-induced muscle activation, suggesting that our chip supports the formation of functional neuromuscular junctions. We anticipate that our accessible protocol will enable researchers to fabricate reproducible, scalable, and functional models of the neuromuscular interface, thus advancing the state-of-the-art in neuromuscular disease modeling.

Figure 1. Bioinspired Limb-on-a-Chip Design.

Figure 1.

Open in a new tab

Compartmentalization of motor neurons in the spinal cord enables forming independently controllable motor units in a downstream limb. In vivo architecture (left) is mimicked in our in vitro Limb-on-a-Chip device (right).

RESULTS

Limb-on-a-Chip Design and Fabrication

Reproducible fabrication of functional neuromuscular interfaces in 2D requires methods to anisotropically align and mature contractile skeletal muscle monolayers. We have recently developed a technique termed STAMP (simple templating of actuators via micro-topographical patterning)16 that leverages a 3D printed mold to pattern micro-scale grooves in a hydrogel substrate, thus providing a physical template to guide fusion of myoblasts into a monolayer of precisely aligned multinucleated muscle fibers. While our previous study demonstrated that STAMP enables reproducible fabrication of muscles in monoculture, adapting STAMP for 2D neuromuscular organ-on-a-chip devices requires a method to precisely compartmentalize muscles and motor neurons in different chambers, while integrating patterning features that both promote alignment of muscle fibers and guide axonal outgrowth from motor neurons.

We designed a Limb-on-a-Chip STAMP, composed of a mm-scale muscle “limb” compartment containing aligned parallel 25 μm width grooves (dimensions previously demonstrated to promote efficient muscle fiber alignment)16 connected to three separate motor neuron 750 μm diameter compartments (mimicking separate motor units within a spinal cord) via 250 μm long tapered micro-channels (Figure 2a–b). Chip design and dimensions were chosen to ensure clear visual separation of muscle and neuron compartments, enabling seeding myoblasts and motor neurons into segregated chambers. Since our motor neuron spheroids are ~500 μm on diameter on average,21,27 while individual myoblasts are ~13 μm in diameter (Figure S1), micro-channels connecting the muscle and motor neuron compartments were designed to be 10 μm wide at the edge of the muscle chamber to prevent myoblast infiltration into neuron compartments after compartmentalized cell seeding. Fabricated STAMPs were mounted onto a 3D printed holder (Figure 2c), enabling easy handling during the hydrogel chip molding process. STAMP-holder assemblies were sterilized, coated with bovine serum albumin (BSA) to prevent hydrogel adhesion during molding, and used as a template to pattern gelatin methacrylate (GelMA) hydrogels into the desired Limb-on-a-Chip architecture (Figure 2d). Imaging of STAMPs and STAMPed hydrogels showed that the compartmentalized chip architecture, as well as the microscale design features within each chamber, were accurately transferred from the mold to the all-hydrogel chip (Figure 2e–f).

Figure 2. Limb-on-a-Chip Design and Fabrication.

Figure 2.

Open in a new tab

(A) Top-down CAD rendering of Limb-on-a-Chip STAMP (scale bar = 2mm). Inset: Top-down CAD rendering of tapered neuron chamber region that guides neurite growth and connecting 10 μm width channel that prevents muscle infiltration (scale bar = 500 μm). (B) CAD rendering of Limb-on-a-Chip STAMP highlighting the 25 μm grooved features of the muscle chamber region. (C) Comparison of CAD rendering of assembled STAMP and holders (top) to the actual physical device (bottom). Holders enable precisely aligning STAMPs within 24-well plates. (scale bars = 1 cm). (D) Overview of process used to pattern 7.5% GelMA Limb-on-a-Chip hydrogels with STAMPs. (E) Brightfield image of representative Limb-on-a-Chip STAMP surface (scale bar right = 1 mm; scale bars left = 500 μm). (F) Confocal transmitted channel image of representative casted Limb-on-a-Chip hydrogel (scale bar right = 1 mm; scale bars left = 500 μm).

Limb-on-a-Chip Design Validation

Compartmentalized seeding and differentiation of motor neurons and skeletal muscles within separate chambers of the Limb-on-a-Chip first required validating that fabricated devices maintained their precisely patterned geometries over long-term culture in liquid media. In our previous proof-of-concept study of muscle monocultures, we showed that STAMP could be used to pattern the topography of fibrin hydrogels. However, swelling and remodeling of the soft fibrin (0.05 ± 0.009 kPa storage modulus at 10 rad/s, Figure S2) reduced shape fidelity of the patterned hydrogel over time,16 particularly for high-aspect ratio micro-structures, thus precluding the ability to fabricate compartmentalized 2D co-cultures for longitudinal studies (Figure S3). To address this challenge, we fabricated Limb-on-Chip devices from stiffer GelMA hydrogels (0.3 ± 0.016 kPa storage modulus at 10 rad/s, Figure S2) that still matched the mechanical properties of substrates typically used for muscle and motor neuron culture27,29 and then monitored the long-term fidelity of microscale features within the chip.

We obtained 3D confocal microscopy images of micro-grooves within the muscle chamber and used the open-source Segment Anything Model (SAM)30 to obtain a binary image of the channel features (Figure 3a). Comparing the measured hydrogel groove dimensions to the STAMP mold groove dimensions over 1 week showed that fabricated groove width closely matched the nominal value of 25 μm, and remained stable over 1 week while immersed in cell culture media (Figure 3b–e).

Figure 3. Limb on a Chip Design Validation.

Figure 3.

Open in a new tab

(A) Right: representative cross section of grooved regions of the Limb-on-a-Chip hydrogel muscle chamber region. Left: SAM segmented and binarized image of grooves (white) with a traced boundary used for data analysis shown in orange. (scale bars = 100 μm). (B) Average segmented groove dimensions (orange) from Limb-on-a-Chip hydrogels across 0hrs (left; n = 16), 24 hrs (middle; n = 29), and 1 week (right; n = 15) compared to the nominal 25 μm groove dimensions. (C) Representative groove traces from 0hrs after stamping (orange) with identified peak and valley locations (black). (D) Measured groove heights across multiple days after being hydrated in phosphate buffered saline (PBS). No statistical difference was found according to a Kruskal-Wallis test (H(3) = 1.8580; p = 0.6024); (0hrs: n = 16, 24hrs: n = 29, 48hrs: n = 11, 1wk: n = 15). (E) Measured groove widths across multiple days after being hydrated in PBS. No statistical difference was found according to a one-way ANOVA (F(4) = 0.8585; p = 0.4722) (0hrs: n = 16, 24hrs: n = 29, 48hrs: n = 11, 1wk: n = 15). F) Representative confocal image of channel region in a Limb-on-a-Chip hydrogel (top) and the subsequent segmented binarized image highlighting 10 μm width region (w1) g) Histogram of average measured w1 width across n = 12 channels (light blue) with a fit gaussian (gray; μ = 10.864 μm; σ = 6.007 μm) overlay. The average width (gray line) was not found to be significantly different than 10 μm according to a one sample t-test (t(9) = 0.4550; p = 0.6599). (H) Histogram of WT C2C12 diameters measured of n = 2322 cells across 5 images of myoblasts in suspension (mean = 15.08 μm; mode = 13.6713 μm; std. dev = 3.1547 μm). (I) Representative image showing C2C12 myoblasts (labeled with a red fluorescent tag) and mouse embryonic stem cell-derived motor neuron spheroids (labeled with a green fluorescent tag) seeded in the muscle chamber and neuron chambers respectively (scale bar = 1 mm).

