Microfluidic device to study spatial and temporal response of astrocyte networks in response to changes in the biochemical milieu

Abstract

Astrocytes are the most abundant cell type in the central nervous system and are major players in brain homeostasis and inflammation. Astrocytes form a syncytial three-dimensional (3D) network across the brain tissue. Establishing a 3D network allows astrocytes to communicate with each other by a variety of mechanisms, including Ca²⁺ transients. Conversely, spatial and temporal disruption of the astrocyte network can have detrimental effects on brain biology. However, traditional in vitro models have struggled to generate astrocyte networks that can be spatially and temporally perturbed, which limits our capacity to understand astrocyte group dynamics. To address this, we developed a microphysiological platform to investigate both the ways in which 3D astrocyte networks form over time and how they are affected by localized perturbations in the biochemical milieu. We observed that extracellular matrix composition played a critical role in the development of astrocyte network structure, leading to fully interconnected networks within 48 h in optimal conditions. Furthermore, we observed that transient exposure to reactive oxygen species led to long-term disruption of the astrocyte network. This network collapse was accompanied by a decrease in astrocyte redox potential and loss of mitochondrial architecture, which transitioned from an organized filamentous pattern to small and fragmented mitochondria. Additionally, exposure to reactive oxygen species immediately led to disruption of Ca²⁺ transients. Interestingly, even following transient exposure, astrocytes exhibited persistent disruption of the network architecture, with individual cells still exhibiting fragmented mitochondria and Ca²⁺ signaling impairment. These findings highlight how temporary perturbations of the biochemical milieu can result in long-term changes in astrocyte behavior.

Introduction

Astrocytes, the star-shaped¹ glial cells, are ubiquitous within the central nervous system (CNS), representing between 20-50% of the cell population therein². Previously regarded as mere support cells within the brain, astrocytes, with their characteristic spongiform morphology, are now understood to play critical roles in maintaining homeostatic function, neuron survival, synaptogenesis and pruning, and form an integral component of the blood-brain barrier (BBB) and the neurovascular unit (NVU). Astrocytes also mediate the immune response by adopting a reactive phenotype³ that allows them to mount a defensive response to acute viral or bacterial infection or injury, such as ischemic stroke or traumatic brain damage. However, astrocytes are also associated with multiple pathologies like Alzheimer’s disease and Amyotrophic Lateral Sclerosis³ where dysregulation of their pro-inflammatory functions leads to neuron loss and tissue damage. Astrocytes involved in harmful processes are also associated with dysregulation of calcium signaling, mitochondrial alterations, and higher levels of reactive oxygen species (ROS)⁴.

Astrocytes form a functional syncytium⁵, i.e., a network of interconnected astrocytes that maintains biochemical and electrical homeostasis in the brain tissue⁵. As they mature, initially unconnected neonatal astrocytes gradually evolve into a functional network in which astrocytes communicate with one another through a variety of mechanisms, including gap junctions and Ca²⁺ oscillations. Transient calcium elevations are the primary form of excitation-based communication in astrocytes⁶ with crucial roles in modulating synaptic transmission and plasticity, neuronal spiking, gliotransmission, and cerebral blood flow in physiological and pathological states^7,8. Ca²⁺ signaling relies on intricate changes in [Ca²⁺] gradients between the extracellular space and cytosol, and between cytosol and intracellular endoplasmic reticulum and mitochondrial Ca²⁺ stores⁹. Disruption of astrocyte networks can disrupt brain homeostasis and neuron function; in this context, oxidative stress, caused by the accumulation of ROS, represents a critical challenge for astrocyte function. The brain operates at high metabolic rates and is responsible for 20% of the body’s total energy consumption despite accounting for only 2% of body mass. This energy consumption results in elevated levels of free radicals like ROS that may surpass the antioxidant capacity of the CNS in some pathological conditions or during exposure to ROS-inducing agents like alcohol¹⁰. ROS gives rise to a range of adverse effects, including premature cellular stress-induced senescence (SIPS)^11,12. In SIPS, affected cells exhibit similar morphological and functional characteristics to those observed in replicative senescence, including during the natural aging process and disease states¹¹.

As an evolutionary survival mechanism, astrocytes exhibit resilience under mild oxidative stress; however, elevated levels of oxidative stress observed under pathological conditions may exert deleterious effects on both the structure and function of the astrocytic network¹³. Excess H₂O₂ is known to underpin myriad neurodegenerative diseases by inducing oxidative stress, premature senescence, perturbed calcium homeostasis, and mitochondrial dysfunction¹⁴. As astrocytes appear to rapidly sense changes in H₂O₂, they may serve as a first line of defense against CNS injury and modify the cerebral microenvironment¹⁵. Exogenous compounds may also disrupt astrocyte signaling networks. For example, as a widely used neurotoxic compound, the neuropathological effects of alcohol have been extensively studied^16,17,18; nonetheless, the consequences of alcohol exposure on astrocytes remain poorly characterized¹⁸.

Great strides have been made in elucidating the role of the astrocyte syncytium in response to insults using traditional two-dimensional (2D) in vitro cell culture techniques such as Petri dishes and well plates. However, cellular monolayers cultured on a stiff plastic surface are unable to faithfully recapitulate the 3D structure inherent to physiology, cell-cell and cell-extracellular matrix (ECM) interactions, and the in vivo-like microenvironment^19,20. In vivo studies using animal models, primarily rodents, mitigate some of the challenges associated with 2D in vitro techniques; however, they offer their own set of limitations, including decreased ability to monitor how punctual changes in the biochemical environment spatially and temporally affect complex cellular systems like astrocyte networks.

Recent advances in 3D cell culture have demonstrated the utility of microphysiological models for studying astrocyte and neural network behavior. For example, Batenburg et al.²¹ developed a co-culture system of hiPSC-derived neurons and primary human astrocytes to investigate tauopathies, enabling spatially complex interactions within hydrogel matrices. Athanasopoulou et al.²² reviewed biomaterial strategies that support astrocytic phenotypes and mechanobiological responses in engineered 3D environments. Matthiesen et al.²³ introduced peptide-functionalized hydrogels for bioprinting astrocyte networks, facilitating modular studies of astrocyte interactions with tumor cells. These studies underscore the growing relevance of 3D microphysiological platforms (MPS) for modeling astrocyte networks; however, such platforms are often unable to control dynamic spatial and temporal changes within a single device, often requiring complex designs to generate gradients and offering limited capacity for the user to manipulate such gradients on demand.

To address these limitations, we used an MPS²⁴ to facilitate the induction of reversible gradients (i.e., of experimental drugs, toxic agents, metabolites, etc.) and evolution of the astrocyte microenvironment across time and space. Although 3D cell culture in hydrogels is well-established, the ability to generate on-demand reversible gradients introduces new opportunities for studying spatiotemporal dynamics. Unlike conventional platforms such as 384-well plates, which rely on static dosing and lack spatial resolution, this MPS enables the generation of continuous and reversible chemical gradients across a 3D cell-laden hydrogel. This allows for real-time study of spatially distinct astrocyte responses to dynamic stimuli, such as oxidative stress, drug exposure, and ischemic insults. The ability to monitor localized effects, including gradient-driven cell death and mitochondrial changes, offers a more versatile system to model graduated tissue microenvironments, enabling longitudinal studies of recovery, adaptation, or treatment response.

We stimulated astrocytes seeded within our MPS with H₂O₂ and ethanol in both a global and local manner to determine whether there is an effect on the genesis of astrocytic syncytia and on established networks. To quantify these disruptions, we investigated changes to the astrocytic arbor, the mitochondrial network, and astrocyte communication in the form of dysregulated calcium signaling. This permitted us to capture perturbations in astrocyte morphology and function ranging from the macro to the micro scale with precision in space and time.

Results

Microdevice operation

In vivo, astrocytes form complex 3D multicellular networks²⁵. However, in 2D cultures, astrocytes tend to grow as single cells with limited, if any, multicellular association²⁶. We have established a 3D astrocyte culture that generates an interconnected network, which allows us to capture dynamic changes in response to local perturbations of the biochemical milieu.

This platform is based on a 384-well plate to allow compatibility with standard laboratory instruments (i.e., microscopes, plate readers)²⁷. To fabricate the devices, the bottom of 384-well plates was micro milled to generate a series of microculture chambers connected to the wells on top of the plate (Fig. 1a). Diffusion ports located at the base of each well result in connectivity between the chambers and the well array above; these permit both loading of hydrogels and cells into the chamber and the diffusion of media and/or compounds added to the reservoir wells into the gel beneath. This approach produced individual, self-contained chambers, each of which comprised a series of interconnected wells to allow spatial and temporal manipulation of the biochemical environment (Fig. 1b).

Fig. 1: Microdevice fabrication and operation.

a Microdevice fabrication schematic Microchambers and diffusion ports are fabricated with CNC milling and a glass base attached with medical-grade tape. Representative image of a device loaded with food dye for illustrative purposes. b Device loading scheme. Astrocytes are embedded in a basement membrane extract (BME) or collagen hydrogel and loaded into the device via diffusion ports located at the bottom of the well array above. The hydrogels are allowed to polymerize for 15 min and then the cell culture medium is added to the microwell array. Fluorescence microscopy images showing rat cortical astrocytes are labeled with CellTracker Green (CTG) and propidium iodide (PI) to discriminate live and dead cells, respectively. c Representative fluorescence microscopy images of viability and network formation in collagen type I (left) and basement membrane extract (right). d Representative fluorescence microscopy images showing network architecture with the cytoskeleton labeled with phalloidin (red) and nuclei labeled with Hoechst (blue) in collagen (left) and BME (right) hydrogels. Lower images: network architecture analysis with the following vectorial elements: Junction = Node groups forming a bifurcation; Extremity = pixel with a single neighbor; Segment = binary line connected to two junctions; Branch = binary line connected to a single junction and a single extremity; Isolated element Green = branches; Magenta = segments; Orange = master segments; Red surrounded by blue = nodes surrounded by junctions; Junctions surrounded by red = master junctions; Blue = isolated elements; Cyan = small isolated elements; Red surrounded by yellow = extremities. e Comparison of network architecture between collagen and BME astrocyte-laden hydrogels. Radar plot indicating differences across vectorial elements outlined in (d), with BME resulting in an interconnected network in comparison to collagen. f Violin plots showing the statistical significance of vectorial elements shown in (e). *, **, ***, and **** denote p-value < 0.05, <0.01, <0.005, and <0.001, respectively

We established the importance of the ECM composition in astrocyte behavior by culturing the cells in type I collagen hydrogels^24,28 or in Cultrex RGF 3D ECM matrix, a rodent basement membrane extract that includes multiple components such as laminin, collagen IV, entactin, and heparin sulfate proteoglycans (R&D systems). To form a syncytium, astrocytes in vivo are known to extend highly ramified processes to make contact with surrounding cells²⁹. The brain does not contain large amounts of fibrillar collagen I³⁰ and, as we observed that astrocytes cultured in collagen hydrogels adopted a more rounded morphology with fewer processes - and thus fewer opportunities for interconnectivity with other astrocytes²⁹ - we determined that an alternative hydrogel that could generate a superior network was required. To compare the performance of the ability of hydrogels to form a 3D scaffold that can support the generation of astrocytes with numerous processes, and, therefore, an interconnected astrocyte network, we loaded an astrocyte cell suspension in both collagen and basement membrane hydrogels through diffusion ports located at the bottom of the chamber well and allowed the gel to polymerize for 15 min. We then added cell culture medium to the wells above the loaded chambers to provide nutrients to the astrocytes in the gel beneath. Subsequent microscopy analysis showed that both ECM(s) maintained high astrocyte viability; however, visual inspection of cell morphology revealed significant differences in cell morphology and cellular organization (Fig. 1c). We used computational analysis to provide a quantitative comparison between the collagen and basement membrane extract (BME) hydrogels (Fig. 1d). Collagen yielded the formation of a loose network where most of the astrocytes that developed 3D projections remained as single cells, with only a minority establishing multicellular connections. Conversely, culturing astrocytes in the BME hydrogel led to the emergence of a complex multicellular network with more astrocytes developing projections linked to adjacent cells (Fig. 1e) and significant differences across the network parameters studied (Fig. 1f and Supplementary Fig. 1).

Effects of transient biochemical perturbations on astrocyte network viability

We investigated whether biochemical perturbations associated with high levels of oxidative stress or ethanol (EtOH) can affect the astrocyte network. In line with prior literature^31,32,33 we assessed cell viability across the network in response to 2-h exposures to high and low doses of H₂O₂, 880 µM and 200 µM, respectively, and 24-hour exposures to three concentrations of EtOH³⁴ (25 mM, 200 mM, and 400 mM). Previous studies have documented a 2-h H₂O₂ exposure protocol to induce oxidative stress via production of the OH radical arising from activity between H₂O₂ and Fe²⁺ ions - Fenton’s reaction³⁵. EtOH, on the other hand, generates oxidative stress, among other interactions, through more complex mechanisms. We stained astrocytes with CellTracker Green and propidium iodide (red) to monitor live and dead cells, respectively. These stained cells were combined with the BME in the MPS and loaded into the platform (Fig. 2a). We exposed the cells to H₂O₂ or EtOH immediately following gel polymerization to determine whether these abrupt changes in the biochemical milieu impair the genesis of the astrocyte network and whether this effect might arise due to reductions in cell viability or whether more nuanced factors play a role. In both the H₂O₂ and EtOH conditions, we loaded all reservoir wells with cell culture medium containing our agents of interest to ensure that all cells within the chamber were exposed, and in line with previous astrocyte hydrogen peroxide and alcohol exposure studies, we allowed the platform to incubate for 2 h after H₂O₂ exposure and for 24 h after administration of EtOH^36,37.

Fig. 2: Transient biochemical perturbations affect astrocyte network viability.

a Schematic representation of the reagent loading protocol to induce effects across the entire chamber (i.e., added to all wells above the gel contained in the chamber). b Fluorescence microscopy images show viability under control, high (880 µM), and low (200 µM) concentrations of H₂O_2. Live cells labeled with CTG (green) and dead cells with PI (red). c Cell viability in response to H₂O₂ across 48 h. d Fluorescence microscopy images show viability under control, high (400 mM), and low (25 mM) concentrations of EtOH. Live cells labeled with CTG (green) and dead cells with PI (red). e Cell viability in response to EtOH across 48 h. * and *** denote p-value < 0.05 and <0.005, respectively

To determine astrocyte viability following a transient insult, we imaged immediately after administration of H₂O₂ to establish baseline conditions, and again at 24 h and 48 h to monitor changes in the number of live (green) and dead (red) cells in response to H₂O₂ (Fig. 2b). The results indicated transient (2 h) exposure to high oxidative stress (880 µM H₂O₂) caused a moderate but statistically significant decrease in cell viability after 24 h (from 95 to 90%) that became progressively worse at 48 h (reaching 85% viability). Exposure to lower concentrations (200 µM H₂O₂) revealed no significant changes in viability after 24 h, but with a significant decrease after 48 h (reaching 90% viability) (Fig. 2c). This suggests that transient H₂O₂ exposure not only exerts a dose-dependent deleterious effect on astrocyte viability, but that this effect increases over time despite removal of the oxidative conditions.

Transcriptional profiling revealed dynamic changes in interferon-stimulated genes (Supplementary Fig. 2) at 24 h and 120 h following H₂O₂ exposure. While some genes, such as LY6E and MX1, exhibited minimal variation over time, others demonstrated significant changes. Notably, IL18BP showed marked upregulation at 120 h, whereas GBP1 remained consistently downregulated, and ISG15 was significantly downregulated at 120 h in comparison to 24 h. These findings reinforce the concept that astrocytes survived oxidative stress and engaged in a temporally orchestrated transcriptional program.