Similarly, 3D confocal microscopy images of micro-channels separating the neuron and muscle chambers were segmented, measured, and compared to nominal STAMP dimensions (Figure 3f–g). The patterned inlet region closely matched the designed mold dimensions of 10 μm on average (10.864 μm ± 6.007 μm), a value which is similar to the width of individual myoblasts (Figure 3h, Figure S1). Since average channel width may not be representative of the actual width at the muscle chamber interface, we loaded fluorescently labeled C2C12 mouse myoblasts and motor neuron spheroids into the respective chambers of the open-well Limb-on-a-Chip to validate whether our design effectively separates both cell types upon seeding (Figure 3i). Confocal imaging showed that myoblasts remained in the muscle chamber without infiltrating the micro-channels, enabling spatial segregation from the motor neuron spheroids as desired.

Differentiation and Visualization of Contractile Skeletal Muscle within Limb-on-a-Chip

Functional in vitro neuromuscular models require the ability to monitor and quantify skeletal muscle contractile force. We have previously shown that C2C12 mouse myoblasts seeded on micro-patterned fibrin hydrogels containing 25 μm grooves efficiently fuse into globally aligned multinucleated muscle fibers.16,31 To validate compatibility of our established protocols32 with stiffer GelMA hydrogels, we seeded wild-type myoblasts (WT C2C12) into the muscle compartment of a Limb-on-a-Chip. Myoblasts were maintained in growth media containing fetal bovine serum (FBS) until confluence and then transitioned to differentiation media containing horse serum (HS) and human insulin-like growth factor-1 (IGF-1) to promote fusion into mature muscle fibers (Figure 4a). We concurrently seeded and differentiated C2C12 myoblasts engineered to express a genetically encoded calcium sensor (RGECO C2C12) in Limb-on-a-Chip devices using the same protocol. Both WT and RGECO C2C12 myoblasts differentiated into multinucleated muscle fibers with similarly elongated nuclei, fusion index, and fiber width (Figure S4).

Figure 4. Differentiation and Visualization of Contractile Skeletal Muscle within Limb-on-a-Chip.

Figure 4.

Open in a new tab

(A) Representative images for mature WT and RGECO C2C12-derived muscle after 12 days of differentiation. (scale bar = 1 mm). Higher magnification immunostaining images showing myosin-expressing fibers (MF20, pink) composed of multiple fused nuclei (NucB, blue) (scale bar = 200 μm). (B) Traces of WT and RGECO C2C12-derived muscle contractile force in response to 1 Hz electrical stimulation. Unpaired t-tests comparing the (C) maximum contraction force (t(19) = 2.677, p = 0.0149) and (D) time-to-peak-force (t(19) = 3.265, p = 0.0041) between WT (n=10) and RGECO C2C12 (n=11) muscles. e) Principal component analysis plot derived from RNAseq STAR-Salmon alignment and quantification data. (F) Heatmap of normalized muscle-relevant gene expression for WT and RGECO C2C12 displayed on a log2 axis. (G) Normalized gene expression for protein-coding genes of myosin isoforms, ordered by increasing maturity along the myogenic timeline. Two-way ANOVA (F(5,54) = 11.67, p < 0.0001) with post-hoc Šídák’s multiple comparisons tests was performed to find significant differences between groups (p<0.001***, p<0.0001****). Error bars represent standard deviation to mean.

Electrical stimulation of engineered muscle tissues after 1 week in culture (1 Hz, 50 ms pulse width) showed that both WT and RGECO C2C12-derived muscle tissues were contractile in response to stimulation (Figure 4b). Videos of muscle displacement were quantified using our established open-source algorithms27,33 and converted into values of force using measured mechanical properties of the GelMA hydrogel substrate (Figure S2). RGECO C2C12-derived muscles demonstrated reduced contractile force and reduced time required to reach peak force as compared to WT muscle (Figure 4c–d), reflecting slight functional differences between the WT and RGECO C2C12 lines. Calcium sensors in RGECO muscle enabled single cell-resolution in visualizing fiber activation, in addition to tissue-wide functional force readouts (Video S1).

To further quantify differences between WT and RGECO C2C12-derived muscles, we conducted bulk RNA sequencing (RNAseq) of tissues fabricated within our Limb-on-a-Chip devices. While principal component analysis (PCA) showed separate clustering of WT and RGECO C2C12 (Figure 4e), heatmaps of genes that play key functional roles in the muscle contractile apparatus34 showed minimal differences in muscle-relevant gene expression between groups (Figure 4f). Importantly, despite differences in the expression of embryonic and postnatal myogenic markers such as myosin heavy chain 3 and 8 (Myh3, Myh8), muscle tissues derived from both WT and RGECO C2C12 demonstrated similar expression of all mature myosin isoforms including Myh1, Myh2, Myh4, and Myh7 (Figure 4g). Taken together, despite slight differences in function between WT and RGECO C2C12 lines, these experiments validated the ability to fabricate aligned, mature, and contractile skeletal muscle tissues within our Limb-on-a-Chip platform using both WT C2C12 that enable tissue-wide force readouts and RGECO C2C12 that are compatible with both tissue-wide force readouts and single-fiber activation recording. Subsequently, we used RGECO C2C12 for any co-culture experiment that required measurements of calcium activity.

Compartmentalized Co-Culture of Motor Neurons and Muscle in Limb-on-a-Chip

Establishing relevance of our Limb-on-a-Chip platform for neuromuscular co-cultures requires demonstration of compartmentalized long-term co-cultures with muscle tissues. We leveraged established protocols27,35 to differentiate mouse embryonic stem cells (mESCs) into spheroids expressing a green fluorescent marker of Hb9 gene expression, indicating differentiation into a motor neuron lineage (Figure 5a). To promote neuron adhesion onto the GelMA hydrogel, we seeded non-proliferative gamma-irradiated mouse embryonic fibroblasts within the neuron compartments prior to seeding motor neuron spheroids, thus providing a cellular substrate for spheroid adhesion and spreading (Figure 5b). Tissues were maintained in a cocktail of muscle differentiation media 50% (v/v) and neuron differentiation media 50% (v/v) containing glial derived neurotrophic factor (GDNF) and ciliary neurotrophic factor (CNTF). We verified that monocultures of muscles and of motor neurons were able to differentiate in this cocktail medium (Figures S5 and S6), and did not observe significant differences in muscle nuclei circularity, muscle fiber fusion index, neurite length, or number of astrocytes compared to monocultures in muscle-specific or neuron-specific media. Interestingly, we observed a slight significant increase in fiber widths in muscle monocultures grown in the cocktail media (Figure S5b), indicating that this media formulation may promote muscle differentiation.

Figure 5. Compartmentalized Co-Culture of Motor Neurons and Muscle in Limb-on-a-Chip.