Likewise, we also determined a similar trend in astrocytes to high and low doses of EtOH (Fig. 2d), with significant declines in viability in the experimental groups compared to the control. This effect was most pronounced between control conditions and the high exposure group at 48 h (Fig. 2e). We also observed that high doses of EtOH led to heightened cell death when compared to both lower doses and control conditions, and that these changes persisted over time. Taken together, these results suggest that irrespective of whether the disrupting agent remains or is removed from the system, enduring changes that lead to a reduction in viability are evident.

Local astrocyte viability

For this set of experiments, the aim was to determine whether local stimulation would lead to alterations in viability and network architecture in distal regions (i.e., far from the stimulated area) (Fig. 3a). We achieved this by seeding astrocytes in the platform and waiting 24 h for the astrocytic network to form in 3D. Next, we added H₂O₂ for 2 h to one microwell while the other microwells remained in the media. After the 2 h H₂O₂ exposure, we washed all the wells with fresh media three times with 10 minutes between each washing step. This approach allowed us to limit direct exposure to only those astrocytes located underneath the well where we added H₂O₂. We assessed cell viability in the proximal (treated) well, the middle interface region, and the distal region (i.e., the well located furthest from the treated area) (Fig. 3b). At 48 h, we observed statistically lower astrocyte viability in the proximal region in comparison to the interface and distal regions, with these regions showing no changes in viability compared with pre-exposure conditions (Fig. 3c). We then repeated the spatial protocol with EtOH, exposing the cells after the network had formed (Fig. 3d). The results demonstrated that, at 48 h, the percentage of viable astrocytes was significantly lower in the proximal region than the distal (Fig. 3e). However, in contrast to the clear spatial patterning of viability across the proximal, interface, and distal regions in the H₂O₂ condition, EtOH decreased astrocyte viability across all three regions. This potentially occurred due to the volatile nature of EtOH, leading to exposure of distal regions to cytotoxic EtOH vapors³⁸. MPS(s) were placed in a sealed container containing EtOH, as described in Dolganiuc (2009)³⁸, to counteract fluctuations arising from evaporation. However, it’s possible this created an ethanol-saturated environment within the container, thus precluding the ability to maintain a spatial gradient.

Fig. 3: Schematic representation of local stimulation of the network with H₂O₂ and EtOH.

a Experimental agents are added to the proximal (treatment) well to generate a gradient of biochemical perturbation across the proximal, interface, and distal regions. b Fluorescence microscopy images showing viability across proximal, interface, and distal regions in response to local network perturbation with H₂O_2. (880 µM)_. Live cells labeled with CTG (green) and dead cells with PI (red). c Viability in proximal and distal regions at 24 h and 48 h, and viability across proximal, interface, and distal regions at 48 h. d Fluorescence microscopy images showing viability across proximal, interface, and distal regions in response to local network perturbation with EtOH (100 mM)_. Live cells labeled with CTG (green) and dead cells with PI (red). e Viability in proximal and distal regions at 24 h and 48 h, and viability across proximal, interface, and distal regions at 48 h. *, ***, and **** denote p-value < 0.05, <0.005, and <0.001, respectively

To further quantify these effects, we performed a simulation of EtOH diffusion and evaporation based on the administration of 25 mM EtOH. The results indicate that the relative mass loss due to EtOH evaporation accounts for 13.56% of the total EtOH mass, throughout the immediate 2 h of the experiment, once the EtOH was added. The simulation also reveals that the bottom of the well reaches a concentration very close to 25 mM. This suggests that the impact of evaporation on the well availability of EtOH is rather limited.

We also considered EtOH evaporation from the liquid in the reservoir to the air above the wells. Our simulation suggests the EtOH concentration in the air chamber reaches a maximum of ~6.02 mM (at 2 h). This value is ~4-fold smaller than the concentration in the chamber. Additionally, this concentration happens at a very long distance from the adjacent well (8 mm). Therefore, we expect the evaporated EtOH effect to be negligible on the adjacent well as compared to diffusion through the hydrogel.

Importantly, our simulation approach assumes that EtOH will disperse in a “virtually infinite” region of air (i.e., the incubator), which represents a less restrictive (i.e., more evaporative) scenario than the one we simulated. Therefore, our approximation provides an upper boundary on mass loss that reflects a “worst case” scenario in terms of EtOH loss due to evaporation.

However, as the spatial dimension was not reliably replicated in the alcohol condition, we henceforth confined further EtOH experiments to the global format.

Overall, the results demonstrate that although administration of exogenous H₂O₂ and EtOH promoted lasting changes and induced cell death in a dose-dependent manner, viability remained high as the majority of the astrocytes were still alive even after 48 h, a trend which persists even after 7 days post-exposure (Supplementary Fig. 4B, C) and a local decrease in cell viability did not affect astrocyte viability across other regions of the network.

Astrocyte network genesis and collapse in response to stress

Our previous experiments established that astrocyte viability remained high (>80%) following exposure to both H₂O₂ and EtOH, suggesting that changes in network structure could not be explained by cell death alone. Thus, we moved to analyze astrocyte networks quantitatively before and after perturbation with H₂O₂ or EtOH. After culturing astrocytes in the MPS in media or the presence of H₂O₂ or EtOH, we fixed the samples and stained the actin fibers and nuclei with Phalloidin (red) and Hoechst (blue), respectively (Fig. 4a). This approach provided superior imaging quality compared to cell trackers to visualize the astrocyte network. We performed network structure analysis in FIJI image processing software³⁹ to generate multiple quantitative descriptors (i.e., vectorial elements) to define astrocyte syncytium architecture⁴⁰. Quantitative analysis of network architecture revealed that treatment with H₂O₂ decreased the area occupied by the syncytium 10-fold (Fig. 4b). The number of network nodes (defined by cell nuclei) remained constant while the average branch length decreased by >10-fold, suggesting that the decrease in network area was caused by impairment of astrocytes’ capacity to form connections with their neighbors. Similarly, we also observed that in the control groups most astrocytes generated interconnected closed loops (indicated by the higher number of master nodes); however, H₂O₂ treatment led to disorganized and fragmented patches, (indicated by a reduction in the number of master nodes and an increase in dead ends) (Fig. 4c and Supplementary Fig. 5). Overall, although cell viability was not greatly affected, the administration of H₂O₂ led to significant disruption in the genesis of the astrocyte network with a reduction in the vectorial elements (e.g., branches, junctions, meshes) required for an interconnected cellular syncytium and an increase in isolated segments (i.e., unconnected elements).

Fig. 4: Analysis of network architecture genesis and collapse in response to stress.

a Schematic representation of fixing cells in the device and subsequent F-actin phalloidin staining (red) and Hoechst staining of nuclei (blue) protocol. b Fluorescence microscopy images show network genesis under control, low (200 µM), and high (880 µM) concentrations of H₂O₂. c Radar plot indicating differences across network vectorial elements in response to H₂O₂. d Fluorescence microscopy images showing network genesis under control, low (25 mM), and high (400 mM) concentrations of EtOH conditions. e Radar plot indicating differences across network vectorial elements in response to EtOH. f Fluorescence microscopy images showing network stability in response to local stimulation with H₂O₂ (880 µM) across proximal, interface, and distal regions. g Radar plot indicating differences across network vectorial elements in response to local H₂O₂ administration

In the EtOH condition, where a large network comprised of large numbers of nodes and branches was formed under control conditions, we observed a significant reduction in the number of nodes, segments, total length, and pieces observable in the 25 mM alcohol exposed condition (Fig. 4d). We also identified that the network structure in the 400 mM condition was greatly impaired with fewer junctions and segments, reduced segment length, branch length and number, and master segment length and number, reduced total mesh length and area, and total length compared both to the control and 25 mM conditions. We also observed that the branching interval and mesh size were significantly reduced in the 400 mM condition compared to control (Fig. 4e and Supplementary Fig. 6). Unlike cells cultured under control conditions, EtOH-exposed cells were unable to generate an interconnected network and exhibited greater numbers of isolated segments, thus impeding astrocytes’ ability to form a syncytium. Consequently, these findings suggest that both H₂O₂ and EtOH interfere with the ability of astrocytes to develop a network and that the impairment in network formation occurred in a dose-dependent manner, with increased concentrations of H₂O₂ and EtOH leading to greater disruption. Altogether, these results showed that astrocyte network structure is affected by changes in the biochemical environment, such as increased levels of H₂O₂ and EtOH, which impair the capacity of astrocytes to self-assemble in an interconnected network by a variety of mechanisms.