Figure 5.

Open in a new tab

(A) Characterization of neuron differentiation timeline. (B) Schematic of Limb-on-a-Chip seeding for both monoculture and co-culture experiments performed on day 8 of differentiation. (C) (Left) Low magnification images of pseudo-brightfield after stamping, (Middle) after seeding MEFs, and (Right) after neuromuscular seeding. (D) Representative image of neurite outgrowth on day 1 of co-culture. (E) Quantification of neurite outgrowth for both monoculture and co-culture (n=24) spheroids using a two-way ANOVA for total neurite length (F(3,181) = 0.472, P = 0.702) and maximum neurite length (F(3,181) = 1.65, P = 0.180). Post-hoc Šídák’s multiple comparisons tests were performed to compare monoculture to coculture at each timepoint. (F) Imaging the same neurite from days 5 to 12 showcases progressive growth along muscle fibers, with immunostaining revealing the co-localization of neurites (TUJ1), muscle (MF20), and nicotinic acetylcholine receptors (α-bungarotoxin), indicating the formation of a neuromuscular junction. (G) Endpoint immunostaining of myosin demonstrates spatial separation of muscle and neurons.

Co-cultures of motor neuron spheroids and myoblasts demonstrated clear spatial segregation and differentiation of both cell types in separate compartments (Figure 5c–d, Figure 5g). To verify whether the presence of muscle influenced the early stages of neuronal growth, we quantified the total length of all neurites projecting out of the motor neuron spheroid, as well as the length of the longest neurite projection from the spheroid, with or without muscle (i.e. in neuronal monocultures and in co-cultures). Our experiments revealed no significant differences in neurite outgrowth between monocultures and co-cultures (Figure 5e). Using our genetically encoded HB9 fluorescent reporter in the motor neurons and genetically encoded calcium indicator in the muscle tissue, we were able to track neurite progression into the muscle chamber via live cell imaging. As shown in Figure 5f, recording a single neurite over a 48-hour period demonstrated alignment along the muscle fiber and formation of neuromuscular junctions, as visualized via co-localization of nicotinic acetylcholine receptors, neurites, and muscle fibers.

Neuromuscular co-cultures within Limb-on-a-Chip devices were maintained over 10 days before recording videos of muscle activity as indicated by the genetically encoded calcium sensor in RGECO muscle tissues (Figure 6a, Video S2). Maximum and minimum intensity projections were made for each video to identify regions with changes in calcium signaling. We defined activation as being at least 1.5x times brighter (maximum intensity projection) as compared to baseline (minimum intensity projection) (Figure 6b). This image processing workflow enabled tracking the activity of several muscle regions within the tissue. After stimulating the neurons with α-latrotoxin, a compound known to trigger neurotransmitter release,36 we measured a significant increase in the number of activated muscle regions (Figure 6c), while the frequency of muscle contraction remained unchanged (Figure 6d). This suggests that chemical stimulation increases the number of firing neurons without altering their spontaneous firing frequency, indicating functional innervation. Moreover, our Limb-on-a-Chip platform and data analysis pipeline enabled capturing calcium flux dynamics of a single activation area before and after stimulation (Figure 6e) as well as global tissue activation in response to chemical stimulation (Figure 6c and Video S3).

Figure 6. Monitoring Muscle Fiber Activation in Neuromuscular Co-Cultures.

Figure 6.

Open in a new tab

(A) Muscle compartment of Limb-on-a-Chip after 10 days of co-culture; scale bar 1mm. (B) Data analysis pipeline used to analyze calcium flux of RGECO muscle. The change in fluorescence is calculated as the normalized difference between the maximum intensity projection (Max IP) and the minimum intensity projection (Min IP). Pixels that are 1.5X brighter than baseline are defined as “activated” regions. (C) Chemical stimulation on Day 10 of co-culture yielded a significant increase in the number of activated areas in the stimulated group (n=8) compared to controls (U=11, P = 0.034; Mann-Whitney test). Error bars represent standard deviation to mean. (D) A power spectral density analysis via Welch’s method was used to identify dominant frequencies of each activated region and was then averaged for each sample (n = 8). There was no significant difference in the dominant frequency of the samples before and after stimulation. Error bars represent standard deviation to mean. (E) Fluorescence fold change ΔF/Fo of an activated region before and after chemical stimulation.

DISCUSSION

Reproducible in vitro models of the neuromuscular interface that enable quantitative readouts of tissue activity have the potential to advance understanding of the neuromuscular system in health and disease.26 Existing model systems, ranging from organoids to organs-on-a-chip, have provided significant insight into neuromuscular biology.5 While self-assembled 3D organoids and 2D co-cultures enable forming functional neuromuscular junctions (that are more readily visualized in 2D cultures), they are typically not compatible with quantitative force readouts and rely on stochastic self-assembly processes that limit inter-tissue reproducibility. Organ-on-a-chip devices enable more repeatable fabrication, and often even separate motor neuron soma and muscle fibers to better mimic in vivo architecture. However, such compartmentalized devices typically rely on 3D tissues which, while compatible with whole-tissue force readouts, do not enable monitoring the activity of single muscle fibers. Taken together, these challenges emphasize the need for new platforms that address the limitations of current tools.

We have developed a Limb-on-a-Chip device that combines the advantages of compartmentalized cultures, as in 3D organs-on-a-chip, with the benefits of high-resolution imaging compatibility, as in 2D co-cultures. Our platform leverages a 1-step molding method to pattern an all-hydrogel chip composed of a “spinal cord” housing independent motor neuron spheroids connected via micro-channels to a downstream muscle “limb”. This format enables reproducible and compartmentalized co-cultures of motor neurons and skeletal muscle, and is compatible with functional tissue-wide force readouts, similar to 3D organs-on-a-chip. Moreover, because our Limb-on-a-Chip maintains a 2D morphology at the neuromuscular interface, it also enables single-cell resolution in monitoring muscle fiber activity, similar to 2D self-assembled co-cultures. Finally, the open-well format of our device enables culturing engineered neuromuscular tissues in the same media, ensuring biochemical crosstalk between cell types that has been shown to be beneficial.27

In this study, we showcase the ability to accurately fabricate Limb-on-a-Chip devices with precisely defined and long-lasting microscale features (Figures 2–3), and also demonstrate that our devices enable compartmentalized monocultures of muscles or motor neurons, as well as compartmentalized co-cultures of muscles and motor neurons (Figures 4–5). Furthermore, we showcase the ability to quantitatively monitor tissue-wide force (Figure 4) and calcium activity in muscle fibers in response to neuron stimulation (Figure 6).

An advantage of our method is that the 3D printed STAMP molds can be readily adapted to modify chip features, such as the number and size of motor neuron and muscle compartments, or the architecture of the micro-channels that connect the two cell types. A limitation of our study is that we did not screen a wide range of hydrogel formulations for fabricating Limb-on-a-Chip devices, highlighting opportunities to identify substrates other than 7.5% GelMA that better promote muscle and motor neuron differentiation while still enabling compartmentalized co-cultures. However, given the compatibility of our STAMP approach with a range of cross-linkable hydrogels, we anticipate that such optimization can be readily performed and adapted to the needs of specific cell lines. For example, functionalizing hydrogels to promote cell adhesion, or testing stiffer hydrogels that better mimic the native properties of skeletal muscle, could help further explore the links between substrate and tissue properties such as maturation and force generation.

Another limitation of our study is that we used mouse cell lines to derive motor neurons and muscles, which do not fully represent human biology.1 Our previous studies have demonstrated that STAMP is also compatible with human cells,16 highlighting an opportunity to extend the Limb-on-a-Chip platform to human neuromuscular co-cultures in the future. Finally, while the open-well format promotes beneficial crosstalk between motor neurons and skeletal muscle,27 it comes with the additional challenge of optimizing a single culture medium for cell types that might have different needs. Additional design modifications, such as higher walls between compartments, may help accommodate two different media formulations in the chip in cases where a single culture medium fails to meet the requirements of highly divergent cell lines.

In summary, we have developed an accessible protocol that enables researchers to fabricate reproducible, compartmentalized, and scalable models of the neuromuscular interface that are compatible with functional readouts at the tissue- and single cell-scale. Our protocol’s use of 3D printed molds could allow others to make custom modifications to Limb-on-a-Chip architecture as needed, such as changes in micro-patterned feature sizes, or the addition of new chambers to enable co-cultures with other cell types. We hope that future studies will leverage our Limb-on-a-Chip system to advance biological understanding of neuromuscular assembly, maturation, and adaptation in both physiological and pathological conditions.

METHODS

Device Fabrication

STAMP Fabrication

Limb-on-a-Chip STAMPs (Figure 2a–b) were designed using CAD software (Solidworks) and prepared as previously described.16 Briefly, the portion of the STAMP containing the main microfeatures of the Limb-on-a-Chip was printed with UpPhoto resin with the NanoOne 2-photon printer (UpNano GmbH) with a 10x objective after slicing exported .stls with UpNano’s slicer software Think3D (UpNano GmbH). STAMPs were post-processed in 3 subsequent isopropyl alcohol baths for 10 minutes each and then allowed to further photo-cure for at least 12 hours before assembly to holders printed using stereolithography (SLA, see Figure 2d). SLA printed holders were designed in CAD software (Solidworks) and then sliced with Preform slicing software (Formlabs) and printed using a Formlabs 3B Stereolithography printer with Biomed Clear Resin (Formlabs). The holders were post-processed in isopropyl alcohol in the Form Wash for 30 minutes and then cured in the Form Cure for 60 minutes at 60 °C. The holders were then run through a Tuttnauer benchtop line autoclave (2840EL-D, Tuttnauer USA Co Ltd) at 121 °C using the preset plastic cycle and allowed to cool to room temperature. Subsequently, the micropatterned UpPhoto stamps were adhered to the biomed clear holders with a thin layer of superglue (JB-JBW-33120, J-B Weld) to create the whole device.

GelMA Solution

7.5% w/v Gelatin Methacrylate (GelMA) solution with 0.5% w/v Lithium phenyl-2,4,6-trimethylbenzoylphosphinate (LAP) (900889, Sigma-Aldrich) was produced by first reconstituting 0.5% w/v LAP in DMEM (D6429–500ML, Sigma Aldrich) at 37 °C for 30 minutes until fully dissolved. After filtering the LAP solution, 7.5% w/v Bovine 225 bloom GelMA (Cytoink) was reconstituted in the LAP solution at 37 °C until fully dissolved. The 7.5% w/v GelMA solution was stored at 4°C until further use.

Limb-On-A-Chip Micropatterning

STAMPs were sterilized via immersion in 3% H2O2 (FisherScientific) overnight at room temperature. Residual H2O2 was rinsed off the STAMPs by immersion in a PBS bath for at least 30 minutes at room temperature. STAMPs were then coated with 10% bovine serum albumin (BSA) (37525, Thermo Fisher) for at least 1 hour at 37°C. We prepared the GelMA for micropatterning by warming it to 37°C. Glass-bottom 24-well plates (P24–1.5H-N, Cellvis) were plasma treated for 30 seconds per well using a handheld oxygen plasma treater to promote forming a flat layer of gel.

Once GelMA was sufficiently warmed and in a liquid state, we pipetted 350 μL of 7.5% GelMA into a plasma-treated well. We subsequently aspirated excess BSA from the stamps carefully and placed the Limb-on-a-Chip stamps into the filled wells. After ensuring there were no visible bubbles in the gel before polymerization, we photo-crosslinked the GelMA around the Limb-on-a-Chip stamp by resting the plate on top of a UV lamp for 2 minutes at approximately 20 mW power (Thorlabs). Then, the STAMPs were carefully removed, and each micropatterned gel comprising a Limb-on-a-Chip was washed once with PBS and kept hydrated in cell culture media.

To ensure Limb-on-a-Chip STAMPs could be reused with high feature fidelity, a three-part cleaning step was followed. First, STAMPs were soaked in a 0.1 mg/mL Liberase solution (5401119001, Sigma Aldrich) for at least 1 hour at 37°C to degrade any residual GelMA stuck to the small features of the STAMP. Then, STAMPs were re-sterilized and cleaned using H2O2 for 30 minutes at room temperature, followed by a final PBS soak for an additional 30 minutes at room temperature. After cleaning, STAMPs could be coated with BSA, enabling reuse.

Rheology

Rheological measurements were performed using a TA Instruments Discovery HR-2 hybrid rheometer equipped with a 20 mm parallel-plate geometry. All tests were conducted at 25 °C using a solvent trap to minimize water evaporation, with a gap distance between 1.0 and 1.3 mm. Amplitude sweep tests were performed at an angular frequency of 10 rad s−1 over a strain range of 10−2 to 102 % to determine the linear viscoelastic (LVE) region. Subsequently, frequency sweep tests were carried out within the LVE range at a fixed strain of 0.5 %, over an angular frequency range of 10−1 to 102 rad s−1.

Cell Culture

Muscle Culture

C2C12 myoblasts were initially cultured on plastic tissue culture treated flasks using muscle growth medium which contained 10% fetal bovine serum (A5670701, Gibco), 1% penicillin-streptomycin (MT30002CI, Fisher Scientific), and 1% L-glutamine (MT25005CI, Fisher Scientific) in a high glucose formulation of DMEM with 4.5 g/L of glucose and sodium pyruvate (MT10013CV, Fisher Scientific). Myoblasts were passaged prior to reaching 70% confluence, with passage numbers between P14 and P19.