The previous findings demonstrated that H₂O₂ and EtOH affected the formation of a new network. In the next set of experiments, we set out to evaluate whether these agents can derail the structure of a pre-existing network. We assessed this by treating only one well (defined as proximal well) of the three-well device with H₂O₂ to generate a gradient across these proximal, interface, and distal regions (Fig. 4f). We determined that H₂O₂ negatively affects pre-established networks with significant reductions across a range of parameters indicative of astrocyte arborization, such as branch length and branching intervals, observed between proximal and distal regions as well as distal and interface areas (Fig. 4g and Supplementary Fig. 7). These findings suggest that the stability of established astrocyte networks can be eroded in a dose-dependent manner (i.e., the gradient of biochemical disruption) and persist over time (Supplementary Fig. 4C, D).

Overall, both H₂O₂ and EtOH had a detrimental effect on the network parameters analyzed, and this was the case for both the developing and established networks. Taken together, this indicates that even sublethal doses of H₂O₂ and EtOH (which showed >80% cell viability) inflicted considerable damage on both the developing and established astrocyte syncytium.

Transient exposure to oxidative stress disrupts astrocyte mitochondrial architecture

Mitochondria underpin metabolic function, cell survival, proliferation, and energy supply via the generation of adenosine triphosphate (ATP)⁴¹ and the production of other metabolic intermediates essential for cell function. Mitochondria are highly dynamic organelles that, based on relative rates of fission and fusion, undergo structural changes, including shifts from filamentous to punctate states, with enlarged and filamentous mitochondria often showing superior mitochondrial function (e.g., oxygen consumption capacity)⁴² compared with fragmented and isolated mitochodnria⁴³. Inflammation and injury to astrocytes induce changes in metabolic pathways along with associated alterations of mitochondrial dynamics, morphology, and loss of mitochondria from processes⁴¹. To determine whether our exposure protocols affected mitochondria, we stained astrocytes with MitoTracker before adding them to the hydrogel (Fig. 5a). We added H₂O₂ and imaged immediately for 2 h; during this period, we were able to detect real-time morphological changes in cell morphology (i.e., retraction of projections) and changes in mitochondrial architecture (Fig. 5b). More specifically, we identified significant changes in mitochondrial structure within astrocytes located in the proximal area, including reduced mean branch length, reduced aspect ratio, the number of branches and their length, diameter, and total length, mitochondrial perimeter, area, and geodesic path length. In summary, we saw a reduction in mitochondria content caused by fragmentation of filamentous and interconnected mitochondria into single and rounded mitochondria (Fig. 5c) with significant differences across the studied parameters (Fig. 5d). We also observed changes in the distribution of mitochondria, as they rapidly retracted from astrocyte projections and concentrated within the soma - revealing that mitochondria retraction preceded the collapse of the cytoplasmatic projections.

Fig. 5: Mitochondrial architectural changes in response to transient oxidative stress.

a Schematic showing local biochemical perturbation in the proximal (treated) well with H₂O₂. b Representative fluorescence microscopy images showing astrocytic mitochondria labeled with MitoTracker in the distal (upper) and proximal (lower) regions. c Radar plot indicating differences across mitochondrial structural parameters in response to H₂O₂. d Representative violin plots showing statistical significance. *, **, ***, and **** denotes p-value < 0.05, <0.01, <0.005, and <0.001, respectively

Exposure to EtOH alters astrocyte mitochondrial architecture

A very similar pattern to that observed in the H₂O₂ was apparent in astrocytes exposed to EtOH (Fig. 6a). While we initially observed a distribution of mitochondria throughout both the soma and processes and an elongated, filamentous phenotype in the pre-treatment condition, there was a dose-dependent reduction in mitochondria located in the cytoplasmic processes. Moreover, a shift towards a more rounded, punctate appearance was evident following the administration of EtOH (Fig. 6b). Mitochondrial structural analysis (Fig. 6c) confirmed significant differences in astrocytic mitochondrial parameters between control and EtOH conditions (Fig. 6d).

Fig. 6: Exposure to EtOH leads to alterations in mitochondrial architecture.

a Schematic outlining the experimental protocol. b Representative fluorescence microscopy images reveal structural changes in MitoTracker-labeled mitochondria in response to control, low (25 mM), and high (400 mM) EtOH. c Radar plots indicating changes in mitochondrial architectural parameters in response to EtOH. d Representative violin plots showing statistical significance. *, **, ***, and **** denotes p-value < 0.05, <0.01, <0.005, and <0.001, respectively

Transient exposure to H₂O₂ led to long-lasting disruption of Ca²⁺ transients in the astrocyte network

One of the main mitochondrial functions is Ca²⁺ storage, which plays a critical role in multiple astrocyte functions like cell-cell signaling. Having established the capacity of H₂O₂ and EtOH to disrupt mitochondria architecture, we were interested in assessing their potential effects on calcium transients. Although incompletely understood in comparison to neuronal communication⁴⁴, previous studies have established that astrocytes use Ca²⁺ transients to communicate with each other, generating non-linear⁴⁵ transient increases of Ca²⁺ which function as a proxy for astrocyte activity⁴⁶. Ca²⁺ signaling plays a primary functional role in the syncytium and is a key dynamic parameter that can be perturbed under varied conditions, with discrete cytosolic fluctuations in astrocytic Ca²⁺ microdomains occurring across time and space⁴⁴.

To capture changes in Ca²⁺, we stained cells using Fluo-4 AM, a high-affinity, membrane-permeable fluorescent dye used to visualize Ca²⁺ spikes and that can be quantified by measuring changes in fluorescence intensity (Fig. 7a). After allowing the astrocytes to form a network, we used live-cell microscopy to monitor rapid changes in calcium signaling. The data was then analyzed on a single-cell basis using ImageJ/Fiji plugin Spiky⁴⁷, an algorithm that permits detection of peaks arising temporally within imaged samples and is predicated on the range between signal minimum and maximum data points in terms of percentage to generate tri-dimensional analyses (XYT analysis using image stacks) (Fig. 7b). This approach also enabled us to track Ca²⁺ transients traveling inside individual astrocytes (Fig. 7c). Our results showed that most astrocytes that had never been exposed to H₂O₂ exhibit frequent Ca²⁺ transients. Ca²⁺ transients, as indicated by the vector map, exhibited signal propagation heterogeneity across single cells in terms of the speed and direction (norm of the vector)⁴⁷ (Fig. 7d). Ca²⁺ events are evoked both within the soma and the astrocyte processes, representing functional microdomains, rather than occurring as a single, linear signal across a single cell⁴⁸.

Fig. 7: Transient oxidative stress affects Ca²⁺ oscillations.

a Schematic showing the Fluo-4 AM staining protocol. b Representative fluorescence microscopy images of rat cortical astrocytes labeled with Fluo-4 AM at Time 1 (upper image) and Time 2 (lower image). c Isochronal map of astrocyte shown in (b), indicating signal onset delay within the cell. d Vector map indicating direction and speed of signal propagation (norm of the vector). e Contingency plots showing the relative numbers of pulsing and non-pulsing astrocytes in response to H₂O₂ and EtOH. f Representative single-cell tri-dimensional XYT peak analysis plots under control and oxidative stress conditions. g Radar plot revealing differences across three-dimensional parameters. h Representative violin plots indicating statistical significance across parameters shown in (g). *, ***, and **** denote p-value < 0.05, <0.005, and <0.001, respectively

After exposing the MPS to H₂O₂, we first quantified the number of pulsing versus non-pulsing cells (Fig. 7e). We observed a considerable reduction in the number of pulsing cells in the H₂O₂ exposure condition in comparison to controls. Additionally, pulsing astrocytes in H₂O₂ conditions showed significant differences in their profiles (Fig. 7F), including decreased pulse frequency, amplitude, and slope (Fig. 7g). This disruption of Ca²⁺ transients persisted over time, suggesting that transient H₂O₂ exposure led to long-term impairment of astrocytes’ capacity to regulate Ca²⁺ transients (Fig. 7h). Treatment with EtOH eliminated Ca²⁺ transients in most astrocytes (Fig. 7e, lower graph), precluding any kinetic analysis of these transients. While much has been studied about the effects of EtOH on neuron action potentials, these findings highlight the need for more in-depth studies focused on the effects of ethanol on astrocyte networks.

Discussion

Here, to explore the effects of agents that induce cellular changes and to mitigate some of the challenges inherent to in vitro platforms, we leveraged a microphysiological system (MPS) that enables dynamic and reversible chemical gradient formation across cell-laden hydrogels. This platform supports real-time modulation of the astrocyte microenvironment, allowing for non-destructive, spatially resolved responses to both local and global insults affecting astrocyte syncytium formation and degradation, cell viability, metabolic activity, and calcium signaling dynamics in a 3D environment. Unlike traditional formats such as 384-well plates, which rely on static dosing and lack spatial resolution, our system facilitates localized analysis of gradient-driven phenomena within a physiologically relevant 3D context.

Fabricated via CNC micromilling directly into standard polystyrene 384-well plates, our platform avoids the limitations of PDMS-based microfluidics, such as gas permeability, molecular absorption, and fabrication complexity. By designing our device to match the well-plate layout, we ensure full compatibility with high-throughput imaging systems and plate readers, increasing the number of read-outs that can be interrogated in the system. Additionally, the use of polystyrene as a fabrication material reduces the risk of molecular absorption into the bulk materials compared with PDMS-based devices.

Compared to existing microfluidic platforms, which often require complex setups and lack gradient reversibility or spatial mapping capabilities, our design offers enhanced experimental flexibility, resolution, and accessibility. The utility of this approach has been demonstrated in our prior work modeling spatially resolved biological phenomena^24,27,28, and here we extend those capabilities while retaining the practical advantages of standard well-plate formats - making this system particularly well-suited for laboratories investigating spatiotemporal dynamics in complex tissue environments.

We compared two hydrogels - collagen type I and basement membrane extract (BME) - as matrix substrates to evaluate the effect of extracellular matrix (ECM) composition on astrocyte morphology and network formation, since the ECM plays a critical role not only in providing structural support but also in delivering physiologically relevant biochemical and mechanical cues that influence cell-specific behaviors^{46,49,50,51,52}.

Consistent with prior findings, including those by Levy et al.⁵⁰, who reported increased astrocyte branching complexity and protrusion tortuosity, we observed enhanced morphological parameters when astrocytes were cultured in BME compared to collagen. These improvements are likely attributable to the composition of BME, which includes laminin, collagen IV, entactin, and heparan sulfate proteoglycans—components known to mediate cell-matrix interactions and regulate proliferation, adhesion, migration, and topological organization⁴⁷.

However, the use of BME presents challenges due to its inherent batch-to-batch variability. As a complex mixture containing over 1400 peptides and 2000 proteins unique to BME⁴⁸, its composition and consequently its mechanical and biochemical properties - can vary significantly. Additionally, discrepancies in viscosity and the potential for xenogeneic contaminants complicate its application in biomimetic platform development and mean that research into alternatives, including synthetic substitutes, is ongoing⁴⁸.

Following optimization of our MPS, we determined that agents that induce oxidative stress and other maladaptive cellular changes exert catastrophic effects on the genesis of the astrocyte syncytium and lead to profound degeneration of established networks. Our results demonstrated that cell viability is significantly reduced when exposed to stressors such as H₂O₂ and EtOH, and that this damage is exacerbated over time, even once the insult is reversed. However, as astrocyte viability remained high overall, this suggests that the markedly adverse effects on the genesis and subsequent degeneration of the syncytium are not merely because of increased cell death, but arise due to an array of complex, multi-level factors, such as metabolic changes and signaling perturbations⁵³.

Undue oxidative stress, including excessive ROS accumulation, is known to play critical roles in both normal aging and disease, giving rise to a cascade of deleterious events^54,55 and inducing the morphological and biochemical changes characteristic of reactive astrogliosis⁵⁶. Indeed, while moderate increases in ROS promote beneficial activity (i.e., hormesis), aberrant levels of ROS are associated with an array of chronic diseases and unfavorable epigenetic changes⁵⁵. Additionally, ROS-signaling mechanisms are understood to contribute towards mitochondrial remodeling⁵⁷ and Ca²⁺ regulation⁵⁸.

In this context, the observed biphasic regulation of interferon-stimulated genes (ISGs) following H₂O₂ exposure aligns with the central role of astrocytes in maintaining CNS homeostasis. Astrocytes act as metabolic and structural sentinels, buffering oxidative stress and preventing neurotoxic cascades⁵⁹. Astrocytes also acquire immune cell-like characteristics in response to CNS injury^60,61 and undergo an array of molecular changes during the recovery and repair period⁶¹. These data revealed a temporally dynamic astrocytic response to hydrogen peroxide-induced oxidative stress, characterized by an early wave of interferon-stimulated gene (ISG) activation followed by further changes. LY6E was upregulated at 24 h post-exposure, suggesting an early stress-induced activation of interferon-responsive pathways, potentially triggered by oxidative damage rather than pathogen-associated molecular patterns. By 120 h, LY6E was significantly downregulated, indicating resolution or suppression of this response. OASL and MX1 showed non-significant increases, consistent with no or low-level activation of antiviral-like programs that may be secondary to oxidative stress. Conversely, the minimal increase in ISG15 at 24 h post-exposure is consistent with previous reports demonstrating that oxidative stress can activate type I interferon signaling pathways in glial cells⁵⁴. The subsequent decrease by 120 h suggests a transient response, potentially reflecting the resolution of acute stress and restoration of homeostasis⁶². GBP1, a GTPase involved in immune defense and inflammasome regulation, remained slightly lower than baseline across timepoints and may indicate a subtle shift in astrocytic function away from pro-inflammatory signaling, as observed in models of neuroinflammation and oxidative injury⁶³. IL18BP, a potent antagonist of IL-18, may serve to limit excessive inflammation, as IL-18 itself exerts both pro- and anti-inflammatory effects depending on context⁶⁴. The biphasic expression of IL18BP observed here - initial suppression at 24 h followed by marked induction at 120 h - may reflect a dynamic feedback mechanism that first permits IL-18 activity to initiate early immune signaling and later restrains it to prevent chronic inflammation and tissue injury⁶⁵. Collectively, these findings highlight the complexity of astrocyte-mediated immune modulation and underscore the importance of temporal resolution in studying glial responses to oxidative stress. Our study analyzed a limited number of genes; future studies could leverage technologies like RNA-seq to explore gene expression changes across thousands of genes, allowing a more comprehensive understanding of how astrocytes respond to spatial-temporal perturbations of the biochemical milieu.