Neuronal Culture

We followed established protocols to differentiate mouse embryonic stem cells into spinal motor neurons.25,35 Briefly, prior to beginning neuronal culture, we seeded gamma-irradiated mouse embryonic fibroblasts onto a 6-well plate to serve as a feeder layer for our mouse embryonic stem cell (mESC) culture. The following day, we seeded mESC onto the feeder layer using ES-LIF media, made up of EmbryoMax ES DMEM (SLM-220, Sigma-Aldrich) with 15% fetal bovine serum, 1% penicillin–streptomycin, 1% L-glutamine, 1% Embryo-Max nucleosides (ES-008, Sigma-Aldrich), 1% MEM non-essential amino acids (11140050, Thermofisher Scientific), 0.1 mM β-mercaptoethanol (21985023, Thermofisher), and 0.1% v/v ESGRO mLIF (26050070, Fisher Scientific). After 48 hours, we added our base neuronal medium (NDM) for 1 hour and then trypsinized the wellplate, seeding onto a tissue culture-treated 10 cm dish. Our NDM consisted of an initial half-and-half mixture of Advanced DMEM/F12(12634028, Thermofisher) and Neurobasal medium (21103049, Thermofisher) with 10% KnockOut Serum Replacement (10828010, Thermofisher), 1% pencillin-streptomycin, 1% L-glutamine, and 0.1 mM of β-mercaptoethanol. After 24 hours, we collected the unadhered neuronal cells and moved them to a low adhesion 10 cm petri dish. On day 3 of neuronal differentiation, we collected the medium and spun it down to aspirate the base neuronal medium, adding fresh NDM media which then contained 1 μM of retinoic acid (R2625, Sigma-Aldrich) and purmorphamine (540220, Sigma-Aldrich) (NDM+). We observed the culture and added 3 mL of the neuronal formulation on day 4 of neuronal differentiation. On day 5, we collect the spheroids and refresh the medium, now containing 10 ng/mL of ciliary neurotrophic factor (C3710, Sigma-Aldrich) and glial-derived neurotrophic factors (PR27022, Neuromics) (NDM++).

Limb-On-A-Chip Seeding

Muscle Monoculture Seeding

For mono-culture studies, C2C12s were prepared at a concentration of 3×106 cells/mL and seeded into the Limb-on-a-Chip muscle chamber using 10 μL of cell solution added to 5 μL of co-culture media for a total of 30k cells suspended in a total of 15 μL in the muscle chamber. Myoblasts were maintained in the chip in growth medium containing 1 mg/mL of aminocaproic acid (A2504, Sigma Aldrich). Upon reaching confluence in the chips, myoblasts were switched to differentiation medium to induce myoblast fusion and maturation. Muscle differentiation media consisted of 10% horse serum (26050–088, Gibco), 1% penicillin-streptomycin, 1% L-glutamine, and 50 ng/mL of human insulin-like growth factor-1 (I1146, Sigma-Aldrich) in a high glucose formulation of DMEM. Media was changed daily throughout the course of the experiments.

Neuron Monoculture Seeding

Mouse embryonic fibroblasts were seeded at a concentration of 9.5e3 cells/mm2 into the neuron compartments. After the irradiated fibroblasts adhered to the surface, motor neuron spheroids (Day 8 in NDM++) were seeded into the three neuron wells of each Limb-on-a-Chip.

Neuromuscular Co-culture Seeding

Prior to neuron spheroid seeding, mouse embryonic fibroblasts are seeded into the neuron compartments utilizing the same protocol as the neuron monoculture experiments. To ensure that the muscle chamber was free of residual media, we used a gel loading pipette tip to aspirate any remaining liquid. Next, we pipetted 2.5 μL of co-culture media, consisting of 50% muscle differentiation medium and 50% NDM++, into the muscle chamber. C2C12s were prepared at a concentration of 3×106 cells/mL and seeded into the chamber using 5 μL of cell solution added to 2.5 μL of co-culture media for a total of 30k cells suspended in a total of 7.5 μL in the muscle chamber. Media volume was optimized to prevent fluid flow into the neuron wells. After 45 minutes, once the myoblasts adhered to the hydrogel substrate, wells were filled with co-culture media. Finally, the neuronal spheroids (on day 8 of neuronal differentiation) were pipetted into the microwells generated by STAMP micropatterning.

Muscle Electrical Stimulation

Electrical stimulation was performed 14 days after being switched to differentiation media during muscle-only characterization studies. Cultures were placed in 2 mL of DMEM and supplied with 10 Vpp 5% duty cycle square waves at a frequency of 1 Hz using a Siglent SDG1032X function generator connected to electrodes made of platinum wires (010286.BY, Thermo Fisher Scientific) suspended approximately 7 mm apart and 9 mm above the culture. Muscle twitch videos were recorded in brightfield with a 10x objective at 30 fps with a Zeiss Primovert Microscope.

Bulk RNA Sequencing

RNA extraction was performed on day 14 of muscle differentiation using the Qiagen (74104, Hilden, Germany) RNeasy Mini Kit following the manufacturer’s instructions. The muscle monolayers were treated with the lysis buffer for 15 minutes before being spun down in QIAshredder spin columns (79656, Qiagen) to homogenize the tissue. Sequencing libraries were generated using the NEBNext® Ultra™ II Directional RNA preparation with poly(A) selection kit. Library sizes were quantified and verified via qPCR and Fragment Analyzer before being loaded on the Singular G4 platform in a 50-base paired-end configuration with 8+8 nucleotide indexes. The reads were then mapped against mm39 (GRCMm39) using the BWA-MEM algorithm. Quality control metrics such as mapping rates, unique 20-mers and fraction of ribosomal RNAs were calculated using bedtools version 2.30.0. FASTQ files were processed using the nf-core/rnaseq version 3.17.0 pipeline. We utilized the GRCm39 reference genome and ENSEMBL GRCm39 murine annotations. Differential gene expression was done in DESeq2 in R using the apeglm method to provide Bayesian shrinkage estimators for effect sizes. We reported log2 fold changes and Benjamini-Hochberg adjusted p-values. DESeq2 calculates differentially expressed genes using a negative binomial model on raw RNAseq counts and controls the false discovery rate using adjusted p-values. The data discussed in this publication have been deposited in NCBI’s Gene Expression Omnibus37 and are accessible through GEO Series accession number GSE309796 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE309796).

Immunofluorescence Staining

Samples were fixed in 4% paraformaldehyde for 15 minutes then washed with PBS. Then, the samples were permeabilized using a 0.5% solution of Triton X-100 (A16046.AE, Thermofisher) in PBS for another 15 minutes. Followed by an hour incubation in a blocking buffer of 10% BSA in PBS. For myosin-4 staining, the samples were left at 4C overnight with the blocking buffer previously mentioned with 0.5% Triton X-100, 1:200 dilution of conjugated anti-myosin 4(MF20) antibody to eFluor 660 (50–6503-82, eBioscience), and 2 drops per mL of NucBlue Live ReadyProbes (R37605, Thermofisher). For neuromuscular junction imaging, samples were left overnight at 4C with unconjugated antibodies for β-3 tubulin (TUJ1 1:200, #302 302, Synaptic Systems, Goettingen, Germany), MF20(1:200) and α-bungarotoxin (1:500, B354450, Thermofisher) conjugated to Alexa Fluor 647. The next day, samples were washed with PBS and incubated for an hour at room temperature with the secondary antibodies for TUJ1 and MF20, Alexa Fluor+ 488 and 555 respectively, and NucBlue Live ReadyProbes. Finally, the samples were washed with PBS prior to imaging and submerged in Fluoromount aqueous mounting medium (F4680, Sigma-Aldrich).

Chemical Stimulation of the Motor Neurons

To stimulate the release of neurotransmitters from the motor neurons, we added either 0.1 mM of L-glutamic acid (G8415, Sigma-Aldrich) or 1nM of α-latrotoxin (#LSP-130, Alomone Labs, Israel) to the samples and incubated for 30 min before recording muscle activity (contraction or calcium signaling).

Image Acquisition

Daily fluorescence imaging was performed on a Nikon AXR confocal microscope, utilizing a 4X and 10X lens. Brightfield video recording of muscle contraction and stimulation was performed on a Zeiss inverted microscope. Stereoscope brightfield images of Limb-on-a-Chip STAMP molds (Figure 2e) were captured using a Leica Stereoscope.

Data Analysis

Device Characterization

A Nikon AXR confocal microscope was used to image the autofluorescent regions of the GelMA Limb-on-a-Chip hydrogels created in this work and to characterize the dimensions of the grooves and microchannels of the hydrogel in comparison to modeled STAMP dimensions, as shown in Figure 3a–g. Groove z-stack images from multiple days within the same device were captured with a 20x objective lens over a range defined by the minimum and maximum heights of the grooves and then filtered with Nikon’s Denoise.AI feature. A cross-section of the z-stacks was then processed using ImageJ’s Orthogonal Views feature and subsequently segmented using the Segment Anything Model (SAM) (Meta AI) (see Figure 3a). Groove dimensions were then extracted by taking the boundary of the segmented image and using MATLAB’s findpeaks() function to identify groove boundaries by locating bounding valley regions (see Figure 3c). Average dimensions were calculated by aligning grooves at their first valley point and calculating average values at each location along the groove’s width. Heights were calculated as the distance between a groove’s first valley point and its peak value, while widths were calculated as the distance between two adjacent valley points.

Channel dimensions were characterized with 20x objective confocal images captured in the regions of the channels across multiple chips just after casting. After filtering the images with Nikon’s Denoise.AI feature, the images were further processed with a Gaussian filter to reduce peppering and then binarized with a global threshold of 0.5 using MATLAB’s imbinarize() function (see Figure 3f). 10 μm regions of the channels were cropped, and channel widths were calculated by skeletonizing each channel region and calculating the shortest distance to the edge of the segmented fiber from each skeleton point, using the MATLAB functions bwskel() and bwdist(). The final reported width for each channel section was then calculated by doubling all the measured distances from the edge of the binarized channel to the skeleton points and taking the average width value for each imaged channel (Figure 3g).

Cell Culture Maturity Data Analysis

Neurite quantification was performed utilizing the daily immunofluorescent imaging of our HB9:GFP neurons. The max length and total neurite length were recorded for each sample and day using the ImageJ plugin NeuronJ. Skeletal muscle morphological maturity metrics (Figures S4 and S5) were quantified by segmenting 40x Nikon AXR confocal images of stained skeletal muscle fibers as described previously.16

Displacement Dynamics

Brightfield videos of muscle contraction were analyzed utilizing an open-source full-field displacement and strain tracking code from our prior published work.33 The code extracts spatial displacement maps and calculates mean average displacement values (MAD) metrics for each frame. We extracted max MAD values by taking the vertical distance between a valley and the subsequent peak as found by MATLAB’s findpeaks() function. Contractile shear force measurements were made as described previously,27 with a shear modulus of 1kPa for 7.5% GelMA, an area of the muscle well at 13 mm2, and the thickness of the GelMA assumed to be a nominal dimension of 0.5 mm. Time to peak metrics were found by taking the time between an identified valley and the next adjacent identified peak.

Quantification of Calcium Signaling

Calcium flux videos were acquired on a Nikon confocal microscope at a ~30 fps acquisition speed. Following video recording of spontaneous twitch, α-latrotoxin was added to each well. The samples were then incubated for 30 minutes at 37 °C before recording another video of the same tissue region using the same acquisition parameters. To extract regions that had meaningful calcium transience, maximum and minimum intensity projects were made for each video. We defined activation as being at least 1.5x times brighter (maxip) as compared to baseline (minip). A custom Matlab script was used to identify all unique, or separated, objects larger than the area of a single myoblast (~10μm diameter). Power spectral density analysis was performed via Welch’s method on each activated region to quantify contractile frequency before and after stimulation.

Statistical Analysis

Quantitative results are reported as mean ± standard deviation (SD). Details of sample sizes and statistical tests for each figure are included in the corresponding caption. For Figures 3d–e, One-way ANOVA and Kruskal-Wallis tests were performed using Python’s scipy.stats package. For Figure 3e, one-way t-tests were performed using MATLAB’s ttest() function, comparing designed CAD model values to measured mean widths. The remaining statistical tests were performed using Graphpad Prism 10. For muscle-only studies, when comparing metrics across both cell lines, a two-way ANOVA was performed with post-hoc Šídák’s multiple comparisons tests. Unpaired t-tests were used to compare contraction dynamics.

Supplementary Material

1

Document S1. Supplemental methods, Figures S1–S6

2

Video S1. RGECO C2C12 Electrical Stimulation.

Download video file (4.8MB, mov)
3

Video S2. Co-Culture Calcium Flux Activity.

Download video file (3.9MB, mp4)
4

Video S3. Brightfield video of muscle contraction in response to L-glutamic acid.

Download video file (754.4KB, mp4)

HIGHLIGHTS.

  • A repeatable, one step method for patterning functional neuromuscular co-cultures

  • Co-cultured skeletal muscle and motor neurons show functional innervation

  • Limb-on-a-Chip is compatible with measurements of single-cell calcium signaling

  • Limb-on-a-Chip is compatible with tissue-wide force readouts

THE BIGGER PICTURE.

Diseases and injuries that affect the neuromuscular system, namely skeletal muscles and the motor neurons that control their activity, have a severe negative impact on human health, mobility, and quality of life. There is a critical need for creating model systems that allow researchers to understand the biological mechanisms of neuromuscular diseases, motivating tissue engineers to develop in vitro neuromuscular models that enable investigating neuromuscular growth and function in healthy and diseased states. Existing in vitro models have demonstrated successful co-cultures of skeletal muscle tissue and motor neurons, but struggle to balance reproducible fabrication, high-throughput imaging compatibility, and functional readouts. To address these issues, this work introduces an accessible, one-step fabrication method that uses 3D printed micro-molds to pattern an all-hydrogel platform for spatially compartmentalized culture of motor neurons in a “spinal cord” and muscle tissue in a downstream “limb”. The resulting Limb-on-a-Chip enables formation of functional neuromuscular junctions, and interrogation of downstream muscle activity in response to upstream neuron stimulation. Specifically, the work demonstrates compatibility with both single cell-readouts of calcium activity and tissue-wide readouts of contractile force. This model system aims to provide researchers with a reproducible and scalable method for fabricating functional neuromuscular tissues, as well as other compartmentalized co-cultures, for applications including disease modeling and drug discovery.