EtOH is a known teratogenic substance that can induce widespread structural and functional impairments in the developing brain and can activate neuroinflammatory pathways and accelerate age-associated macro and microstructural changes in adolescent and adult individuals⁶⁶. We observed that astrocytes exposed to EtOH undergo atrophic changes, consistent with the observation that tissues extracted from patients with alcohol use disorder show reduced astrocyte density; this pattern is observed as well in animal models of both chronic and binge drinking³. Intoxication is linked to increases in pro-inflammatory mediators, secretion of chemokines, interleukin-1 β (IL-1β), tumor necrosis factor α (TNF-α), and oxidative stress; thus, future research should examine these factors to provide a more comprehensive picture of the effects of alcohol⁴⁵. Moreover, evidence suggests that the prime targets of alcohol-induced oxidative stress, and subsequent withdrawal are mitochondria in the brain and that astrocytes may be especially susceptible to alcohol toxicity during proliferation⁶⁷. Consequently, astrocytic mitochondrial dysfunction appears to play a principal role in the pathogenic response – suggesting that therapeutics developed to target mitochondrial changes may mitigate damage arising from alcohol misuse⁶⁷.

Both oxidative stress and EtOH exposure induce changes in bioenergetic metabolism beyond which the human brain is evolutionarily equipped, leading to harmful effects on astrocytic mitochondrial morphology and function^68,69 and the subsequent dysregulation of calcium transients⁷⁰. Moreover, in contrast to the classical understanding that astrocytic mitochondria are confined to the cell body, astrocytes are now understood to contain abundant mitochondria in their distal processes, which constitute non-overlapping, i.e., tessellating, specialized domains throughout the brain⁴³, and support local calcium signaling within astrocytes. Although astrocyte cytoarchitecture is yet to be fully mapped⁷¹, recent evidence⁷² has revealed the presence of extensive tubular mitochondrial networks throughout the soma, branches, branchlets, and end feet processes. Furthermore, despite being considered energy-conserving, electrically silent cells, astrocytes are now believed to contain comparable mitochondria volume ratios as those observed in electrically excitable neurons. Although their mitochondrial dynamics are less fully characterized than in neuronal cells⁴¹, current evidence suggests that homeostatic and reactive actions are potentially more energy-demanding than previously recognized and that astrocytic bioenergetics may play a greater role in both normative and pathological states than previously believed⁴³. This is noteworthy as the astrocyte response to cerebral insults mediates the extent of CNS damage and influences the likelihood of repair versus lasting impairments⁴³.

Astrocytic calcium signaling is now emerging as an increasingly prominent player within normal and abnormal functioning. Changes in calcium homeostasis and the production of oxidative stress have been implicated in the early stages of numerous pathologies of the CNS, including neurodegenerative diseases and ischemic stroke, while their synergistic activity with mitochondria has been associated with the opening of mitochondrial permeability transition pores and, ultimately, apoptosis, ferroptosis, and necrosis⁶⁸. Moreover, the slower timescale of astrocytic communication compared to neuronal action potentials may contribute to myriad roles within the CNS. The slower kinetics of Ca²⁺ signaling, heterogeneity of events, and the potential for microdomain expansion from focal to coordinated amplified activity with greater amplitude, duration, and spatiotemporal reach could give rise to astrocyte remodeling, gliotransmission, neuronal circuitry feedback, and dynamic communication with non-neuronal cells within the extracellular milieu¹. Nonetheless, the minutiae of this elegant interplay between astrocytes and their microenvironment are yet to be fully elucidated⁸. The current study focused upon signaling within individual cells that comprise the syncytium; further experiments could not only delineate the nanoscale differences in Ca² elevations within discrete regions of the soma and astrocytic arbor but could also investigate propagation of signals throughout the network. Astrocytic Ca²⁺ activity appears to be intimately related to both homeostasis in the healthy CNS and the cascade of pathological events intrinsic to numerous diseased or injured states, thus making them an attractive therapeutic target. However, clinical applications have, as yet, failed to live up to their promise in spite of vast technical improvements in capturing transient Ca²⁺ events and deciphering their heterogeneous roles⁸.

Ca²⁺ transients may be related to either intrinsic excitability or evoked by external stimuli, including synaptic inputs⁴⁵. Future studies should include an astrocyte-neuronal cell co-culture within the MPS to explore the reciprocal relationship between cell types and the effects of gliotransmission on neural activity, as well as the ways in which neurotransmission may disrupt Ca²⁺ homeostasis in astrocytes. Astrocytes also display sodium transients evoked by neurons that spread throughout the syncytium to modulate GABA and glutamate uptake and spatiotemporal patterns of gliotransmission^45,73, but their functional roles are yet to be fully characterized. Additionally, astrocytes are intimately involved in the formation and ongoing function of neuronal circuitry and play a significant role in remodeling synaptic connectivity⁷⁴. Plastic structural changes in the astrocyte-neuron synapse can lead to reduced network and neurovascular unit coverage within brain tissue⁷⁵ and are understood to play a role in the development and progression of neurodegenerative and psychiatric disorders. Nonetheless, the influence of these changes on disease phenotype and the temporal window for perturbation and prolonged changes remains unknown and warrants clarification⁷⁶. Additional studies incorporating neurons and other glial cell types into the astrocyte network and stimulation with agents mimicking neurodegenerative disease, infection, or injury may be illuminating in the context of understanding intricate 3D astrocyte function in the CNS⁷⁷.

Overall, the findings of this study indicate that transient perturbations of the biochemical milieu result in persistent changes to network genesis and stability and exert an array of deleterious structural and functional effects. These reductions in astrocyte syncytial integrity are observed during aging, injury, and disease⁷⁴ and are implicated in the etiology of a wide range of pathologies⁷⁷; thus, elucidating plastic changes in response to even brief disruption may help identify novel therapeutic interventions for some of the most prevalent diseases of an increasingly aged population^78,79.

Materials and methods

Astrocyte culture

Commercial rat cortical astrocytes (Ref: R882A-05N: Merck) were seeded in T-75 cell culture flasks (Ref: 156499; Thermo Scientific) and supplemented with Rat Astrocyte Growth Media (RAGM; Ref: R821-500; Cell Applications, Inc.). The flasks were incubated within a humidified incubator at 37 °C and 5% CO₂. Experiments were performed with cells between passages 1–14 with 75–90% confluency. Astrocytes were collected using Accutase (Ref: A69694; Sigma Aldrich) and resuspended at the appropriate cell density prior to seeding in the MPS.

Device fabrication

Our MPS was designed using Fusion 360® (Autodesk; v.16.13.0.2489) and fabricated using CNC micro milling (Tormach Personal CNC 770). Chambers and diffusion ports were generated by removing material from the bottom of a 384 polystyrene-based well plate (Ref: 3701; Corning) and subsequently cleaned using an air gun and isopropyl alcohol (IPA; MilliporeSigma). To ensure chamber sealing and optimal visualization for microscopy, a new glass base was attached to the MPS using double-sided tape (Ref: 59069831, MSC), cut to the appropriate specifications using a laser cutter (Automation Technology Inc.). To promote hydrophilicity, thus enhancing hydrogel flow across the chamber, the MPS was treated with oxygen-rich plasma (Femto Version 1, Electronic Diener) and sterilized in an ultraviolet (UV) light chamber (Benchmark UV Clave) for 20 min immediately prior to use.

3D astrocyte culture within the MPS

For initial hydrogel comparisons a collagen gel was prepared; for each 100 µL of hydrogel: 25 μL of collagen type I (Ref: 354249, Corning Collagen Type I Rat Tail High Concentration 8.43 mg mL⁻¹), 0.65 μL of 1 M NaOH, 5 μL of 10× phosphate-buffered PBS (PBS; Ref: 82027-604, Life Technologies), 19.35 μL of deionized (DI) water, and 50 μL of cultured astrocyte cell suspension were mixed giving a total volume of 100 μL. A Cultrex hydrogel was used for all remaining experiments. For each 100 µL of hydrogel, 70 µL of Cultrex to which 30 µL of astrocyte cell suspension was added and mixed thoroughly. All hydrogels were prepared on ice to prevent polymerization. Volumes can be scaled up as required. All hydrogels were allowed to polymerize for 15 min at room temperature after seeding. 60 µL of RAGM was then added to each well, and devices were incubated at 37 °C until required.