ACKNOWLEDGMENTS

The authors thank Prof. Rashid Bashir from the University of Illinois at Urbana-Champaign for the gift of muscle cells expressing genetically encoded calcium ion sensors that were used in this study. The authors acknowledge funding support from the DoD Office of Naval Research Young Investigator Program (N00014-24-1-2060, awarded to R.R.), DoD Army Research Office Early Career Program and PECASE (W911NF-22-1-0126, awarded to R.R.), the NSF CAREER Program (2238715, awarded to R.R.), the DoD DURIP Program (W911NF-24-1-0106, awarded to R.R.), the NSF Graduate Research Fellowship Program (awarded to A.B.), the NIH T32 Neurobiological Engineering Training Program (awarded to L.S.), and the la Caixa Foundation Postdoctoral Fellowship (awarded to T.R.).

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

DECLARATION OF INTERESTS

The authors declare no competing interests.

DECLARATION OF GENERATIVE AI AND AI-ASSISTED TECHNOLOGIES

During the preparation of this work, the author(s) used no Generative AI or AI-Assisted Technologies for the formatting of figures or for writing the main and supplementary text.

RESOURCE AVAILABILITY

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Ritu Raman (ritur@mit.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

RNASeq data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE309796(https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE309796)

.xlsx and .mat data files for select figures have been deposited in Zenedo alongside relevant scripts and are publicly available as of the date of publication at https://doi.org/10.5281/zenodo.19381453

All original code has been deposited in Zenedo and is publicly available at https://doi.org/10.5281/zenodo.19381453 as of the date of publication.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