Network genesis disruption assay

To determine the effects of biochemical perturbations on the formation of the network, we seeded astrocytes in our MPS as described previously. Following polymerization, we then added either H₂O₂ or EtOH at the appropriate concentration to all wells. For H₂O_2, we added H₂O₂ to RAGM for a final concentration of 200 mM and 880 mM and incubated at 37 °C for 2 h. This was followed by three 1× PBS washes, 10 min apart. For EtOH, 200 proof EtOH was added to RAGM for a final concentration of 25 mM, 200 mM, and 400 mM and incubated at 37 °C overnight. This was followed by three 1× PBS washes, 10 min apart.

Network degradation assay

To determine whether H₂O₂ and EtOH affect the stability of an established network, we followed our standard seeding protocol of adding RAGM immediately following polymerization and incubating at 37 °C for 24 h to allow the network to form. For H₂O_2, following 24 h incubation with RAGM, we aspirated the cell culture medium from the proximal well and replaced it with RAGM with H₂O₂ at a concentration of 880 mM. The MPS was then incubated at 37 °C for 2 h. This was followed by three 1× PBS washes, 10 min apart. For EtOH, 200 proof EtOH was added to RAGM for a final concentration of 100 mM and incubated at 37 °C overnight. To counteract evaporation, the MPS were placed in a sealed container containing twice the desired EtOH cell culture concentration. This was followed by three 1× PBS washes, 10 min apart.

Cell viability assay

Astrocyte viability in the MPS was assessed by staining live and dead cells with green and red fluorescence dyes, respectively. Live cells were stained with 10 mM CellTracker Green (CMFDA; Ref: C7025, Thermo Fisher Scientific) and 2 mg mL⁻¹ propidium iodide (PI; Ref: P4170-25MG, MilliporeSigma). Stock solutions were prepared in DMSO (Dimethyl Sulfoxide; Ref: 67-68-5, Santa Cruz Biotechnology) for CellTracker Green and distilled water for propidium iodide. Working solutions were prepared by diluting stock solutions in 1× PBS (CellTracker 1:1000 and propidium iodide at 1:500). Astrocytes were incubated with CellTracker Green for 15 min at 37 °C and washed twice with 1× PBS to remove excess dye and resuspended at the desired seeding density. The stained cell suspension was added to the hydrogel, and propidium iodide was added at a concentration of 0.1 g/100 µL of gel and polymerized and incubated as described above. Viable cells were defined as CellTracker Green-positive and PI-negative, while non-viable cells were PI-positive. The MPS was imaged at the selected time points, and the percentage of live and dead cells was quantified using FIJI image processing software.

Modeling and simulation of EtOH diffusion and evaporation

In this section, we present the modeling process for EtOH diffusion and evaporation. Our goal was double: (1) to quantify the loss of EtOH due to evaporation and (2) to estimate the risk of EtOH absorption through air in the side chambers (i.e., wells not seeded with EtOH), compared to EtOH reaching via passive diffusion.

We therefore modeled two phenomena: EtOH diffusion through media (after being seeded into the chamber) and EtOH evaporation. In the following sections, we present (A) the general equation, (B) specifics on the diffusion term, and (C) specifics on the evaporation term.

EtOH dynamics

To model EtOH dynamics, we established a partial differential equation containing the classical diffusion terms, plus an additional evaporation term.

This equation has the following form:

$$\frac{\partial C({\boldsymbol{x}},t)}{\partial t}=D\,\nabla C\left({\boldsymbol{x}},t\right)-{k}_{{evap}}\left({\boldsymbol{x}}\right)\,C({\boldsymbol{x}},t)$$

with boundary conditions:

$${\left.\frac{\partial C}{\partial n}\right|}_{\partial \Omega }=0$$

Where:

• $C\left({\boldsymbol{x}},t\right)$ is the concentration at position ${\boldsymbol{x}}$ and time $t$.

• ${D}$ is the diffusion coefficient of our compound (in this case ${D={D}}_{{water}}$).

• ${\nabla }^{2}C$ is the Laplacian (diffusion term).

• ${k}_{{evap}}$ is the constant evaporation rate.

• $\frac{\partial C}{\partial n}$ is the derivative of $C$ normal to the boundary $\partial \Omega$

We considered the equation in 2-dimensions (x-axis and z-axis), modeling one cross-section of the microfluidic device. The behavior along the y-axis is expected to be analogous to the x-axis.

EtOH diffusion

As the matrix that forms the core of our device has very large pores, we assumed that compounds diffuse at rates similar to water. Therefore, we applied normal diffusion conditions with a diffusion coefficient for EtOH of:

$${D}_{{water}}=1.33\,\times {10}^{-3}\,{{mm}}^{2}/s$$

This coefficient at 25 °C is obtained from the classical textbook Bergman et al.⁸⁰ and adjusted to its value at 37 °C.

EtOH evaporation

To model EtOH evaporation in bulk concentration, we utilized the classical approach described in Reid et al.⁸¹, in which the evaporation phase is described as a thin layer in which the concentration in the liquid transitions gradually towards the concentration in air ${C}_{\infty }$. Therefore, the mass loss rate ${k}_{{evap}}$ can be determined by the following equation:

$${k}_{{evap}}=\frac{{D}_{{air}}}{\delta }$$

Where ${D}_{{air}}$ is the diffusion coefficient of EtOH in air, and $\delta$ is the thickness of the interchange layer.

Prior studies⁸², estimate the thickness of this layer as approximately 1 cm. However, in our setting, the size of this layer is limited to a maximum of ~0.5 cm due to the cover placed on top of the wells. This situation differs from the conventional scenario in which the vapor can escape to the surroundings without limitation. In general, our setting will be more restrictive, and we expect the actual evaporation rates to be lower. We simulated this unfavorable scenario as an upper bound, showing that even in such a case, the impact of evaporation is insufficient to alter our experiments, therefore taking δ = 0.5 × 10⁻² mm/s.

With the previous values, under non-convective conditions at 37 °C, we assume a gas-side mass transfer coefficient for ethanol of approximately ${k}_{{evap}}\sim 1.2\times {10}^{-3}\,\mathrm{mm}/{\rm{s}}$, consistent with estimates based on molecular diffusion in air⁸³. This value reflects purely diffusive transport across a ~2 cm gas layer with ethanol vapor diffusivity ${D}_{{air}}\sim 1.2\times {10}^{1}$ mm²/s.

Software for numerical simulation

To solve the equations in our setting, we use a Crank–Nicolson semi-implicit method⁸⁴ implemented in Python. Our numerical approach has been shown to have a good balance between accuracy and efficiency for diffusion problems⁸⁵.

We run simulations using an in-house software tool built in Python 3.13. Our implementation is based on standard Python numerical libraries (i.e., numpy). Plot generation is done with the help of the matplotlib-pyplot library.

3D diffusion assay

The 2-NBDG (2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose) (Ref: N13195, Invitrogen), a fluorescent glucose analog, was used to validate glucose diffusion across cell-laden hydrogels. As before, devices were seeded with cell-laden hydrogels, and the 2-NBDG at a concentration was added to the proximal well. Diffusion was monitored over time by acquiring images using a fluorescence microscope (Thunder Imager, Leica DMi8). Microscopy data was analyzed using FIJI image processing software (https://imagej.net/Fiji/Downloads) to quantify fluorescence intensity across regions of the device at 15 min, 90 min, and 24 h.

Reverse transcriptase quantitative polymerase chain reaction

RT-qPCR was performed to evaluate gene profiles of inflammation-related pathways in response to H₂O₂ exposure. Pre-treatment and 24 h and 5 days post-exposure cells were collected, and their mRNA was isolated using the Dynabeads mRNA DIRECT Purification kit (Ref: 61012; Thermo Fisher). Upon mRNA isolation, complementary DNA (cDNA) was synthesized using the High-Capacity cDNA Reverse Transcription kit (4368814; Thermo Fisher). The cDNA obtained was amplified using the SsoAdvanced PreAmp Supermix (Ref: 6777; Bio-Rad) to ensure sufficient DNA to conduct RT-qPCR. Finally, RT-qPCR was run on the LightCycler 480 II instrument (05015278001; Roche). Relative fold change in gene expression was analyzed using the 2ΔΔCt method and normalized to housekeeping genes as an internal control. Primers for genes of interest were obtained from Thermo Fisher.

Phalloidin staining

We aspirated the media from all reservoir wells and replaced it with a 4% PFA solution (Ref: AA433689L; Thermo Fisher Scientific) and refrigerated it for 48 h. We then washed with Tween 20% (Ref: 50-176-4983, Sigma Aldrich) to remove the PFA and added Triton X-100 (Ref: 9036-19-5; Sigma Aldrich) to induce cell permeability. The MPPs were incubated at room temperature for 30 min before washing with 1× PBS to remove the Triton X-100. We then added phalloidin (Ref: A34055; Invitrogen) (1:500 in 1× PBS) and Hoechst (Ref: 62249: Thermo Fisher Scientific) (1:1000 in 1× PBS) and refrigerated for 48 h before washing with PBS three times and imaging. Devices covered with PBS to prevent dehydration and protected from light can be stored at 4 °C for several weeks.

Network analysis

The phalloidin-stained images were imaged with the Leica fluorescence microscope and analyzed using FIJI image processing software. The number of extremities in the analyzed area, number of nodes, number of junctions, number of master junctions, number of master segments, master segments length (i.e., the sum of the length of the detected master segments), number of meshes, the total meshes area (i.e., the sum of mesh areas detected), the number of pieces (i.e., the sum of number of segments, isolated elements and branches), number of segments, number of branches, number of isolated elements, total length (i.e., sum of length of segments, isolated elements and branches), total branching length (i.e., sum of length of the trees composed from segments and branches), total segments length (i.e., sum of length of the segments), total branches length (i.e., sum of length of the branches), total isolated branches length (i.e., sum of length of the isolated elements), branching interval (i.e., mean distance separating two branches in the trees in the analyzed area – total segments length/number of branches), mesh index (i.e., mean distance separating two master junctions in the trees in the analyzed area – total master segments length/number of master segments), and mean mesh size were analyzed. This program operates by creating a binary mask that is segmented to produce a skeleton from which extremities, junctions, and nodes can be identified. The detected bifurcations then permit branches and segments to be categorized and, from there, an overall mesh analysis is generated⁸⁶.

Mitochondria architecture assay

Mitochondria were stained with a fluorescent dye prior to adding cells to the hydrogel. MitoTracker CMXRos (Ref: 9082S; Cell Signaling Technology) is a lipophilic cationic stain that accumulates selectively in mitochondria. A 1 mM stock solution in DMSO (Dimethyl Sulfoxide; Ref: 67-68-5, Santa Cruz Biotechnology) was diluted to 50 nM in 1× PBS, added to the astrocytes, and incubated at 37 °C for 15 min. After two 1× PBS washes, astrocytes were resuspended in RAGM at the desired density and added to the hydrogel as previously described.

For H₂O_2, following 24 h incubation with RAGM, we aspirated the cell culture medium from the proximal well and replaced it with RAGM with H₂O₂ at a concentration of 880 mM and began imaging immediately for 2 h to capture changes occurring in real time during the H₂O₂ exposure period.

For EtOH, 200 proof EtOH was added to RAGM for a final concentration of 25 mM and 400 mM and incubated at 37 °C overnight. This was followed by three 1× PBS washes, 10 min apart. We then imaged to capture changes in the mitochondria architecture.

Following imaging with a Leica fluorescence microscope (Thunder Imager, Leica DMi8), we used FIJI image processing software to quantify changes in mitochondrial architecture, including aspect ratio, branch length, and area in the post-treatment proximal and distal regions.

Mitochondrial architecture analysis

Fluorescence microscopy images of astrocytes under each experimental condition were acquired, and representative cells were selected for mitochondrial morphology assessment. Images were processed using Fiji software (Version: 2.14.0/1.54f; Build: c89e8500e4) and converted to 8-bit format. To minimize background interference, regions of interest (ROIs) encompassing individual astrocytes were manually outlined, and non-cellular background signals were subtracted. Mitochondrial architecture was quantified using the open-source plugin Mitochondria Analyzer (Version: MitochondriaAnalyzer-V2.3.1)⁴⁰. Images underwent preprocessing via 2D thresholding with default parameters: background subtraction (rolling ball radius: 1.25 µm), sigma filtering (σ = 0.63), local contrast enhancement (kernel size: 1.8), gamma adjustment (γ = 0.80), weighted mean thresholding (block size: 1.25 µm, C-value: 10), and post-processing steps (de-speckling and outlier removal). Mitochondria were analyzed on a per-organelle basis using the plugin’s 2D analyzer module to extract morphometric parameters, including branch count, branch endpoints, total branch length, mean branch diameter, mean branch length, geodesic (longest shortest) path, and branch junctions. Numerical output data were exported to Microsoft Excel for statistical analysis and graphical representation.

Calcium signaling

We loaded unstained cells suspended in a Cultrex hydrogel into the chambers and allowed them to form a network over 48 h. We added H₂O₂ and, after two hours, performed the washing steps; for EtOH, we added the desired concentrations and incubated overnight. We then used 1 mM stock solution of Fluo-4 AM in DMSO (Dimethyl Sulfoxide; Ref: 67-68-5, Santa Cruz Biotechnology) and prepared a working solution of 5 µM diluted in 1× PBS. We aspirated the RAGM, replaced it with the 5 µM Fluo-4AM solution, and incubated the devices at 37 °C for 2 h. This was followed by additional incubation at room temperature for 30 min to allow de-esterification of the dye, and then three 1× PBS washes were performed to remove any excess dye from the system.

Devices were imaged using a Leica fluorescence microscope (Thunder Imager, Leica DMi8) every 3 s over a total span of 120 s per replicate to capture changes over time. Due to the rapid photobleaching of the dye, the imaging duration was limited. Although acquisition parameters were optimized to minimize photobleaching, the dye’s instability still constrained the maximum imaging time. The data was then analyzed on a manually segmented single-cell basis using an ImageJ/Fiji plugin, Spiky, an algorithm that permits detection of peaks arising temporally within imaged samples and is predicated on the range between signal and maximum data points in terms of percentage. Changes in fluorescence intensity was used to quantify parameters including Tmax (i.e., the time at which the peak has been detected with the value adjusted from a Gaussian fit of the two points around the maximum), APeak (i.e., the absolute value of the amplitude of the peak), Baseline Bl (i.e., mean of the two consecutive minimum values around the peak), amplitude (i.e., the relative value of the amplitude of the peak from the baseline (APeak - Baseline), SlopeMax2Pk (i.e., absolute maximum slope from peak start to the peak maximum), SlopeMax2Bl (i.e., absolute maximum slope from peak maximum to next baseline), PkArea (i.e., the integrate of the peak - area under the peak from baseline), AreaBl2Pk i.e., integrate from start to peak maximum) and AreaPk2Bl (i.e., integrate from peak maximum to peak end and back to baseline).

Image acquisition and analysis

Brightfield and fluorescence microscopy were acquired with a fluorescence microscope (Thunder Imager, Leica DMi8) to image 3D astrocyte viability, network formation and stability, mitochondrial architecture, and calcium signaling. Microscopy data was analyzed using FIJI image processing software (https://imagej.net/Fiji/Downloads).

Statistical analysis

All experiments included at least three biological replicates. Statistical analysis was performed using GraphPad Prism 9 and R (4.2.2). One-way ANOVA with Dunn’s or Tukey multiple comparisons tests and unpaired t-tests were used for parametric comparisons, whilst Mann–Whitney U and Kruskal–Wallis with Dunn’s multiple comparisons test were used for non-parametric comparisons. All analyses were performed at a set significance level of p < 0.05.