REFERENCES

  • 1.Gromova A, and Spada ARL (2020). Harmony Lost: Cell–Cell Communication at the Neuromuscular Junction in Motor Neuron Disease. Trends in Neurosciences 43, 709–724. 10.1080/10643389.2012.728825. [DOI] [PubMed] [Google Scholar]
  • 2.Putten MV, Hmeljak J, Aartsma-rus A, and Dowling JJ (2020). Moving neuromuscular disorders research forward : from novel models to clinical studies. Disease Models & Mechanisms. 10.1242/dmm.044370. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Deenen JCW, Horlings CGC, Verschuuren JJGM, Verbeek ALM, and Van Engelen BGM (2015). The Epidemiology of Neuromuscular Disorders: A Comprehensive Overview of the Literature. Journal of Neuromuscular Diseases 2, 73–85. 10.3233/JND-140045. [DOI] [PubMed] [Google Scholar]
  • 4.Lynch E, Peek E, Reilly M, FitzGibbons C, Robertson S, and Suzuki M (2021). Current Progress in the Creation, Characterization, and Application of Human Stem Cell-derived in Vitro Neuromuscular Junction Models. Stem Cell Reviews and Reports. 10.1007/s12015-021-10201-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Leng Y, Li X, Zheng F, Liu H, Wang C, Wang X, Liao Y, Liu J, Meng K, Yu J, et al. (2023). Advances in In Vitro Models of Neuromuscular Junction: Focusing on Organ-on-a-Chip, Organoids, and Biohybrid Robotics. Advanced Materials 35, 2211059. 10.1002/adma.202211059. [DOI] [PubMed] [Google Scholar]
  • 6.Cedillo-Servin G, Al-Jehani EAA, Rossy T, Teixeira SPB, Sage F, Domingues RMA, Raman R, and Castilho M (2025). Meta-adaptive biomaterials: multiscale, spatiotemporal organization and actuation in engineered tissues. Trends in Biotechnology, S0167779925001702. 10.1016/j.tibtech.2025.05.004. [DOI] [PubMed] [Google Scholar]
  • 7.Yang J, Qian S, Chen M, Wang L, Wang Y, Liu J, Xi C, Yang Y, Li Y, Gao C, et al. (2025). Skeletal muscle, neuromuscular organoids and assembloids: a scoping review. eBioMedicine 118, 105825. 10.1016/j.ebiom.2025.105825. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Martins JF, Fischer C, Urzi A, Spuler S, and Sauer S (2020). Self-Organizing 3D Human Trunk Neuromuscular Organoids. Cell Stem Cell 26, 1–15. 10.1016/j.stem.2019.12.007. [DOI] [PubMed] [Google Scholar]
  • 9.Pereira JD, DuBreuil DM, Devlin A-C, Held A, Sapir Y, Berezovski E, Hawrot J, Dorfman K, Chander V, and Wainger BJ (2021). Human sensorimotor organoids derived from healthy and amyotrophic lateral sclerosis stem cells form neuromuscular junctions. Nature Communications 12. 10.1038/s41467-021-24776-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Andersen J, Revah O, Miura Y, Thorn N, Amin ND, Kelley KW, Singh M, Chen X, Thete MV, Walcxak EM, et al. (2020). Generation of Functional Human 3D Cortico-Motor Assembloids Generation of Functional Human 3D Cortico-Motor Assembloids. Cell, 1–17. 10.1016/j.cell.2020.11.017. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Yu H, Yang S, Chen Y, Wu C, Xu J, Yang Y, Wu R, Guo Y, Chen Z, Ding Y, et al. (2025). Construction of a rodent neural network-skeletal muscle assembloid that simulate the postnatal development of spinal cord motor neuronal network. Sci Rep 15, 3635. 10.1038/s41598-025-88292-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Urzi A, Lahmann I, Nguyen LVN, Rost BR, García-Pérez A, Lelievre N, Merritt-Garza ME, Phan HC, Bassell GJ, Rossoll W, et al. (2023). Efficient generation of a self-organizing neuromuscular junction model from human pluripotent stem cells. Nat Commun 14, 8043. 10.1038/s41467-023-43781-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Huang K-Y, Upadhyay G, Ahn Y, Sakakura M, Pagan-Diaz GJ, Cho Y, Weiss AC, Huang C, Mitchell JW, Li J, et al. (2024). Neuronal innervation regulates the secretion of neurotrophic myokines and exosomes from skeletal muscle. Proc. Natl. Acad. Sci. U.S.A 121, e2313590121. 10.1073/pnas.2313590121. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Cheesbrough A, Harley P, Riccio F, Wu L, Song W, and Lieberam I (2023). A scalable human iPSC-based neuromuscular disease model on suspended biobased elastomer nanofiber scaffolds. Biofabrication 15, 045020. 10.1088/1758-5090/acf39e. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Guo X, Gonzalez M, Stancescu M, Vandenburgh HH, and Hickman JJ (2011). Neuromuscular junction formation between human stem cell-derived motoneurons and human skeletal muscle in a defined system. Biomaterials 32, 9602–9611. 10.1016/j.biomaterials.2011.09.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Rossy T, Schwendeman L, Kohli S, Bawa M, Umashankar P, Habba R, Tchaicheeyan O, Lesman A, and Raman R (2025). Leveraging microtopography to pattern multi-oriented muscle actuators. Biomater. Sci, 10.1039.D4BM01017E. 10.1039/D4BM01017E. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17.Duffy RM, Sun Y, and Feinberg AW (2016). Understanding the Role of ECM Protein Composition and Geometric Micropatterning for Engineering Human Skeletal Muscle. Ann Biomed Eng 44, 2076–2089. 10.1007/s10439-016-1592-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Bajaj P, Reddy B, Millet L, Wei C, Zorlutuna P, Bao G, and Bashir R (2011). Patterning the differentiation of C2C12 skeletal myoblasts. Integrative Biology 3, 897–909. 10.1039/c1ib00058f. [DOI] [PubMed] [Google Scholar]
  • 19.Jangir H, and Hickman JJ (2023). Mimicking the Tendon Microenvironment to Enhance Skeletal Muscle Adhesion and Longevity in a Functional Microcantilever Platform. ACS Biomater. Sci. Eng 9, 4698–4708. 10.1021/acsbiomaterials.3c00235. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Zhang W, Yu L, Pan J, Deng J, Tong X, Zhao B, Liu W, Sun L, Zhang M, Han X, et al. (2025). Geometrically-engineered human motor assembloids-on-a-chip for neuromuscular interaction readout and hypoxia-driven disease modeling. Nat Commun 16, 8693. 10.1038/s41467-025-63736-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Cvetkovic C, Rich MH, Raman R, Kong H, and Bashir R (2017). A 3D-printed platform for modular neuromuscular motor units. Microsystems and Nanoengineering 3, 1–9. 10.1038/micronano.2017.15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Aydin O, Passaro AP, Elhebeary M, Pagan-Diaz GJ, Fan A, Nuethong S, Bashir R, Stice SL, and Saif MTA (2020). Development of 3D neuromuscular bioactuators. APL Bioengineering 4, 016107. 10.1063/1.5134477. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Osaki T, Uzel SGM, and Kamm RD (2020). On-chip 3D neuromuscular model for drug screening and precision medicine in neuromuscular disease. Nature Protocols. 10.1038/s41596-019-0248-1. [DOI] [PubMed] [Google Scholar]
  • 24.Vila OF, Chavez M, Ma SP, Yeager K, Zholudeva LV, Qu Y, Nash TR, Lai C, Col JM, Feliciano CM, et al. (2021). Bioengineered optogenetic model of human neuromuscular junction. Biomaterials 276. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Uzel SGM, Platt RJ, Subramanian V, Pearl TM, Rowlands CJ, Chan V, Boyer LA, So PTC, and Kamm RD (2016). Microfluidic platform for the formation of optically excitable, three-dimensional, compartmentalized motor units. Science Advances. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Fralish Z, Lotz EM, Chavez T, Khodabukus A, and Bursac N (2021). Neuromuscular Development and Disease: Learning From in vitro and in vivo Models. Front. Cell Dev. Biol 9, 764732. 10.3389/fcell.2021.764732. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Bu A, Afghah F, Castro N, Bawa M, Kohli S, Shah K, Rios B, Butty V, and Raman R (2024). Actuating Extracellular Matrices Decouple the Mechanical and Biochemical Effects of Muscle Contraction on Motor Neurons. Adv Healthcare Materials, 2403712. 10.1002/adhm.202403712. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Aydin O, Passaro AP, Raman R, Spellicy SE, Weinberg RP, Kamm RD, Sample M, Truskey GA, Zartman J, Dar RD, et al. (2022). Principles for the design of multicellular engineered living systems. APL Bioengineering 6, 010903. 10.1063/5.0076635. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Roth JG, Huang MS, Navarro RS, Akram JT, LeSavage BL, and Heilshorn SC (2023). Tunable hydrogel viscoelasticity modulates human neural maturation. Sci. Adv 9, eadh8313. 10.1126/sciadv.adh8313. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Kirillov A, Mintun E, Ravi N, Mao H, Rolland C, Gustafson L, Xiao T, Whitehead S, Berg AC, Lo W-Y, et al. (2023). Segment Anything. In 2023 IEEE/CVF International Conference on Computer Vision (ICCV) (IEEE), pp. 3992–4003. 10.1109/ICCV51070.2023.00371. [DOI] [Google Scholar]
  • 31.Heisser RH, Bu A, Schwendeman L, Rossy T, Umashankar P, Butty V, and Raman R (2025). Physiological and functional characterization for high-throughput optogenetic skeletal muscle exercise assays. Bioengineering & Transla Med, e70101. 10.1002/btm2.70101. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Raman R, Cvetkovic C, and Bashir R (2017). A modular approach to the design, fabrication, and characterization of muscle-powered biological machines. Nature Protocols 12, 519–533. 10.1038/nprot.2016.185. [DOI] [PubMed] [Google Scholar]
  • 33.Rios B, Bu A, Sheehan T, Kobeissi H, Kohli S, Shah K, Lejeune E, and Raman R (2023). Mechanically programming anisotropy in engineered muscle with actuating extracellular matrices. Device 1, 100097. 10.1016/j.device.2023.100097. [DOI] [Google Scholar]
  • 34.Van Der Wal E, Iuliano A, In ‘T Groen SLM, Bholasing AP, Priesmann D, Sharma P, Den Hamer B, Saggiomo V, Krüger M, Pijnappel WWMP, et al. (2023). Highly contractile 3D tissue engineered skeletal muscles from human iPSCs reveal similarities with primary myoblast-derived tissues. Stem Cell Reports, S2213671123003132. 10.1016/j.stemcr.2023.08.014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Wichterle H, and Peljto M (2008). Differentiation of mouse embryonic stem cells to spinal motor neurons. Current Protocols in Stem Cell Biology, 1–9. 10.1002/9780470151808.sc01h01s5. [DOI] [PubMed] [Google Scholar]
  • 36.Deak F, Liu X, Khvotchev M, Li G, Kavalali ET, Sugita S, and Sudhof TC (2009). α-Latrotoxin Stimulates a Novel Pathway of Ca2+-Dependent Synaptic Exocytosis Independent of the Classical Synaptic Fusion Machinery. Journal of Neuroscience 29, 8639–8648. 10.1523/JNEUROSCI.0898-09.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Edgar R (2002). Gene Expression Omnibus: NCBI gene expression and hybridization array data repository. Nucleic Acids Research 30, 207–210. 10.1093/nar/30.1.207. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1

Document S1. Supplemental methods, Figures S1–S6

NIHMS2174474-supplement-1.pdf (840KB, pdf)
2

Video S1. RGECO C2C12 Electrical Stimulation.

Download video file (4.8MB, mov)
3

Video S2. Co-Culture Calcium Flux Activity.

Download video file (3.9MB, mp4)
4

Video S3. Brightfield video of muscle contraction in response to L-glutamic acid.

Download video file (754.4KB, mp4)

Data Availability Statement

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Ritu Raman (ritur@mit.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

RNASeq data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE309796(https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE309796)

.xlsx and .mat data files for select figures have been deposited in Zenedo alongside relevant scripts and are publicly available as of the date of publication at https://doi.org/10.5281/zenodo.19381453

All original code has been deposited in Zenedo and is publicly available at https://doi.org/10.5281/zenodo.19381453 as of the date of publication.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

RNASeq data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE309796(https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE309796)

.xlsx and .mat data files for select figures have been deposited in Zenedo alongside relevant scripts and are publicly available as of the date of publication at https://doi.org/10.5281/zenodo.19381453

All original code has been deposited in Zenedo and is publicly available at https://doi.org/10.5281/zenodo.19381453 as of the date of publication.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

RESOURCES